K/V control prints (for debug only) O:-)

- Do not try to allocate more DB blocks in an inode block until it's "confirmed" and "locked" by the first write
- Do not recheck for new zero DB blocks on first write into an inode block - a CAS failure means someone else is already writing into it
- Throw new allocation blocks away regardless of whether the known_version is 0 on a CAS failure

CMakeLists.txt

@@ -2,6 +2,6 @@ cmake_minimum_required(VERSION 2.8.12)
 
 project(vitastor)
 
-set(VERSION "1.0.0")
+set(VERSION "1.2.0")
 
 add_subdirectory(src)

README-ru.md

@@ -50,6 +50,7 @@ Vitastor поддерживает QEMU-драйвер, протоколы NBD и
   - Параметры
     - [Общие](docs/config/common.ru.md)
     - [Сетевые](docs/config/network.ru.md)
+    - [Клиентский код](docs/config/client.en.md)
     - [Глобальные дисковые параметры](docs/config/layout-cluster.ru.md)
     - [Дисковые параметры OSD](docs/config/layout-osd.ru.md)
     - [Прочие параметры OSD](docs/config/osd.ru.md)

README.md

@@ -50,6 +50,7 @@ Read more details below in the documentation.
   - Parameter Reference
     - [Common](docs/config/common.en.md)
     - [Network](docs/config/network.en.md)
+    - [Client](docs/config/client.en.md)
     - [Global Disk Layout](docs/config/layout-cluster.en.md)
     - [OSD Disk Layout](docs/config/layout-osd.en.md)
     - [OSD Runtime Parameters](docs/config/osd.en.md)

csi/Makefile

@@ -1,4 +1,4 @@
-VERSION ?= v1.0.0
+VERSION ?= v1.2.0
 
 all: build push
 

csi/deploy/004-csi-nodeplugin.yaml

@@ -49,7 +49,7 @@ spec:
             capabilities:
               add: ["SYS_ADMIN"]
             allowPrivilegeEscalation: true
-          image: vitalif/vitastor-csi:v1.0.0
+          image: vitalif/vitastor-csi:v1.2.0
           args:
             - "--node=$(NODE_ID)"
             - "--endpoint=$(CSI_ENDPOINT)"

csi/deploy/005-csi-provisioner-rbac.yaml

@@ -35,10 +35,13 @@ rules:
     verbs: ["get", "list", "watch"]
   - apiGroups: ["snapshot.storage.k8s.io"]
     resources: ["volumesnapshots"]
-    verbs: ["get", "list"]
+    verbs: ["get", "list", "patch"]
+  - apiGroups: ["snapshot.storage.k8s.io"]
+    resources: ["volumesnapshots/status"]
+    verbs: ["get", "list", "patch"]
   - apiGroups: ["snapshot.storage.k8s.io"]
     resources: ["volumesnapshotcontents"]
-    verbs: ["create", "get", "list", "watch", "update", "delete"]
+    verbs: ["create", "get", "list", "watch", "update", "delete", "patch"]
   - apiGroups: ["snapshot.storage.k8s.io"]
     resources: ["volumesnapshotclasses"]
     verbs: ["get", "list", "watch"]
@@ -53,7 +56,7 @@ rules:
     verbs: ["get", "list", "watch"]
   - apiGroups: ["snapshot.storage.k8s.io"]
     resources: ["volumesnapshotcontents/status"]
-    verbs: ["update"]
+    verbs: ["update", "patch"]
   - apiGroups: [""]
     resources: ["configmaps"]
     verbs: ["get"]

csi/deploy/007-csi-provisioner.yaml

@@ -23,6 +23,11 @@ metadata:
   name: csi-vitastor-provisioner
 spec:
   replicas: 3
+  strategy:
+    type: RollingUpdate
+    rollingUpdate:
+      maxUnavailable: 1
+      maxSurge: 0
   selector:
     matchLabels:
       app: csi-vitastor-provisioner
@@ -46,7 +51,7 @@ spec:
       priorityClassName: system-cluster-critical
       containers:
         - name: csi-provisioner
-          image: k8s.gcr.io/sig-storage/csi-provisioner:v2.2.0
+          image: k8s.gcr.io/sig-storage/csi-provisioner:v3.0.0
           args:
             - "--csi-address=$(ADDRESS)"
             - "--v=5"
@@ -116,7 +121,7 @@ spec:
             privileged: true
             capabilities:
               add: ["SYS_ADMIN"]
-          image: vitalif/vitastor-csi:v1.0.0
+          image: vitalif/vitastor-csi:v1.2.0
           args:
             - "--node=$(NODE_ID)"
             - "--endpoint=$(CSI_ENDPOINT)"

csi/deploy/009-storage-class.yaml

@@ -17,3 +17,4 @@ parameters:
   # multiple etcdUrls may be specified, delimited by comma
   #etcdUrl: "http://192.168.7.2:2379"
   #etcdPrefix: "/vitastor"
+allowVolumeExpansion: true

csi/deploy/example-snapshot-class.yaml

@@ -0,0 +1,7 @@
+apiVersion: snapshot.storage.k8s.io/v1
+kind: VolumeSnapshotClass
+metadata:
+  name: vitastor-snapclass
+driver: csi.vitastor.io
+deletionPolicy: Delete
+parameters:

csi/deploy/example-snapshot-clone.yaml

@@ -0,0 +1,16 @@
+---
+apiVersion: v1
+kind: PersistentVolumeClaim
+metadata:
+  name: test-vitastor-clone
+spec:
+  storageClassName: vitastor
+  dataSource:
+    name: snap1
+    kind: VolumeSnapshot
+    apiGroup: snapshot.storage.k8s.io
+  accessModes:
+    - ReadWriteOnce
+  resources:
+    requests:
+      storage: 10Gi

csi/deploy/example-snapshot.yaml

@@ -0,0 +1,8 @@
+apiVersion: snapshot.storage.k8s.io/v1
+kind: VolumeSnapshot
+metadata:
+  name: snap1
+spec:
+  volumeSnapshotClassName: vitastor-snapclass
+  source:
+    persistentVolumeClaimName: test-vitastor-pvc

csi/go.mod

@@ -9,6 +9,7 @@ require (
 	golang.org/x/net v0.0.0-20201202161906-c7110b5ffcbb
 	golang.org/x/xerrors v0.0.0-20200804184101-5ec99f83aff1 // indirect
 	google.golang.org/grpc v1.33.1
+	google.golang.org/protobuf v1.24.0
 	k8s.io/klog v1.0.0
 	k8s.io/utils v0.0.0-20210305010621-2afb4311ab10
 )

csi/src/config.go

@@ -5,7 +5,7 @@ package vitastor
 
 const (
     vitastorCSIDriverName    = "csi.vitastor.io"
-    vitastorCSIDriverVersion = "1.0.0"
+    vitastorCSIDriverVersion = "1.2.0"
 )
 
 // Config struct fills the parameters of request or user input

csi/src/controllerserver.go

@@ -20,6 +20,7 @@ import (
 
     "google.golang.org/grpc/codes"
     "google.golang.org/grpc/status"
+    "google.golang.org/protobuf/types/known/timestamppb"
 
     "github.com/container-storage-interface/spec/lib/go/csi"
 )
@@ -45,6 +46,7 @@ type InodeConfig struct
     ParentPool uint64 `json:"parent_pool,omitempty"`
     ParentId uint64 `json:"parent_id,omitempty"`
     Readonly bool `json:"readonly,omitempty"`
+    CreateTs uint64 `json:"create_ts,omitempty"`
 }
 
 type ControllerServer struct
@@ -178,27 +180,43 @@ func (cs *ControllerServer) CreateVolume(ctx context.Context, req *csi.CreateVol
         return nil, status.Error(codes.InvalidArgument, "no etcdUrl in storage class configuration and no etcd_address in vitastor.conf")
     }
 
+    args := []string{ "create", volName, "-s", fmt.Sprintf("%v", volSize), "--pool", fmt.Sprintf("%v", poolId) }
+
+    // Support creation from snapshot
+    var src *csi.VolumeContentSource
+    if (req.VolumeContentSource.GetSnapshot() != nil)
+    {
+        snapId := req.VolumeContentSource.GetSnapshot().GetSnapshotId()
+        if (snapId != "")
+        {
+            snapVars := make(map[string]string)
+            err := json.Unmarshal([]byte(snapId), &snapVars)
+            if (err != nil)
+            {
+                return nil, status.Error(codes.Internal, "volume ID not in JSON format")
+            }
+            args = append(args, "--parent", snapVars["name"]+"@"+snapVars["snapshot"])
+            src = &csi.VolumeContentSource{
+                Type: &csi.VolumeContentSource_Snapshot{
+                    Snapshot: &csi.VolumeContentSource_SnapshotSource{
+                        SnapshotId: snapId,
+                    },
+                },
+            }
+        }
+    }
+
     // Create image using vitastor-cli
-    _, err := invokeCLI(ctxVars, []string{ "create", volName, "-s", fmt.Sprintf("%v", volSize), "--pool", fmt.Sprintf("%v", poolId) })
+    _, err := invokeCLI(ctxVars, args)
     if (err != nil)
     {
         if (strings.Index(err.Error(), "already exists") > 0)
         {
-            stat, err := invokeCLI(ctxVars, []string{ "ls", "--json", volName })
+            inodeCfg, err := invokeList(ctxVars, volName, true)
             if (err != nil)
             {
                 return nil, err
             }
-            var inodeCfg []InodeConfig
-            err = json.Unmarshal(stat, &inodeCfg)
-            if (err != nil)
-            {
-                return nil, status.Error(codes.Internal, "Invalid JSON in vitastor-cli ls: "+err.Error())
-            }
-            if (len(inodeCfg) == 0)
-            {
-                return nil, status.Error(codes.Internal, "vitastor-cli create said that image already exists, but ls can't find it")
-            }
             if (inodeCfg[0].Size < uint64(volSize))
             {
                 return nil, status.Error(codes.Internal, "image "+volName+" is already created, but size is less than expected")
@@ -217,6 +235,7 @@ func (cs *ControllerServer) CreateVolume(ctx context.Context, req *csi.CreateVol
             // Ugly, but VolumeContext isn't passed to DeleteVolume :-(
             VolumeId: string(volumeIdJson),
             CapacityBytes: volSize,
+            ContentSource: src,
         },
     }, nil
 }
@@ -230,15 +249,15 @@ func (cs *ControllerServer) DeleteVolume(ctx context.Context, req *csi.DeleteVol
         return nil, status.Error(codes.InvalidArgument, "request cannot be empty")
     }
 
-    ctxVars := make(map[string]string)
-    err := json.Unmarshal([]byte(req.VolumeId), &ctxVars)
+    volVars := make(map[string]string)
+    err := json.Unmarshal([]byte(req.VolumeId), &volVars)
     if (err != nil)
     {
         return nil, status.Error(codes.Internal, "volume ID not in JSON format")
     }
-    volName := ctxVars["name"]
+    volName := volVars["name"]
 
-    ctxVars, _, _ = GetConnectionParams(ctxVars)
+    ctxVars, _, _ := GetConnectionParams(volVars)
 
     _, err = invokeCLI(ctxVars, []string{ "rm", volName })
     if (err != nil)
@@ -344,6 +363,8 @@ func (cs *ControllerServer) ControllerGetCapabilities(ctx context.Context, req *
         csi.ControllerServiceCapability_RPC_LIST_VOLUMES,
         csi.ControllerServiceCapability_RPC_EXPAND_VOLUME,
         csi.ControllerServiceCapability_RPC_CREATE_DELETE_SNAPSHOT,
+        csi.ControllerServiceCapability_RPC_LIST_SNAPSHOTS,
+        // TODO: csi.ControllerServiceCapability_RPC_CLONE_VOLUME,
     } {
         controllerServerCapabilities = append(controllerServerCapabilities, functionControllerServerCapabilities(capability))
     }
@@ -353,28 +374,214 @@ func (cs *ControllerServer) ControllerGetCapabilities(ctx context.Context, req *
     }, nil
 }
 
+func invokeList(ctxVars map[string]string, pattern string, expectExist bool) ([]InodeConfig, error)
+{
+    stat, err := invokeCLI(ctxVars, []string{ "ls", "--json", pattern })
+    if (err != nil)
+    {
+        return nil, err
+    }
+    var inodeCfg []InodeConfig
+    err = json.Unmarshal(stat, &inodeCfg)
+    if (err != nil)
+    {
+        return nil, status.Error(codes.Internal, "Invalid JSON in vitastor-cli ls: "+err.Error())
+    }
+    if (expectExist && len(inodeCfg) == 0)
+    {
+        return nil, status.Error(codes.Internal, "Can't find expected image "+pattern+" via vitastor-cli ls")
+    }
+    return inodeCfg, nil
+}
+
 // CreateSnapshot create snapshot of an existing PV
 func (cs *ControllerServer) CreateSnapshot(ctx context.Context, req *csi.CreateSnapshotRequest) (*csi.CreateSnapshotResponse, error)
 {
-    return nil, status.Error(codes.Unimplemented, "")
+    klog.Infof("received controller create snapshot request %+v", protosanitizer.StripSecrets(req))
+    if (req == nil)
+    {
+        return nil, status.Errorf(codes.InvalidArgument, "request cannot be empty")
+    }
+    if (req.SourceVolumeId == "" || req.Name == "")
+    {
+        return nil, status.Error(codes.InvalidArgument, "source volume ID and snapshot name are required fields")
+    }
+
+    // snapshot name
+    snapName := req.Name
+
+    // req.VolumeId is an ugly json string in our case :)
+    ctxVars := make(map[string]string)
+    err := json.Unmarshal([]byte(req.SourceVolumeId), &ctxVars)
+    if (err != nil)
+    {
+        return nil, status.Error(codes.Internal, "volume ID not in JSON format")
+    }
+    volName := ctxVars["name"]
+
+    // Create image using vitastor-cli
+    _, err = invokeCLI(ctxVars, []string{ "create", "--snapshot", snapName, volName })
+    if (err != nil && strings.Index(err.Error(), "already exists") <= 0)
+    {
+        return nil, err
+    }
+
+    // Check created snapshot
+    inodeCfg, err := invokeList(ctxVars, volName+"@"+snapName, true)
+    if (err != nil)
+    {
+        return nil, err
+    }
+
+    // Use ugly JSON snapshot ID again, DeleteSnapshot doesn't have context :-(
+    ctxVars["snapshot"] = snapName
+    snapIdJson, _ := json.Marshal(ctxVars)
+    return &csi.CreateSnapshotResponse{
+        Snapshot: &csi.Snapshot{
+            SizeBytes: int64(inodeCfg[0].Size),
+            SnapshotId: string(snapIdJson),
+            SourceVolumeId: req.SourceVolumeId,
+            CreationTime: &timestamppb.Timestamp{ Seconds: int64(inodeCfg[0].CreateTs) },
+            ReadyToUse: true,
+        },
+    }, nil
 }
 
 // DeleteSnapshot delete provided snapshot of a PV
 func (cs *ControllerServer) DeleteSnapshot(ctx context.Context, req *csi.DeleteSnapshotRequest) (*csi.DeleteSnapshotResponse, error)
 {
-    return nil, status.Error(codes.Unimplemented, "")
+    klog.Infof("received controller delete snapshot request %+v", protosanitizer.StripSecrets(req))
+    if (req == nil)
+    {
+        return nil, status.Errorf(codes.InvalidArgument, "request cannot be empty")
+    }
+    if (req.SnapshotId == "")
+    {
+        return nil, status.Error(codes.InvalidArgument, "snapshot ID is a required field")
+    }
+
+    volVars := make(map[string]string)
+    err := json.Unmarshal([]byte(req.SnapshotId), &volVars)
+    if (err != nil)
+    {
+        return nil, status.Error(codes.Internal, "snapshot ID not in JSON format")
+    }
+    volName := volVars["name"]
+    snapName := volVars["snapshot"]
+
+    ctxVars, _, _ := GetConnectionParams(volVars)
+
+    _, err = invokeCLI(ctxVars, []string{ "rm", volName+"@"+snapName })
+    if (err != nil)
+    {
+        return nil, err
+    }
+
+    return &csi.DeleteSnapshotResponse{}, nil
 }
 
 // ListSnapshots list the snapshots of a PV
 func (cs *ControllerServer) ListSnapshots(ctx context.Context, req *csi.ListSnapshotsRequest) (*csi.ListSnapshotsResponse, error)
 {
-    return nil, status.Error(codes.Unimplemented, "")
+    klog.Infof("received controller list snapshots request %+v", protosanitizer.StripSecrets(req))
+    if (req == nil)
+    {
+        return nil, status.Error(codes.InvalidArgument, "request cannot be empty")
+    }
+
+    volVars := make(map[string]string)
+    err := json.Unmarshal([]byte(req.SourceVolumeId), &volVars)
+    if (err != nil)
+    {
+        return nil, status.Error(codes.Internal, "volume ID not in JSON format")
+    }
+    volName := volVars["name"]
+    ctxVars, _, _ := GetConnectionParams(volVars)
+
+    inodeCfg, err := invokeList(ctxVars, volName+"@*", false)
+    if (err != nil)
+    {
+        return nil, err
+    }
+
+    resp := &csi.ListSnapshotsResponse{}
+    for _, ino := range inodeCfg
+    {
+        snapName := ino.Name[len(volName)+1:]
+        if (len(req.StartingToken) > 0 && snapName < req.StartingToken)
+        {
+        }
+        else if (req.MaxEntries == 0 || len(resp.Entries) < int(req.MaxEntries))
+        {
+            volVars["snapshot"] = snapName
+            snapIdJson, _ := json.Marshal(volVars)
+            resp.Entries = append(resp.Entries, &csi.ListSnapshotsResponse_Entry{
+                Snapshot: &csi.Snapshot{
+                    SizeBytes: int64(ino.Size),
+                    SnapshotId: string(snapIdJson),
+                    SourceVolumeId: req.SourceVolumeId,
+                    CreationTime: &timestamppb.Timestamp{ Seconds: int64(ino.CreateTs) },
+                    ReadyToUse: true,
+                },
+            })
+        }
+        else
+        {
+            resp.NextToken = snapName
+            break
+        }
+    }
+
+    return resp, nil
 }
 
-// ControllerExpandVolume resizes a volume
+// ControllerExpandVolume increases the size of a volume
 func (cs *ControllerServer) ControllerExpandVolume(ctx context.Context, req *csi.ControllerExpandVolumeRequest) (*csi.ControllerExpandVolumeResponse, error)
 {
-    return nil, status.Error(codes.Unimplemented, "")
+    klog.Infof("received controller expand volume request %+v", protosanitizer.StripSecrets(req))
+    if (req == nil)
+    {
+        return nil, status.Error(codes.InvalidArgument, "request cannot be empty")
+    }
+    if (req.VolumeId == "" || req.CapacityRange == nil || req.CapacityRange.RequiredBytes == 0)
+    {
+        return nil, status.Error(codes.InvalidArgument, "VolumeId, CapacityRange and RequiredBytes are required fields")
+    }
+
+    volVars := make(map[string]string)
+    err := json.Unmarshal([]byte(req.VolumeId), &volVars)
+    if (err != nil)
+    {
+        return nil, status.Error(codes.Internal, "volume ID not in JSON format")
+    }
+    volName := volVars["name"]
+    ctxVars, _, _ := GetConnectionParams(volVars)
+
+    inodeCfg, err := invokeList(ctxVars, volName, true)
+    if (err != nil)
+    {
+        return nil, err
+    }
+
+    if (req.CapacityRange.RequiredBytes > 0 && inodeCfg[0].Size < uint64(req.CapacityRange.RequiredBytes))
+    {
+        sz := ((req.CapacityRange.RequiredBytes+4095)/4096)*4096
+        _, err := invokeCLI(ctxVars, []string{ "modify", "--inc_size", "1", "--resize", fmt.Sprintf("%d", sz), volName })
+        if (err != nil)
+        {
+            return nil, err
+        }
+        inodeCfg, err = invokeList(ctxVars, volName, true)
+        if (err != nil)
+        {
+            return nil, err
+        }
+    }
+
+    return &csi.ControllerExpandVolumeResponse{
+        CapacityBytes: int64(inodeCfg[0].Size),
+        NodeExpansionRequired: false,
+    }, nil
 }
 
 // ControllerGetVolume get volume info

csi/src/identityserver.go

@@ -49,6 +49,13 @@ func (is *IdentityServer) GetPluginCapabilities(ctx context.Context, req *csi.Ge
                     },
                 },
             },
+            {
+                Type: &csi.PluginCapability_VolumeExpansion_{
+                    VolumeExpansion: &csi.PluginCapability_VolumeExpansion{
+                        Type: csi.PluginCapability_VolumeExpansion_OFFLINE,
+                    },
+                },
+            },
         },
     }, nil
 }

csi/src/nodeserver.go

@@ -70,10 +70,10 @@ func (ns *NodeServer) NodePublishVolume(ctx context.Context, req *csi.NodePublis
     isBlock := req.GetVolumeCapability().GetBlock() != nil
 
     // Check that it's not already mounted
-    _, error := mount.IsNotMountPoint(ns.mounter, targetPath)
-    if (error != nil)
+    _, err := mount.IsNotMountPoint(ns.mounter, targetPath)
+    if (err != nil)
     {
-        if (os.IsNotExist(error))
+        if (os.IsNotExist(err))
         {
             if (isBlock)
             {
@@ -102,12 +102,12 @@ func (ns *NodeServer) NodePublishVolume(ctx context.Context, req *csi.NodePublis
         }
         else
         {
-            return nil, status.Error(codes.Internal, error.Error())
+            return nil, status.Error(codes.Internal, err.Error())
         }
     }
 
     ctxVars := make(map[string]string)
-    err := json.Unmarshal([]byte(req.VolumeId), &ctxVars)
+    err = json.Unmarshal([]byte(req.VolumeId), &ctxVars)
     if (err != nil)
     {
         return nil, status.Error(codes.Internal, "volume ID not in JSON format")
@@ -147,70 +147,74 @@ func (ns *NodeServer) NodePublishVolume(ctx context.Context, req *csi.NodePublis
     }
     devicePath := strings.TrimSpace(stdoutStr)
 
-    // Check existing format
     diskMounter := &mount.SafeFormatAndMount{Interface: ns.mounter, Exec: utilexec.New()}
-    existingFormat, err := diskMounter.GetDiskFormat(devicePath)
-    if (err != nil)
-    {
-        klog.Errorf("failed to get disk format for path %s, error: %v", err)
-        // unmap NBD device
-        unmapOut, unmapErr := exec.Command("/usr/bin/vitastor-nbd", "unmap", devicePath).CombinedOutput()
-        if (unmapErr != nil)
-        {
-            klog.Errorf("failed to unmap NBD device %s: %s, error: %v", devicePath, unmapOut, unmapErr)
-        }
-        return nil, err
-    }
-
-    // Format the device (ext4 or xfs)
-    fsType := req.GetVolumeCapability().GetMount().GetFsType()
-    opt := req.GetVolumeCapability().GetMount().GetMountFlags()
-    opt = append(opt, "_netdev")
-    if ((req.VolumeCapability.AccessMode.Mode == csi.VolumeCapability_AccessMode_MULTI_NODE_READER_ONLY ||
-        req.VolumeCapability.AccessMode.Mode == csi.VolumeCapability_AccessMode_SINGLE_NODE_READER_ONLY) &&
-        !Contains(opt, "ro"))
-    {
-        opt = append(opt, "ro")
-    }
-    if (fsType == "xfs")
-    {
-        opt = append(opt, "nouuid")
-    }
-    readOnly := Contains(opt, "ro")
-    if (existingFormat == "" && !readOnly)
-    {
-        args := []string{}
-        switch fsType
-        {
-            case "ext4":
-                args = []string{"-m0", "-Enodiscard,lazy_itable_init=1,lazy_journal_init=1", devicePath}
-            case "xfs":
-                args = []string{"-K", devicePath}
-        }
-        if (len(args) > 0)
-        {
-            cmdOut, cmdErr := diskMounter.Exec.Command("mkfs."+fsType, args...).CombinedOutput()
-            if (cmdErr != nil)
-            {
-                klog.Errorf("failed to run mkfs error: %v, output: %v", cmdErr, string(cmdOut))
-                // unmap NBD device
-                unmapOut, unmapErr := exec.Command("/usr/bin/vitastor-nbd", "unmap", devicePath).CombinedOutput()
-                if (unmapErr != nil)
-                {
-                    klog.Errorf("failed to unmap NBD device %s: %s, error: %v", devicePath, unmapOut, unmapErr)
-                }
-                return nil, status.Error(codes.Internal, cmdErr.Error())
-            }
-        }
-    }
     if (isBlock)
     {
-        opt = append(opt, "bind")
-        err = diskMounter.Mount(devicePath, targetPath, fsType, opt)
+        err = diskMounter.Mount(devicePath, targetPath, "", []string{"bind"})
     }
     else
     {
+        // Check existing format
+        existingFormat, err := diskMounter.GetDiskFormat(devicePath)
+        if (err != nil)
+        {
+            klog.Errorf("failed to get disk format for path %s, error: %v", err)
+            goto unmap
+        }
+
+        // Format the device (ext4 or xfs)
+        fsType := req.GetVolumeCapability().GetMount().GetFsType()
+        opt := req.GetVolumeCapability().GetMount().GetMountFlags()
+        opt = append(opt, "_netdev")
+        if ((req.VolumeCapability.AccessMode.Mode == csi.VolumeCapability_AccessMode_MULTI_NODE_READER_ONLY ||
+            req.VolumeCapability.AccessMode.Mode == csi.VolumeCapability_AccessMode_SINGLE_NODE_READER_ONLY) &&
+            !Contains(opt, "ro"))
+        {
+            opt = append(opt, "ro")
+        }
+        if (fsType == "xfs")
+        {
+            opt = append(opt, "nouuid")
+        }
+        readOnly := Contains(opt, "ro")
+        if (existingFormat == "" && !readOnly)
+        {
+            var cmdOut []byte
+            switch fsType
+            {
+                case "ext4":
+                    args := []string{"-m0", "-Enodiscard,lazy_itable_init=1,lazy_journal_init=1", devicePath}
+                    cmdOut, err = diskMounter.Exec.Command("mkfs.ext4", args...).CombinedOutput()
+                case "xfs":
+                    cmdOut, err = diskMounter.Exec.Command("mkfs.xfs", "-K", devicePath).CombinedOutput()
+            }
+            if (err != nil)
+            {
+                klog.Errorf("failed to run mkfs error: %v, output: %v", err, string(cmdOut))
+                goto unmap
+            }
+        }
+
         err = diskMounter.FormatAndMount(devicePath, targetPath, fsType, opt)
+
+        // Try to run online resize on mount.
+        // FIXME: Implement online resize. It requires online resize support in vitastor-nbd.
+        if (err == nil && existingFormat != "" && !readOnly)
+        {
+            var cmdOut []byte
+            switch (fsType)
+            {
+                case "ext4":
+                    cmdOut, err = diskMounter.Exec.Command("resize2fs", devicePath).CombinedOutput()
+                case "xfs":
+                    cmdOut, err = diskMounter.Exec.Command("xfs_growfs", devicePath).CombinedOutput()
+            }
+            if (err != nil)
+            {
+                klog.Errorf("failed to run resizefs error: %v, output: %v", err, string(cmdOut))
+                goto unmap
+            }
+        }
     }
     if (err != nil)
     {
@@ -218,15 +222,18 @@ func (ns *NodeServer) NodePublishVolume(ctx context.Context, req *csi.NodePublis
             "failed to mount device path (%s) to path (%s) for volume (%s) error: %s",
             devicePath, targetPath, volName, err,
         )
-        // unmap NBD device
-        unmapOut, unmapErr := exec.Command("/usr/bin/vitastor-nbd", "unmap", devicePath).CombinedOutput()
-        if (unmapErr != nil)
-        {
-            klog.Errorf("failed to unmap NBD device %s: %s, error: %v", devicePath, unmapOut, unmapErr)
-        }
-        return nil, status.Error(codes.Internal, err.Error())
+        goto unmap
     }
     return &csi.NodePublishVolumeResponse{}, nil
+
+unmap:
+    // unmap NBD device
+    unmapOut, unmapErr := exec.Command("/usr/bin/vitastor-nbd", "unmap", devicePath).CombinedOutput()
+    if (unmapErr != nil)
+    {
+        klog.Errorf("failed to unmap NBD device %s: %s, error: %v", devicePath, unmapOut, unmapErr)
+    }
+    return nil, status.Error(codes.Internal, err.Error())
 }
 
 // NodeUnpublishVolume unmounts the volume from the target path

debian/changelog

@@ -1,10 +1,10 @@
-vitastor (1.0.0-1) unstable; urgency=medium
+vitastor (1.2.0-1) unstable; urgency=medium
 
   * Bugfixes
 
  -- Vitaliy Filippov <vitalif@yourcmc.ru>  Fri, 03 Jun 2022 02:09:44 +0300
 
-vitastor (1.0.0-1) unstable; urgency=medium
+vitastor (1.2.0-1) unstable; urgency=medium
 
   * Implement NFS proxy
   * Add documentation

debian/control

@@ -2,7 +2,7 @@ Source: vitastor
 Section: admin
 Priority: optional
 Maintainer: Vitaliy Filippov <vitalif@yourcmc.ru>
-Build-Depends: debhelper, liburing-dev (>= 0.6), g++ (>= 8), libstdc++6 (>= 8), linux-libc-dev, libgoogle-perftools-dev, libjerasure-dev, libgf-complete-dev, libibverbs-dev, libisal-dev
+Build-Depends: debhelper, liburing-dev (>= 0.6), g++ (>= 8), libstdc++6 (>= 8), linux-libc-dev, libgoogle-perftools-dev, libjerasure-dev, libgf-complete-dev, libibverbs-dev, libisal-dev, cmake, pkg-config
 Standards-Version: 4.5.0
 Homepage: https://vitastor.io/
 Rules-Requires-Root: no

debian/patched-qemu.Dockerfile

@@ -54,7 +54,8 @@ RUN set -e; \
     quilt add block/vitastor.c; \
     cp /root/vitastor/src/qemu_driver.c block/vitastor.c; \
     quilt refresh; \
-    V=$(head -n1 debian/changelog | perl -pe 's/^.*\((.*?)(~bpo[\d\+]*)?\).*$/$1/')+vitastor3; \
+    V=$(head -n1 debian/changelog | perl -pe 's/5\.2\+dfsg-9/5.2+dfsg-11/; s/^.*\((.*?)(~bpo[\d\+]*)?\).*$/$1/')+vitastor4; \
+    if [ "$REL" = bullseye ]; then V=${V}bullseye; fi; \
     DEBEMAIL="Vitaliy Filippov <vitalif@yourcmc.ru>" dch -D $REL -v $V 'Plug Vitastor block driver'; \
     DEB_BUILD_OPTIONS=nocheck dpkg-buildpackage --jobs=auto -sa; \
     rm -rf /root/packages/qemu-$REL/qemu-*/

debian/vitastor.Dockerfile

@@ -35,8 +35,8 @@ RUN set -e -x; \
     mkdir -p /root/packages/vitastor-$REL; \
     rm -rf /root/packages/vitastor-$REL/*; \
     cd /root/packages/vitastor-$REL; \
-    cp -r /root/vitastor vitastor-1.0.0; \
-    cd vitastor-1.0.0; \
+    cp -r /root/vitastor vitastor-1.2.0; \
+    cd vitastor-1.2.0; \
     ln -s /root/fio-build/fio-*/ ./fio; \
     FIO=$(head -n1 fio/debian/changelog | perl -pe 's/^.*\((.*?)\).*$/$1/'); \
     ls /usr/include/linux/raw.h || cp ./debian/raw.h /usr/include/linux/raw.h; \
@@ -49,8 +49,8 @@ RUN set -e -x; \
     rm -rf a b; \
     echo "dep:fio=$FIO" > debian/fio_version; \
     cd /root/packages/vitastor-$REL; \
-    tar --sort=name --mtime='2020-01-01' --owner=0 --group=0 --exclude=debian -cJf vitastor_1.0.0.orig.tar.xz vitastor-1.0.0; \
-    cd vitastor-1.0.0; \
+    tar --sort=name --mtime='2020-01-01' --owner=0 --group=0 --exclude=debian -cJf vitastor_1.2.0.orig.tar.xz vitastor-1.2.0; \
+    cd vitastor-1.2.0; \
     V=$(head -n1 debian/changelog | perl -pe 's/^.*\((.*?)\).*$/$1/'); \
     DEBFULLNAME="Vitaliy Filippov <vitalif@yourcmc.ru>" dch -D $REL -v "$V""$REL" "Rebuild for $REL"; \
     DEB_BUILD_OPTIONS=nocheck dpkg-buildpackage --jobs=auto -sa; \

docs/config.en.md

@@ -33,6 +33,7 @@ In the future, additional configuration methods may be added:
 
 - [Common](config/common.en.md)
 - [Network](config/network.en.md)
+- [Client](config/client.en.md)
 - [Global Disk Layout](config/layout-cluster.en.md)
 - [OSD Disk Layout](config/layout-osd.en.md)
 - [OSD Runtime Parameters](config/osd.en.md)

docs/config.ru.md

@@ -36,6 +36,7 @@
 
 - [Общие](config/common.ru.md)
 - [Сеть](config/network.ru.md)
+- [Клиентский код](config/client.ru.md)
 - [Глобальные дисковые параметры](config/layout-cluster.ru.md)
 - [Дисковые параметры OSD](config/layout-osd.ru.md)
 - [Прочие параметры OSD](config/osd.ru.md)

docs/config/client.en.md

@@ -0,0 +1,103 @@
+[Documentation](../../README.md#documentation) → [Configuration](../config.en.md) → Client Parameters
+
+-----
+
+[Читать на русском](client.ru.md)
+
+# Client Parameters
+
+These parameters apply only to clients and affect their interaction with
+the cluster.
+
+- [client_max_dirty_bytes](#client_max_dirty_bytes)
+- [client_max_dirty_ops](#client_max_dirty_ops)
+- [client_enable_writeback](#client_enable_writeback)
+- [client_max_buffered_bytes](#client_max_buffered_bytes)
+- [client_max_buffered_ops](#client_max_buffered_ops)
+- [client_max_writeback_iodepth](#client_max_writeback_iodepth)
+
+## client_max_dirty_bytes
+
+- Type: integer
+- Default: 33554432
+- Can be changed online: yes
+
+Without [immediate_commit](layout-cluster.en.md#immediate_commit)=all this parameter sets the limit of "dirty"
+(not committed by fsync) data allowed by the client before forcing an
+additional fsync and committing the data. Also note that the client always
+holds a copy of uncommitted data in memory so this setting also affects
+RAM usage of clients.
+
+## client_max_dirty_ops
+
+- Type: integer
+- Default: 1024
+- Can be changed online: yes
+
+Same as client_max_dirty_bytes, but instead of total size, limits the number
+of uncommitted write operations.
+
+## client_enable_writeback
+
+- Type: boolean
+- Default: false
+- Can be changed online: yes
+
+This parameter enables client-side write buffering. This means that write
+requests are accumulated in memory for a short time before being sent to
+a Vitastor cluster which allows to send them in parallel and increase
+performance of some applications. Writes are buffered until client forces
+a flush with fsync() or until the amount of buffered writes exceeds the
+limit.
+
+Write buffering significantly increases performance of some applications,
+for example, CrystalDiskMark under Windows (LOL :-D), but also any other
+applications if they do writes in one of two non-optimal ways: either if
+they do a lot of small (4 kb or so) sequential writes, or if they do a lot
+of small random writes, but without any parallelism or asynchrony, and also
+without calling fsync().
+
+With write buffering enabled, you can expect around 22000 T1Q1 random write
+iops in QEMU more or less regardless of the quality of your SSDs, and this
+number is in fact bound by QEMU itself rather than Vitastor (check it
+yourself by adding a "driver=null-co" disk in QEMU). Without write
+buffering, the current record is 9900 iops, but the number is usually
+even lower with non-ideal hardware, for example, it may be 5000 iops.
+
+Even when this parameter is enabled, write buffering isn't enabled until
+the client explicitly allows it, because enabling it without the client
+being aware of the fact that his writes may be buffered may lead to data
+loss. Because of this, older versions of clients don't support write
+buffering at all, newer versions of the QEMU driver allow write buffering
+only if it's enabled in disk settings with `-blockdev cache.direct=false`,
+and newer versions of FIO only allow write buffering if you don't specify
+`-direct=1`. NBD and NFS drivers allow write buffering by default.
+
+You can overcome this restriction too with the `client_writeback_allowed`
+parameter, but you shouldn't do that unless you **really** know what you
+are doing.
+
+## client_max_buffered_bytes
+
+- Type: integer
+- Default: 33554432
+- Can be changed online: yes
+
+Maximum total size of buffered writes which triggers write-back when reached.
+
+## client_max_buffered_ops
+
+- Type: integer
+- Default: 1024
+- Can be changed online: yes
+
+Maximum number of buffered writes which triggers write-back when reached.
+Multiple consecutive modified data regions are counted as 1 write here.
+
+## client_max_writeback_iodepth
+
+- Type: integer
+- Default: 256
+- Can be changed online: yes
+
+Maximum number of parallel writes when flushing buffered data to the server.

docs/config/client.ru.md

@@ -0,0 +1,103 @@
+[Документация](../../README-ru.md#документация) → [Конфигурация](../config.ru.md) → Параметры клиентского кода
+
+-----
+
+[Read in English](client.en.md)
+
+# Параметры клиентского кода
+
+Данные параметры применяются только к клиентам Vitastor (QEMU, fio, NBD) и
+затрагивают логику их работы с кластером.
+
+- [client_max_dirty_bytes](#client_max_dirty_bytes)
+- [client_max_dirty_ops](#client_max_dirty_ops)
+- [client_enable_writeback](#client_enable_writeback)
+- [client_max_buffered_bytes](#client_max_buffered_bytes)
+- [client_max_buffered_ops](#client_max_buffered_ops)
+- [client_max_writeback_iodepth](#client_max_writeback_iodepth)
+
+## client_max_dirty_bytes
+
+- Тип: целое число
+- Значение по умолчанию: 33554432
+- Можно менять на лету: да
+
+При работе без [immediate_commit](layout-cluster.ru.md#immediate_commit)=all - это лимит объёма "грязных" (не
+зафиксированных fsync-ом) данных, при достижении которого клиент будет
+принудительно вызывать fsync и фиксировать данные. Также стоит иметь в виду,
+что в этом случае до момента fsync клиент хранит копию незафиксированных
+данных в памяти, то есть, настройка влияет на потребление памяти клиентами.
+
+## client_max_dirty_ops
+
+- Тип: целое число
+- Значение по умолчанию: 1024
+- Можно менять на лету: да
+
+Аналогично client_max_dirty_bytes, но ограничивает количество
+незафиксированных операций записи вместо их общего объёма.
+
+## client_enable_writeback
+
+- Тип: булево (да/нет)
+- Значение по умолчанию: false
+- Можно менять на лету: да
+
+Данный параметр разрешает включать буферизацию записи в памяти. Буферизация
+означает, что операции записи отправляются на кластер Vitastor не сразу, а
+могут небольшое время накапливаться в памяти и сбрасываться сразу пакетами,
+до тех пор, пока либо не будет превышен лимит неотправленных записей, либо
+пока клиент не вызовет fsync.
+
+Буферизация значительно повышает производительность некоторых приложений,
+например, CrystalDiskMark в Windows (ха-ха :-D), но также и любых других,
+которые пишут на диск неоптимально: либо последовательно, но мелкими блоками
+(например, по 4 кб), либо случайно, но без параллелизма и без fsync - то
+есть, например, отправляя 128 операций записи в разные места диска, но не
+все сразу с помощью асинхронного I/O, а по одной.
+
+В QEMU с буферизацией записи можно ожидать показателя примерно 22000
+операций случайной записи в секунду в 1 поток и с глубиной очереди 1 (T1Q1)
+без fsync, почти вне зависимости от того, насколько хороши ваши диски - эта
+цифра упирается в сам QEMU. Без буферизации рекорд пока что - 9900 операций
+в секунду, но на железе похуже может быть и поменьше, например, 5000 операций
+в секунду.
+
+При этом, даже если данный параметр включён, буферизация не включается, если
+явно не разрешена клиентом, т.к. если клиент не знает, что запросы записи
+буферизуются, это может приводить к потере данных. Поэтому в старых версиях
+клиентских драйверов буферизация записи не включается вообще, в новых
+версиях QEMU-драйвера включается, только если разрешена опцией диска
+`-blockdev cache.direct=false`, а в fio - только если нет опции `-direct=1`.
+В NBD и NFS драйверах буферизация записи разрешена по умолчанию.
+
+Можно обойти и это ограничение с помощью параметра `client_writeback_allowed`,
+но делать так не надо, если только вы не уверены в том, что делаете, на все
+100%. :-)
+
+## client_max_buffered_bytes
+
+- Тип: целое число
+- Значение по умолчанию: 33554432
+- Можно менять на лету: да
+
+Максимальный общий размер буферизованных записей, при достижении которого
+начинается процесс сброса данных на сервер.
+
+## client_max_buffered_ops
+
+- Тип: целое число
+- Значение по умолчанию: 1024
+- Можно менять на лету: да
+
+Максимальное количество буферизованных записей, при достижении которого
+начинается процесс сброса данных на сервер. При этом несколько
+последовательных изменённых областей здесь считаются 1 записью.
+
+## client_max_writeback_iodepth
+
+- Тип: целое число
+- Значение по умолчанию: 256
+- Можно менять на лету: да
+
+Максимальное число параллельных операций записи при сбросе буферов на сервер.

docs/config/network.en.md

@@ -20,6 +20,7 @@ between clients, OSDs and etcd.
 - [rdma_max_msg](#rdma_max_msg)
 - [rdma_max_recv](#rdma_max_recv)
 - [rdma_max_send](#rdma_max_send)
+- [rdma_odp](#rdma_odp)
 - [peer_connect_interval](#peer_connect_interval)
 - [peer_connect_timeout](#peer_connect_timeout)
 - [osd_idle_timeout](#osd_idle_timeout)
@@ -30,7 +31,6 @@ between clients, OSDs and etcd.
 - [etcd_slow_timeout](#etcd_slow_timeout)
 - [etcd_keepalive_timeout](#etcd_keepalive_timeout)
 - [etcd_ws_keepalive_timeout](#etcd_ws_keepalive_timeout)
-- [client_dirty_limit](#client_dirty_limit)
 
 ## tcp_header_buffer_size
 
@@ -69,11 +69,14 @@ but they are not connected to the cluster.
 - Type: string
 
 RDMA device name to use for Vitastor OSD communications (for example,
-"rocep5s0f0"). Please note that Vitastor RDMA requires Implicit On-Demand
-Paging (Implicit ODP) and Scatter/Gather (SG) support from the RDMA device
-to work. For example, Mellanox ConnectX-3 and older adapters don't have
-Implicit ODP, so they're unsupported by Vitastor. Run `ibv_devinfo -v` as
-root to list available RDMA devices and their features.
+"rocep5s0f0"). Now Vitastor supports all adapters, even ones without
+ODP support, like Mellanox ConnectX-3 and non-Mellanox cards.
+
+Versions up to Vitastor 1.2.0 required ODP which is only present in
+Mellanox ConnectX >= 4. See also [rdma_odp](#rdma_odp).
+
+Run `ibv_devinfo -v` as root to list available RDMA devices and their
+features.
 
 Remember that you also have to configure your network switches if you use
 RoCE/RoCEv2, otherwise you may experience unstable performance. Refer to
@@ -148,6 +151,28 @@ less than `rdma_max_recv` so the receiving side doesn't run out of buffers.
 Doesn't affect memory usage - additional memory isn't allocated for send
 operations.
 
+## rdma_odp
+
+- Type: boolean
+- Default: false
+
+Use RDMA with On-Demand Paging. ODP is currently only available on Mellanox
+ConnectX-4 and newer adapters. ODP allows to not register memory explicitly
+for RDMA adapter to be able to use it. This, in turn, allows to skip memory
+copying during sending. One would think this should improve performance, but
+**in reality** RDMA performance with ODP is **drastically** worse. Example
+3-node cluster with 8 NVMe in each node and 2*25 GBit/s ConnectX-6 RDMA network
+without ODP pushes 3950000 read iops, but only 239000 iops with ODP...
+
+This happens because Mellanox ODP implementation seems to be based on
+message retransmissions when the adapter doesn't know about the buffer yet -
+it likely uses standard "RNR retransmissions" (RNR = receiver not ready)
+which is generally slow in RDMA/RoCE networks. Here's a presentation about
+it from ISPASS-2021 conference: https://tkygtr6.github.io/pub/ISPASS21_slides.pdf
+
+ODP support is retained in the code just in case a good ODP implementation
+appears one day.
+
 ## peer_connect_interval
 
 - Type: seconds
@@ -240,17 +265,3 @@ etcd_report_interval to guarantee that keepalive actually works.
 
 etcd websocket ping interval required to keep the connection alive and
 detect disconnections quickly.
-
-## client_dirty_limit
-
-- Type: integer
-- Default: 33554432
-- Can be changed online: yes
-
-Without immediate_commit=all this parameter sets the limit of "dirty"
-(not committed by fsync) data allowed by the client before forcing an
-additional fsync and committing the data. Also note that the client always
-holds a copy of uncommitted data in memory so this setting also affects
-RAM usage of clients.
-
-This parameter doesn't affect OSDs themselves.

docs/config/network.ru.md

@@ -20,6 +20,7 @@
 - [rdma_max_msg](#rdma_max_msg)
 - [rdma_max_recv](#rdma_max_recv)
 - [rdma_max_send](#rdma_max_send)
+- [rdma_odp](#rdma_odp)
 - [peer_connect_interval](#peer_connect_interval)
 - [peer_connect_timeout](#peer_connect_timeout)
 - [osd_idle_timeout](#osd_idle_timeout)
@@ -30,7 +31,6 @@
 - [etcd_slow_timeout](#etcd_slow_timeout)
 - [etcd_keepalive_timeout](#etcd_keepalive_timeout)
 - [etcd_ws_keepalive_timeout](#etcd_ws_keepalive_timeout)
-- [client_dirty_limit](#client_dirty_limit)
 
 ## tcp_header_buffer_size
 
@@ -72,12 +72,15 @@ RDMA может быть нужно только если у клиентов е
 - Тип: строка
 
 Название RDMA-устройства для связи с Vitastor OSD (например, "rocep5s0f0").
-Имейте в виду, что поддержка RDMA в Vitastor требует функций устройства
-Implicit On-Demand Paging (Implicit ODP) и Scatter/Gather (SG). Например,
-адаптеры Mellanox ConnectX-3 и более старые не поддерживают Implicit ODP и
-потому не поддерживаются в Vitastor. Запустите `ibv_devinfo -v` от имени
-суперпользователя, чтобы посмотреть список доступных RDMA-устройств, их
-параметры и возможности.
+Сейчас Vitastor поддерживает все модели адаптеров, включая те, у которых
+нет поддержки ODP, то есть вы можете использовать RDMA с ConnectX-3 и
+картами производства не Mellanox.
+
+Версии Vitastor до 1.2.0 включительно требовали ODP, который есть только
+на Mellanox ConnectX 4 и более новых. См. также [rdma_odp](#rdma_odp).
+
+Запустите `ibv_devinfo -v` от имени суперпользователя, чтобы посмотреть
+список доступных RDMA-устройств, их параметры и возможности.
 
 Обратите внимание, что если вы используете RoCE/RoCEv2, вам также необходимо
 правильно настроить для него коммутаторы, иначе вы можете столкнуться с
@@ -156,6 +159,29 @@ OSD в любом случае согласовывают реальное зн
 Не влияет на потребление памяти - дополнительная память на операции отправки
 не выделяется.
 
+## rdma_odp
+
+- Тип: булево (да/нет)
+- Значение по умолчанию: false
+
+Использовать RDMA с On-Demand Paging. ODP - функция, доступная пока что
+исключительно на адаптерах Mellanox ConnectX-4 и более новых. ODP позволяет
+не регистрировать память для её использования RDMA-картой. Благодаря этому
+можно не копировать данные при отправке их в сеть и, казалось бы, это должно
+улучшать производительность - но **по факту** получается так, что
+производительность только ухудшается, причём сильно. Пример - на 3-узловом
+кластере с 8 NVMe в каждом узле и сетью 2*25 Гбит/с на чтение с RDMA без ODP
+удаётся снять 3950000 iops, а с ODP - всего 239000 iops...
+
+Это происходит из-за того, что реализация ODP у Mellanox неоптимальная и
+основана на повторной передаче сообщений, когда карте не известен буфер -
+вероятно, на стандартных "RNR retransmission" (RNR = receiver not ready).
+А данные повторные передачи в RDMA/RoCE - всегда очень медленная штука.
+Презентация на эту тему с конференции ISPASS-2021: https://tkygtr6.github.io/pub/ISPASS21_slides.pdf
+
+Возможность использования ODP сохранена в коде на случай, если вдруг в один
+прекрасный день появится хорошая реализация ODP.
+
 ## peer_connect_interval
 
 - Тип: секунды
@@ -251,17 +277,3 @@ etcd_report_interval, чтобы keepalive гарантированно рабо
 - Можно менять на лету: да
 
 Интервал проверки живости вебсокет-подключений к etcd.
-
-## client_dirty_limit
-
-- Тип: целое число
-- Значение по умолчанию: 33554432
-- Можно менять на лету: да
-
-При работе без immediate_commit=all - это лимит объёма "грязных" (не
-зафиксированных fsync-ом) данных, при достижении которого клиент будет
-принудительно вызывать fsync и фиксировать данные. Также стоит иметь в виду,
-что в этом случае до момента fsync клиент хранит копию незафиксированных
-данных в памяти, то есть, настройка влияет на потребление памяти клиентами.
-
-Параметр не влияет на сами OSD.

docs/config/osd.en.md

@@ -11,6 +11,7 @@ initialization and can be changed - either with an OSD restart or, for some of
 them, even without restarting by updating configuration in etcd.
 
 - [etcd_report_interval](#etcd_report_interval)
+- [etcd_stats_interval](#etcd_stats_interval)
 - [run_primary](#run_primary)
 - [osd_network](#osd_network)
 - [bind_address](#bind_address)
@@ -56,11 +57,21 @@ them, even without restarting by updating configuration in etcd.
 - Type: seconds
 - Default: 5
 
-Interval at which OSDs report their state to etcd. Affects OSD lease time
+Interval at which OSDs report their liveness to etcd. Affects OSD lease time
 and thus the failover speed. Lease time is equal to this parameter value
 plus max_etcd_attempts * etcd_quick_timeout because it should be guaranteed
 that every OSD always refreshes its lease in time.
 
+## etcd_stats_interval
+
+- Type: seconds
+- Default: 30
+
+Interval at which OSDs report their statistics to etcd. Highly affects the
+imposed load on etcd, because statistics include a key for every OSD and
+for every PG. At the same time, low statistic intervals make `vitastor-cli`
+statistics more responsive.
+
 ## run_primary
 
 - Type: boolean

docs/config/osd.ru.md

@@ -12,6 +12,7 @@
 изменения конфигурации в etcd.
 
 - [etcd_report_interval](#etcd_report_interval)
+- [etcd_stats_interval](#etcd_stats_interval)
 - [run_primary](#run_primary)
 - [osd_network](#osd_network)
 - [bind_address](#bind_address)
@@ -57,11 +58,21 @@
 - Тип: секунды
 - Значение по умолчанию: 5
 
-Интервал, с которым OSD обновляет своё состояние в etcd. Значение параметра
-влияет на время резервации (lease) OSD и поэтому на скорость переключения
+Интервал, с которым OSD сообщает о том, что жив, в etcd. Значение параметра
+влияет на время резервации (lease) OSD и поэтому - на скорость переключения
 при падении OSD. Время lease равняется значению этого параметра плюс
 max_etcd_attempts * etcd_quick_timeout.
 
+## etcd_stats_interval
+
+- Тип: секунды
+- Значение по умолчанию: 30
+
+Интервал, с которым OSD обновляет свою статистику в etcd. Сильно влияет на
+создаваемую нагрузку на etcd, потому что статистика содержит по ключу на
+каждый OSD и на каждую PG. В то же время низкий интервал делает
+статистику, печатаемую `vitastor-cli`, отзывчивей.
+
 ## run_primary
 
 - Тип: булево (да/нет)

docs/config/src/client.en.md

@@ -0,0 +1,4 @@
+# Client Parameters
+
+These parameters apply only to clients and affect their interaction with
+the cluster.

docs/config/src/client.ru.md

@@ -0,0 +1,4 @@
+# Параметры клиентского кода
+
+Данные параметры применяются только к клиентам Vitastor (QEMU, fio, NBD) и
+затрагивают логику их работы с кластером.

docs/config/src/client.yml

@@ -0,0 +1,124 @@
+- name: client_max_dirty_bytes
+  type: int
+  default: 33554432
+  online: true
+  info: |
+    Without [immediate_commit](layout-cluster.en.md#immediate_commit)=all this parameter sets the limit of "dirty"
+    (not committed by fsync) data allowed by the client before forcing an
+    additional fsync and committing the data. Also note that the client always
+    holds a copy of uncommitted data in memory so this setting also affects
+    RAM usage of clients.
+  info_ru: |
+    При работе без [immediate_commit](layout-cluster.ru.md#immediate_commit)=all - это лимит объёма "грязных" (не
+    зафиксированных fsync-ом) данных, при достижении которого клиент будет
+    принудительно вызывать fsync и фиксировать данные. Также стоит иметь в виду,
+    что в этом случае до момента fsync клиент хранит копию незафиксированных
+    данных в памяти, то есть, настройка влияет на потребление памяти клиентами.
+- name: client_max_dirty_ops
+  type: int
+  default: 1024
+  online: true
+  info: |
+    Same as client_max_dirty_bytes, but instead of total size, limits the number
+    of uncommitted write operations.
+  info_ru: |
+    Аналогично client_max_dirty_bytes, но ограничивает количество
+    незафиксированных операций записи вместо их общего объёма.
+- name: client_enable_writeback
+  type: bool
+  default: false
+  online: true
+  info: |
+    This parameter enables client-side write buffering. This means that write
+    requests are accumulated in memory for a short time before being sent to
+    a Vitastor cluster which allows to send them in parallel and increase
+    performance of some applications. Writes are buffered until client forces
+    a flush with fsync() or until the amount of buffered writes exceeds the
+    limit.
+
+    Write buffering significantly increases performance of some applications,
+    for example, CrystalDiskMark under Windows (LOL :-D), but also any other
+    applications if they do writes in one of two non-optimal ways: either if
+    they do a lot of small (4 kb or so) sequential writes, or if they do a lot
+    of small random writes, but without any parallelism or asynchrony, and also
+    without calling fsync().
+
+    With write buffering enabled, you can expect around 22000 T1Q1 random write
+    iops in QEMU more or less regardless of the quality of your SSDs, and this
+    number is in fact bound by QEMU itself rather than Vitastor (check it
+    yourself by adding a "driver=null-co" disk in QEMU). Without write
+    buffering, the current record is 9900 iops, but the number is usually
+    even lower with non-ideal hardware, for example, it may be 5000 iops.
+
+    Even when this parameter is enabled, write buffering isn't enabled until
+    the client explicitly allows it, because enabling it without the client
+    being aware of the fact that his writes may be buffered may lead to data
+    loss. Because of this, older versions of clients don't support write
+    buffering at all, newer versions of the QEMU driver allow write buffering
+    only if it's enabled in disk settings with `-blockdev cache.direct=false`,
+    and newer versions of FIO only allow write buffering if you don't specify
+    `-direct=1`. NBD and NFS drivers allow write buffering by default.
+
+    You can overcome this restriction too with the `client_writeback_allowed`
+    parameter, but you shouldn't do that unless you **really** know what you
+    are doing.
+  info_ru: |
+    Данный параметр разрешает включать буферизацию записи в памяти. Буферизация
+    означает, что операции записи отправляются на кластер Vitastor не сразу, а
+    могут небольшое время накапливаться в памяти и сбрасываться сразу пакетами,
+    до тех пор, пока либо не будет превышен лимит неотправленных записей, либо
+    пока клиент не вызовет fsync.
+
+    Буферизация значительно повышает производительность некоторых приложений,
+    например, CrystalDiskMark в Windows (ха-ха :-D), но также и любых других,
+    которые пишут на диск неоптимально: либо последовательно, но мелкими блоками
+    (например, по 4 кб), либо случайно, но без параллелизма и без fsync - то
+    есть, например, отправляя 128 операций записи в разные места диска, но не
+    все сразу с помощью асинхронного I/O, а по одной.
+
+    В QEMU с буферизацией записи можно ожидать показателя примерно 22000
+    операций случайной записи в секунду в 1 поток и с глубиной очереди 1 (T1Q1)
+    без fsync, почти вне зависимости от того, насколько хороши ваши диски - эта
+    цифра упирается в сам QEMU. Без буферизации рекорд пока что - 9900 операций
+    в секунду, но на железе похуже может быть и поменьше, например, 5000 операций
+    в секунду.
+
+    При этом, даже если данный параметр включён, буферизация не включается, если
+    явно не разрешена клиентом, т.к. если клиент не знает, что запросы записи
+    буферизуются, это может приводить к потере данных. Поэтому в старых версиях
+    клиентских драйверов буферизация записи не включается вообще, в новых
+    версиях QEMU-драйвера включается, только если разрешена опцией диска
+    `-blockdev cache.direct=false`, а в fio - только если нет опции `-direct=1`.
+    В NBD и NFS драйверах буферизация записи разрешена по умолчанию.
+
+    Можно обойти и это ограничение с помощью параметра `client_writeback_allowed`,
+    но делать так не надо, если только вы не уверены в том, что делаете, на все
+    100%. :-)
+- name: client_max_buffered_bytes
+  type: int
+  default: 33554432
+  online: true
+  info: |
+    Maximum total size of buffered writes which triggers write-back when reached.
+  info_ru: |
+    Максимальный общий размер буферизованных записей, при достижении которого
+    начинается процесс сброса данных на сервер.
+- name: client_max_buffered_ops
+  type: int
+  default: 1024
+  online: true
+  info: |
+    Maximum number of buffered writes which triggers write-back when reached.
+    Multiple consecutive modified data regions are counted as 1 write here.
+  info_ru: |
+    Максимальное количество буферизованных записей, при достижении которого
+    начинается процесс сброса данных на сервер. При этом несколько
+    последовательных изменённых областей здесь считаются 1 записью.
+- name: client_max_writeback_iodepth
+  type: int
+  default: 256
+  online: true
+  info: |
+    Maximum number of parallel writes when flushing buffered data to the server.
+  info_ru: |
+    Максимальное число параллельных операций записи при сбросе буферов на сервер.

docs/config/src/included.en.md

@@ -28,6 +28,8 @@
 
 {{../../config/network.en.md|indent=2}}
 
+{{../../config/client.en.md|indent=2}}
+
 {{../../config/layout-cluster.en.md|indent=2}}
 
 {{../../config/layout-osd.en.md|indent=2}}

docs/config/src/included.ru.md

@@ -28,6 +28,8 @@
 
 {{../../config/network.ru.md|indent=2}}
 
+{{../../config/client.ru.md|indent=2}}
+
 {{../../config/layout-cluster.ru.md|indent=2}}
 
 {{../../config/layout-osd.ru.md|indent=2}}

docs/config/src/network.yml

@@ -48,11 +48,14 @@
   type: string
   info: |
     RDMA device name to use for Vitastor OSD communications (for example,
-    "rocep5s0f0"). Please note that Vitastor RDMA requires Implicit On-Demand
-    Paging (Implicit ODP) and Scatter/Gather (SG) support from the RDMA device
-    to work. For example, Mellanox ConnectX-3 and older adapters don't have
-    Implicit ODP, so they're unsupported by Vitastor. Run `ibv_devinfo -v` as
-    root to list available RDMA devices and their features.
+    "rocep5s0f0"). Now Vitastor supports all adapters, even ones without
+    ODP support, like Mellanox ConnectX-3 and non-Mellanox cards.
+
+    Versions up to Vitastor 1.2.0 required ODP which is only present in
+    Mellanox ConnectX >= 4. See also [rdma_odp](#rdma_odp).
+
+    Run `ibv_devinfo -v` as root to list available RDMA devices and their
+    features.
 
     Remember that you also have to configure your network switches if you use
     RoCE/RoCEv2, otherwise you may experience unstable performance. Refer to
@@ -61,12 +64,15 @@
     PFC (Priority Flow Control) and ECN (Explicit Congestion Notification).
   info_ru: |
     Название RDMA-устройства для связи с Vitastor OSD (например, "rocep5s0f0").
-    Имейте в виду, что поддержка RDMA в Vitastor требует функций устройства
-    Implicit On-Demand Paging (Implicit ODP) и Scatter/Gather (SG). Например,
-    адаптеры Mellanox ConnectX-3 и более старые не поддерживают Implicit ODP и
-    потому не поддерживаются в Vitastor. Запустите `ibv_devinfo -v` от имени
-    суперпользователя, чтобы посмотреть список доступных RDMA-устройств, их
-    параметры и возможности.
+    Сейчас Vitastor поддерживает все модели адаптеров, включая те, у которых
+    нет поддержки ODP, то есть вы можете использовать RDMA с ConnectX-3 и
+    картами производства не Mellanox.
+
+    Версии Vitastor до 1.2.0 включительно требовали ODP, который есть только
+    на Mellanox ConnectX 4 и более новых. См. также [rdma_odp](#rdma_odp).
+
+    Запустите `ibv_devinfo -v` от имени суперпользователя, чтобы посмотреть
+    список доступных RDMA-устройств, их параметры и возможности.
 
     Обратите внимание, что если вы используете RoCE/RoCEv2, вам также необходимо
     правильно настроить для него коммутаторы, иначе вы можете столкнуться с
@@ -160,6 +166,45 @@
     у принимающей стороны в процессе работы не заканчивались буферы на приём.
     Не влияет на потребление памяти - дополнительная память на операции отправки
     не выделяется.
+- name: rdma_odp
+  type: bool
+  default: false
+  online: false
+  info: |
+    Use RDMA with On-Demand Paging. ODP is currently only available on Mellanox
+    ConnectX-4 and newer adapters. ODP allows to not register memory explicitly
+    for RDMA adapter to be able to use it. This, in turn, allows to skip memory
+    copying during sending. One would think this should improve performance, but
+    **in reality** RDMA performance with ODP is **drastically** worse. Example
+    3-node cluster with 8 NVMe in each node and 2*25 GBit/s ConnectX-6 RDMA network
+    without ODP pushes 3950000 read iops, but only 239000 iops with ODP...
+
+    This happens because Mellanox ODP implementation seems to be based on
+    message retransmissions when the adapter doesn't know about the buffer yet -
+    it likely uses standard "RNR retransmissions" (RNR = receiver not ready)
+    which is generally slow in RDMA/RoCE networks. Here's a presentation about
+    it from ISPASS-2021 conference: https://tkygtr6.github.io/pub/ISPASS21_slides.pdf
+
+    ODP support is retained in the code just in case a good ODP implementation
+    appears one day.
+  info_ru: |
+    Использовать RDMA с On-Demand Paging. ODP - функция, доступная пока что
+    исключительно на адаптерах Mellanox ConnectX-4 и более новых. ODP позволяет
+    не регистрировать память для её использования RDMA-картой. Благодаря этому
+    можно не копировать данные при отправке их в сеть и, казалось бы, это должно
+    улучшать производительность - но **по факту** получается так, что
+    производительность только ухудшается, причём сильно. Пример - на 3-узловом
+    кластере с 8 NVMe в каждом узле и сетью 2*25 Гбит/с на чтение с RDMA без ODP
+    удаётся снять 3950000 iops, а с ODP - всего 239000 iops...
+
+    Это происходит из-за того, что реализация ODP у Mellanox неоптимальная и
+    основана на повторной передаче сообщений, когда карте не известен буфер -
+    вероятно, на стандартных "RNR retransmission" (RNR = receiver not ready).
+    А данные повторные передачи в RDMA/RoCE - всегда очень медленная штука.
+    Презентация на эту тему с конференции ISPASS-2021: https://tkygtr6.github.io/pub/ISPASS21_slides.pdf
+
+    Возможность использования ODP сохранена в коде на случай, если вдруг в один
+    прекрасный день появится хорошая реализация ODP.
 - name: peer_connect_interval
   type: sec
   min: 1
@@ -259,23 +304,3 @@
     detect disconnections quickly.
   info_ru: |
     Интервал проверки живости вебсокет-подключений к etcd.
-- name: client_dirty_limit
-  type: int
-  default: 33554432
-  online: true
-  info: |
-    Without immediate_commit=all this parameter sets the limit of "dirty"
-    (not committed by fsync) data allowed by the client before forcing an
-    additional fsync and committing the data. Also note that the client always
-    holds a copy of uncommitted data in memory so this setting also affects
-    RAM usage of clients.
-
-    This parameter doesn't affect OSDs themselves.
-  info_ru: |
-    При работе без immediate_commit=all - это лимит объёма "грязных" (не
-    зафиксированных fsync-ом) данных, при достижении которого клиент будет
-    принудительно вызывать fsync и фиксировать данные. Также стоит иметь в виду,
-    что в этом случае до момента fsync клиент хранит копию незафиксированных
-    данных в памяти, то есть, настройка влияет на потребление памяти клиентами.
-
-    Параметр не влияет на сами OSD.

docs/config/src/osd.yml

@@ -2,15 +2,28 @@
   type: sec
   default: 5
   info: |
-    Interval at which OSDs report their state to etcd. Affects OSD lease time
+    Interval at which OSDs report their liveness to etcd. Affects OSD lease time
     and thus the failover speed. Lease time is equal to this parameter value
     plus max_etcd_attempts * etcd_quick_timeout because it should be guaranteed
     that every OSD always refreshes its lease in time.
   info_ru: |
-    Интервал, с которым OSD обновляет своё состояние в etcd. Значение параметра
-    влияет на время резервации (lease) OSD и поэтому на скорость переключения
+    Интервал, с которым OSD сообщает о том, что жив, в etcd. Значение параметра
+    влияет на время резервации (lease) OSD и поэтому - на скорость переключения
     при падении OSD. Время lease равняется значению этого параметра плюс
     max_etcd_attempts * etcd_quick_timeout.
+- name: etcd_stats_interval
+  type: sec
+  default: 30
+  info: |
+    Interval at which OSDs report their statistics to etcd. Highly affects the
+    imposed load on etcd, because statistics include a key for every OSD and
+    for every PG. At the same time, low statistic intervals make `vitastor-cli`
+    statistics more responsive.
+  info_ru: |
+    Интервал, с которым OSD обновляет свою статистику в etcd. Сильно влияет на
+    создаваемую нагрузку на etcd, потому что статистика содержит по ключу на
+    каждый OSD и на каждую PG. В то же время низкий интервал делает
+    статистику, печатаемую `vitastor-cli`, отзывчивей.
 - name: run_primary
   type: bool
   default: true

docs/installation/kubernetes.en.md

@@ -17,4 +17,15 @@ and apply all `NNN-*.yaml` manifests to your Kubernetes installation:
 for i in ./???-*.yaml; do kubectl apply -f $i; done
 ```
 
-After that you'll be able to create PersistentVolumes. See example in [csi/deploy/example-pvc.yaml](../../csi/deploy/example-pvc.yaml).
+After that you'll be able to create PersistentVolumes.
+
+## Features
+
+Vitastor CSI supports:
+- Kubernetes starting with 1.20 (or 1.17 for older vitastor-csi <= 1.1.0)
+- Filesystem RWO (ReadWriteOnce) volumes. Example: [PVC](../../csi/deploy/example-pvc.yaml), [pod](../../csi/deploy/example-test-pod.yaml)
+- Raw block RWX (ReadWriteMany) volumes. Example: [PVC](../../csi/deploy/example-pvc-block.yaml), [pod](../../csi/deploy/example-test-pod-block.yaml)
+- Volume expansion
+- Volume snapshots. Example: [snapshot class](../../csi/deploy/example-snapshot-class.yaml), [snapshot](../../csi/deploy/example-snapshot.yaml), [clone](../../csi/deploy/example-snapshot-clone.yaml)
+
+Remember that to use snapshots with CSI you also have to install [Snapshot Controller and CRDs](https://kubernetes-csi.github.io/docs/snapshot-controller.html#deployment).

docs/installation/kubernetes.ru.md

@@ -17,4 +17,15 @@
 for i in ./???-*.yaml; do kubectl apply -f $i; done
 ```
 
-После этого вы сможете создавать PersistentVolume. Пример смотрите в файле [csi/deploy/example-pvc.yaml](../../csi/deploy/example-pvc.yaml).
+После этого вы сможете создавать PersistentVolume.
+
+## Возможности
+
+CSI-плагин Vitastor поддерживает:
+- Версии Kubernetes, начиная с 1.20 (или с 1.17 для более старых vitastor-csi <= 1.1.0)
+- Файловые RWO (ReadWriteOnce) тома. Пример: [PVC](../../csi/deploy/example-pvc.yaml), [под](../../csi/deploy/example-test-pod.yaml)
+- Сырые блочные RWX (ReadWriteMany) тома. Пример: [PVC](../../csi/deploy/example-pvc-block.yaml), [под](../../csi/deploy/example-test-pod-block.yaml)
+- Расширение размера томов
+- Снимки томов. Пример: [класс снимков](../../csi/deploy/example-snapshot-class.yaml), [снимок](../../csi/deploy/example-snapshot.yaml), [клон снимка](../../csi/deploy/example-snapshot-clone.yaml)
+
+Не забывайте, что для использования снимков нужно сначала установить [контроллер снимков и CRD](https://kubernetes-csi.github.io/docs/snapshot-controller.html#deployment).

docs/intro/features.en.md

@@ -31,6 +31,7 @@
 - [RDMA/RoCEv2 support via libibverbs](../config/network.en.md#rdma_device)
 - [Scrubbing](../config/osd.en.md#auto_scrub) (verification of copies)
 - [Checksums](../config/layout-osd.en.md#data_csum_type)
+- [Client write-back cache](../config/client.en.md#client_enable_writeback)
 
 ## Plugins and tools
 
@@ -50,13 +51,15 @@
 
 The following features are planned for the future:
 
+- File system
+- Control plane optimisation
 - Other administrative tools
 - Web GUI
 - OpenNebula plugin
-- iSCSI proxy
+- iSCSI and NVMeoF gateways
 - Multi-threaded client
 - Faster failover
+- S3
 - Tiered storage (SSD caching)
 - NVDIMM support
 - Compression (possibly)
-- Read caching using system page cache (possibly)

docs/intro/features.ru.md

@@ -33,6 +33,7 @@
 - [Поддержка RDMA/RoCEv2 через libibverbs](../config/network.ru.md#rdma_device)
 - [Фоновая проверка целостности](../config/osd.ru.md#auto_scrub) (сверка копий)
 - [Контрольные суммы](../config/layout-osd.ru.md#data_csum_type)
+- [Буферизация записи на стороне клиента](../config/client.ru.md#client_enable_writeback)
 
 ## Драйверы и инструменты
 
@@ -50,12 +51,15 @@
 
 ## Планы развития
 
+- Файловая система
+- Оптимизация слоя управления
 - Другие инструменты администрирования
 - Web-интерфейс
 - Плагин для OpenNebula
-- iSCSI-прокси
+- iSCSI и NVMeoF прокси
 - Многопоточный клиент
 - Более быстрое переключение при отказах
+- S3
 - Поддержка SSD-кэширования (tiered storage)
 - Поддержка NVDIMM
 - Возможно, сжатие

docs/usage/qemu.en.md

@@ -34,6 +34,20 @@ qemu-system-x86_64 -enable-kvm -m 1024 \
     -vnc 0.0.0.0:0
 ```
 
+With a separate I/O thread:
+
+```
+qemu-system-x86_64 -enable-kvm -m 1024 \
+    -object iothread,id=vitastor1 \
+    -blockdev '{"node-name":"drive-virtio-disk0","driver":"vitastor","image":"debian9",
+        "cache":{"direct":true,"no-flush":false},"auto-read-only":true,"discard":"unmap"}' \
+    -device 'virtio-blk-pci,iothread=vitastor1,scsi=off,bus=pci.0,addr=0x5,drive=drive-virtio-disk0,
+        id=virtio-disk0,bootindex=1,write-cache=off' \
+    -vnc 0.0.0.0:0
+```
+
+You can also specify inode ID, pool and size manually instead of `:image=<IMAGE>` option: `:pool=<POOL>:inode=<INODE>:size=<SIZE>`.
+
 ## qemu-img
 
 For qemu-img, you should use `vitastor:etcd_host=<HOST>:image=<IMAGE>` as filename.
@@ -84,25 +98,75 @@ This can be used for backups. Just note that exporting an image that is currentl
 is of course unsafe and doesn't produce a consistent result, so only export snapshots if you do this
 on a live VM.
 
+## vhost-user-blk
+
+QEMU, starting with 6.0, includes support for attaching disks via a separate
+userspace worker process, called `vhost-user-blk`. It usually has slightly (20-30 us)
+lower latency.
+
+Example commands to use it with Vitastor:
+
+```
+qemu-storage-daemon \
+    --daemonize \
+    --blockdev '{"node-name":"drive-virtio-disk1","driver":"vitastor","image":"testosd1","cache":{"direct":true,"no-flush":false},"auto-read-only":true,"discard":"unmap"}' \
+    --export type=vhost-user-blk,id=vitastor1,node-name=drive-virtio-disk1,addr.type=unix,addr.path=/run/vitastor1-user-blk.sock,writable=on,num-queues=1
+
+qemu-system-x86_64 -enable-kvm -m 2048 -M accel=kvm,memory-backend=mem \
+    -object memory-backend-memfd,id=mem,size=2G,share=on \
+    -chardev socket,id=vitastor1,reconnect=1,path=/run/vitastor1-user-blk.sock \
+    -device vhost-user-blk-pci,chardev=vitastor1,num-queues=1,config-wce=off \
+    -vnc 0.0.0.0:0
+```
+
+memfd memory-backend is crucial, vhost-user-blk does not work without it.
+
 ## VDUSE
 
 Linux kernel, starting with version 5.15, supports a new interface for attaching virtual disks
 to the host - VDUSE (vDPA Device in Userspace). QEMU, starting with 7.2, has support for
 exporting QEMU block devices over this protocol using qemu-storage-daemon.
 
-VDUSE has the same problem as other FUSE-like interfaces in Linux: if a userspace process hangs,
-for example, if it loses connectivity with Vitastor cluster - active processes doing I/O may
-hang in the D state (uninterruptible sleep) and you won't be able to kill them even with kill -9.
-In this case reboot will be the only way to remove VDUSE devices from system.
+VDUSE is currently the best interface to attach Vitastor disks as kernel devices because:
+- It avoids data copies and thus achieves much better performance than [NBD](nbd.en.md)
+- It doesn't have NBD timeout problem - the device doesn't die if an operation executes for too long
+- It doesn't have hung device problem - if the userspace process dies it can be restarted (!)
+  and block device will continue operation
+- It doesn't seem to have the device number limit
 
-On the other hand, VDUSE is faster than [NBD](nbd.en.md), so you may prefer to use it if
-performance is important for you. Approximate performance numbers:
-direct fio benchmark - 115000 iops, NBD - 60000 iops, VDUSE - 90000 iops.
+Example performance comparison:
+
+|                      | direct fio  | NBD         | VDUSE       |
+|----------------------|-------------|-------------|-------------|
+| linear write         | 3.85 GB/s   | 1.12 GB/s   | 3.85 GB/s   |
+| 4k random write Q128 | 240000 iops | 120000 iops | 178000 iops |
+| 4k random write Q1   | 9500 iops   | 7620 iops   | 7640 iops   |
+| linear read          | 4.3 GB/s    | 1.8 GB/s    | 2.85 GB/s   |
+| 4k random read Q128  | 287000 iops | 140000 iops | 189000 iops |
+| 4k random read Q1    | 9600 iops   | 7640 iops   | 7780 iops   |
 
 To try VDUSE you need at least Linux 5.15, built with VDUSE support
-(CONFIG_VIRTIO_VDPA=m and CONFIG_VDPA_USER=m). Debian Linux kernels have these options
-disabled by now, so if you want to try it on Debian, use a kernel from Ubuntu
-[kernel-ppa/mainline](https://kernel.ubuntu.com/~kernel-ppa/mainline/) or Proxmox.
+(CONFIG_VIRTIO_VDPA=m, CONFIG_VDPA_USER=m, CONFIG_VIRTIO_VDPA=m).
+
+Debian Linux kernels have these options disabled by now, so if you want to try it on Debian,
+use a kernel from Ubuntu [kernel-ppa/mainline](https://kernel.ubuntu.com/~kernel-ppa/mainline/), Proxmox,
+or build modules for Debian kernel manually:
+
+```
+mkdir build
+cd build
+apt-get install linux-headers-`uname -r`
+apt-get build-dep linux-image-`uname -r`-unsigned
+apt-get source linux-image-`uname -r`-unsigned
+cd linux*/drivers/vdpa
+make -C /lib/modules/`uname -r`/build M=$PWD CONFIG_VDPA=m CONFIG_VDPA_USER=m CONFIG_VIRTIO_VDPA=m -j8 modules modules_install
+cat Module.symvers >> /lib/modules/`uname -r`/build/Module.symvers
+cd ../virtio
+make -C /lib/modules/`uname -r`/build M=$PWD CONFIG_VDPA=m CONFIG_VDPA_USER=m CONFIG_VIRTIO_VDPA=m -j8 modules modules_install
+depmod -a
+```
+
+You also need `vdpa` tool from the `iproute2` package.
 
 Commands to attach Vitastor image as a VDUSE device:
 
@@ -115,7 +179,7 @@ qemu-storage-daemon --daemonize --blockdev '{"node-name":"test1","driver":"vitas
 vdpa dev add name test1 mgmtdev vduse
 ```
 
-After running these commands /dev/vda device will appear in the system and you'll be able to
+After running these commands, `/dev/vda` device will appear in the system and you'll be able to
 use it as a normal disk.
 
 To remove the device:

docs/usage/qemu.ru.md

@@ -36,6 +36,18 @@ qemu-system-x86_64 -enable-kvm -m 1024 \
     -vnc 0.0.0.0:0
 ```
 
+С отдельным потоком ввода-вывода:
+
+```
+qemu-system-x86_64 -enable-kvm -m 1024 \
+    -object iothread,id=vitastor1 \
+    -blockdev '{"node-name":"drive-virtio-disk0","driver":"vitastor","image":"debian9",
+        "cache":{"direct":true,"no-flush":false},"auto-read-only":true,"discard":"unmap"}' \
+    -device 'virtio-blk-pci,iothread=vitastor1,scsi=off,bus=pci.0,addr=0x5,drive=drive-virtio-disk0,
+        id=virtio-disk0,bootindex=1,write-cache=off' \
+    -vnc 0.0.0.0:0
+```
+
 Вместо `:image=<IMAGE>` также можно указывать номер инода, пул и размер: `:pool=<POOL>:inode=<INODE>:size=<SIZE>`.
 
 ## qemu-img
@@ -88,25 +100,76 @@ qemu-img rebase -u -b '' testimg.qcow2
 в то же время идёт запись, небезопасно - результат чтения не будет целостным. Так что если вы работаете
 с активными виртуальными машинами, экспортируйте только их снимки, но не сам образ.
 
+## vhost-user-blk
+
+QEMU, начиная с 6.0, позволяет подключать диски через отдельный рабочий процесс.
+Этот метод подключения называется `vhost-user-blk` и обычно имеет чуть меньшую
+задержку (ниже на 20-30 микросекунд, чем при обычном методе).
+
+Пример команд для использования vhost-user-blk с Vitastor:
+
+```
+qemu-storage-daemon \
+    --daemonize \
+    --blockdev '{"node-name":"drive-virtio-disk1","driver":"vitastor","image":"testosd1","cache":{"direct":true,"no-flush":false},"auto-read-only":true,"discard":"unmap"}' \
+    --export type=vhost-user-blk,id=vitastor1,node-name=drive-virtio-disk1,addr.type=unix,addr.path=/run/vitastor1-user-blk.sock,writable=on,num-queues=1
+
+qemu-system-x86_64 -enable-kvm -m 2048 -M accel=kvm,memory-backend=mem \
+    -object memory-backend-memfd,id=mem,size=2G,share=on \
+    -chardev socket,id=vitastor1,reconnect=1,path=/run/vitastor1-user-blk.sock \
+    -device vhost-user-blk-pci,chardev=vitastor1,num-queues=1,config-wce=off \
+    -vnc 0.0.0.0:0
+```
+
+Здесь критична опция memory-backend-memfd, vhost-user-blk без неё не работает.
+
 ## VDUSE
 
 В Linux, начиная с версии ядра 5.15, доступен новый интерфейс для подключения виртуальных дисков
 к системе - VDUSE (vDPA Device in Userspace), а в QEMU, начиная с версии 7.2, есть поддержка
 экспорта блочных устройств QEMU по этому протоколу через qemu-storage-daemon.
 
-VDUSE страдает общей проблемой FUSE-подобных интерфейсов в Linux: если пользовательский процесс
-подвиснет, например, если будет потеряна связь с кластером Vitastor - читающие/пишущие в кластер
-процессы могут "залипнуть" в состоянии D (непрерываемый сон) и их будет невозможно убить даже
-через kill -9. В этом случае удалить из системы устройство можно только перезагрузившись.
+VDUSE - на данный момент лучший интерфейс для подключения дисков Vitastor в виде блочных
+устройств на уровне ядра, ибо:
+- VDUSE не копирует данные и поэтому достигает значительно лучшей производительности, чем [NBD](nbd.ru.md)
+- Также оно не имеет проблемы NBD-таймаута - устройство не умирает, если операция выполняется слишком долго
+- Также оно не имеет проблемы подвисающих устройств - если процесс-обработчик умирает, его можно
+  перезапустить (!) и блочное устройство продолжит работать
+- По-видимому, у него нет предела числа подключаемых в систему устройств
 
-С другой стороны, VDUSE быстрее по сравнению с [NBD](nbd.ru.md), поэтому его может
-быть предпочтительно использовать там, где производительность важнее. Порядок показателей:
-прямое тестирование через fio - 115000 iops, NBD - 60000 iops, VDUSE - 90000 iops.
+Пример сравнения производительности:
 
-Чтобы использовать VDUSE, вам нужно ядро Linux версии хотя бы 5.15, собранное с поддержкой
-VDUSE (CONFIG_VIRTIO_VDPA=m и CONFIG_VDPA_USER=m). В ядрах в Debian Linux поддержка пока
-отключена - если хотите попробовать эту функцию на Debian, поставьте ядро из Ubuntu
-[kernel-ppa/mainline](https://kernel.ubuntu.com/~kernel-ppa/mainline/) или из Proxmox.
+|                          | Прямой fio  | NBD         | VDUSE       |
+|--------------------------|-------------|-------------|-------------|
+| линейная запись          | 3.85 GB/s   | 1.12 GB/s   | 3.85 GB/s   |
+| 4k случайная запись Q128 | 240000 iops | 120000 iops | 178000 iops |
+| 4k случайная запись Q1   | 9500 iops   | 7620 iops   | 7640 iops   |
+| линейное чтение          | 4.3 GB/s    | 1.8 GB/s    | 2.85 GB/s   |
+| 4k случайное чтение Q128 | 287000 iops | 140000 iops | 189000 iops |
+| 4k случайное чтение Q1   | 9600 iops   | 7640 iops   | 7780 iops   |
+
+Чтобы попробовать VDUSE, вам нужно ядро Linux как минимум версии 5.15, собранное с поддержкой
+VDUSE (CONFIG_VIRTIO_VDPA=m, CONFIG_VDPA_USER=m, CONFIG_VIRTIO_VDPA=m).
+
+В ядрах в Debian Linux поддержка пока отключена по умолчанию, так что чтобы попробовать VDUSE
+на Debian, поставьте ядро из Ubuntu [kernel-ppa/mainline](https://kernel.ubuntu.com/~kernel-ppa/mainline/),
+из Proxmox или соберите модули для ядра Debian вручную:
+
+```
+mkdir build
+cd build
+apt-get install linux-headers-`uname -r`
+apt-get build-dep linux-image-`uname -r`-unsigned
+apt-get source linux-image-`uname -r`-unsigned
+cd linux*/drivers/vdpa
+make -C /lib/modules/`uname -r`/build M=$PWD CONFIG_VDPA=m CONFIG_VDPA_USER=m CONFIG_VIRTIO_VDPA=m -j8 modules modules_install
+cat Module.symvers >> /lib/modules/`uname -r`/build/Module.symvers
+cd ../virtio
+make -C /lib/modules/`uname -r`/build M=$PWD CONFIG_VDPA=m CONFIG_VDPA_USER=m CONFIG_VIRTIO_VDPA=m -j8 modules modules_install
+depmod -a
+```
+
+Также вам понадобится консольная утилита `vdpa` из пакета `iproute2`.
 
 Команды для подключения виртуального диска через VDUSE:
 
@@ -119,7 +182,7 @@ qemu-storage-daemon --daemonize --blockdev '{"node-name":"test1","driver":"vitas
 vdpa dev add name test1 mgmtdev vduse
 ```
 
-После этого в системе появится устройство /dev/vda, которое можно будет использовать как
+После этого в системе появится устройство `/dev/vda`, которое можно будет использовать как
 обычный диск.
 
 Для удаления устройства из системы:

mon/90-vitastor.rules

@@ -3,5 +3,5 @@ SUBSYSTEM=="block", ENV{ID_PART_ENTRY_TYPE}=="e7009fac-a5a1-4d72-af72-53de130599
     IMPORT{program}="/usr/bin/vitastor-disk udev $devnode", \
     SYMLINK+="vitastor/$env{VITASTOR_ALIAS}"
 
-ENV{VITASTOR_OSD_NUM}!="", ACTION=="add", RUN{program}+="/usr/bin/systemctl enable --now vitastor-osd@$env{VITASTOR_OSD_NUM}"
-ENV{VITASTOR_OSD_NUM}!="", ACTION=="remove", RUN{program}+="/usr/bin/systemctl disable --now vitastor-osd@$env{VITASTOR_OSD_NUM}"
+ENV{VITASTOR_OSD_NUM}!="", ACTION=="add", RUN{program}+="/usr/bin/systemctl enable --now --no-block vitastor-osd@$env{VITASTOR_OSD_NUM}"
+ENV{VITASTOR_OSD_NUM}!="", ACTION=="remove", RUN{program}+="/usr/bin/systemctl disable --now --no-block vitastor-osd@$env{VITASTOR_OSD_NUM}"

mon/mon.js

@@ -99,6 +99,7 @@ const etcd_tree = {
             etcd_ws_keepalive_interval: 30, // seconds
             // osd
             etcd_report_interval: 5, // seconds
+            etcd_stats_interval: 30, // seconds
             run_primary: true,
             osd_network: null, // "192.168.7.0/24" or an array of masks
             bind_address: "0.0.0.0",
@@ -396,12 +397,13 @@ class Mon
         this.etcd_prefix = this.etcd_prefix.replace(/\/\/+/g, '/').replace(/^\/?(.*[^\/])\/?$/, '/$1');
         this.etcd_start_timeout = (config.etcd_start_timeout || 5) * 1000;
         this.state = JSON.parse(JSON.stringify(this.constructor.etcd_tree));
+        this.prev_stats = { osd_stats: {}, osd_diff: {} };
         this.signals_set = false;
-        this.stat_time = Date.now();
         this.ws = null;
         this.ws_alive = false;
         this.ws_keepalive_timer = null;
         this.on_stop_cb = () => this.on_stop(0).catch(console.error);
+        this.recheck_pgs_active = false;
     }
 
     parse_etcd_addresses(addrs)
@@ -551,9 +553,9 @@ class Mon
             const cur_addr = this.pick_next_etcd();
             const base = 'ws'+cur_addr.substr(4);
             let now = Date.now();
-            if (tried[base] && now-tried[base] < timeout)
+            if (tried[base] && now-tried[base] < this.etcd_start_timeout)
             {
-                await new Promise(ok => setTimeout(ok, timeout-(now-tried[base])));
+                await new Promise(ok => setTimeout(ok, this.etcd_start_timeout-(now-tried[base])));
                 now = Date.now();
             }
             tried[base] = now;
@@ -691,8 +693,27 @@ class Mon
         });
     }
 
+    // Schedule save_last_clean() to to run after a small timeout (1s) (to not spam etcd)
+    schedule_save_last_clean()
+    {
+        if (!this.save_last_clean_timer)
+        {
+            this.save_last_clean_timer = setTimeout(() =>
+            {
+                this.save_last_clean_timer = null;
+                this.save_last_clean().catch(this.die);
+            }, this.config.mon_change_timeout || 1000);
+        }
+    }
+
     async save_last_clean()
     {
+        if (this.save_last_clean_running)
+        {
+            this.schedule_save_last_clean();
+            return;
+        }
+        this.save_last_clean_running = true;
         // last_clean_pgs is used to avoid extra data move when observing a series of changes in the cluster
         const new_clean_pgs = { items: {} };
     next_pool:
@@ -729,6 +750,7 @@ class Mon
                 value: b64(JSON.stringify(this.state.history.last_clean_pgs))
             } } ],
         }, this.etcd_start_timeout, 0);
+        this.save_last_clean_running = false;
     }
 
     get_mon_state()
@@ -1202,6 +1224,12 @@ class Mon
 
     async recheck_pgs()
     {
+        if (this.recheck_pgs_active)
+        {
+            this.schedule_recheck();
+            return;
+        }
+        this.recheck_pgs_active = true;
         // Take configuration and state, check it against the stored configuration hash
         // Recalculate PGs and save them to etcd if the configuration is changed
         // FIXME: Do not change anything if the distribution is good and random enough and no PGs are degraded
@@ -1223,6 +1251,7 @@ class Mon
                     // Pool deleted. Delete all PGs, but first stop them.
                     if (!await this.stop_all_pgs(pool_id))
                     {
+                        this.recheck_pgs_active = false;
                         this.schedule_recheck();
                         return;
                     }
@@ -1291,9 +1320,16 @@ class Mon
                         // PG count changed. Need to bring all PGs down.
                         if (!await this.stop_all_pgs(pool_id))
                         {
+                            this.recheck_pgs_active = false;
                             this.schedule_recheck();
                             return;
                         }
+                    }
+                    if (prev_pgs.length != pool_cfg.pg_count)
+                    {
+                        // Scale PG count
+                        // Do it even if old_pg_count is already equal to pool_cfg.pg_count,
+                        // because last_clean_pgs may still contain the old number of PGs
                         const new_pg_history = [];
                         PGUtil.scale_pg_count(prev_pgs, real_prev_pgs, pg_history, new_pg_history, pool_cfg.pg_count);
                         pg_history = new_pg_history;
@@ -1395,6 +1431,7 @@ class Mon
                 await this.save_pg_config(new_config_pgs);
             }
         }
+        this.recheck_pgs_active = false;
     }
 
     async save_pg_config(new_config_pgs, etcd_request = { compare: [], success: [] })
@@ -1444,7 +1481,6 @@ class Mon
     }
 
     // Schedule a recheck to run after a small timeout (1s)
-    // If already scheduled, cancel previous timer and schedule it again
     // This is required for multiple change events to trigger at most 1 recheck in 1s
     schedule_recheck()
     {
@@ -1458,15 +1494,15 @@ class Mon
         }
     }
 
-    derive_osd_stats(st, prev)
+    derive_osd_stats(st, prev, prev_diff)
     {
         const zero_stats = { op: { bps: 0n, iops: 0n, lat: 0n }, subop: { iops: 0n, lat: 0n }, recovery: { bps: 0n, iops: 0n } };
-        const diff = { op_stats: {}, subop_stats: {}, recovery_stats: {} };
-        if (!st || !st.time || prev && (prev.time || this.stat_time/1000) >= st.time)
+        const diff = { op_stats: {}, subop_stats: {}, recovery_stats: {}, inode_stats: {} };
+        if (!st || !st.time || !prev || !prev.time || prev.time >= st.time)
         {
-            return diff;
+            return prev_diff || diff;
         }
-        const timediff = BigInt(st.time*1000 - (prev && prev.time*1000 || this.stat_time));
+        const timediff = BigInt(st.time*1000 - prev.time*1000);
         for (const op in st.op_stats||{})
         {
             const pr = prev && prev.op_stats && prev.op_stats[op];
@@ -1498,25 +1534,47 @@ class Mon
             if (n > 0)
                 diff.recovery_stats[op] = { ...c, bps: b*1000n/timediff, iops: n*1000n/timediff };
         }
+        for (const pool_id in st.inode_stats||{})
+        {
+            const pool_diff = diff.inode_stats[pool_id] = {};
+            for (const inode_num in st.inode_stats[pool_id])
+            {
+                const inode_diff = diff.inode_stats[pool_id][inode_num] = {};
+                for (const op of [ 'read', 'write', 'delete' ])
+                {
+                    const c = st.inode_stats[pool_id][inode_num][op];
+                    const pr = prev && prev.inode_stats && prev.inode_stats[pool_id] &&
+                        prev.inode_stats[pool_id][inode_num] && prev.inode_stats[pool_id][inode_num][op];
+                    const n = BigInt(c.count||0) - BigInt(pr && pr.count||0);
+                    inode_diff[op] = {
+                        bps: (BigInt(c.bytes||0) - BigInt(pr && pr.bytes||0))*1000n/timediff,
+                        iops: n*1000n/timediff,
+                        lat: (BigInt(c.usec||0) - BigInt(pr && pr.usec||0))/(n || 1n),
+                    };
+                }
+            }
+        }
         return diff;
     }
 
-    sum_op_stats(timestamp, prev_stats)
+    sum_op_stats()
     {
-        const sum_diff = { op_stats: {}, subop_stats: {}, recovery_stats: {} };
-        if (!prev_stats || prev_stats.timestamp >= timestamp)
+        for (const osd in this.state.osd.stats)
         {
-            return sum_diff;
+            const cur = { ...this.state.osd.stats[osd], inode_stats: this.state.osd.inodestats[osd]||{} };
+            this.prev_stats.osd_diff[osd] = this.derive_osd_stats(
+                cur, this.prev_stats.osd_stats[osd], this.prev_stats.osd_diff[osd]
+            );
+            this.prev_stats.osd_stats[osd] = cur;
         }
-        const tm = BigInt(timestamp - (prev_stats.timestamp || 0));
+        const sum_diff = { op_stats: {}, subop_stats: {}, recovery_stats: {} };
         // Sum derived values instead of deriving summed
         for (const osd in this.state.osd.stats)
         {
-            const derived = this.derive_osd_stats(this.state.osd.stats[osd],
-                this.prev_stats && this.prev_stats.osd_stats && this.prev_stats.osd_stats[osd]);
-            for (const type in derived)
+            const derived = this.prev_stats.osd_diff[osd];
+            for (const type in sum_diff)
             {
-                for (const op in derived[type])
+                for (const op in derived[type]||{})
                 {
                     for (const k in derived[type][op])
                     {
@@ -1573,14 +1631,14 @@ class Mon
         return { object_counts, object_bytes };
     }
 
-    sum_inode_stats(prev_stats, timestamp, prev_timestamp)
+    sum_inode_stats()
     {
         const inode_stats = {};
         const inode_stub = () => ({
             raw_used: 0n,
-            read: { count: 0n, usec: 0n, bytes: 0n },
-            write: { count: 0n, usec: 0n, bytes: 0n },
-            delete: { count: 0n, usec: 0n, bytes: 0n },
+            read: { count: 0n, usec: 0n, bytes: 0n, bps: 0n, iops: 0n, lat: 0n },
+            write: { count: 0n, usec: 0n, bytes: 0n, bps: 0n, iops: 0n, lat: 0n },
+            delete: { count: 0n, usec: 0n, bytes: 0n, bps: 0n, iops: 0n, lat: 0n },
         });
         const seen_pools = {};
         for (const pool_id in this.state.config.pools)
@@ -1632,11 +1690,25 @@ class Mon
                 }
             }
         }
-        if (prev_stats && prev_timestamp >= timestamp)
+        for (const osd in this.prev_stats.osd_diff)
         {
-            prev_stats = null;
+            for (const pool_id in this.prev_stats.osd_diff[osd].inode_stats)
+            {
+                for (const inode_num in this.prev_stats.osd_diff[osd].inode_stats[pool_id])
+                {
+                    inode_stats[pool_id][inode_num] = inode_stats[pool_id][inode_num] || inode_stub();
+                    for (const op of [ 'read', 'write', 'delete' ])
+                    {
+                        const op_diff = this.prev_stats.osd_diff[osd].inode_stats[pool_id][inode_num][op] || {};
+                        const op_st = inode_stats[pool_id][inode_num][op];
+                        op_st.bps += op_diff.bps;
+                        op_st.iops += op_diff.iops;
+                        op_st.lat += op_diff.lat;
+                        op_st.n_osd = (op_st.n_osd || 0) + 1;
+                    }
+                }
+            }
         }
-        const tm = prev_stats ? BigInt(timestamp - prev_timestamp) : 0;
         for (const pool_id in inode_stats)
         {
             for (const inode_num in inode_stats[pool_id])
@@ -1645,11 +1717,12 @@ class Mon
                 for (const op of [ 'read', 'write', 'delete' ])
                 {
                     const op_st = inode_stats[pool_id][inode_num][op];
-                    const prev_st = prev_stats && prev_stats[pool_id] && prev_stats[pool_id][inode_num] && prev_stats[pool_id][inode_num][op];
-                    op_st.bps = prev_st ? (op_st.bytes - prev_st.bytes) * 1000n / tm : 0;
-                    op_st.iops = prev_st ? (op_st.count - prev_st.count) * 1000n / tm : 0;
-                    op_st.lat = prev_st ? (op_st.usec - prev_st.usec) / ((op_st.count - prev_st.count) || 1n) : 0;
-                    if (op_st.bps > 0 || op_st.iops > 0 || op_st.lat > 0)
+                    if (op_st.n_osd)
+                    {
+                        op_st.lat /= BigInt(op_st.n_osd);
+                        delete op_st.n_osd;
+                    }
+                    if (op_st.bps > 0 || op_st.iops > 0)
                         nonzero = true;
                 }
                 if (!nonzero && (!this.state.config.inode[pool_id] || !this.state.config.inode[pool_id][inode_num]))
@@ -1682,15 +1755,9 @@ class Mon
     async update_total_stats()
     {
         const txn = [];
-        const timestamp = Date.now();
         const { object_counts, object_bytes } = this.sum_object_counts();
-        let stats = this.sum_op_stats(timestamp, this.prev_stats);
-        let { inode_stats, seen_pools } = this.sum_inode_stats(
-            this.prev_stats ? this.prev_stats.inode_stats : null,
-            timestamp, this.prev_stats ? this.prev_stats.timestamp : null
-        );
-        this.prev_stats = { timestamp, inode_stats, osd_stats: { ...this.state.osd.stats } };
-        this.stat_time = Date.now();
+        let stats = this.sum_op_stats();
+        let { inode_stats, seen_pools } = this.sum_inode_stats();
         stats.object_counts = object_counts;
         stats.object_bytes = object_bytes;
         stats = this.serialize_bigints(stats);

mon/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vitastor-mon",
-  "version": "1.0.0",
+  "version": "1.2.0",
   "description": "Vitastor SDS monitor service",
   "main": "mon-main.js",
   "scripts": {

patches/cinder-vitastor.py

@@ -50,7 +50,7 @@ from cinder.volume import configuration
 from cinder.volume import driver
 from cinder.volume import volume_utils
 
-VERSION = '1.0.0'
+VERSION = '1.2.0'
 
 LOG = logging.getLogger(__name__)
 

rpm/build-tarball.sh

@@ -24,4 +24,4 @@ rm fio
 mv fio-copy fio
 FIO=`rpm -qi fio | perl -e 'while(<>) { /^Epoch[\s:]+(\S+)/ && print "$1:"; /^Version[\s:]+(\S+)/ && print $1; /^Release[\s:]+(\S+)/ && print "-$1"; }'`
 perl -i -pe 's/(Requires:\s*fio)([^\n]+)?/$1 = '$FIO'/' $VITASTOR/rpm/vitastor-el$EL.spec
-tar --transform 's#^#vitastor-1.0.0/#' --exclude 'rpm/*.rpm' -czf $VITASTOR/../vitastor-1.0.0$(rpm --eval '%dist').tar.gz *
+tar --transform 's#^#vitastor-1.2.0/#' --exclude 'rpm/*.rpm' -czf $VITASTOR/../vitastor-1.2.0$(rpm --eval '%dist').tar.gz *

rpm/vitastor-el7.Dockerfile

@@ -35,7 +35,7 @@ ADD . /root/vitastor
 RUN set -e; \
     cd /root/vitastor/rpm; \
     sh build-tarball.sh; \
-    cp /root/vitastor-1.0.0.el7.tar.gz ~/rpmbuild/SOURCES; \
+    cp /root/vitastor-1.2.0.el7.tar.gz ~/rpmbuild/SOURCES; \
     cp vitastor-el7.spec ~/rpmbuild/SPECS/vitastor.spec; \
     cd ~/rpmbuild/SPECS/; \
     rpmbuild -ba vitastor.spec; \

rpm/vitastor-el7.spec

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        1.0.0
+Version:        1.2.0
 Release:        1%{?dist}
 Summary:        Vitastor, a fast software-defined clustered block storage
 
 License:        Vitastor Network Public License 1.1
 URL:            https://vitastor.io/
-Source0:        vitastor-1.0.0.el7.tar.gz
+Source0:        vitastor-1.2.0.el7.tar.gz
 
 BuildRequires:  liburing-devel >= 0.6
 BuildRequires:  gperftools-devel

rpm/vitastor-el8.Dockerfile

@@ -35,7 +35,7 @@ ADD . /root/vitastor
 RUN set -e; \
     cd /root/vitastor/rpm; \
     sh build-tarball.sh; \
-    cp /root/vitastor-1.0.0.el8.tar.gz ~/rpmbuild/SOURCES; \
+    cp /root/vitastor-1.2.0.el8.tar.gz ~/rpmbuild/SOURCES; \
     cp vitastor-el8.spec ~/rpmbuild/SPECS/vitastor.spec; \
     cd ~/rpmbuild/SPECS/; \
     rpmbuild -ba vitastor.spec; \

rpm/vitastor-el8.spec

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        1.0.0
+Version:        1.2.0
 Release:        1%{?dist}
 Summary:        Vitastor, a fast software-defined clustered block storage
 
 License:        Vitastor Network Public License 1.1
 URL:            https://vitastor.io/
-Source0:        vitastor-1.0.0.el8.tar.gz
+Source0:        vitastor-1.2.0.el8.tar.gz
 
 BuildRequires:  liburing-devel >= 0.6
 BuildRequires:  gperftools-devel

rpm/vitastor-el9.Dockerfile

@@ -18,7 +18,7 @@ ADD . /root/vitastor
 RUN set -e; \
     cd /root/vitastor/rpm; \
     sh build-tarball.sh; \
-    cp /root/vitastor-1.0.0.el9.tar.gz ~/rpmbuild/SOURCES; \
+    cp /root/vitastor-1.2.0.el9.tar.gz ~/rpmbuild/SOURCES; \
     cp vitastor-el9.spec ~/rpmbuild/SPECS/vitastor.spec; \
     cd ~/rpmbuild/SPECS/; \
     rpmbuild -ba vitastor.spec; \

rpm/vitastor-el9.spec

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        1.0.0
+Version:        1.2.0
 Release:        1%{?dist}
 Summary:        Vitastor, a fast software-defined clustered block storage
 
 License:        Vitastor Network Public License 1.1
 URL:            https://vitastor.io/
-Source0:        vitastor-1.0.0.el9.tar.gz
+Source0:        vitastor-1.2.0.el9.tar.gz
 
 BuildRequires:  liburing-devel >= 0.6
 BuildRequires:  gperftools-devel

src/CMakeLists.txt

@@ -16,10 +16,11 @@ if("${CMAKE_INSTALL_PREFIX}" MATCHES "^/usr/local/?$")
 	set(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_LIBDIR}")
 endif()
 
-add_definitions(-DVERSION="1.0.0")
-add_definitions(-Wall -Wno-sign-compare -Wno-comment -Wno-parentheses -Wno-pointer-arith -fdiagnostics-color=always -I ${CMAKE_SOURCE_DIR}/src)
+add_definitions(-DVERSION="1.2.0")
+add_definitions(-Wall -Wno-sign-compare -Wno-comment -Wno-parentheses -Wno-pointer-arith -fdiagnostics-color=always -fno-omit-frame-pointer -I ${CMAKE_SOURCE_DIR}/src)
+add_link_options(-fno-omit-frame-pointer)
 if (${WITH_ASAN})
-	add_definitions(-fsanitize=address -fno-omit-frame-pointer)
+	add_definitions(-fsanitize=address)
 	add_link_options(-fsanitize=address -fno-omit-frame-pointer)
 endif (${WITH_ASAN})
 
@@ -180,6 +181,25 @@ target_link_libraries(vitastor-nbd
 	vitastor_client
 )
 
+# vitastor-kv
+add_executable(vitastor-kv
+	kv_cli.cpp
+	kv_db.cpp
+	kv_db.h
+)
+target_link_libraries(vitastor-kv
+	vitastor_client
+)
+
+add_executable(vitastor-kv-stress
+	kv_stress.cpp
+	kv_db.cpp
+	kv_db.h
+)
+target_link_libraries(vitastor-kv-stress
+	vitastor_client
+)
+
 # vitastor-nfs
 add_executable(vitastor-nfs
 	nfs_proxy.cpp

src/blockstore.cpp

@@ -82,8 +82,3 @@ uint32_t blockstore_t::get_bitmap_granularity()
 {
     return impl->get_bitmap_granularity();
 }
-
-bool blockstore_t::wants_fsync()
-{
-    return impl->wants_fsync();
-}

src/blockstore.h

@@ -226,7 +226,4 @@ public:
     uint64_t get_journal_size();
 
     uint32_t get_bitmap_granularity();
-
-    // Returns true if writing can stall due to a lack of fsync
-    bool wants_fsync();
 };

src/blockstore_flush.cpp

@@ -1372,7 +1372,8 @@ bool journal_flusher_co::trim_journal(int wait_base)
                     ? (uint32_t)JE_START_V1_SIZE : (uint32_t)JE_START_V2_SIZE),
                 .reserved = 0,
                 .journal_start = new_trim_pos,
-                .version = JOURNAL_VERSION_V2,
+                .version = (uint64_t)(!bs->dsk.data_csum_type && ((journal_entry_start*)flusher->journal_superblock)->version == JOURNAL_VERSION_V1
+                    ? JOURNAL_VERSION_V1 : JOURNAL_VERSION_V2),
                 .data_csum_type = bs->dsk.data_csum_type,
                 .csum_block_size = bs->dsk.csum_block_size,
             };

src/blockstore_impl.cpp

@@ -167,7 +167,7 @@ void blockstore_impl_t::loop()
                 // wait for all big writes to complete, submit data device fsync
                 // wait for the data device fsync to complete, then submit journal writes for big writes
                 // then submit an fsync operation
-                if (0 && has_writes)
+                if (has_writes)
                 {
                     // Can't submit SYNC before previous writes
                     continue;
@@ -734,15 +734,3 @@ void blockstore_impl_t::disk_error_abort(const char *op, int retval, int expecte
     fprintf(stderr, "Disk %s failed: result is %d, expected %d. Can't continue, sorry :-(\n", op, retval, expected);
     exit(1);
 }
-
-bool blockstore_impl_t::wants_fsync()
-{
-    if (!unstable_writes.size())
-    {
-        return false;
-    }
-    uint64_t journal_free_space = journal.next_free < journal.used_start
-        ? (journal.used_start - journal.next_free)
-        : (journal.len - journal.next_free + journal.used_start - journal.block_size);
-    return journal_fsync_feedback_limit > 0 && journal.len-journal_free_space >= journal_fsync_feedback_limit;
-}

src/blockstore_impl.h

@@ -264,8 +264,6 @@ class blockstore_impl_t
     int throttle_threshold_us = 50;
     // Maximum writes between automatically added fsync operations
     uint64_t autosync_writes = 128;
-    // Maximum free space in the journal in bytes to start sending fsync feedback to primary OSDs
-    uint64_t journal_fsync_feedback_limit = 0;
     /******* END OF OPTIONS *******/
 
     struct ring_consumer_t ring_consumer;
@@ -276,7 +274,7 @@ class blockstore_impl_t
     blockstore_dirty_db_t dirty_db;
     std::vector<blockstore_op_t*> submit_queue;
     std::vector<obj_ver_id> unsynced_big_writes, unsynced_small_writes;
-    int unsynced_big_write_count = 0;
+    int unsynced_big_write_count = 0, unstable_unsynced = 0;
     int unsynced_queued_ops = 0;
     allocator *data_alloc = NULL;
     uint8_t *zero_object;
@@ -435,6 +433,4 @@ public:
     inline uint64_t get_free_block_count() { return data_alloc->get_free_count(); }
     inline uint32_t get_bitmap_granularity() { return dsk.disk_alignment; }
     inline uint64_t get_journal_size() { return dsk.journal_len; }
-
-    bool wants_fsync();
 };

src/blockstore_init.cpp

@@ -553,7 +553,7 @@ resume_1:
         }
         if (je_start->size == JE_START_V0_SIZE ||
             (je_start->version != JOURNAL_VERSION_V1 || je_start->size != JE_START_V1_SIZE) &&
-            (je_start->version != JOURNAL_VERSION_V2 || je_start->size != JE_START_V2_SIZE))
+            (je_start->version != JOURNAL_VERSION_V2 || je_start->size != JE_START_V2_SIZE && je_start->size != JE_START_V1_SIZE))
         {
             fprintf(
                 stderr, "The code only supports journal versions 2 and 1, but it is %lu on disk."
@@ -562,7 +562,8 @@ resume_1:
             );
             exit(1);
         }
-        if (je_start->version == JOURNAL_VERSION_V1)
+        if (je_start->version == JOURNAL_VERSION_V1 ||
+            je_start->version == JOURNAL_VERSION_V2 && je_start->size == JE_START_V1_SIZE)
         {
             je_start->data_csum_type = 0;
             je_start->csum_block_size = 0;

src/blockstore_journal.cpp

@@ -145,6 +145,7 @@ journal_entry* prefill_single_journal_entry(journal_t & journal, uint16_t type,
         journal.sector_info[journal.cur_sector].offset = journal.next_free;
         journal.in_sector_pos = 0;
         journal.next_free = (journal.next_free+journal.block_size) < journal.len ? journal.next_free + journal.block_size : journal.block_size;
+        assert(journal.next_free != journal.used_start);
         memset(journal.inmemory
             ? (uint8_t*)journal.buffer + journal.sector_info[journal.cur_sector].offset
             : (uint8_t*)journal.sector_buf + journal.block_size*journal.cur_sector, 0, journal.block_size);

src/blockstore_journal.h

@@ -13,12 +13,6 @@
 #define JOURNAL_BUFFER_SIZE 4*1024*1024
 #define JOURNAL_ENTRY_HEADER_SIZE 16
 
-// We reserve some extra space for future stabilize requests during writes
-// FIXME: This value should be dynamic i.e. Blockstore ideally shouldn't allow
-// writing more than can be stabilized afterwards
-#define JOURNAL_STABILIZE_RESERVATION 65536
-#define JOURNAL_INSTANT_RESERVATION 131072
-
 // Journal entries
 // Journal entries are linked to each other by their crc32 value
 // The journal is almost a blockchain, because object versions constantly increase

src/blockstore_open.cpp

@@ -4,25 +4,6 @@
 #include <sys/file.h>
 #include "blockstore_impl.h"
 
-static uint64_t parse_fsync_feedback(blockstore_config_t & config, uint64_t journal_len)
-{
-    uint64_t journal_fsync_feedback_limit = 0;
-    if (config.find("journal_min_free_bytes") == config.end() &&
-        config.find("journal_min_free_percent") == config.end())
-    {
-        journal_fsync_feedback_limit = 90 * journal_len / 100;
-    }
-    else
-    {
-        journal_fsync_feedback_limit = strtoull(config["journal_min_free_bytes"].c_str(), NULL, 10);
-        if (!journal_fsync_feedback_limit)
-        {
-            journal_fsync_feedback_limit = strtoull(config["journal_min_free_percent"].c_str(), NULL, 10) * journal_len / 100;
-        }
-    }
-    return journal_fsync_feedback_limit;
-}
-
 void blockstore_impl_t::parse_config(blockstore_config_t & config, bool init)
 {
     // Online-configurable options:
@@ -72,8 +53,6 @@ void blockstore_impl_t::parse_config(blockstore_config_t & config, bool init)
     }
     if (!init)
     {
-        // has to be parsed after dsk.parse_config(), thus repeated here for online update
-        journal_fsync_feedback_limit = parse_fsync_feedback(config, journal.len);
         return;
     }
     // Offline-configurable options:
@@ -117,7 +96,6 @@ void blockstore_impl_t::parse_config(blockstore_config_t & config, bool init)
         config["journal_no_same_sector_overwrites"] == "1" || config["journal_no_same_sector_overwrites"] == "yes";
     journal.inmemory = config["inmemory_journal"] != "false" && config["inmemory_journal"] != "0" &&
         config["inmemory_journal"] != "no";
-    journal_fsync_feedback_limit = parse_fsync_feedback(config, journal.len);
     // Validate
     if (journal.sector_count < 2)
     {

src/blockstore_sync.cpp

@@ -16,6 +16,7 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op)
 {
     if (immediate_commit == IMMEDIATE_ALL)
     {
+        // We can return immediately because sync is only dequeued after all previous writes
         op->retval = 0;
         FINISH_OP(op);
         return 2;
@@ -85,14 +86,15 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op)
                 auto & dirty_entry = dirty_db.at(sbw);
                 uint64_t dyn_size = dsk.dirty_dyn_size(dirty_entry.offset, dirty_entry.len);
                 if (!space_check.check_available(op, 1, sizeof(journal_entry_big_write) + dyn_size,
-                    left == 0 ? JOURNAL_STABILIZE_RESERVATION : 0))
+                    (unstable_writes.size()+unstable_unsynced)*journal.block_size))
                 {
                     return 0;
                 }
             }
         }
         else if (!space_check.check_available(op, PRIV(op)->sync_big_writes.size(),
-            sizeof(journal_entry_big_write) + dsk.clean_entry_bitmap_size, JOURNAL_STABILIZE_RESERVATION))
+            sizeof(journal_entry_big_write) + dsk.clean_entry_bitmap_size,
+            (unstable_writes.size()+unstable_unsynced)*journal.block_size))
         {
             return 0;
         }
@@ -183,6 +185,11 @@ void blockstore_impl_t::ack_sync(blockstore_op_t *op)
         {
             mark_stable(dirty_it->first);
         }
+        else
+        {
+            unstable_unsynced--;
+            assert(unstable_unsynced >= 0);
+        }
         dirty_it++;
         while (dirty_it != dirty_db.end() && dirty_it->first.oid == it->oid)
         {
@@ -213,6 +220,11 @@ void blockstore_impl_t::ack_sync(blockstore_op_t *op)
             {
                 mark_stable(*it);
             }
+            else
+            {
+                unstable_unsynced--;
+                assert(unstable_unsynced >= 0);
+            }
         }
     }
     op->retval = 0;

src/blockstore_write.cpp

@@ -21,7 +21,7 @@ bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
         dyn = calloc_or_die(1, dyn_size+sizeof(int));
         *((int*)dyn) = 1;
     }
-    uint8_t *dyn_ptr = (uint8_t*)(alloc_dyn_data ? dyn+sizeof(int) : &dyn);
+    uint8_t *dyn_ptr = (alloc_dyn_data ? (uint8_t*)dyn+sizeof(int) : (uint8_t*)&dyn);
     uint64_t version = 1;
     if (dirty_db.size() > 0)
     {
@@ -289,13 +289,18 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
         printf("Restoring %lx:%lx version: v%lu -> v%lu\n", op->oid.inode, op->oid.stripe, op->version, PRIV(op)->real_version);
 #endif
         auto prev_it = dirty_it;
-        prev_it--;
-        if (prev_it->first.oid == op->oid && prev_it->first.version >= PRIV(op)->real_version)
+        if (prev_it != dirty_db.begin())
         {
-            // Original version is still invalid
-            // All subsequent writes to the same object must be canceled too
-            cancel_all_writes(op, dirty_it, -EEXIST);
-            return 2;
+            prev_it--;
+            if (prev_it->first.oid == op->oid && prev_it->first.version >= PRIV(op)->real_version)
+            {
+                // Original version is still invalid
+                // All subsequent writes to the same object must be canceled too
+                printf("Tried to write %lx:%lx v%lu after delete (old version v%lu), but already have v%lu\n",
+                    op->oid.inode, op->oid.stripe, PRIV(op)->real_version, op->version, prev_it->first.version);
+                cancel_all_writes(op, dirty_it, -EEXIST);
+                return 2;
+            }
         }
         op->version = PRIV(op)->real_version;
         PRIV(op)->real_version = 0;
@@ -315,7 +320,7 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
         blockstore_journal_check_t space_check(this);
         if (!space_check.check_available(op, unsynced_big_write_count + 1,
             sizeof(journal_entry_big_write) + dsk.clean_dyn_size,
-            (dirty_it->second.state & BS_ST_INSTANT) ? JOURNAL_INSTANT_RESERVATION : JOURNAL_STABILIZE_RESERVATION))
+            (unstable_writes.size()+unstable_unsynced)*journal.block_size))
         {
             return 0;
         }
@@ -381,6 +386,10 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
             sqe, dsk.data_fd, PRIV(op)->iov_zerofill, vcnt, dsk.data_offset + (loc << dsk.block_order) + op->offset - stripe_offset
         );
         PRIV(op)->pending_ops = 1;
+        if (immediate_commit != IMMEDIATE_ALL && !(dirty_it->second.state & BS_ST_INSTANT))
+        {
+            unstable_unsynced++;
+        }
         if (immediate_commit != IMMEDIATE_ALL)
         {
             // Increase the counter, but don't save into unsynced_writes yet (can't sync until the write is finished)
@@ -403,7 +412,7 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
                 sizeof(journal_entry_big_write) + dsk.clean_dyn_size, 0)
             || !space_check.check_available(op, 1,
                 sizeof(journal_entry_small_write) + dyn_size,
-                op->len + ((dirty_it->second.state & BS_ST_INSTANT) ? JOURNAL_INSTANT_RESERVATION : JOURNAL_STABILIZE_RESERVATION)))
+                (unstable_writes.size()+unstable_unsynced)*journal.block_size))
         {
             return 0;
         }
@@ -494,6 +503,11 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
         if (journal.next_free >= journal.len)
         {
             journal.next_free = dsk.journal_block_size;
+            assert(journal.next_free != journal.used_start);
+        }
+        if (immediate_commit == IMMEDIATE_NONE && !(dirty_it->second.state & BS_ST_INSTANT))
+        {
+            unstable_unsynced++;
         }
         if (!PRIV(op)->pending_ops)
         {
@@ -533,7 +547,7 @@ resume_2:
         uint64_t dyn_size = dsk.dirty_dyn_size(op->offset, op->len);
         blockstore_journal_check_t space_check(this);
         if (!space_check.check_available(op, 1, sizeof(journal_entry_big_write) + dyn_size,
-            ((dirty_it->second.state & BS_ST_INSTANT) ? JOURNAL_INSTANT_RESERVATION : JOURNAL_STABILIZE_RESERVATION)))
+            (unstable_writes.size()+unstable_unsynced)*journal.block_size))
         {
             return 0;
         }
@@ -577,14 +591,20 @@ resume_4:
 #endif
         bool is_big = (dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_BIG_WRITE;
         bool imm = is_big ? (immediate_commit == IMMEDIATE_ALL) : (immediate_commit != IMMEDIATE_NONE);
+        bool is_instant = ((dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_DELETE || (dirty_it->second.state & BS_ST_INSTANT));
         if (imm)
         {
             auto & unstab = unstable_writes[op->oid];
             unstab = unstab < op->version ? op->version : unstab;
         }
+        else if (!is_instant)
+        {
+            unstable_unsynced--;
+            assert(unstable_unsynced >= 0);
+        }
         dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK)
             | (imm ? BS_ST_SYNCED : BS_ST_WRITTEN);
-        if (imm && ((dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_DELETE || (dirty_it->second.state & BS_ST_INSTANT)))
+        if (imm && is_instant)
         {
             // Deletions and 'instant' operations are treated as immediately stable
             mark_stable(dirty_it->first);
@@ -730,7 +750,7 @@ int blockstore_impl_t::dequeue_del(blockstore_op_t *op)
     });
     assert(dirty_it != dirty_db.end());
     blockstore_journal_check_t space_check(this);
-    if (!space_check.check_available(op, 1, sizeof(journal_entry_del), JOURNAL_INSTANT_RESERVATION))
+    if (!space_check.check_available(op, 1, sizeof(journal_entry_del), (unstable_writes.size()+unstable_unsynced)*journal.block_size))
     {
         return 0;
     }

src/cli.cpp

@@ -17,7 +17,7 @@
 static const char *exe_name = NULL;
 
 static const char* help_text =
-    "Vitastor command-line tool\n"
+    "Vitastor command-line tool " VERSION "\n"
     "(c) Vitaliy Filippov, 2019+ (VNPL-1.1)\n"
     "\n"
     "COMMANDS:\n"
@@ -331,7 +331,7 @@ static int run(cli_tool_t *p, json11::Json::object cfg)
     {
         // Create client
         json11::Json cfg_j = cfg;
-        p->ringloop = new ring_loop_t(512);
+        p->ringloop = new ring_loop_t(RINGLOOP_DEFAULT_SIZE);
         p->epmgr = new epoll_manager_t(p->ringloop);
         p->cli = new cluster_client_t(p->ringloop, p->epmgr->tfd, cfg_j);
         // Smaller timeout by default for more interactiveness

src/cli_df.cpp

@@ -109,7 +109,7 @@ resume_1:
             }
             for (auto pg_per_pair: pg_per_osd)
             {
-                uint64_t pg_free = osd_free[pg_per_pair.first] * pool_cfg.pg_count / pg_per_pair.second;
+                uint64_t pg_free = osd_free[pg_per_pair.first] * pool_cfg.real_pg_count / pg_per_pair.second;
                 if (pool_avail > pg_free)
                 {
                     pool_avail = pg_free;
@@ -124,8 +124,10 @@ resume_1:
                 pool_avail *= (pool_cfg.pg_size - pool_cfg.parity_chunks);
             }
             pool_stats[pool_cfg.id] = json11::Json::object {
+                { "id", (uint64_t)pool_cfg.id },
                 { "name", pool_cfg.name },
                 { "pg_count", pool_cfg.pg_count },
+                { "real_pg_count", pool_cfg.real_pg_count },
                 { "scheme", pool_cfg.scheme == POOL_SCHEME_REPLICATED ? "replicated" : "ec" },
                 { "scheme_name", pool_cfg.scheme == POOL_SCHEME_REPLICATED
                     ? std::to_string(pool_cfg.pg_size)+"/"+std::to_string(pool_cfg.pg_minsize)
@@ -176,7 +178,7 @@ resume_1:
             { "title", "SCHEME" },
         });
         cols.push_back(json11::Json::object{
-            { "key", "pg_count" },
+            { "key", "pg_count_fmt" },
             { "title", "PGS" },
         });
         cols.push_back(json11::Json::object{
@@ -205,6 +207,9 @@ resume_1:
             double raw_to = kv.second["raw_to_usable"].number_value();
             if (raw_to < 0.000001 && raw_to > -0.000001)
                 raw_to = 1;
+            kv.second["pg_count_fmt"] = kv.second["real_pg_count"] == kv.second["pg_count"]
+                ? kv.second["real_pg_count"].as_string()
+                : kv.second["real_pg_count"].as_string()+"->"+kv.second["pg_count"].as_string();
             kv.second["total_fmt"] = format_size(kv.second["total_raw"].uint64_value() / raw_to);
             kv.second["used_fmt"] = format_size(kv.second["used_raw"].uint64_value() / raw_to);
             kv.second["max_avail_fmt"] = format_size(kv.second["max_available"].uint64_value());

src/cli_ls.cpp

@@ -174,7 +174,7 @@ resume_1:
                     { "size", 0 },
                     { "readonly", false },
                     { "pool_id", (uint64_t)INODE_POOL(inode_num) },
-                    { "pool_name", pool_it == parent->cli->st_cli.pool_config.end()
+                    { "pool_name", pool_it != parent->cli->st_cli.pool_config.end()
                         ? (pool_it->second.name == "" ? "<Unnamed>" : pool_it->second.name) : "?" },
                     { "inode_num", INODE_NO_POOL(inode_num) },
                     { "inode_id", inode_num },

src/cli_status.cpp

@@ -158,12 +158,7 @@ resume_2:
         for (auto & pool_pair: parent->cli->st_cli.pool_config)
         {
             auto & pool_cfg = pool_pair.second;
-            bool active = true;
-            if (pool_cfg.pg_config.size() != pool_cfg.pg_count)
-            {
-                active = false;
-                pgs_by_state["offline"] += pool_cfg.pg_count-pool_cfg.pg_config.size();
-            }
+            bool active = pool_cfg.real_pg_count > 0;
             pool_count++;
             for (auto pg_it = pool_cfg.pg_config.begin(); pg_it != pool_cfg.pg_config.end(); pg_it++)
             {

src/cluster_client.cpp

@@ -6,7 +6,7 @@
 #include "cluster_client_impl.h"
 #include "http_client.h" // json_is_true
 
-cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json & config)
+cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json config)
 {
     wb = new writeback_cache_t();
 
@@ -64,7 +64,7 @@ cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd
 
 cluster_client_t::~cluster_client_t()
 {
-    msgr.repeer_pgs = [this](osd_num_t){};
+    msgr.repeer_pgs = [](osd_num_t){};
     if (ringloop)
     {
         ringloop->unregister_consumer(&consumer);
@@ -169,46 +169,52 @@ void cluster_client_t::calc_wait(cluster_op_t *op)
 
 void cluster_client_t::inc_wait(uint64_t opcode, uint64_t flags, cluster_op_t *next, int inc)
 {
-    if (opcode == OSD_OP_WRITE)
+    if (opcode != OSD_OP_WRITE && opcode != OSD_OP_SYNC)
     {
-        while (next)
-        {
-            auto n2 = next->next;
-            if (next->opcode == OSD_OP_SYNC && (!(flags & OP_IMMEDIATE_COMMIT) || enable_writeback) ||
-                next->opcode == OSD_OP_WRITE && (flags & OP_FLUSH_BUFFER) && !(next->flags & OP_FLUSH_BUFFER))
-            {
-                next->prev_wait += inc;
-                assert(next->prev_wait >= 0);
-                if (!next->prev_wait)
-                {
-                    if (next->opcode == OSD_OP_SYNC)
-                        continue_sync(next);
-                    else
-                        continue_rw(next);
-                }
-            }
-            next = n2;
-        }
+        return;
     }
-    else if (opcode == OSD_OP_SYNC)
+    cluster_op_t *bh_ops_local[32], **bh_ops = bh_ops_local;
+    int bh_op_count = 0, bh_op_max = 32;
+    while (next)
     {
-        while (next)
+        auto n2 = next->next;
+        if (opcode == OSD_OP_WRITE
+            ? (next->opcode == OSD_OP_SYNC && (!(flags & OP_IMMEDIATE_COMMIT) || enable_writeback) ||
+                next->opcode == OSD_OP_WRITE && (flags & OP_FLUSH_BUFFER) && !(next->flags & OP_FLUSH_BUFFER))
+            : (next->opcode == OSD_OP_SYNC || next->opcode == OSD_OP_WRITE))
         {
-            auto n2 = next->next;
-            if (next->opcode == OSD_OP_SYNC || next->opcode == OSD_OP_WRITE)
+            next->prev_wait += inc;
+            assert(next->prev_wait >= 0);
+            if (!next->prev_wait)
             {
-                next->prev_wait += inc;
-                assert(next->prev_wait >= 0);
-                if (!next->prev_wait)
+                // Kind of std::vector with local "small vector optimisation"
+                if (bh_op_count >= bh_op_max)
                 {
-                    if (next->opcode == OSD_OP_SYNC)
-                        continue_sync(next);
-                    else
-                        continue_rw(next);
+                    bh_op_max *= 2;
+                    cluster_op_t **n = (cluster_op_t**)malloc_or_die(sizeof(cluster_op_t*) * bh_op_max);
+                    memcpy(n, bh_ops, sizeof(cluster_op_t*) * bh_op_count);
+                    if (bh_ops != bh_ops_local)
+                    {
+                        free(bh_ops);
+                    }
+                    bh_ops = n;
                 }
+                bh_ops[bh_op_count++] = next;
             }
-            next = n2;
         }
+        next = n2;
+    }
+    for (int i = 0; i < bh_op_count; i++)
+    {
+        cluster_op_t *next = bh_ops[i];
+        if (next->opcode == OSD_OP_SYNC)
+            continue_sync(next);
+        else
+            continue_rw(next);
+    }
+    if (bh_ops != bh_ops_local)
+    {
+        free(bh_ops);
     }
 }
 
@@ -448,7 +454,7 @@ bool cluster_client_t::flush()
             wb->start_writebacks(this, 0);
             cluster_op_t *sync = new cluster_op_t;
             sync->opcode = OSD_OP_SYNC;
-            sync->callback = [this](cluster_op_t *sync)
+            sync->callback = [](cluster_op_t *sync)
             {
                 delete sync;
             };
@@ -459,7 +465,7 @@ bool cluster_client_t::flush()
     bool sync_done = false;
     cluster_op_t *sync = new cluster_op_t;
     sync->opcode = OSD_OP_SYNC;
-    sync->callback = [this, &sync_done](cluster_op_t *sync)
+    sync->callback = [&sync_done](cluster_op_t *sync)
     {
         delete sync;
         sync_done = true;
@@ -526,7 +532,7 @@ void cluster_client_t::execute_internal(cluster_op_t *op)
         return;
     }
     if (op->opcode == OSD_OP_WRITE && enable_writeback && !(op->flags & OP_FLUSH_BUFFER) &&
-        !op->version /* FIXME no CAS writeback */)
+        !op->version /* no CAS writeback */)
     {
         if (wb->writebacks_active >= client_max_writeback_iodepth)
         {
@@ -547,7 +553,7 @@ void cluster_client_t::execute_internal(cluster_op_t *op)
     }
     if (op->opcode == OSD_OP_WRITE && !(op->flags & OP_IMMEDIATE_COMMIT))
     {
-        if (!(op->flags & OP_FLUSH_BUFFER))
+        if (!(op->flags & OP_FLUSH_BUFFER) && !op->version /* no CAS write-repeat */)
         {
             wb->copy_write(op, CACHE_WRITTEN);
         }
@@ -1146,7 +1152,7 @@ void cluster_client_t::handle_op_part(cluster_op_part_t *part)
                 osd_op_names[part->op.req.hdr.opcode], part->osd_num, part->op.reply.hdr.retval, expected
             );
         }
-        else
+        else if (log_level > 0)
         {
             fprintf(
                 stderr, "%s operation failed on OSD %lu: retval=%ld (expected %d)\n",

src/cluster_client.h

@@ -121,7 +121,7 @@ public:
     json11::Json::object cli_config, file_config, etcd_global_config;
     json11::Json::object config;
 
-    cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json & config);
+    cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json config);
     ~cluster_client_t();
     void execute(cluster_op_t *op);
     void execute_raw(osd_num_t osd_num, osd_op_t *op);

src/cluster_client_wb.cpp

@@ -263,7 +263,7 @@ void writeback_cache_t::flush_buffers(cluster_client_t *cli, dirty_buf_it_t from
     }
     assert(calc_len == op->len);
     writebacks_active++;
-    op->callback = [this, cli, flush_id](cluster_op_t* op)
+    op->callback = [this, flush_id](cluster_op_t* op)
     {
         // Buffer flushes should be always retried, regardless of the error,
         // so they should never result in an error here
@@ -383,7 +383,7 @@ static void copy_to_op(cluster_op_t *op, uint64_t offset, uint8_t *buf, uint64_t
             auto begin = (cur_offset < offset ? offset : cur_offset);
             auto end = (cur_offset+v.iov_len > offset+len ? offset+len : cur_offset+v.iov_len);
             memcpy(
-                v.iov_base + begin - cur_offset,
+                (uint8_t*)v.iov_base + begin - cur_offset,
                 buf + (cur_offset <= offset ? 0 : cur_offset-offset),
                 end - begin
             );

src/disk_tool.cpp

@@ -5,7 +5,7 @@
 #include "str_util.h"
 
 static const char *help_text =
-    "Vitastor disk management tool\n"
+    "Vitastor disk management tool " VERSION "\n"
     "(c) Vitaliy Filippov, 2022+ (VNPL-1.1)\n"
     "\n"
     "COMMANDS:\n"
@@ -229,7 +229,7 @@ int main(int argc, char *argv[])
         {
             self.options["allow_data_loss"] = "1";
         }
-        else if (argv[i][0] == '-' && argv[i][1] == '-')
+        else if (argv[i][0] == '-' && argv[i][1] == '-' && i < argc-1)
         {
             char *key = argv[i]+2;
             self.options[key] = argv[++i];

src/disk_tool_journal.cpp

@@ -320,7 +320,7 @@ void disk_tool_t::dump_journal_entry(int num, journal_entry *je, bool json)
         if (journal_calc_data_pos != sw.data_offset)
         {
             printf(json ? ",\"bad_loc\":true,\"calc_loc\":\"0x%lx\""
-                : " (mismatched, calculated = %lu)", journal_pos);
+                : " (mismatched, calculated = %08lx)", journal_pos);
         }
         uint32_t data_csum_size = (!je_start.csum_block_size
             ? 0

src/disk_tool_resize.cpp

@@ -245,7 +245,7 @@ int disk_tool_t::resize_copy_data()
     {
         iodepth = 32;
     }
-    ringloop = new ring_loop_t(iodepth < 512 ? 512 : iodepth);
+    ringloop = new ring_loop_t(iodepth < RINGLOOP_DEFAULT_SIZE ? RINGLOOP_DEFAULT_SIZE : iodepth);
     dsk.data_fd = open(dsk.data_device.c_str(), O_DIRECT|O_RDWR);
     if (dsk.data_fd < 0)
     {

src/epoll_manager.cpp

@@ -23,19 +23,24 @@ epoll_manager_t::epoll_manager_t(ring_loop_t *ringloop)
 
     tfd = new timerfd_manager_t([this](int fd, bool wr, std::function<void(int, int)> handler) { set_fd_handler(fd, wr, handler); });
 
-    consumer.loop = [this]()
+    if (ringloop)
     {
-        if (pending)
-            handle_epoll_events();
-    };
-    ringloop->register_consumer(&consumer);
-
-    handle_epoll_events();
+        consumer.loop = [this]()
+        {
+            if (pending)
+                handle_uring_event();
+        };
+        ringloop->register_consumer(&consumer);
+        handle_uring_event();
+    }
 }
 
 epoll_manager_t::~epoll_manager_t()
 {
-    ringloop->unregister_consumer(&consumer);
+    if (ringloop)
+    {
+        ringloop->unregister_consumer(&consumer);
+    }
     if (tfd)
     {
         delete tfd;
@@ -44,6 +49,11 @@ epoll_manager_t::~epoll_manager_t()
     close(epoll_fd);
 }
 
+int epoll_manager_t::get_fd()
+{
+    return epoll_fd;
+}
+
 void epoll_manager_t::set_fd_handler(int fd, bool wr, std::function<void(int, int)> handler)
 {
     if (handler != NULL)
@@ -75,7 +85,7 @@ void epoll_manager_t::set_fd_handler(int fd, bool wr, std::function<void(int, in
     }
 }
 
-void epoll_manager_t::handle_epoll_events()
+void epoll_manager_t::handle_uring_event()
 {
     io_uring_sqe *sqe = ringloop->get_sqe();
     if (!sqe)
@@ -95,14 +105,20 @@ void epoll_manager_t::handle_epoll_events()
         {
             throw std::runtime_error(std::string("epoll failed: ") + strerror(-data->res));
         }
-        handle_epoll_events();
+        handle_uring_event();
     };
     ringloop->submit();
+    handle_events(0);
+}
+
+void epoll_manager_t::handle_events(int timeout)
+{
     int nfds;
     epoll_event events[MAX_EPOLL_EVENTS];
     do
     {
-        nfds = epoll_wait(epoll_fd, events, MAX_EPOLL_EVENTS, 0);
+        nfds = epoll_wait(epoll_fd, events, MAX_EPOLL_EVENTS, timeout);
+        timeout = 0;
         for (int i = 0; i < nfds; i++)
         {
             auto cb_it = epoll_handlers.find(events[i].data.fd);

src/epoll_manager.h

@@ -15,11 +15,14 @@ class epoll_manager_t
     ring_consumer_t consumer;
     ring_loop_t *ringloop;
     std::map<int, std::function<void(int, int)>> epoll_handlers;
+
+    void handle_uring_event();
 public:
     epoll_manager_t(ring_loop_t *ringloop);
     ~epoll_manager_t();
+    int get_fd();
     void set_fd_handler(int fd, bool wr, std::function<void(int, int)> handler);
-    void handle_epoll_events();
+    void handle_events(int timeout);
 
     timerfd_manager_t *tfd;
 };

src/fio_cluster.cpp

@@ -32,6 +32,7 @@
 struct sec_data
 {
     vitastor_c *cli = NULL;
+    bool epoll_based = false;
     void *watch = NULL;
     bool last_sync = false;
     /* The list of completed io_u structs. */
@@ -58,6 +59,7 @@ struct sec_options
     int rdma_port_num = 0;
     int rdma_gid_index = 0;
     int rdma_mtu = 0;
+    int no_io_uring = 0;
 };
 
 static struct fio_option options[] = {
@@ -193,6 +195,16 @@ static struct fio_option options[] = {
         .category = FIO_OPT_C_ENGINE,
         .group  = FIO_OPT_G_FILENAME,
     },
+    {
+        .name   = "no_io_uring",
+        .lname  = "Disable io_uring",
+        .type   = FIO_OPT_BOOL,
+        .off1   = offsetof(struct sec_options, no_io_uring),
+        .help   = "Use epoll and plain sendmsg/recvmsg instead of io_uring (slower)",
+        .def    = "0",
+        .category = FIO_OPT_C_ENGINE,
+        .group  = FIO_OPT_G_FILENAME,
+    },
     {
         .name = NULL,
     },
@@ -281,7 +293,17 @@ static int sec_setup(struct thread_data *td)
         opt_push(options, "log_level", std::to_string(o->cluster_log).c_str());
     // allow writeback caching if -direct is not set
     opt_push(options, "client_writeback_allowed", td->o.odirect ? "0" : "1");
-    bsd->cli = vitastor_c_create_uring_json((const char**)options.data(), options.size());
+    bsd->cli = o->no_io_uring ? NULL : vitastor_c_create_uring_json((const char**)options.data(), options.size());
+    bsd->epoll_based = false;
+    if (!bsd->cli)
+    {
+        if (o->no_io_uring)
+            fprintf(stderr, "vitastor: io_uring disabled - I/O will be slower\n");
+        else
+            fprintf(stderr, "vitastor: failed to create io_uring: %s - I/O will be slower\n", strerror(errno));
+        bsd->cli = vitastor_c_create_epoll_json((const char**)options.data(), options.size());
+        bsd->epoll_based = true;
+    }
     for (auto opt: options)
         free(opt);
     options.clear();
@@ -289,12 +311,24 @@ static int sec_setup(struct thread_data *td)
     {
         bsd->watch = NULL;
         vitastor_c_watch_inode(bsd->cli, o->image, watch_callback, bsd);
-        while (true)
+        if (!bsd->epoll_based)
         {
-            vitastor_c_uring_handle_events(bsd->cli);
-            if (bsd->watch)
-                break;
-            vitastor_c_uring_wait_events(bsd->cli);
+            while (true)
+            {
+                vitastor_c_uring_handle_events(bsd->cli);
+                if (bsd->watch)
+                    break;
+                vitastor_c_uring_wait_events(bsd->cli);
+            }
+        }
+        else
+        {
+            while (true)
+            {
+                if (bsd->watch)
+                    break;
+                vitastor_c_epoll_handle_events(bsd->cli, 1000);
+            }
         }
         td->files[0]->real_file_size = vitastor_c_inode_get_size(bsd->watch);
         if (!vitastor_c_inode_get_num(bsd->watch) ||
@@ -437,12 +471,24 @@ static enum fio_q_status sec_queue(struct thread_data *td, struct io_u *io)
 static int sec_getevents(struct thread_data *td, unsigned int min, unsigned int max, const struct timespec *t)
 {
     sec_data *bsd = (sec_data*)td->io_ops_data;
-    while (true)
+    if (!bsd->epoll_based)
     {
-        vitastor_c_uring_handle_events(bsd->cli);
-        if (bsd->completed.size() >= min)
-            break;
-        vitastor_c_uring_wait_events(bsd->cli);
+        while (true)
+        {
+            vitastor_c_uring_handle_events(bsd->cli);
+            if (bsd->completed.size() >= min)
+                break;
+            vitastor_c_uring_wait_events(bsd->cli);
+        }
+    }
+    else
+    {
+        while (true)
+        {
+            if (bsd->completed.size() >= min)
+                break;
+            vitastor_c_epoll_handle_events(bsd->cli, 1000);
+        }
     }
     return bsd->completed.size();
 }

src/fio_engine.cpp

@@ -130,7 +130,7 @@ static int bs_init(struct thread_data *td)
                 config[p.first] = p.second.dump();
         }
     }
-    bsd->ringloop = new ring_loop_t(512);
+    bsd->ringloop = new ring_loop_t(RINGLOOP_DEFAULT_SIZE);
     bsd->epmgr = new epoll_manager_t(bsd->ringloop);
     bsd->bs = new blockstore_t(config, bsd->ringloop, bsd->epmgr->tfd);
     while (1)

src/kv_cli.cpp

@@ -0,0 +1,401 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+//
+// Vitastor shared key/value database test CLI
+
+#define _XOPEN_SOURCE
+#include <limits.h>
+
+#include <netinet/tcp.h>
+#include <sys/epoll.h>
+#include <unistd.h>
+#include <fcntl.h>
+//#include <signal.h>
+
+#include "epoll_manager.h"
+#include "str_util.h"
+#include "kv_db.h"
+
+const char *exe_name = NULL;
+
+class kv_cli_t
+{
+public:
+    kv_dbw_t *db = NULL;
+    ring_loop_t *ringloop = NULL;
+    epoll_manager_t *epmgr = NULL;
+    cluster_client_t *cli = NULL;
+    bool interactive = false;
+    int in_progress = 0;
+    char *cur_cmd = NULL;
+    int cur_cmd_size = 0, cur_cmd_alloc = 0;
+    bool finished = false, eof = false;
+    json11::Json::object cfg;
+
+    ~kv_cli_t();
+
+    static json11::Json::object parse_args(int narg, const char *args[]);
+    void run(const json11::Json::object & cfg);
+    void read_cmd();
+    void next_cmd();
+    void handle_cmd(const std::string & cmd, std::function<void()> cb);
+};
+
+kv_cli_t::~kv_cli_t()
+{
+    if (cur_cmd)
+    {
+        free(cur_cmd);
+        cur_cmd = NULL;
+    }
+    cur_cmd_alloc = 0;
+    if (db)
+        delete db;
+    if (cli)
+    {
+        cli->flush();
+        delete cli;
+    }
+    if (epmgr)
+        delete epmgr;
+    if (ringloop)
+        delete ringloop;
+}
+
+json11::Json::object kv_cli_t::parse_args(int narg, const char *args[])
+{
+    json11::Json::object cfg;
+    for (int i = 1; i < narg; i++)
+    {
+        if (!strcmp(args[i], "-h") || !strcmp(args[i], "--help"))
+        {
+            printf(
+                "Vitastor Key/Value CLI\n"
+                "(c) Vitaliy Filippov, 2023+ (VNPL-1.1)\n"
+                "\n"
+                "USAGE: %s [--etcd_address ADDR] [OTHER OPTIONS]\n",
+                exe_name
+            );
+            exit(0);
+        }
+        else if (args[i][0] == '-' && args[i][1] == '-')
+        {
+            const char *opt = args[i]+2;
+            cfg[opt] = !strcmp(opt, "json") || i == narg-1 ? "1" : args[++i];
+        }
+    }
+    return cfg;
+}
+
+void kv_cli_t::run(const json11::Json::object & cfg)
+{
+    // Create client
+    ringloop = new ring_loop_t(512);
+    epmgr = new epoll_manager_t(ringloop);
+    cli = new cluster_client_t(ringloop, epmgr->tfd, cfg);
+    db = new kv_dbw_t(cli);
+    // Load image metadata
+    while (!cli->is_ready())
+    {
+        ringloop->loop();
+        if (cli->is_ready())
+            break;
+        ringloop->wait();
+    }
+    // Run
+    fcntl(0, F_SETFL, fcntl(0, F_GETFL, 0) | O_NONBLOCK);
+    try
+    {
+        epmgr->tfd->set_fd_handler(0, false, [this](int fd, int events)
+        {
+            if (events & EPOLLIN)
+            {
+                read_cmd();
+            }
+            if (events & EPOLLRDHUP)
+            {
+                epmgr->tfd->set_fd_handler(0, false, NULL);
+                finished = true;
+            }
+        });
+        interactive = true;
+        printf("> ");
+    }
+    catch (std::exception & e)
+    {
+        // Can't add to epoll, STDIN is probably a file
+        read_cmd();
+    }
+    while (!finished)
+    {
+        ringloop->loop();
+        if (!finished)
+            ringloop->wait();
+    }
+    // Destroy the client
+    delete db;
+    db = NULL;
+    cli->flush();
+    delete cli;
+    delete epmgr;
+    delete ringloop;
+    cli = NULL;
+    epmgr = NULL;
+    ringloop = NULL;
+}
+
+void kv_cli_t::read_cmd()
+{
+    if (!cur_cmd_alloc)
+    {
+        cur_cmd_alloc = 65536;
+        cur_cmd = (char*)malloc_or_die(cur_cmd_alloc);
+    }
+    while (cur_cmd_size < cur_cmd_alloc)
+    {
+        int r = read(0, cur_cmd+cur_cmd_size, cur_cmd_alloc-cur_cmd_size);
+        if (r < 0 && errno != EAGAIN)
+            fprintf(stderr, "Error reading from stdin: %s\n", strerror(errno));
+        if (r > 0)
+            cur_cmd_size += r;
+        if (r == 0)
+            eof = true;
+        if (r <= 0)
+            break;
+    }
+    next_cmd();
+}
+
+void kv_cli_t::next_cmd()
+{
+    if (in_progress > 0)
+    {
+        return;
+    }
+    int pos = 0;
+    for (; pos < cur_cmd_size; pos++)
+    {
+        if (cur_cmd[pos] == '\n' || cur_cmd[pos] == '\r')
+        {
+            auto cmd = trim(std::string(cur_cmd, pos));
+            pos++;
+            memmove(cur_cmd, cur_cmd+pos, cur_cmd_size-pos);
+            cur_cmd_size -= pos;
+            in_progress++;
+            handle_cmd(cmd, [this]()
+            {
+                in_progress--;
+                if (interactive)
+                    printf("> ");
+                next_cmd();
+                if (!in_progress)
+                    read_cmd();
+            });
+            break;
+        }
+    }
+    if (eof && !in_progress)
+    {
+        finished = true;
+    }
+}
+
+void kv_cli_t::handle_cmd(const std::string & cmd, std::function<void()> cb)
+{
+    if (cmd == "")
+    {
+        cb();
+        return;
+    }
+    auto pos = cmd.find_first_of(" \t");
+    if (pos != std::string::npos)
+    {
+        while (pos < cmd.size()-1 && (cmd[pos+1] == ' ' || cmd[pos+1] == '\t'))
+            pos++;
+    }
+    auto opname = strtolower(pos == std::string::npos ? cmd : cmd.substr(0, pos));
+    if (opname == "open")
+    {
+        uint64_t pool_id = 0;
+        inode_t inode_id = 0;
+        uint32_t kv_block_size = 0;
+        int scanned = sscanf(cmd.c_str() + pos+1, "%lu %lu %u", &pool_id, &inode_id, &kv_block_size);
+        if (scanned == 2)
+        {
+            kv_block_size = 4096;
+        }
+        if (scanned < 2 || !pool_id || !inode_id || !kv_block_size || (kv_block_size & (kv_block_size-1)) != 0)
+        {
+            fprintf(stderr, "Usage: open <pool_id> <inode_id> [block_size]. Block size must be a power of 2. Default is 4096.\n");
+            cb();
+            return;
+        }
+        cfg["kv_block_size"] = (uint64_t)kv_block_size;
+        db->open(INODE_WITH_POOL(pool_id, inode_id), cfg, [=](int res)
+        {
+            if (res < 0)
+                fprintf(stderr, "Error opening index: %s (code %d)\n", strerror(-res), res);
+            else
+                printf("Index opened. Current size: %lu bytes\n", db->get_size());
+            cb();
+        });
+    }
+    else if (opname == "config")
+    {
+        auto pos2 = cmd.find_first_of(" \t", pos+1);
+        if (pos2 == std::string::npos)
+        {
+            fprintf(stderr, "Usage: config <property> <value>\n");
+            cb();
+            return;
+        }
+        auto key = trim(cmd.substr(pos+1, pos2-pos-1));
+        auto value = parse_size(trim(cmd.substr(pos2+1)));
+        if (key != "kv_memory_limit" &&
+            key != "kv_allocate_blocks" &&
+            key != "kv_evict_max_misses" &&
+            key != "kv_evict_attempts_per_level" &&
+            key != "kv_evict_unused_age" &&
+            key != "kv_log_level")
+        {
+            fprintf(
+                stderr, "Allowed properties: kv_memory_limit, kv_allocate_blocks,"
+                " kv_evict_max_misses, kv_evict_attempts_per_level, kv_evict_unused_age, kv_log_level\n"
+            );
+        }
+        else
+        {
+            cfg[key] = value;
+            db->set_config(cfg);
+        }
+        cb();
+    }
+    else if (opname == "get" || opname == "set" || opname == "del")
+    {
+        if (opname == "get" || opname == "del")
+        {
+            if (pos == std::string::npos)
+            {
+                fprintf(stderr, "Usage: %s <key>\n", opname.c_str());
+                cb();
+                return;
+            }
+            auto key = trim(cmd.substr(pos+1));
+            if (opname == "get")
+            {
+                db->get(key, [this, cb](int res, const std::string & value)
+                {
+                    if (res < 0)
+                        fprintf(stderr, "Error: %s (code %d)\n", strerror(-res), res);
+                    else
+                    {
+                        write(1, value.c_str(), value.size());
+                        write(1, "\n", 1);
+                    }
+                    cb();
+                });
+            }
+            else
+            {
+                db->del(key, [this, cb](int res)
+                {
+                    if (res < 0)
+                        fprintf(stderr, "Error: %s (code %d)\n", strerror(-res), res);
+                    else
+                        printf("OK\n");
+                    cb();
+                });
+            }
+        }
+        else
+        {
+            auto pos2 = cmd.find_first_of(" \t", pos+1);
+            if (pos2 == std::string::npos)
+            {
+                fprintf(stderr, "Usage: set <key> <value>\n");
+                cb();
+                return;
+            }
+            auto key = trim(cmd.substr(pos+1, pos2-pos-1));
+            auto value = trim(cmd.substr(pos2+1));
+            db->set(key, value, [this, cb](int res)
+            {
+                if (res < 0)
+                    fprintf(stderr, "Error: %s (code %d)\n", strerror(-res), res);
+                else
+                    printf("OK\n");
+                cb();
+            });
+        }
+    }
+    else if (opname == "list")
+    {
+        std::string start, end;
+        if (pos != std::string::npos)
+        {
+            auto pos2 = cmd.find_first_of(" \t", pos+1);
+            if (pos2 != std::string::npos)
+            {
+                start = trim(cmd.substr(pos+1, pos2-pos-1));
+                end = trim(cmd.substr(pos2+1));
+            }
+            else
+            {
+                start = trim(cmd.substr(pos+1));
+            }
+        }
+        void *handle = db->list_start(start);
+        db->list_next(handle, [=](int res, const std::string & key, const std::string & value)
+        {
+            if (res < 0)
+            {
+                if (res != -ENOENT)
+                {
+                    fprintf(stderr, "Error: %s (code %d)\n", strerror(-res), res);
+                }
+                db->list_close(handle);
+                cb();
+            }
+            else
+            {
+                printf("%s = %s\n", key.c_str(), value.c_str());
+                db->list_next(handle, NULL);
+            }
+        });
+    }
+    else if (opname == "close")
+    {
+        db->close([=]()
+        {
+            printf("Index closed\n");
+            cb();
+        });
+    }
+    else if (opname == "quit" || opname == "q")
+    {
+        ::close(0);
+        finished = true;
+    }
+    else
+    {
+        fprintf(
+            stderr, "Unknown operation: %s. Supported operations:\n"
+            "open <pool_id> <inode_id> [block_size]\n"
+            "config <property> <value>\n"
+            "get <key>\nset <key> <value>\ndel <key>\nlist [<start> [end]]\n"
+            "close\nquit\n", opname.c_str()
+        );
+        cb();
+    }
+}
+
+int main(int narg, const char *args[])
+{
+    setvbuf(stdout, NULL, _IONBF, 0);
+    setvbuf(stderr, NULL, _IONBF, 0);
+    exe_name = args[0];
+    kv_cli_t *p = new kv_cli_t();
+    p->run(kv_cli_t::parse_args(narg, args));
+    delete p;
+    return 0;
+}

src/kv_db.h

@@ -0,0 +1,36 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+//
+// Vitastor shared key/value database
+// Parallel optimistic B-Tree O:-)
+
+#pragma once
+
+#include "cluster_client.h"
+
+struct kv_db_t;
+
+struct kv_dbw_t
+{
+    kv_dbw_t(cluster_client_t *cli);
+    ~kv_dbw_t();
+
+    void open(inode_t inode_id, json11::Json cfg, std::function<void(int)> cb);
+    void set_config(json11::Json cfg);
+    void close(std::function<void()> cb);
+
+    uint64_t get_size();
+
+    void get(const std::string & key, std::function<void(int res, const std::string & value)> cb,
+        bool allow_old_cached = false);
+    void set(const std::string & key, const std::string & value, std::function<void(int res)> cb,
+        std::function<bool(int res, const std::string & value)> cas_compare = NULL);
+    void del(const std::string & key, std::function<void(int res)> cb,
+        std::function<bool(int res, const std::string & value)> cas_compare = NULL);
+
+    void* list_start(const std::string & start);
+    void list_next(void *handle, std::function<void(int res, const std::string & key, const std::string & value)> cb);
+    void list_close(void *handle);
+
+    kv_db_t *db;
+};

src/kv_stress.cpp

@@ -0,0 +1,697 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+//
+// Vitastor shared key/value database stress tester / benchmark
+
+#define _XOPEN_SOURCE
+#include <limits.h>
+
+#include <netinet/tcp.h>
+#include <sys/epoll.h>
+#include <unistd.h>
+#include <fcntl.h>
+//#include <signal.h>
+
+#include "epoll_manager.h"
+#include "str_util.h"
+#include "kv_db.h"
+
+const char *exe_name = NULL;
+
+struct kv_test_listing_t
+{
+    uint64_t count = 0, done = 0;
+    void *handle = NULL;
+    std::string next_after;
+    std::set<std::string> inflights;
+    timespec tv_begin;
+    bool error = false;
+};
+
+struct kv_test_lat_t
+{
+    const char *name = NULL;
+    uint64_t usec = 0, count = 0;
+};
+
+struct kv_test_stat_t
+{
+    kv_test_lat_t get, add, update, del, list;
+    uint64_t list_keys = 0;
+};
+
+class kv_test_t
+{
+public:
+    // Config
+    json11::Json::object kv_cfg;
+    std::string key_prefix, key_suffix;
+    uint64_t inode_id = 0;
+    uint64_t op_count = 1000000;
+    uint64_t runtime_sec = 0;
+    uint64_t parallelism = 4;
+    uint64_t reopen_prob = 1;
+    uint64_t get_prob = 30000;
+    uint64_t add_prob = 20000;
+    uint64_t update_prob = 20000;
+    uint64_t del_prob = 5000;
+    uint64_t list_prob = 300;
+    uint64_t min_key_len = 10;
+    uint64_t max_key_len = 70;
+    uint64_t min_value_len = 50;
+    uint64_t max_value_len = 300;
+    uint64_t min_list_count = 10;
+    uint64_t max_list_count = 1000;
+    uint64_t print_stats_interval = 1;
+    bool json_output = false;
+    uint64_t log_level = 1;
+    bool trace = false;
+    bool stop_on_error = false;
+    // FIXME: Multiple clients
+    kv_test_stat_t stat, prev_stat;
+    timespec prev_stat_time, start_stat_time;
+
+    // State
+    kv_dbw_t *db = NULL;
+    ring_loop_t *ringloop = NULL;
+    epoll_manager_t *epmgr = NULL;
+    cluster_client_t *cli = NULL;
+    ring_consumer_t consumer;
+    bool finished = false;
+    uint64_t total_prob = 0;
+    uint64_t ops_sent = 0, ops_done = 0;
+    int stat_timer_id = -1;
+    int in_progress = 0;
+    bool reopening = false;
+    std::set<kv_test_listing_t*> listings;
+    std::set<std::string> changing_keys;
+    std::map<std::string, std::string> values;
+
+    ~kv_test_t();
+
+    static json11::Json::object parse_args(int narg, const char *args[]);
+    void parse_config(json11::Json cfg);
+    void run(json11::Json cfg);
+    void loop();
+    void print_stats(kv_test_stat_t & prev_stat, timespec & prev_stat_time);
+    void print_total_stats();
+    void start_change(const std::string & key);
+    void stop_change(const std::string & key);
+    void add_stat(kv_test_lat_t & stat, timespec tv_begin);
+};
+
+kv_test_t::~kv_test_t()
+{
+    if (db)
+        delete db;
+    if (cli)
+    {
+        cli->flush();
+        delete cli;
+    }
+    if (epmgr)
+        delete epmgr;
+    if (ringloop)
+        delete ringloop;
+}
+
+json11::Json::object kv_test_t::parse_args(int narg, const char *args[])
+{
+    json11::Json::object cfg;
+    for (int i = 1; i < narg; i++)
+    {
+        if (!strcmp(args[i], "-h") || !strcmp(args[i], "--help"))
+        {
+            printf(
+                "Vitastor Key/Value DB stress tester / benchmark\n"
+                "(c) Vitaliy Filippov, 2023+ (VNPL-1.1)\n"
+                "\n"
+                "USAGE: %s --pool_id POOL_ID --inode_id INODE_ID [OPTIONS]\n"
+                "  --op_count 1000000\n"
+                "    Total operations to run during test. 0 means unlimited\n"
+                "  --key_prefix \"\"\n"
+                "    Prefix for all keys read or written (to avoid collisions)\n"
+                "  --key_suffix \"\"\n"
+                "    Suffix for all keys read or written (to avoid collisions, but scan all DB)\n"
+                "  --runtime 0\n"
+                "    Run for this number of seconds. 0 means unlimited\n"
+                "  --parallelism 4\n"
+                "    Run this number of operations in parallel\n"
+                "  --get_prob 30000\n"
+                "    Fraction of key retrieve operations\n"
+                "  --add_prob 20000\n"
+                "    Fraction of key addition operations\n"
+                "  --update_prob 20000\n"
+                "    Fraction of key update operations\n"
+                "  --del_prob 30000\n"
+                "    Fraction of key delete operations\n"
+                "  --list_prob 300\n"
+                "    Fraction of listing operations\n"
+                "  --min_key_len 10\n"
+                "    Minimum key size in bytes\n"
+                "  --max_key_len 70\n"
+                "    Maximum key size in bytes\n"
+                "  --min_value_len 50\n"
+                "    Minimum value size in bytes\n"
+                "  --max_value_len 300\n"
+                "    Maximum value size in bytes\n"
+                "  --min_list_count 10\n"
+                "    Minimum number of keys read in listing (0 = all keys)\n"
+                "  --max_list_count 1000\n"
+                "    Maximum number of keys read in listing\n"
+                "  --print_stats 1\n"
+                "    Print operation statistics every this number of seconds\n"
+                "  --json\n"
+                "    JSON output\n"
+                "  --stop_on_error 0\n"
+                "    Stop on first execution error, mismatch, lost key or extra key during listing\n"
+                "  --kv_memory_limit 128M\n"
+                "    Maximum memory to use for vitastor-kv index cache\n"
+                "  --kv_allocate_blocks 4\n"
+                "    Number of PG blocks used for new tree block allocation in parallel\n"
+                "  --kv_evict_max_misses 10\n"
+                "    Eviction algorithm parameter: retry eviction from another random spot\n"
+                "    if this number of keys is used currently or was used recently\n"
+                "  --kv_evict_attempts_per_level 3\n"
+                "    Retry eviction at most this number of times per tree level, starting\n"
+                "    with bottom-most levels\n"
+                "  --kv_evict_unused_age 1000\n"
+                "    Evict only keys unused during this number of last operations\n"
+                "  --kv_log_level 1\n"
+                "    Log level. 0 = errors, 1 = warnings, 10 = trace operations\n",
+                exe_name
+            );
+            exit(0);
+        }
+        else if (args[i][0] == '-' && args[i][1] == '-')
+        {
+            const char *opt = args[i]+2;
+            cfg[opt] = !strcmp(opt, "json") || i == narg-1 ? "1" : args[++i];
+        }
+    }
+    return cfg;
+}
+
+void kv_test_t::parse_config(json11::Json cfg)
+{
+    inode_id = INODE_WITH_POOL(cfg["pool_id"].uint64_value(), cfg["inode_id"].uint64_value());
+    if (cfg["op_count"].uint64_value() > 0)
+        op_count = cfg["op_count"].uint64_value();
+    key_prefix = cfg["key_prefix"].string_value();
+    key_suffix = cfg["key_suffix"].string_value();
+    if (cfg["runtime"].uint64_value() > 0)
+        runtime_sec = cfg["runtime"].uint64_value();
+    if (cfg["parallelism"].uint64_value() > 0)
+        parallelism = cfg["parallelism"].uint64_value();
+    if (!cfg["reopen_prob"].is_null())
+        reopen_prob = cfg["reopen_prob"].uint64_value();
+    if (!cfg["get_prob"].is_null())
+        get_prob = cfg["get_prob"].uint64_value();
+    if (!cfg["add_prob"].is_null())
+        add_prob = cfg["add_prob"].uint64_value();
+    if (!cfg["update_prob"].is_null())
+        update_prob = cfg["update_prob"].uint64_value();
+    if (!cfg["del_prob"].is_null())
+        del_prob = cfg["del_prob"].uint64_value();
+    if (!cfg["list_prob"].is_null())
+        list_prob = cfg["list_prob"].uint64_value();
+    if (!cfg["min_key_len"].is_null())
+        min_key_len = cfg["min_key_len"].uint64_value();
+    if (cfg["max_key_len"].uint64_value() > 0)
+        max_key_len = cfg["max_key_len"].uint64_value();
+    if (!cfg["min_value_len"].is_null())
+        min_value_len = cfg["min_value_len"].uint64_value();
+    if (cfg["max_value_len"].uint64_value() > 0)
+        max_value_len = cfg["max_value_len"].uint64_value();
+    if (!cfg["min_list_count"].is_null())
+        min_list_count = cfg["min_list_count"].uint64_value();
+    if (!cfg["max_list_count"].is_null())
+        max_list_count = cfg["max_list_count"].uint64_value();
+    if (!cfg["print_stats"].is_null())
+        print_stats_interval = cfg["print_stats"].uint64_value();
+    if (!cfg["json"].is_null())
+        json_output = true;
+    if (!cfg["stop_on_error"].is_null())
+        stop_on_error = cfg["stop_on_error"].bool_value();
+    if (!cfg["kv_memory_limit"].is_null())
+        kv_cfg["kv_memory_limit"] = cfg["kv_memory_limit"];
+    if (!cfg["kv_allocate_blocks"].is_null())
+        kv_cfg["kv_allocate_blocks"] = cfg["kv_allocate_blocks"];
+    if (!cfg["kv_evict_max_misses"].is_null())
+        kv_cfg["kv_evict_max_misses"] = cfg["kv_evict_max_misses"];
+    if (!cfg["kv_evict_attempts_per_level"].is_null())
+        kv_cfg["kv_evict_attempts_per_level"] = cfg["kv_evict_attempts_per_level"];
+    if (!cfg["kv_evict_unused_age"].is_null())
+        kv_cfg["kv_evict_unused_age"] = cfg["kv_evict_unused_age"];
+    if (!cfg["kv_log_level"].is_null())
+    {
+        log_level = cfg["kv_log_level"].uint64_value();
+        trace = log_level >= 10;
+        kv_cfg["kv_log_level"] = cfg["kv_log_level"];
+    }
+    total_prob = reopen_prob+get_prob+add_prob+update_prob+del_prob+list_prob;
+    stat.get.name = "get";
+    stat.add.name = "add";
+    stat.update.name = "update";
+    stat.del.name = "del";
+    stat.list.name = "list";
+}
+
+void kv_test_t::run(json11::Json cfg)
+{
+    srand48(time(NULL));
+    parse_config(cfg);
+    // Create client
+    ringloop = new ring_loop_t(512);
+    epmgr = new epoll_manager_t(ringloop);
+    cli = new cluster_client_t(ringloop, epmgr->tfd, cfg);
+    db = new kv_dbw_t(cli);
+    // Load image metadata
+    while (!cli->is_ready())
+    {
+        ringloop->loop();
+        if (cli->is_ready())
+            break;
+        ringloop->wait();
+    }
+    // Run
+    reopening = true;
+    db->open(inode_id, kv_cfg, [this](int res)
+    {
+        reopening = false;
+        if (res < 0)
+        {
+            fprintf(stderr, "ERROR: Open index: %d (%s)\n", res, strerror(-res));
+            exit(1);
+        }
+        if (trace)
+            printf("Index opened\n");
+        ringloop->wakeup();
+    });
+    consumer.loop = [this]() { loop(); };
+    ringloop->register_consumer(&consumer);
+    if (print_stats_interval)
+        stat_timer_id = epmgr->tfd->set_timer(print_stats_interval*1000, true, [this](int) { print_stats(prev_stat, prev_stat_time); });
+    clock_gettime(CLOCK_REALTIME, &start_stat_time);
+    prev_stat_time = start_stat_time;
+    while (!finished)
+    {
+        ringloop->loop();
+        if (!finished)
+            ringloop->wait();
+    }
+    if (stat_timer_id >= 0)
+        epmgr->tfd->clear_timer(stat_timer_id);
+    ringloop->unregister_consumer(&consumer);
+    // Print total stats
+    print_total_stats();
+    // Destroy the client
+    delete db;
+    db = NULL;
+    cli->flush();
+    delete cli;
+    delete epmgr;
+    delete ringloop;
+    cli = NULL;
+    epmgr = NULL;
+    ringloop = NULL;
+}
+
+static const char *base64_chars = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789@+/";
+
+std::string random_str(int len)
+{
+    std::string str;
+    str.resize(len);
+    for (int i = 0; i < len; i++)
+    {
+        str[i] = base64_chars[lrand48() % 64];
+    }
+    return str;
+}
+
+void kv_test_t::loop()
+{
+    if (reopening)
+    {
+        return;
+    }
+    if (ops_done >= op_count)
+    {
+        finished = true;
+    }
+    while (!finished && ops_sent < op_count && in_progress < parallelism)
+    {
+        uint64_t dice = (lrand48() % total_prob);
+        if (dice < reopen_prob)
+        {
+            reopening = true;
+            db->close([this]()
+            {
+                if (trace)
+                    printf("Index closed\n");
+                db->open(inode_id, kv_cfg, [this](int res)
+                {
+                    reopening = false;
+                    if (res < 0)
+                    {
+                        fprintf(stderr, "ERROR: Reopen index: %d (%s)\n", res, strerror(-res));
+                        finished = true;
+                        return;
+                    }
+                    if (trace)
+                        printf("Index reopened\n");
+                    ringloop->wakeup();
+                });
+            });
+            return;
+        }
+        else if (dice < reopen_prob+get_prob)
+        {
+            // get existing
+            auto key = random_str(max_key_len);
+            auto k_it = values.lower_bound(key);
+            if (k_it == values.end())
+                continue;
+            key = k_it->first;
+            if (changing_keys.find(key) != changing_keys.end())
+                continue;
+            in_progress++;
+            ops_sent++;
+            if (trace)
+                printf("get %s\n", key.c_str());
+            timespec tv_begin;
+            clock_gettime(CLOCK_REALTIME, &tv_begin);
+            db->get(key, [this, key, tv_begin](int res, const std::string & value)
+            {
+                add_stat(stat.get, tv_begin);
+                ops_done++;
+                in_progress--;
+                auto it = values.find(key);
+                if (res != (it == values.end() ? -ENOENT : 0))
+                {
+                    fprintf(stderr, "ERROR: get %s: %d (%s)\n", key.c_str(), res, strerror(-res));
+                    if (stop_on_error)
+                        exit(1);
+                }
+                else if (it != values.end() && value != it->second)
+                {
+                    fprintf(stderr, "ERROR: get %s: mismatch: %s vs %s\n", key.c_str(), value.c_str(), it->second.c_str());
+                    if (stop_on_error)
+                        exit(1);
+                }
+                ringloop->wakeup();
+            });
+        }
+        else if (dice < reopen_prob+get_prob+add_prob+update_prob)
+        {
+            bool is_add = false;
+            std::string key;
+            if (dice < reopen_prob+get_prob+add_prob)
+            {
+                // add
+                is_add = true;
+                uint64_t key_len = min_key_len + (max_key_len > min_key_len ? lrand48() % (max_key_len-min_key_len) : 0);
+                key = key_prefix + random_str(key_len) + key_suffix;
+            }
+            else
+            {
+                // update
+                key = random_str(max_key_len);
+                auto k_it = values.lower_bound(key);
+                if (k_it == values.end())
+                    continue;
+                key = k_it->first;
+            }
+            if (changing_keys.find(key) != changing_keys.end())
+                continue;
+            uint64_t value_len = min_value_len + (max_value_len > min_value_len ? lrand48() % (max_value_len-min_value_len) : 0);
+            auto value = random_str(value_len);
+            start_change(key);
+            ops_sent++;
+            in_progress++;
+            if (trace)
+                printf("set %s = %s\n", key.c_str(), value.c_str());
+            timespec tv_begin;
+            clock_gettime(CLOCK_REALTIME, &tv_begin);
+            db->set(key, value, [this, key, value, tv_begin, is_add](int res)
+            {
+                add_stat(is_add ? stat.add : stat.update, tv_begin);
+                stop_change(key);
+                ops_done++;
+                in_progress--;
+                if (res != 0)
+                {
+                    fprintf(stderr, "ERROR: set %s = %s: %d (%s)\n", key.c_str(), value.c_str(), res, strerror(-res));
+                    if (stop_on_error)
+                        exit(1);
+                }
+                else
+                {
+                    values[key] = value;
+                }
+                ringloop->wakeup();
+            }, NULL);
+        }
+        else if (dice < reopen_prob+get_prob+add_prob+update_prob+del_prob)
+        {
+            // delete
+            auto key = random_str(max_key_len);
+            auto k_it = values.lower_bound(key);
+            if (k_it == values.end())
+                continue;
+            key = k_it->first;
+            if (changing_keys.find(key) != changing_keys.end())
+                continue;
+            start_change(key);
+            ops_sent++;
+            in_progress++;
+            if (trace)
+                printf("del %s\n", key.c_str());
+            timespec tv_begin;
+            clock_gettime(CLOCK_REALTIME, &tv_begin);
+            db->del(key, [this, key, tv_begin](int res)
+            {
+                add_stat(stat.del, tv_begin);
+                stop_change(key);
+                ops_done++;
+                in_progress--;
+                if (res != 0)
+                {
+                    fprintf(stderr, "ERROR: del %s: %d (%s)\n", key.c_str(), res, strerror(-res));
+                    if (stop_on_error)
+                        exit(1);
+                }
+                else
+                {
+                    values.erase(key);
+                }
+                ringloop->wakeup();
+            }, NULL);
+        }
+        else if (dice < reopen_prob+get_prob+add_prob+update_prob+del_prob+list_prob)
+        {
+            // list
+            ops_sent++;
+            in_progress++;
+            auto key = random_str(max_key_len);
+            auto lst = new kv_test_listing_t;
+            auto k_it = values.lower_bound(key);
+            lst->count = min_list_count + (max_list_count > min_list_count ? lrand48() % (max_list_count-min_list_count) : 0);
+            lst->handle = db->list_start(k_it == values.begin() ? key_prefix : key);
+            lst->next_after = k_it == values.begin() ? key_prefix : key;
+            lst->inflights = changing_keys;
+            listings.insert(lst);
+            if (trace)
+                printf("list from %s\n", key.c_str());
+            clock_gettime(CLOCK_REALTIME, &lst->tv_begin);
+            db->list_next(lst->handle, [this, lst](int res, const std::string & key, const std::string & value)
+            {
+                if (log_level >= 11)
+                    printf("list: %s = %s\n", key.c_str(), value.c_str());
+                if (res >= 0 && key_prefix.size() && (key.size() < key_prefix.size() ||
+                    key.substr(0, key_prefix.size()) != key_prefix))
+                {
+                    // stop at this key
+                    res = -ENOENT;
+                }
+                if (res < 0 || (lst->count > 0 && lst->done >= lst->count))
+                {
+                    add_stat(stat.list, lst->tv_begin);
+                    if (res == 0)
+                    {
+                        // ok (done >= count)
+                    }
+                    else if (res != -ENOENT)
+                    {
+                        fprintf(stderr, "ERROR: list: %d (%s)\n", res, strerror(-res));
+                        lst->error = true;
+                    }
+                    else
+                    {
+                        auto k_it = lst->next_after == "" ? values.begin() : values.upper_bound(lst->next_after);
+                        while (k_it != values.end())
+                        {
+                            while (k_it != values.end() && lst->inflights.find(k_it->first) != lst->inflights.end())
+                                k_it++;
+                            if (k_it != values.end())
+                            {
+                                fprintf(stderr, "ERROR: list: missing key %s\n", (k_it++)->first.c_str());
+                                lst->error = true;
+                            }
+                        }
+                    }
+                    if (lst->error && stop_on_error)
+                        exit(1);
+                    ops_done++;
+                    in_progress--;
+                    db->list_close(lst->handle);
+                    delete lst;
+                    listings.erase(lst);
+                    ringloop->wakeup();
+                }
+                else
+                {
+                    stat.list_keys++;
+                    // Do not check modified keys in listing
+                    // Listing may return their old or new state
+                    if ((!key_suffix.size() || key.size() >= key_suffix.size() &&
+                        key.substr(key.size()-key_suffix.size()) == key_suffix) &&
+                        lst->inflights.find(key) == lst->inflights.end())
+                    {
+                        lst->done++;
+                        auto k_it = lst->next_after == "" ? values.begin() : values.upper_bound(lst->next_after);
+                        while (true)
+                        {
+                            while (k_it != values.end() && lst->inflights.find(k_it->first) != lst->inflights.end())
+                            {
+                                k_it++;
+                            }
+                            if (k_it == values.end() || k_it->first > key)
+                            {
+                                fprintf(stderr, "ERROR: list: extra key %s\n", key.c_str());
+                                lst->error = true;
+                                break;
+                            }
+                            else if (k_it->first < key)
+                            {
+                                fprintf(stderr, "ERROR: list: missing key %s\n", k_it->first.c_str());
+                                lst->error = true;
+                                lst->next_after = k_it->first;
+                                k_it++;
+                            }
+                            else
+                            {
+                                if (k_it->second != value)
+                                {
+                                    fprintf(stderr, "ERROR: list: mismatch: %s = %s but should be %s\n",
+                                        key.c_str(), value.c_str(), k_it->second.c_str());
+                                    lst->error = true;
+                                }
+                                lst->next_after = k_it->first;
+                                break;
+                            }
+                        }
+                    }
+                    db->list_next(lst->handle, NULL);
+                }
+            });
+        }
+    }
+}
+
+void kv_test_t::add_stat(kv_test_lat_t & stat, timespec tv_begin)
+{
+    timespec tv_end;
+    clock_gettime(CLOCK_REALTIME, &tv_end);
+    int64_t usec = (tv_end.tv_sec - tv_begin.tv_sec)*1000000 +
+        (tv_end.tv_nsec - tv_begin.tv_nsec)/1000;
+    if (usec > 0)
+    {
+        stat.usec += usec;
+        stat.count++;
+    }
+}
+
+void kv_test_t::print_stats(kv_test_stat_t & prev_stat, timespec & prev_stat_time)
+{
+    timespec cur_stat_time;
+    clock_gettime(CLOCK_REALTIME, &cur_stat_time);
+    int64_t usec = (cur_stat_time.tv_sec - prev_stat_time.tv_sec)*1000000 +
+        (cur_stat_time.tv_nsec - prev_stat_time.tv_nsec)/1000;
+    if (usec > 0)
+    {
+        kv_test_lat_t *lats[] = { &stat.get, &stat.add, &stat.update, &stat.del, &stat.list };
+        kv_test_lat_t *prev[] = { &prev_stat.get, &prev_stat.add, &prev_stat.update, &prev_stat.del, &prev_stat.list };
+        if (!json_output)
+        {
+            char buf[128] = { 0 };
+            for (int i = 0; i < sizeof(lats)/sizeof(lats[0]); i++)
+            {
+                snprintf(buf, sizeof(buf)-1, "%.1f %s/s (%lu us)", (lats[i]->count-prev[i]->count)*1000000.0/usec,
+                    lats[i]->name, (lats[i]->usec-prev[i]->usec)/(lats[i]->count-prev[i]->count > 0 ? lats[i]->count-prev[i]->count : 1));
+                int k;
+                for (k = strlen(buf); k < strlen(lats[i]->name)+21; k++)
+                    buf[k] = ' ';
+                buf[k] = 0;
+                printf("%s", buf);
+            }
+            printf("\n");
+        }
+        else
+        {
+            int64_t runtime = (cur_stat_time.tv_sec - start_stat_time.tv_sec)*1000000 +
+                (cur_stat_time.tv_nsec - start_stat_time.tv_nsec)/1000;
+            printf("{\"runtime\":%.1f", (double)runtime/1000000.0);
+            for (int i = 0; i < sizeof(lats)/sizeof(lats[0]); i++)
+            {
+                if (lats[i]->count > prev[i]->count)
+                {
+                    printf(
+                        ",\"%s\":{\"avg\":{\"iops\":%.1f,\"usec\":%lu},\"total\":{\"count\":%lu,\"usec\":%lu}}",
+                        lats[i]->name, (lats[i]->count-prev[i]->count)*1000000.0/usec,
+                        (lats[i]->usec-prev[i]->usec)/(lats[i]->count-prev[i]->count),
+                        lats[i]->count, lats[i]->usec
+                    );
+                }
+            }
+            printf("}\n");
+        }
+    }
+    prev_stat = stat;
+    prev_stat_time = cur_stat_time;
+}
+
+void kv_test_t::print_total_stats()
+{
+    if (!json_output)
+        printf("Total:\n");
+    kv_test_stat_t start_stats;
+    timespec start_stat_time = this->start_stat_time;
+    print_stats(start_stats, start_stat_time);
+}
+
+void kv_test_t::start_change(const std::string & key)
+{
+    changing_keys.insert(key);
+    for (auto lst: listings)
+    {
+        lst->inflights.insert(key);
+    }
+}
+
+void kv_test_t::stop_change(const std::string & key)
+{
+    changing_keys.erase(key);
+}
+
+int main(int narg, const char *args[])
+{
+    setvbuf(stdout, NULL, _IONBF, 0);
+    setvbuf(stderr, NULL, _IONBF, 0);
+    exe_name = args[0];
+    kv_test_t *p = new kv_test_t();
+    p->run(kv_test_t::parse_args(narg, args));
+    delete p;
+    return 0;
+}

src/messenger.cpp

@@ -22,7 +22,7 @@ void osd_messenger_t::init()
     {
         rdma_context = msgr_rdma_context_t::create(
             rdma_device != "" ? rdma_device.c_str() : NULL,
-            rdma_port_num, rdma_gid_index, rdma_mtu, log_level
+            rdma_port_num, rdma_gid_index, rdma_mtu, rdma_odp, log_level
         );
         if (!rdma_context)
         {
@@ -167,6 +167,7 @@ void osd_messenger_t::parse_config(const json11::Json & config)
     this->rdma_max_msg = config["rdma_max_msg"].uint64_value();
     if (!this->rdma_max_msg || this->rdma_max_msg > 128*1024*1024)
         this->rdma_max_msg = 129*1024;
+    this->rdma_odp = config["rdma_odp"].bool_value();
 #endif
     this->receive_buffer_size = (uint32_t)config["tcp_header_buffer_size"].uint64_value();
     if (!this->receive_buffer_size || this->receive_buffer_size > 1024*1024*1024)
@@ -395,27 +396,24 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
             },
         },
     };
-    json11::Json::object payload;
-    if (this->osd_num)
-    {
-        payload["osd_num"] = this->osd_num;
-    }
 #ifdef WITH_RDMA
     if (rdma_context)
     {
         cl->rdma_conn = msgr_rdma_connection_t::create(rdma_context, rdma_max_send, rdma_max_recv, rdma_max_sge, rdma_max_msg);
         if (cl->rdma_conn)
         {
-            payload["connect_rdma"] = cl->rdma_conn->addr.to_string();
-            payload["rdma_max_msg"] = cl->rdma_conn->max_msg;
+            json11::Json payload = json11::Json::object {
+                { "connect_rdma", cl->rdma_conn->addr.to_string() },
+                { "rdma_max_msg", cl->rdma_conn->max_msg },
+            };
+            std::string payload_str = payload.dump();
+            op->req.show_conf.json_len = payload_str.size();
+            op->buf = malloc_or_die(payload_str.size());
+            op->iov.push_back(op->buf, payload_str.size());
+            memcpy(op->buf, payload_str.c_str(), payload_str.size());
         }
     }
 #endif
-    std::string payload_str = json11::Json(payload).dump();
-    op->req.show_conf.json_len = payload_str.size();
-    op->buf = malloc_or_die(payload_str.size());
-    op->iov.push_back(op->buf, payload_str.size());
-    memcpy(op->buf, payload_str.c_str(), payload_str.size());
     op->callback = [this, cl](osd_op_t *op)
     {
         std::string json_err;
@@ -493,7 +491,14 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
                     fprintf(stderr, "Connected to OSD %lu using RDMA\n", cl->osd_num);
                 }
                 cl->peer_state = PEER_RDMA;
-                tfd->set_fd_handler(cl->peer_fd, false, NULL);
+                tfd->set_fd_handler(cl->peer_fd, false, [this](int peer_fd, int epoll_events)
+                {
+                    // Do not miss the disconnection!
+                    if (epoll_events & EPOLLRDHUP)
+                    {
+                        handle_peer_epoll(peer_fd, epoll_events);
+                    }
+                });
                 // Add the initial receive request
                 try_recv_rdma(cl);
             }

src/messenger.h

@@ -131,6 +131,7 @@ protected:
     msgr_rdma_context_t *rdma_context = NULL;
     uint64_t rdma_max_sge = 0, rdma_max_send = 0, rdma_max_recv = 0;
     uint64_t rdma_max_msg = 0;
+    bool rdma_odp = false;
 #endif
 
     std::vector<int> read_ready_clients;
@@ -197,7 +198,9 @@ protected:
     void handle_reply_ready(osd_op_t *op);
 
 #ifdef WITH_RDMA
-    bool try_send_rdma(osd_client_t *cl);
+    void try_send_rdma(osd_client_t *cl);
+    void try_send_rdma_odp(osd_client_t *cl);
+    void try_send_rdma_nodp(osd_client_t *cl);
     bool try_recv_rdma(osd_client_t *cl);
     void handle_rdma_events();
 #endif

src/msgr_rdma.cpp

@@ -47,11 +47,29 @@ msgr_rdma_connection_t::~msgr_rdma_connection_t()
     if (qp)
         ibv_destroy_qp(qp);
     if (recv_buffers.size())
+    {
         for (auto b: recv_buffers)
-            free(b);
+        {
+            if (b.mr)
+                ibv_dereg_mr(b.mr);
+            free(b.buf);
+        }
+        recv_buffers.clear();
+    }
+    if (send_out.mr)
+    {
+        ibv_dereg_mr(send_out.mr);
+        send_out.mr = NULL;
+    }
+    if (send_out.buf)
+    {
+        free(send_out.buf);
+        send_out.buf = NULL;
+    }
+    send_out_size = 0;
 }
 
-msgr_rdma_context_t *msgr_rdma_context_t::create(const char *ib_devname, uint8_t ib_port, uint8_t gid_index, uint32_t mtu, int log_level)
+msgr_rdma_context_t *msgr_rdma_context_t::create(const char *ib_devname, uint8_t ib_port, uint8_t gid_index, uint32_t mtu, bool odp, int log_level)
 {
     int res;
     ibv_device **dev_list = NULL;
@@ -136,21 +154,27 @@ msgr_rdma_context_t *msgr_rdma_context_t::create(const char *ib_devname, uint8_t
             fprintf(stderr, "Couldn't query RDMA device for its features\n");
             goto cleanup;
         }
-        if (!(ctx->attrx.odp_caps.general_caps & IBV_ODP_SUPPORT) ||
+        ctx->odp = odp;
+        if (ctx->odp &&
+            (!(ctx->attrx.odp_caps.general_caps & IBV_ODP_SUPPORT) ||
             !(ctx->attrx.odp_caps.general_caps & IBV_ODP_SUPPORT_IMPLICIT) ||
             !(ctx->attrx.odp_caps.per_transport_caps.rc_odp_caps & IBV_ODP_SUPPORT_SEND) ||
-            !(ctx->attrx.odp_caps.per_transport_caps.rc_odp_caps & IBV_ODP_SUPPORT_RECV))
+            !(ctx->attrx.odp_caps.per_transport_caps.rc_odp_caps & IBV_ODP_SUPPORT_RECV)))
         {
-            fprintf(stderr, "The RDMA device isn't implicit ODP (On-Demand Paging) capable or does not support RC send and receive with ODP\n");
-            goto cleanup;
+            ctx->odp = false;
+            if (log_level > 0)
+                fprintf(stderr, "The RDMA device isn't implicit ODP (On-Demand Paging) capable, disabling it\n");
         }
     }
 
-    ctx->mr = ibv_reg_mr(ctx->pd, NULL, SIZE_MAX, IBV_ACCESS_LOCAL_WRITE | IBV_ACCESS_ON_DEMAND);
-    if (!ctx->mr)
+    if (ctx->odp)
     {
-        fprintf(stderr, "Couldn't register RDMA memory region\n");
-        goto cleanup;
+        ctx->mr = ibv_reg_mr(ctx->pd, NULL, SIZE_MAX, IBV_ACCESS_LOCAL_WRITE | IBV_ACCESS_ON_DEMAND);
+        if (!ctx->mr)
+        {
+            fprintf(stderr, "Couldn't register RDMA memory region\n");
+            goto cleanup;
+        }
     }
 
     ctx->channel = ibv_create_comp_channel(ctx->context);
@@ -365,12 +389,34 @@ static void try_send_rdma_wr(osd_client_t *cl, ibv_sge *sge, int op_sge)
     cl->rdma_conn->cur_send++;
 }
 
-bool osd_messenger_t::try_send_rdma(osd_client_t *cl)
+static int try_send_rdma_copy(osd_client_t *cl, uint8_t *dst, int dst_len)
+{
+    auto rc = cl->rdma_conn;
+    int total_dst_len = dst_len;
+    while (dst_len > 0 && rc->send_pos < cl->send_list.size())
+    {
+        iovec & iov = cl->send_list[rc->send_pos];
+        uint32_t len = (uint32_t)(iov.iov_len-rc->send_buf_pos < dst_len
+            ? iov.iov_len-rc->send_buf_pos : dst_len);
+        memcpy(dst, iov.iov_base+rc->send_buf_pos, len);
+        dst += len;
+        dst_len -= len;
+        rc->send_buf_pos += len;
+        if (rc->send_buf_pos >= iov.iov_len)
+        {
+            rc->send_pos++;
+            rc->send_buf_pos = 0;
+        }
+    }
+    return total_dst_len-dst_len;
+}
+
+void osd_messenger_t::try_send_rdma_odp(osd_client_t *cl)
 {
     auto rc = cl->rdma_conn;
     if (!cl->send_list.size() || rc->cur_send >= rc->max_send)
     {
-        return true;
+        return;
     }
     uint64_t op_size = 0, op_sge = 0;
     ibv_sge sge[rc->max_sge];
@@ -408,15 +454,70 @@ bool osd_messenger_t::try_send_rdma(osd_client_t *cl)
         rc->send_sizes.push_back(op_size);
         try_send_rdma_wr(cl, sge, op_sge);
     }
-    return true;
 }
 
-static void try_recv_rdma_wr(osd_client_t *cl, void *buf)
+void osd_messenger_t::try_send_rdma_nodp(osd_client_t *cl)
+{
+    auto rc = cl->rdma_conn;
+    if (!rc->send_out_size)
+    {
+        // Allocate send ring buffer, if not yet
+        rc->send_out_size = rc->max_msg*rdma_max_send;
+        rc->send_out.buf = malloc_or_die(rc->send_out_size);
+        if (!rdma_context->odp)
+        {
+            rc->send_out.mr = ibv_reg_mr(rdma_context->pd, rc->send_out.buf, rc->send_out_size, 0);
+            if (!rc->send_out.mr)
+            {
+                fprintf(stderr, "Failed to register RDMA memory region: %s\n", strerror(errno));
+                exit(1);
+            }
+        }
+    }
+    // Copy data into the buffer and send it
+    uint8_t *dst = NULL;
+    int dst_len = 0;
+    int copied = 1;
+    while (!rc->send_out_full && copied > 0 && rc->cur_send < rc->max_send)
+    {
+        dst = (uint8_t*)rc->send_out.buf + rc->send_out_pos;
+        dst_len = (rc->send_out_pos < rc->send_out_size ? rc->send_out_size-rc->send_out_pos : rc->send_done_pos-rc->send_out_pos);
+        if (dst_len > rc->max_msg)
+            dst_len = rc->max_msg;
+        copied = try_send_rdma_copy(cl, dst, dst_len);
+        if (copied > 0)
+        {
+            rc->send_out_pos += copied;
+            if (rc->send_out_pos == rc->send_out_size)
+                rc->send_out_pos = 0;
+            assert(rc->send_out_pos < rc->send_out_size);
+            if (rc->send_out_pos >= rc->send_done_pos)
+                rc->send_out_full = true;
+            ibv_sge sge = {
+                .addr = (uintptr_t)dst,
+                .length = (uint32_t)copied,
+                .lkey = rdma_context->odp ? rdma_context->mr->lkey : rc->send_out.mr->lkey,
+            };
+            try_send_rdma_wr(cl, &sge, 1);
+            rc->send_sizes.push_back(copied);
+        }
+    }
+}
+
+void osd_messenger_t::try_send_rdma(osd_client_t *cl)
+{
+    if (rdma_context->odp)
+        try_send_rdma_odp(cl);
+    else
+        try_send_rdma_nodp(cl);
+}
+
+static void try_recv_rdma_wr(osd_client_t *cl, msgr_rdma_buf_t b)
 {
     ibv_sge sge = {
-        .addr = (uintptr_t)buf,
+        .addr = (uintptr_t)b.buf,
         .length = (uint32_t)cl->rdma_conn->max_msg,
-        .lkey = cl->rdma_conn->ctx->mr->lkey,
+        .lkey = cl->rdma_conn->ctx->odp ? cl->rdma_conn->ctx->mr->lkey : b.mr->lkey,
     };
     ibv_recv_wr *bad_wr = NULL;
     ibv_recv_wr wr = {
@@ -438,9 +539,19 @@ bool osd_messenger_t::try_recv_rdma(osd_client_t *cl)
     auto rc = cl->rdma_conn;
     while (rc->cur_recv < rc->max_recv)
     {
-        void *buf = malloc_or_die(rc->max_msg);
-        rc->recv_buffers.push_back(buf);
-        try_recv_rdma_wr(cl, buf);
+        msgr_rdma_buf_t b;
+        b.buf = malloc_or_die(rc->max_msg);
+        if (!rdma_context->odp)
+        {
+            b.mr = ibv_reg_mr(rdma_context->pd, b.buf, rc->max_msg, IBV_ACCESS_LOCAL_WRITE);
+            if (!b.mr)
+            {
+                fprintf(stderr, "Failed to register RDMA memory region: %s\n", strerror(errno));
+                exit(1);
+            }
+        }
+        rc->recv_buffers.push_back(b);
+        try_recv_rdma_wr(cl, b);
     }
     return true;
 }
@@ -492,7 +603,7 @@ void osd_messenger_t::handle_rdma_events()
             if (!is_send)
             {
                 rc->cur_recv--;
-                if (!handle_read_buffer(cl, rc->recv_buffers[rc->next_recv_buf], wc[i].byte_len))
+                if (!handle_read_buffer(cl, rc->recv_buffers[rc->next_recv_buf].buf, wc[i].byte_len))
                 {
                     // handle_read_buffer may stop the client
                     continue;
@@ -505,6 +616,14 @@ void osd_messenger_t::handle_rdma_events()
                 rc->cur_send--;
                 uint64_t sent_size = rc->send_sizes.at(0);
                 rc->send_sizes.erase(rc->send_sizes.begin(), rc->send_sizes.begin()+1);
+                if (!rdma_context->odp)
+                {
+                    rc->send_done_pos += sent_size;
+                    rc->send_out_full = false;
+                    if (rc->send_done_pos == rc->send_out_size)
+                        rc->send_done_pos = 0;
+                    assert(rc->send_done_pos < rc->send_out_size);
+                }
                 int send_pos = 0, send_buf_pos = 0;
                 while (sent_size > 0)
                 {

src/msgr_rdma.h

@@ -23,6 +23,7 @@ struct msgr_rdma_context_t
     ibv_device *dev = NULL;
     ibv_device_attr_ex attrx;
     ibv_pd *pd = NULL;
+    bool odp = false;
     ibv_mr *mr = NULL;
     ibv_comp_channel *channel = NULL;
     ibv_cq *cq = NULL;
@@ -35,10 +36,16 @@ struct msgr_rdma_context_t
     int max_cqe = 0;
     int used_max_cqe = 0;
 
-    static msgr_rdma_context_t *create(const char *ib_devname, uint8_t ib_port, uint8_t gid_index, uint32_t mtu, int log_level);
+    static msgr_rdma_context_t *create(const char *ib_devname, uint8_t ib_port, uint8_t gid_index, uint32_t mtu, bool odp, int log_level);
     ~msgr_rdma_context_t();
 };
 
+struct msgr_rdma_buf_t
+{
+    void *buf = NULL;
+    ibv_mr *mr = NULL;
+};
+
 struct msgr_rdma_connection_t
 {
     msgr_rdma_context_t *ctx = NULL;
@@ -50,8 +57,11 @@ struct msgr_rdma_connection_t
 
     int send_pos = 0, send_buf_pos = 0;
     int next_recv_buf = 0;
-    std::vector<void*> recv_buffers;
+    std::vector<msgr_rdma_buf_t> recv_buffers;
     std::vector<uint64_t> send_sizes;
+    msgr_rdma_buf_t send_out;
+    int send_out_pos = 0, send_done_pos = 0, send_out_size = 0;
+    bool send_out_full = false;
 
     ~msgr_rdma_connection_t();
     static msgr_rdma_connection_t *create(msgr_rdma_context_t *ctx, uint32_t max_send, uint32_t max_recv, uint32_t max_sge, uint32_t max_msg);

src/msgr_send.cpp

@@ -3,6 +3,7 @@
 
 #define _XOPEN_SOURCE
 #include <limits.h>
+#include <sys/epoll.h>
 
 #include "messenger.h"
 
@@ -119,9 +120,9 @@ void osd_messenger_t::outbox_push(osd_op_t *cur_op)
             try_send(cl);
         }
     }
-    else if (cl->write_msg.msg_iovlen > 0 || !try_send(cl))
+    else
     {
-        if (cl->write_state == 0)
+        if ((cl->write_msg.msg_iovlen > 0 || !try_send(cl)) && (cl->write_state == 0))
         {
             cl->write_state = CL_WRITE_READY;
             write_ready_clients.push_back(cur_op->peer_fd);
@@ -283,7 +284,14 @@ void osd_messenger_t::handle_send(int result, osd_client_t *cl)
                 fprintf(stderr, "Successfully connected with client %d using RDMA\n", cl->peer_fd);
             }
             cl->peer_state = PEER_RDMA;
-            tfd->set_fd_handler(cl->peer_fd, false, NULL);
+            tfd->set_fd_handler(cl->peer_fd, false, [this](int peer_fd, int epoll_events)
+            {
+                // Do not miss the disconnection!
+                if (epoll_events & EPOLLRDHUP)
+                {
+                    handle_peer_epoll(peer_fd, epoll_events);
+                }
+            });
             // Add the initial receive request
             try_recv_rdma(cl);
         }

src/nbd_proxy.cpp

@@ -216,8 +216,16 @@ public:
         {
             nbd_timeout = cfg["nbd_timeout"].uint64_value();
         }
+        if (cfg["client_writeback_allowed"].is_null())
+        {
+            // NBD is always aware of fsync, so we allow write-back cache
+            // by default if it's enabled
+            auto obj = cfg.object_items();
+            obj["client_writeback_allowed"] = true;
+            cfg = obj;
+        }
         // Create client
-        ringloop = new ring_loop_t(512);
+        ringloop = new ring_loop_t(RINGLOOP_DEFAULT_SIZE);
         epmgr = new epoll_manager_t(ringloop);
         cli = new cluster_client_t(ringloop, epmgr->tfd, cfg);
         if (!inode)

src/nfs_proxy.cpp

@@ -65,8 +65,9 @@ json11::Json::object nfs_proxy_t::parse_args(int narg, const char *args[])
                 "  --pool <POOL>     use <POOL> as default pool for new files (images)\n"
                 "  --foreground 1    stay in foreground, do not daemonize\n"
                 "\n"
-                "NFS proxy is stateless if you use immediate_commit=all in your cluster, so\n"
-                "you can freely use multiple NFS proxies with L3 load balancing in this case.\n"
+                "NFS proxy is stateless if you use immediate_commit=all in your cluster and if\n"
+                "you do not use client_enable_writeback=true, so you can freely use multiple\n"
+                "NFS proxies with L3 load balancing in this case.\n"
                 "\n"
                 "Example start and mount commands for a custom NFS port:\n"
                 "  %s --etcd_address 192.168.5.10:2379 --portmap 0 --port 2050 --pool testpool\n"
@@ -114,8 +115,16 @@ void nfs_proxy_t::run(json11::Json cfg)
         if (name_prefix.size())
             name_prefix += "/";
     }
+    if (cfg["client_writeback_allowed"].is_null())
+    {
+        // NFS is always aware of fsync, so we allow write-back cache
+        // by default if it's enabled
+        auto obj = cfg.object_items();
+        obj["client_writeback_allowed"] = true;
+        cfg = obj;
+    }
     // Create client
-    ringloop = new ring_loop_t(512);
+    ringloop = new ring_loop_t(RINGLOOP_DEFAULT_SIZE);
     epmgr = new epoll_manager_t(ringloop);
     cli = new cluster_client_t(ringloop, epmgr->tfd, cfg);
     cmd = new cli_tool_t();

src/osd.cpp

@@ -160,6 +160,9 @@ void osd_t::parse_config(bool init)
         etcd_report_interval = config["etcd_report_interval"].uint64_value();
         if (etcd_report_interval <= 0)
             etcd_report_interval = 5;
+        etcd_stats_interval = config["etcd_stats_interval"].uint64_value();
+        if (etcd_stats_interval <= 0)
+            etcd_stats_interval = 30;
         readonly = json_is_true(config["readonly"]);
         run_primary = !json_is_false(config["run_primary"]);
         allow_test_ops = json_is_true(config["allow_test_ops"]);
@@ -184,14 +187,6 @@ void osd_t::parse_config(bool init)
         // Allow to set it to 0
         autosync_writes = config["autosync_writes"].uint64_value();
     }
-    if (!config["fsync_feedback_repeat_interval"].is_null())
-    {
-        fsync_feedback_repeat_interval = config["fsync_feedback_repeat_interval"].uint64_value();
-    }
-    if (!fsync_feedback_repeat_interval)
-    {
-        fsync_feedback_repeat_interval = 500; // ms
-    }
     if (!config["client_queue_depth"].is_null())
     {
         client_queue_depth = config["client_queue_depth"].uint64_value();
@@ -546,11 +541,15 @@ void osd_t::print_slow()
                 }
                 else if (op->req.hdr.opcode == OSD_OP_SEC_STABILIZE || op->req.hdr.opcode == OSD_OP_SEC_ROLLBACK)
                 {
-                    for (uint64_t i = 0; i < op->req.sec_stab.len; i += sizeof(obj_ver_id))
+                    for (uint64_t i = 0; i < op->req.sec_stab.len && i < sizeof(obj_ver_id)*12; i += sizeof(obj_ver_id))
                     {
                         obj_ver_id *ov = (obj_ver_id*)((uint8_t*)op->buf + i);
                         bufprintf(i == 0 ? " %lx:%lx v%lu" : ", %lx:%lx v%lu", ov->oid.inode, ov->oid.stripe, ov->version);
                     }
+                    if (op->req.sec_stab.len > sizeof(obj_ver_id)*12)
+                    {
+                        bufprintf(", ... (%lu items)", op->req.sec_stab.len/sizeof(obj_ver_id));
+                    }
                 }
                 else if (op->req.hdr.opcode == OSD_OP_SEC_LIST)
                 {

src/osd.h

@@ -93,6 +93,7 @@ class osd_t
 
     json11::Json::object cli_config, file_config, etcd_global_config, etcd_osd_config, config;
     int etcd_report_interval = 5;
+    int etcd_stats_interval = 30;
 
     bool readonly = false;
     osd_num_t osd_num = 1; // OSD numbers start with 1
@@ -122,7 +123,6 @@ class osd_t
     uint32_t scrub_list_limit = 1000;
     bool scrub_find_best = true;
     uint64_t scrub_ec_max_bruteforce = 100;
-    uint64_t fsync_feedback_repeat_interval = 500;
 
     // cluster state
 
@@ -167,8 +167,6 @@ class osd_t
     uint64_t unstable_write_count = 0;
     std::map<osd_object_id_t, uint64_t> unstable_writes;
     std::deque<osd_op_t*> syncs_in_progress;
-    std::map<int, timespec> unstable_write_osds;
-    int fsync_feedback_timer_id = -1;
 
     // client & peer I/O
 
@@ -260,7 +258,6 @@ class osd_t
     void exec_show_config(osd_op_t *cur_op);
     void exec_secondary(osd_op_t *cur_op);
     void secondary_op_callback(osd_op_t *cur_op);
-    void fsync_feedback();
 
     // primary ops
     void autosync();

src/osd_cluster.cpp

@@ -429,14 +429,18 @@ void osd_t::acquire_lease()
         create_osd_state();
     });
     printf(
-        "[OSD %lu] reporting to etcd at %s every %d seconds\n", this->osd_num,
+        "[OSD %lu] reporting to etcd at %s every %d seconds (statistics every %d seconds)\n", this->osd_num,
         (config["etcd_address"].is_string() ? config["etcd_address"].string_value() : config["etcd_address"].dump()).c_str(),
-        etcd_report_interval
+        etcd_report_interval, etcd_stats_interval
     );
     tfd->set_timer(etcd_report_interval*1000, true, [this](int timer_id)
     {
         renew_lease(false);
     });
+    tfd->set_timer(etcd_stats_interval*1000, true, [this](int timer_id)
+    {
+        report_statistics();
+    });
 }
 
 // Report "up" state once, then keep it alive using the lease
@@ -541,7 +545,6 @@ void osd_t::renew_lease(bool reload)
         else
         {
             etcd_failed_attempts = 0;
-            report_statistics();
             // Reload PGs
             if (reload && run_primary)
             {

src/osd_main.cpp

@@ -19,6 +19,14 @@ static void handle_sigint(int sig)
     exit(0);
 }
 
+static const char* help_text =
+    "Vitastor OSD (block object storage daemon) " VERSION "\n"
+    "(c) Vitaliy Filippov, 2019+ (VNPL-1.1)\n"
+    "\n"
+    "OSDs are usually started by vitastor-disk.\n"
+    "Manual usage: vitastor-osd [--option value] ...\n"
+;
+
 int main(int narg, char *args[])
 {
     setvbuf(stdout, NULL, _IONBF, 0);
@@ -37,10 +45,20 @@ int main(int narg, char *args[])
             char *opt = args[i]+2;
             config[std::string(opt)] = std::string(args[++i]);
         }
+        else if (!strcmp(args[i], "--help"))
+        {
+            printf("%s", help_text);
+            return 0;
+        }
+    }
+    if (!config.size())
+    {
+        printf("%s", help_text);
+        return 1;
     }
     signal(SIGINT, handle_sigint);
     signal(SIGTERM, handle_sigint);
-    ring_loop_t *ringloop = new ring_loop_t(512);
+    ring_loop_t *ringloop = new ring_loop_t(RINGLOOP_DEFAULT_SIZE);
     osd = new osd_t(config, ringloop);
     while (1)
     {

src/osd_rmw.cpp

@@ -239,8 +239,9 @@ static void* get_jerasure_decoding_matrix(osd_rmw_stripe_t *stripes, int pg_size
 {
     int edd = 0;
     int erased[pg_size];
+    // we should distinguish stripes which are not at all and missing stripes
     for (int i = 0; i < pg_size; i++)
-        erased[i] = (stripes[i].read_end == 0 || stripes[i].missing ? 1 : 0);
+        erased[i] = (stripes[i].read_end == 0 ? 2 : (stripes[i].missing ? 1 : 0));
     for (int i = 0; i < pg_minsize; i++)
         if (stripes[i].read_end != 0 && stripes[i].missing)
             edd++;
@@ -253,7 +254,7 @@ static void* get_jerasure_decoding_matrix(osd_rmw_stripe_t *stripes, int pg_size
 #ifdef WITH_ISAL
         int smrow = 0;
         uint8_t *submatrix = (uint8_t*)malloc_or_die(pg_minsize*pg_minsize*2);
-        for (int i = 0; i < pg_size; i++)
+        for (int i = 0; i < pg_size && smrow < pg_minsize; i++)
         {
             if (!erased[i])
             {
@@ -279,7 +280,7 @@ static void* get_jerasure_decoding_matrix(osd_rmw_stripe_t *stripes, int pg_size
         smrow = 0;
         for (int i = 0; i < pg_minsize; i++)
         {
-            if (erased[i])
+            if (erased[i] == 1)
             {
                 memcpy(submatrix + pg_minsize*smrow, submatrix + (pg_minsize+i)*pg_minsize, pg_minsize);
                 smrow++;

src/osd_rmw_test.cpp

@@ -29,6 +29,7 @@ void test15(bool second);
 void test16();
 void test_recover_22_d2();
 void test_ec43_error_bruteforce();
+void test_recover_53_d5();
 
 int main(int narg, char *args[])
 {
@@ -67,6 +68,8 @@ int main(int narg, char *args[])
     test_recover_22_d2();
     // Error bruteforce
     test_ec43_error_bruteforce();
+    // Test 19
+    test_recover_53_d5();
     // End
     printf("all ok\n");
     return 0;
@@ -1112,7 +1115,7 @@ void test_recover_22_d2()
 
 /***
 
-EC 4+2 error location bruteforce
+18. EC 4+2 error location bruteforce
 
 ***/
 
@@ -1178,3 +1181,66 @@ void test_ec43_error_bruteforce()
     free(write_buf);
     use_ec(7, 4, false);
 }
+
+/***
+
+19. EC 5+3 recover 5th data block but not 4th
+
+***/
+
+void test_recover_53_d5()
+{
+    const int bmp = 128*1024 / 4096 / 8;
+    use_ec(8, 5, true);
+    osd_num_t osd_set[8] = { 1, 2, 3, 0, 0, 6, 7, 8 };
+    osd_rmw_stripe_t stripes[8] = {};
+    unsigned bitmaps[8] = { 0 };
+    // Read 512+128K
+    split_stripes(5, 128*1024, 512*1024, 128*1024, stripes);
+    assert(stripes[0].req_start == 0 && stripes[0].req_end == 0);
+    assert(stripes[1].req_start == 0 && stripes[1].req_end == 0);
+    assert(stripes[2].req_start == 0 && stripes[2].req_end == 0);
+    assert(stripes[3].req_start == 0 && stripes[3].req_end == 0);
+    assert(stripes[4].req_start == 0 && stripes[4].req_end == 128*1024);
+    uint8_t *data_buf = (uint8_t*)malloc_or_die(128*1024*8);
+    for (int i = 0; i < 8; i++)
+    {
+        stripes[i].read_start = stripes[i].req_start;
+        stripes[i].read_end = stripes[i].req_end;
+        stripes[i].read_buf = data_buf + i*128*1024;
+        stripes[i].bmp_buf = bitmaps + i;
+    }
+    // Read using parity
+    assert(extend_missing_stripes(stripes, osd_set, 5, 8) == 0);
+    assert(stripes[0].read_start == 0 && stripes[0].read_end == 128*1024);
+    assert(stripes[1].read_start == 0 && stripes[1].read_end == 128*1024);
+    assert(stripes[2].read_start == 0 && stripes[2].read_end == 128*1024);
+    assert(stripes[3].read_start == 0 && stripes[3].read_end == 0);
+    assert(stripes[4].read_start == 0 && stripes[4].read_end == 128*1024);
+    assert(stripes[5].read_start == 0 && stripes[5].read_end == 128*1024);
+    assert(stripes[6].read_start == 0 && stripes[6].read_end == 128*1024);
+    assert(stripes[7].read_start == 0 && stripes[7].read_end == 0);
+    bitmaps[0] = 0xffffffff;
+    bitmaps[1] = 0xffffffff;
+    bitmaps[2] = 0xffffffff;
+    bitmaps[3] = 0;
+    bitmaps[4] = 0;
+    bitmaps[5] = 0xffffffff;
+    bitmaps[6] = 0x64646464;
+    bitmaps[7] = 0;
+    set_pattern(stripes[0].read_buf, 128*1024, 0x70a549add9a2280a);
+    set_pattern(stripes[1].read_buf, 128*1024, 0xa70a549add9a2280);
+    set_pattern(stripes[2].read_buf, 128*1024, 0x0a70a549add9a228);
+    set_pattern(stripes[3].read_buf, 128*1024, 0); // 0x80a70a549add9a22
+    set_pattern(stripes[4].read_buf, 128*1024, 0); // 0x280a70a549add9a2
+    set_pattern(stripes[5].read_buf, 128*1024, 0x7572c28f7a91eb22); // xor
+    set_pattern(stripes[6].read_buf, 128*1024, 0xb4542b32a560fe26); // 2nd EC chunk
+    set_pattern(stripes[7].read_buf, 128*1024, 0);
+    // Reconstruct
+    reconstruct_stripes_ec(stripes, 8, 5, bmp);
+    check_pattern(stripes[4].read_buf, 128*1024, 0x280a70a549add9a2);
+    assert(bitmaps[4] == 0xFFFFFFFF);
+    free(data_buf);
+    // Done
+    use_ec(8, 5, false);
+}

src/osd_secondary.cpp

@@ -29,23 +29,6 @@ void osd_t::secondary_op_callback(osd_op_t *op)
         if (op->bs_op->retval > 0)
             op->iov.push_back(op->buf, op->bs_op->retval);
     }
-    else if (op->req.hdr.opcode == OSD_OP_SEC_WRITE ||
-        op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE)
-    {
-#ifndef OSD_STUB
-        fsync_feedback();
-#endif
-        if (op->req.hdr.opcode == OSD_OP_SEC_WRITE)
-        {
-            auto & u = unstable_write_osds[op->peer_fd];
-            u = u;
-        }
-    }
-    else if (op->req.hdr.opcode == OSD_OP_SEC_SYNC)
-    {
-        // FIXME It would be more correct to track STABILIZE ops, not just reset on SYNC
-        unstable_write_osds.erase(op->peer_fd);
-    }
     else if (op->req.hdr.opcode == OSD_OP_SEC_LIST)
     {
         // allocated by blockstore
@@ -62,71 +45,6 @@ void osd_t::secondary_op_callback(osd_op_t *op)
     finish_op(op, retval);
 }
 
-void osd_t::fsync_feedback()
-{
-    if (!unstable_write_osds.size() || !bs->wants_fsync())
-    {
-        return;
-    }
-    bool postpone = false;
-    // Broadcast fsync feedback
-    timespec now;
-    clock_gettime(CLOCK_REALTIME, &now);
-    for (auto up_it = unstable_write_osds.begin(); up_it != unstable_write_osds.end(); )
-    {
-        auto & peer_fd = up_it->first;
-        auto & last_feedback = up_it->second;
-        if (msgr.clients.find(peer_fd) == msgr.clients.end() ||
-            !msgr.clients.at(peer_fd)->osd_num)
-        {
-            unstable_write_osds.erase(up_it++);
-            continue;
-        }
-        auto diff = (now.tv_sec-last_feedback.tv_sec)*1000 + (now.tv_nsec-last_feedback.tv_nsec)/1000000;
-        if (diff > fsync_feedback_repeat_interval)
-        {
-            last_feedback = now;
-            // Request fsync from the primary OSD
-            // Note: primary OSD should NOT divide syncs by clients or this logic will break
-            osd_op_t *fb_op = new osd_op_t();
-            fb_op->op_type = OSD_OP_OUT;
-            fb_op->req = (osd_any_op_t){
-                .sync = {
-                    .header = {
-                        .magic = SECONDARY_OSD_OP_MAGIC,
-                        .id = msgr.next_subop_id++,
-                        .opcode = OSD_OP_SYNC,
-                    },
-                },
-            };
-            fb_op->callback = [this](osd_op_t *op)
-            {
-                delete op;
-            };
-            fb_op->peer_fd = peer_fd;
-            msgr.outbox_push(fb_op);
-        }
-        else
-        {
-            postpone = true;
-        }
-        up_it++;
-    }
-    if (fsync_feedback_timer_id >= 0)
-    {
-        tfd->clear_timer(fsync_feedback_timer_id);
-        fsync_feedback_timer_id = -1;
-    }
-    if (postpone)
-    {
-        fsync_feedback_timer_id = tfd->set_timer(fsync_feedback_repeat_interval, false, [this](int timer_id)
-        {
-            fsync_feedback_timer_id = -1;
-            fsync_feedback();
-        });
-    }
-}
-
 void osd_t::exec_secondary(osd_op_t *cur_op)
 {
     if (cur_op->req.hdr.opcode == OSD_OP_SEC_READ_BMP)
@@ -240,7 +158,6 @@ void osd_t::exec_show_config(osd_op_t *cur_op)
     json11::Json req_json = cur_op->req.show_conf.json_len > 0
         ? json11::Json::parse(std::string((char *)cur_op->buf), json_err)
         : json11::Json();
-    msgr.clients.at(cur_op->peer_fd)->osd_num = req_json["osd_num"].uint64_value();
     // Expose sensitive configuration values so peers can check them
     json11::Json::object wire_config = json11::Json::object {
         { "osd_num", osd_num },

src/qemu_driver.c

@@ -388,6 +388,7 @@ static void vitastor_aio_set_fd_handler(void *vcli, int fd, int unused1, IOHandl
     );
 }
 
+#if defined VITASTOR_C_API_VERSION && VITASTOR_C_API_VERSION >= 2
 typedef struct str_array
 {
     const char **items;
@@ -424,6 +425,7 @@ static void strarray_free(str_array *a)
     a->items = NULL;
     a->len = a->alloc = 0;
 }
+#endif
 
 static int vitastor_file_open(BlockDriverState *bs, QDict *options, int flags, Error **errp)
 {
@@ -443,6 +445,7 @@ static int vitastor_file_open(BlockDriverState *bs, QDict *options, int flags, E
     client->rdma_gid_index = qdict_get_try_int(options, "rdma-gid-index", 0);
     client->rdma_mtu = qdict_get_try_int(options, "rdma-mtu", 0);
     client->ctx = bdrv_get_aio_context(bs);
+#if defined VITASTOR_C_API_VERSION && VITASTOR_C_API_VERSION >= 2
     str_array opt = {};
     strarray_push_kv(&opt, "config_path", qdict_get_try_str(options, "config-path"));
     strarray_push_kv(&opt, "etcd_address", qdict_get_try_str(options, "etcd-host"));
@@ -472,40 +475,15 @@ static int vitastor_file_open(BlockDriverState *bs, QDict *options, int flags, E
         // Writeback cache is unusable without io_uring because the client can't correctly flush on exit
         fprintf(stderr, "vitastor: failed to create io_uring: %s - I/O will be slower%s\n",
             strerror(errno), (flags & BDRV_O_NOCACHE ? "" : " and writeback cache will be disabled"));
+#endif
         client->uring_eventfd = -1;
-#if defined VITASTOR_C_API_VERSION && VITASTOR_C_API_VERSION >= 2
-        client->proxy = vitastor_c_create_qemu_uring(
-            vitastor_aio_set_fd_handler, client, client->config_path, client->etcd_host, client->etcd_prefix,
-            client->use_rdma, client->rdma_device, client->rdma_port_num, client->rdma_gid_index, client->rdma_mtu, 0
-        );
-        if (!client->proxy)
-        {
-            fprintf(stderr, "vitastor: failed to create io_uring: %s - I/O will be slower\n", strerror(errno));
-            client->uring_eventfd = -1;
-            client->proxy = vitastor_c_create_qemu(
-                vitastor_aio_set_fd_handler, client, client->config_path, client->etcd_host, client->etcd_prefix,
-                client->use_rdma, client->rdma_device, client->rdma_port_num, client->rdma_gid_index, client->rdma_mtu, 0
-            );
-        }
-        else
-        {
-            client->uring_eventfd = vitastor_c_uring_register_eventfd(client->proxy);
-            if (client->uring_eventfd < 0)
-            {
-                fprintf(stderr, "vitastor: failed to create io_uring eventfd: %s\n", strerror(errno));
-                error_setg(errp, "failed to create io_uring eventfd");
-                vitastor_close(bs);
-                return -1;
-            }
-            universal_aio_set_fd_handler(client->ctx, client->uring_eventfd, vitastor_uring_handler, NULL, client);
-        }
-#else
         client->proxy = vitastor_c_create_qemu(
             vitastor_aio_set_fd_handler, client, client->config_path, client->etcd_host, client->etcd_prefix,
             client->use_rdma, client->rdma_device, client->rdma_port_num, client->rdma_gid_index, client->rdma_mtu, 0
         );
-#endif
+#if defined VITASTOR_C_API_VERSION && VITASTOR_C_API_VERSION >= 2
     }
+#endif
     image = client->image = g_strdup(qdict_get_try_str(options, "image"));
     client->readonly = (flags & BDRV_O_RDWR) ? 1 : 0;
     // Get image metadata (size and readonly flag) or just wait until the client is ready

src/ringloop.cpp

@@ -17,7 +17,7 @@ ring_loop_t::ring_loop_t(int qd)
     {
         throw std::runtime_error(std::string("io_uring_queue_init: ") + strerror(-ret));
     }
-    free_ring_data_ptr = *ring.cq.kring_entries;
+    free_ring_data_ptr = *ring.sq.kring_entries;
     ring_datas = (struct ring_data_t*)calloc(free_ring_data_ptr, sizeof(ring_data_t));
     free_ring_data = (int*)malloc(sizeof(int) * free_ring_data_ptr);
     if (!ring_datas || !free_ring_data)

src/ringloop.h

@@ -15,6 +15,8 @@
 #include <functional>
 #include <vector>
 
+#define RINGLOOP_DEFAULT_SIZE 1024
+
 static inline void my_uring_prep_rw(int op, struct io_uring_sqe *sqe, int fd, const void *addr, unsigned len, off_t offset)
 {
     // Prepare a read/write operation without clearing user_data
@@ -139,11 +141,9 @@ public:
         if (free_ring_data_ptr == 0)
             return NULL;
         struct io_uring_sqe* sqe = io_uring_get_sqe(&ring);
-        if (sqe)
-        {
-            *sqe = { 0 };
-            io_uring_sqe_set_data(sqe, ring_datas + free_ring_data[--free_ring_data_ptr]);
-        }
+        assert(sqe);
+        *sqe = { 0 };
+        io_uring_sqe_set_data(sqe, ring_datas + free_ring_data[--free_ring_data_ptr]);
         return sqe;
     }
     inline void set_immediate(const std::function<void()> cb)

src/stub_uring_osd.cpp

@@ -30,7 +30,7 @@ void stub_exec_op(osd_messenger_t *msgr, osd_op_t *op);
 int main(int narg, char *args[])
 {
     ring_consumer_t looper;
-    ring_loop_t *ringloop = new ring_loop_t(512);
+    ring_loop_t *ringloop = new ring_loop_t(RINGLOOP_DEFAULT_SIZE);
     epoll_manager_t *epmgr = new epoll_manager_t(ringloop);
     osd_messenger_t *msgr = new osd_messenger_t();
     msgr->osd_num = 1351;