Dump only actual part of the journal by default

In fact, w=8 looks better than w=32, so it may be changed in the future

README.md

@@ -16,7 +16,8 @@ breaking changes in the future. However, the following is implemented:
 
 - Basic part: highly-available block storage with symmetric clustering and no SPOF
 - Performance ;-D
-- Two redundancy schemes: Replication and XOR n+1 (simplest case of EC)
+- Multiple redundancy schemes: Replication, XOR n+1, Reed-Solomon erasure codes
+  based on jerasure library with any number of data and parity drives in a group
 - Configuration via simple JSON data structures in etcd
 - Automatic data distribution over OSDs, with support for:
   - Mathematical optimization for better uniformity and less data movement
@@ -39,8 +40,6 @@ breaking changes in the future. However, the following is implemented:
 - OSD creation tool (OSDs currently have to be created by hand)
 - Other administrative tools
 - Per-inode I/O and space usage statistics
-- jerasure EC support with any number of data and parity drives in a group
-- Parallel usage of multiple network interfaces
 - Proxmox and OpenNebula plugins
 - iSCSI proxy
 - Inode metadata storage in etcd
@@ -50,6 +49,7 @@ breaking changes in the future. However, the following is implemented:
 - Checksums
 - SSD+HDD optimizations, possibly including tiered storage and soft journal flushes
 - RDMA and NVDIMM support
+- Web GUI
 - Compression (possibly)
 - Read caching using system page cache (possibly)
 
@@ -353,6 +353,7 @@ and calculate disk offsets almost by hand. This will be fixed in near future.
 - Create global configuration in etcd: `etcdctl --endpoints=... put /vitastor/config/global '{"immediate_commit":"all"}'`
   (if all your drives have capacitors).
 - Create pool configuration in etcd: `etcdctl --endpoints=... put /vitastor/config/pools '{"1":{"name":"testpool","scheme":"replicated","pg_size":2,"pg_minsize":1,"pg_count":256,"failure_domain":"host"}}'`.
+  For jerasure pools the configuration should look like the following: `2:{"name":"ecpool","scheme":"jerasure","pg_size":4,"parity_chunks":2,"pg_minsize":2,"pg_count":256,"failure_domain":"host"}`.
 - Calculate offsets for your drives with `node /usr/lib/vitastor/mon/simple-offsets.js --device /dev/sdX`.
 - Make systemd units for your OSDs. Look at `/usr/lib/vitastor/mon/make-units.sh` for example.
   Notable configuration variables from the example:
@@ -398,10 +399,7 @@ and calculate disk offsets almost by hand. This will be fixed in near future.
 
 - Object deletion requests may currently lead to 'incomplete' objects if your OSDs crash during
   deletion because proper handling of object cleanup in a cluster should be "three-phase"
-  and it's currently not implemented. Inode removal tool currently can't handle unclean
+  and it's currently not implemented. Just to repeat the removal again in this case.
-  objects, so incomplete objects become undeletable. This will be fixed in near future
-  by allowing the inode removal tool to delete unclean objects. With this problem fixed
-  you'll be able just to repeat the removal again.
 
 ## Implementation Principles
 

debian/control

@@ -9,7 +9,7 @@ Rules-Requires-Root: no
 
 Package: vitastor
 Architecture: amd64
-Depends: ${shlibs:Depends}, ${misc:Depends}, fio (= ${dep:fio}), qemu (= ${dep:qemu}), nodejs (>= 10), node-sprintf-js, node-ws (>= 7), libjerasure2
+Depends: ${shlibs:Depends}, ${misc:Depends}, fio (= ${dep:fio}), qemu (= ${dep:qemu}), nodejs (>= 10), node-sprintf-js, node-ws (>= 7), libjerasure2, lp-solve
 Description: Vitastor, a fast software-defined clustered block storage
  Vitastor is a small, simple and fast clustered block storage (storage for VM drives),
  architecturally similar to Ceph which means strong consistency, primary-replication,

dump_journal.cpp

@@ -26,23 +26,32 @@ struct journal_dump_t
     uint64_t journal_offset;
     uint64_t journal_len;
     uint64_t journal_pos;
+    bool all;
+    bool started;
     int fd;
+    uint32_t crc32_last;
 
-    void dump_block(void *buf);
+    int dump_block(void *buf);
 };
 
 int main(int argc, char *argv[])
 {
-    if (argc < 5)
+    journal_dump_t self = { 0 };
+    int b = 1;
+    if (argc >= 2 && !strcmp(argv[1], "--all"))
     {
-        printf("USAGE: %s <journal_file> <journal_block_size> <offset> <size>\n", argv[0]);
+        self.all = true;
+        b = 2;
+    }
+    if (argc < b+4)
+    {
+        printf("USAGE: %s [--all] <journal_file> <journal_block_size> <offset> <size>\n", argv[0]);
         return 1;
     }
-    journal_dump_t self;
+    self.journal_device = argv[b];
-    self.journal_device = argv[1];
+    self.journal_block = strtoul(argv[b+1], NULL, 10);
-    self.journal_block = strtoul(argv[2], NULL, 10);
+    self.journal_offset = strtoull(argv[b+2], NULL, 10);
-    self.journal_offset = strtoull(argv[3], NULL, 10);
+    self.journal_len = strtoull(argv[b+3], NULL, 10);
-    self.journal_len = strtoull(argv[4], NULL, 10);
     if (self.journal_block < MEM_ALIGNMENT || (self.journal_block % MEM_ALIGNMENT) ||
         self.journal_block > 128*1024)
     {
@@ -57,30 +66,64 @@ int main(int argc, char *argv[])
     }
     void *data = memalign(MEM_ALIGNMENT, self.journal_block);
     self.journal_pos = 0;
-    while (self.journal_pos < self.journal_len)
+    if (self.all)
+    {
+        while (self.journal_pos < self.journal_len)
+        {
+            int r = pread(self.fd, data, self.journal_block, self.journal_offset+self.journal_pos);
+            assert(r == self.journal_block);
+            uint64_t s;
+            for (s = 0; s < self.journal_block; s += 8)
+            {
+                if (*((uint64_t*)(data+s)) != 0)
+                    break;
+            }
+            if (s == self.journal_block)
+            {
+                printf("offset %08lx: zeroes\n", self.journal_pos);
+                self.journal_pos += self.journal_block;
+            }
+            else if (((journal_entry*)data)->magic == JOURNAL_MAGIC)
+            {
+                printf("offset %08lx:\n", self.journal_pos);
+                self.dump_block(data);
+            }
+            else
+            {
+                printf("offset %08lx: no magic in the beginning, looks like random data (pattern=%lx)\n", self.journal_pos, *((uint64_t*)data));
+                self.journal_pos += self.journal_block;
+            }
+        }
+    }
+    else
     {
         int r = pread(self.fd, data, self.journal_block, self.journal_offset+self.journal_pos);
         assert(r == self.journal_block);
-        uint64_t s;
+        journal_entry *je = (journal_entry*)(data);
-        for (s = 0; s < self.journal_block; s += 8)
+        if (je->magic != JOURNAL_MAGIC || je->type != JE_START || je_crc32(je) != je->crc32)
         {
-            if (*((uint64_t*)(data+s)) != 0)
+            printf("offset %08lx: journal superblock is invalid\n", self.journal_pos);
-                break;
-        }
-        if (s == self.journal_block)
-        {
-            printf("offset %08lx: zeroes\n", self.journal_pos);
-            self.journal_pos += self.journal_block;
-        }
-        else if (((journal_entry*)data)->magic == JOURNAL_MAGIC)
-        {
-            printf("offset %08lx:\n", self.journal_pos);
-            self.dump_block(data);
         }
         else
         {
-            printf("offset %08lx: no magic in the beginning, looks like random data (pattern=%lx)\n", self.journal_pos, *((uint64_t*)data));
+            printf("offset %08lx:\n", self.journal_pos);
-            self.journal_pos += self.journal_block;
+            self.dump_block(data);
+            self.started = false;
+            self.journal_pos = je->start.journal_start;
+            while (1)
+            {
+                if (self.journal_pos >= self.journal_len)
+                    self.journal_pos = self.journal_block;
+                r = pread(self.fd, data, self.journal_block, self.journal_offset+self.journal_pos);
+                assert(r == self.journal_block);
+                printf("offset %08lx:\n", self.journal_pos);
+                r = self.dump_block(data);
+                if (r <= 0)
+                {
+                    printf("end of the journal\n");
+                    break;
+                }
+            }
         }
     }
     free(data);
@@ -88,7 +131,7 @@ int main(int argc, char *argv[])
     return 0;
 }
 
-void journal_dump_t::dump_block(void *buf)
+int journal_dump_t::dump_block(void *buf)
 {
     uint32_t pos = 0;
     journal_pos += journal_block;
@@ -97,12 +140,19 @@ void journal_dump_t::dump_block(void *buf)
     while (pos < journal_block)
     {
         journal_entry *je = (journal_entry*)(buf + pos);
-        if (je->magic != JOURNAL_MAGIC || je->type < JE_MIN || je->type > JE_MAX)
+        if (je->magic != JOURNAL_MAGIC || je->type < JE_MIN || je->type > JE_MAX ||
+            !all && started && je->crc32_prev != crc32_last)
         {
             break;
         }
-        const char *crc32_valid = je_crc32(je) == je->crc32 ? "(valid)" : "(invalid)";
+        bool crc32_valid = je_crc32(je) == je->crc32;
-        printf("entry % 3d: crc32=%08x %s prev=%08x ", entry, je->crc32, crc32_valid, je->crc32_prev);
+        if (!all && !crc32_valid)
+        {
+            break;
+        }
+        started = true;
+        crc32_last = je->crc32;
+        printf("entry % 3d: crc32=%08x %s prev=%08x ", entry, je->crc32, (crc32_valid ? "(valid)" : "(invalid)"), je->crc32_prev);
         if (je->type == JE_START)
         {
             printf("je_start start=%08lx\n", je->start.journal_start);
@@ -170,4 +220,5 @@ void journal_dump_t::dump_block(void *buf)
     {
         journal_pos = journal_len;
     }
+    return entry;
 }

epoll_manager.cpp

@@ -84,8 +84,12 @@ void epoll_manager_t::handle_epoll_events()
         nfds = epoll_wait(epoll_fd, events, MAX_EPOLL_EVENTS, 0);
         for (int i = 0; i < nfds; i++)
         {
-            auto & cb = epoll_handlers[events[i].data.fd];
+            auto cb_it = epoll_handlers.find(events[i].data.fd);
-            cb(events[i].data.fd, events[i].events);
+            if (cb_it != epoll_handlers.end())
+            {
+                auto & cb = cb_it->second;
+                cb(events[i].data.fd, events[i].events);
+            }
         }
     } while (nfds == MAX_EPOLL_EVENTS);
 }

messenger.cpp

@@ -348,7 +348,14 @@ void osd_messenger_t::stop_client(int peer_fd)
     }
     if (cl->read_op)
     {
-        delete cl->read_op;
+        if (cl->read_op->callback)
+        {
+            cancel_op(cl->read_op);
+        }
+        else
+        {
+            delete cl->read_op;
+        }
         cl->read_op = NULL;
     }
     for (auto rit = read_ready_clients.begin(); rit != read_ready_clients.end(); rit++)

osd_peering_pg.h

@@ -94,7 +94,7 @@ struct pg_t
     std::vector<osd_num_t> cur_set;
     // same thing in state_dict-like format
     pg_osd_set_t cur_loc_set;
-    // moved object map. by default, each object is considered to reside on the cur_set.
+    // moved object map. by default, each object is considered to reside on cur_set.
     // this map stores all objects that differ.
     // it may consume up to ~ (raw storage / object size) * 24 bytes in the worst case scenario
     // which is up to ~192 MB per 1 TB in the worst case scenario

osd_rmw.cpp

@@ -11,6 +11,8 @@
 #include "osd_rmw.h"
 #include "malloc_or_die.h"
 
+#define OSD_JERASURE_W 32
+
 static inline void extend_read(uint32_t start, uint32_t end, osd_rmw_stripe_t & stripe)
 {
     if (stripe.read_end == 0)
@@ -158,7 +160,7 @@ void use_jerasure(int pg_size, int pg_minsize, bool use)
         {
             return;
         }
-        int *matrix = reed_sol_vandermonde_coding_matrix(pg_minsize, pg_size-pg_minsize, 32);
+        int *matrix = reed_sol_vandermonde_coding_matrix(pg_minsize, pg_size-pg_minsize, OSD_JERASURE_W);
         matrices[key] = (reed_sol_matrix_t){
             .refs = 0,
             .data = matrix,
@@ -214,8 +216,8 @@ int* get_jerasure_decoding_matrix(osd_rmw_stripe_t *stripes, int pg_size, int pg
         int *decoding_matrix = dm_ids + pg_minsize;
         if (!dm_ids)
             throw std::bad_alloc();
-        // we always use row_k_ones=1 and w=32
+        // we always use row_k_ones=1 and w=8 (OSD_JERASURE_W)
-        if (jerasure_make_decoding_matrix(pg_minsize, pg_size-pg_minsize, 32, matrix->data, erased, decoding_matrix, dm_ids) < 0)
+        if (jerasure_make_decoding_matrix(pg_minsize, pg_size-pg_minsize, OSD_JERASURE_W, matrix->data, erased, decoding_matrix, dm_ids) < 0)
         {
             free(dm_ids);
             throw std::runtime_error("jerasure_make_decoding_matrix() failed");
@@ -252,7 +254,7 @@ void reconstruct_stripes_jerasure(osd_rmw_stripe_t *stripes, int pg_size, int pg
             }
             data_ptrs[role] = (char*)stripes[role].read_buf;
             jerasure_matrix_dotprod(
-                pg_minsize, 32, decoding_matrix+(role*pg_minsize), dm_ids, role,
+                pg_minsize, OSD_JERASURE_W, decoding_matrix+(role*pg_minsize), dm_ids, role,
                 data_ptrs, data_ptrs+pg_minsize, stripes[role].read_end - stripes[role].read_start
             );
         }
@@ -694,7 +696,7 @@ void calc_rmw_parity_jerasure(osd_rmw_stripe_t *stripes, int pg_size, int pg_min
                     }
                 }
                 jerasure_matrix_encode(
-                    pg_minsize, pg_size-pg_minsize, 32, matrix->data,
+                    pg_minsize, pg_size-pg_minsize, OSD_JERASURE_W, matrix->data,
                     (char**)data_ptrs, (char**)data_ptrs+pg_minsize, next_end-pos
                 );
                 pos = next_end;

rm_inode.cpp

@@ -6,26 +6,38 @@
  * May be included into a bigger "command-line management interface" in the future
  */
 
+#include <vector>
 #include <algorithm>
 
 #include "epoll_manager.h"
 #include "cluster_client.h"
 #include "pg_states.h"
 
-#define RM_NO_LIST 1
+#define RM_LISTING 1
-#define RM_LIST_SENT 2
+#define RM_REMOVING 2
-#define RM_REMOVING 3
+#define RM_END 3
-#define RM_END 4
 
 const char *exe_name = NULL;
 
+struct rm_pg_t;
+
 struct rm_pg_osd_t
 {
-    pg_num_t pg_num;
+    rm_pg_t *pg = NULL;
     osd_num_t osd_num;
+    bool sent = false;
+};
+
+struct rm_pg_t
+{
+    pg_num_t pg_num;
+    osd_num_t rm_osd_num;
+    std::vector<rm_pg_osd_t> list_osds;
     int state = 0;
-    obj_ver_id *obj_list = NULL;
+    int to_list;
-    uint64_t obj_count = 0, obj_pos = 0, obj_done = 0, obj_prev_done = 0;
+    std::set<object_id> objects;
+    std::set<object_id>::iterator obj_pos;
+    uint64_t obj_count = 0, obj_done = 0, obj_prev_done = 0;
     int in_flight = 0;
 };
 
@@ -41,11 +53,12 @@ protected:
     cluster_client_t *cli = NULL;
     ring_consumer_t consumer;
 
-    std::vector<rm_pg_osd_t*> lists;
+    std::vector<rm_pg_t*> lists;
     uint64_t total_count = 0, total_done = 0, total_prev_pct = 0;
     uint64_t pgs_to_list = 0;
     bool started = false;
     bool progress = true;
+    bool list_first = false;
     int log_level = 0;
 
 public:
@@ -62,7 +75,7 @@ public:
             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];
+                cfg[opt] = !strcmp(opt, "json") || !strcmp(opt, "wait-list") || i == narg-1 ? "1" : args[++i];
             }
         }
         return cfg;
@@ -74,7 +87,7 @@ public:
             "Vitastor inode removal tool\n"
             "(c) Vitaliy Filippov, 2020 (VNPL-1.0)\n\n"
             "USAGE:\n"
-            "  %s --etcd_address <etcd_address> --pool <pool> --inode <inode>\n",
+            "  %s --etcd_address <etcd_address> --pool <pool> --inode <inode> [--wait-list]\n",
             exe_name
         );
         exit(0);
@@ -105,6 +118,7 @@ public:
             parallel_osds = 4;
         log_level = cfg["log_level"].int64_value();
         progress = cfg["progress"].uint64_value() ? true : false;
+        list_first = cfg["wait-list"].uint64_value() ? true : false;
         // Create client
         ringloop = new ring_loop_t(512);
         epmgr = new epoll_manager_t(ringloop);
@@ -137,21 +151,57 @@ public:
         for (auto & pg_item: pool_cfg.pg_config)
         {
             auto & pg = pg_item.second;
-            if (pg.pause || !pg.cur_primary || pg.cur_state != PG_ACTIVE)
+            if (pg.pause || !pg.cur_primary || !(pg.cur_state & PG_ACTIVE))
             {
-                // FIXME Support deletion in non-clean active PGs by introducing a "primary-list" command
+                fprintf(stderr, "PG %u is inactive, skipping\n", pg_item.first);
-                fprintf(stderr, "PG %u is not active+clean, skipping\n", pg_item.first);
                 continue;
             }
-            rm_pg_osd_t *r = new rm_pg_osd_t();
+            rm_pg_t *r = new rm_pg_t();
             r->pg_num = pg_item.first;
-            r->osd_num = pg.cur_primary;
+            r->rm_osd_num = pg.cur_primary;
-            r->state = RM_NO_LIST;
+            r->state = RM_LISTING;
+            if (pg.cur_state != PG_ACTIVE)
+            {
+                std::set<osd_num_t> all_peers;
+                for (osd_num_t pg_osd: pg.target_set)
+                {
+                    if (pg_osd != 0)
+                    {
+                        all_peers.insert(pg_osd);
+                    }
+                }
+                for (osd_num_t pg_osd: pg.all_peers)
+                {
+                    if (pg_osd != 0)
+                    {
+                        all_peers.insert(pg_osd);
+                    }
+                }
+                for (auto & hist_item: pg.target_history)
+                {
+                    for (auto pg_osd: hist_item)
+                    {
+                        if (pg_osd != 0)
+                        {
+                            all_peers.insert(pg_osd);
+                        }
+                    }
+                }
+                for (osd_num_t peer_osd: all_peers)
+                {
+                    r->list_osds.push_back((rm_pg_osd_t){ .pg = r, .osd_num = peer_osd, .sent = false });
+                }
+            }
+            else
+            {
+                r->list_osds.push_back((rm_pg_osd_t){ .pg = r, .osd_num = pg.cur_primary, .sent = false });
+            }
+            r->to_list = r->list_osds.size();
             lists.push_back(r);
         }
-        std::sort(lists.begin(), lists.end(), [](rm_pg_osd_t *a, rm_pg_osd_t *b)
+        std::sort(lists.begin(), lists.end(), [](rm_pg_t *a, rm_pg_t *b)
         {
-            return a->osd_num < b->osd_num ? true : false;
+            return a->rm_osd_num < b->rm_osd_num ? true : false;
         });
         pgs_to_list = lists.size();
         started = true;
@@ -160,6 +210,10 @@ public:
 
     void send_list(rm_pg_osd_t *cur_list)
     {
+        if (cur_list->sent)
+        {
+            return;
+        }
         if (cli->msgr.osd_peer_fds.find(cur_list->osd_num) ==
             cli->msgr.osd_peer_fds.end())
         {
@@ -177,7 +231,7 @@ public:
                     .id = cli->msgr.next_subop_id++,
                     .opcode = OSD_OP_SEC_LIST,
                 },
-                .list_pg = cur_list->pg_num,
+                .list_pg = cur_list->pg->pg_num,
                 .pg_count = (pg_num_t)cli->st_cli.pool_config[pool_id].real_pg_count,
                 .pg_stripe_size = cli->st_cli.pool_config[pool_id].pg_stripe_size,
                 .min_inode = inode,
@@ -186,53 +240,67 @@ public:
         };
         op->callback = [this, cur_list](osd_op_t *op)
         {
-            pgs_to_list--;
+            cur_list->pg->to_list--;
             if (op->reply.hdr.retval < 0)
             {
-                fprintf(stderr, "Failed to get object list from OSD %lu (retval=%ld), skipping the PG\n",
+                fprintf(stderr, "Failed to get PG %u/%u object list from OSD %lu (retval=%ld), skipping\n",
-                    cur_list->osd_num, op->reply.hdr.retval);
+                    pool_id, cur_list->pg->pg_num, cur_list->osd_num, op->reply.hdr.retval);
-                cli->msgr.stop_client(cur_list->osd_num);
-                delete op;
-                cur_list->state = RM_END;
-                continue_delete();
-                return;
             }
-            if (log_level > 0)
+            else
             {
-                printf(
+                if (op->reply.sec_list.stable_count < op->reply.hdr.retval)
-                    "[PG %u/%u] Got inode object list from OSD %lu: %ld object versions\n",
+                {
-                    pool_id, cur_list->pg_num, cur_list->osd_num, op->reply.hdr.retval
+                    // Unstable objects, if present, mean that someone still writes into the inode. Warn the user about it.
-                );
+                    printf(
+                        "[PG %u/%u] Inode still has %lu unstable object versions - is it still open? Not a good idea to delete it.\n",
+                        pool_id, cur_list->pg->pg_num, op->reply.hdr.retval - op->reply.sec_list.stable_count
+                    );
+                }
+                if (log_level > 0)
+                {
+                    printf(
+                        "[PG %u/%u] Got inode object list from OSD %lu: %ld object versions\n",
+                        pool_id, cur_list->pg->pg_num, cur_list->osd_num, op->reply.hdr.retval
+                    );
+                }
+                for (uint64_t i = 0; i < op->reply.hdr.retval; i++)
+                {
+                    object_id oid = ((obj_ver_id*)op->buf)[i].oid;
+                    oid.stripe = oid.stripe & ~STRIPE_MASK;
+                    cur_list->pg->objects.insert(oid);
+                }
             }
-            cur_list->obj_list = (obj_ver_id*)op->buf;
-            cur_list->obj_count = (uint64_t)op->reply.hdr.retval;
-            cur_list->obj_done = cur_list->obj_prev_done = cur_list->obj_pos = 0;
-            total_count += cur_list->obj_count;
-            total_prev_pct = 0;
-            // set op->buf to NULL so it doesn't get freed
-            op->buf = NULL;
             delete op;
-            cur_list->state = RM_REMOVING;
+            if (cur_list->pg->to_list <= 0)
+            {
+                cur_list->pg->obj_done = cur_list->pg->obj_prev_done = 0;
+                cur_list->pg->obj_pos = cur_list->pg->objects.begin();
+                cur_list->pg->obj_count = cur_list->pg->objects.size();
+                total_count += cur_list->pg->obj_count;
+                total_prev_pct = 0;
+                cur_list->pg->state = RM_REMOVING;
+                pgs_to_list--;
+            }
             continue_delete();
         };
-        cur_list->state = RM_LIST_SENT;
         cli->msgr.outbox_push(op);
+        cur_list->sent = true;
     }
 
-    void send_ops(rm_pg_osd_t *cur_list)
+    void send_ops(rm_pg_t *cur_list)
     {
-        if (cli->msgr.osd_peer_fds.find(cur_list->osd_num) ==
+        if (cli->msgr.osd_peer_fds.find(cur_list->rm_osd_num) ==
             cli->msgr.osd_peer_fds.end())
         {
             // Initiate connection
-            cli->msgr.connect_peer(cur_list->osd_num, cli->st_cli.peer_states[cur_list->osd_num]);
+            cli->msgr.connect_peer(cur_list->rm_osd_num, cli->st_cli.peer_states[cur_list->rm_osd_num]);
             return;
         }
-        while (cur_list->in_flight < iodepth && cur_list->obj_pos < cur_list->obj_count)
+        while (cur_list->in_flight < iodepth && cur_list->obj_pos != cur_list->objects.end())
         {
             osd_op_t *op = new osd_op_t();
             op->op_type = OSD_OP_OUT;
-            op->peer_fd = cli->msgr.osd_peer_fds[cur_list->osd_num];
+            op->peer_fd = cli->msgr.osd_peer_fds[cur_list->rm_osd_num];
             op->req = (osd_any_op_t){
                 .rw = {
                     .header = {
@@ -240,8 +308,8 @@ public:
                         .id = cli->msgr.next_subop_id++,
                         .opcode = OSD_OP_DELETE,
                     },
-                    .inode = cur_list->obj_list[cur_list->obj_pos].oid.inode,
+                    .inode = cur_list->obj_pos->inode,
-                    .offset = (cur_list->obj_list[cur_list->obj_pos].oid.stripe & ~STRIPE_MASK),
+                    .offset = (cur_list->obj_pos->stripe & ~STRIPE_MASK),
                     .len = 0,
                 },
             };
@@ -251,7 +319,7 @@ public:
                 if (op->reply.hdr.retval < 0)
                 {
                     fprintf(stderr, "Failed to remove object from PG %u (OSD %lu) (retval=%ld)\n",
-                        cur_list->pg_num, cur_list->osd_num, op->reply.hdr.retval);
+                        cur_list->pg_num, cur_list->rm_osd_num, op->reply.hdr.retval);
                 }
                 delete op;
                 cur_list->obj_done++;
@@ -262,12 +330,10 @@ public:
             cur_list->obj_pos++;
             cur_list->in_flight++;
         }
-        if (!cur_list->in_flight && cur_list->obj_pos >= cur_list->obj_count)
+        if (!cur_list->in_flight && cur_list->obj_pos == cur_list->objects.end())
         {
-            free(cur_list->obj_list);
-            cur_list->obj_list = NULL;
             cur_list->obj_count = 0;
-            cur_list->obj_done = cur_list->obj_prev_done = cur_list->obj_pos = 0;
+            cur_list->obj_done = cur_list->obj_prev_done = 0;
             cur_list->state = RM_END;
         }
     }
@@ -276,6 +342,22 @@ public:
     {
         int par_osd = 0;
         osd_num_t max_seen_osd = 0;
+        bool no_del = false;
+        if (list_first)
+        {
+            int i, n = 0;
+            for (i = 0; i < lists.size(); i++)
+            {
+                if (lists[i]->state == RM_LISTING)
+                {
+                    n++;
+                }
+            }
+            if (n > 0)
+            {
+                no_del = true;
+            }
+        }
         for (int i = 0; i < lists.size(); i++)
         {
             if (lists[i]->state == RM_END)
@@ -284,18 +366,25 @@ public:
                 lists.erase(lists.begin()+i, lists.begin()+i+1);
                 i--;
             }
-            else if (lists[i]->osd_num > max_seen_osd)
+            else if (lists[i]->rm_osd_num > max_seen_osd)
             {
-                if (lists[i]->state == RM_NO_LIST)
+                if (lists[i]->state == RM_LISTING)
                 {
-                    send_list(lists[i]);
+                    for (int j = 0; j < lists[i]->list_osds.size(); j++)
+                    {
+                        send_list(&lists[i]->list_osds[j]);
+                    }
                 }
                 else if (lists[i]->state == RM_REMOVING)
                 {
+                    if (no_del)
+                    {
+                        continue;
+                    }
                     send_ops(lists[i]);
                 }
                 par_osd++;
-                max_seen_osd = lists[i]->osd_num;
+                max_seen_osd = lists[i]->rm_osd_num;
                 if (par_osd >= parallel_osds)
                 {
                     break;

rpm/qemu-kvm-el7.spec.patch

@@ -0,0 +1,257 @@
+--- qemu-kvm.spec.orig	2020-11-09 23:41:03.000000000 +0000
++++ qemu-kvm.spec	2020-12-06 10:44:24.207640963 +0000
+@@ -2,7 +2,7 @@
+ %global SLOF_gittagcommit 899d9883
+ 
+ %global have_usbredir 1
+-%global have_spice    1
++%global have_spice    0
+ %global have_opengl   1
+ %global have_fdt      0
+ %global have_gluster  1
+@@ -56,7 +56,7 @@ Requires: %{name}-block-curl = %{epoch}:
+ Requires: %{name}-block-gluster = %{epoch}:%{version}-%{release} \
+ %endif                                                           \
+ Requires: %{name}-block-iscsi = %{epoch}:%{version}-%{release}   \
+-Requires: %{name}-block-rbd = %{epoch}:%{version}-%{release}     \
++#Requires: %{name}-block-rbd = %{epoch}:%{version}-%{release}     \
+ Requires: %{name}-block-ssh = %{epoch}:%{version}-%{release}
+ 
+ # Macro to properly setup RHEL/RHEV conflict handling
+@@ -67,7 +67,7 @@ Obsoletes: %1-rhev
+ Summary: QEMU is a machine emulator and virtualizer
+ Name: qemu-kvm
+ Version: 4.2.0
+-Release: 29.vitastor%{?dist}.6
++Release: 30.vitastor%{?dist}.6
+ # Epoch because we pushed a qemu-1.0 package. AIUI this can't ever be dropped
+ Epoch: 15
+ License: GPLv2 and GPLv2+ and CC-BY
+@@ -99,8 +99,8 @@ Source30: kvm-s390x.conf
+ Source31: kvm-x86.conf
+ Source32: qemu-pr-helper.service
+ Source33: qemu-pr-helper.socket
+-Source34: 81-kvm-rhel.rules
+-Source35: udev-kvm-check.c
++#Source34: 81-kvm-rhel.rules
++#Source35: udev-kvm-check.c
+ Source36: README.tests
+ 
+ 
+@@ -825,7 +825,9 @@ Patch331: kvm-Drop-bogus-IPv6-messages.p
+ Patch333: kvm-virtiofsd-Whitelist-fchmod.patch
+ # For bz#1883869 - virtiofsd core dump in KATA Container [rhel-8.2.1.z]
+ Patch334: kvm-virtiofsd-avoid-proc-self-fd-tempdir.patch
+-Patch335: qemu-4.2-vitastor.patch
++Patch335: qemu-use-sphinx-1.2.patch
++Patch336: qemu-config-tcmalloc-warning.patch
++Patch337: qemu-4.2-vitastor.patch
+ 
+ BuildRequires: wget
+ BuildRequires: rpm-build
+@@ -842,7 +844,8 @@ BuildRequires: pciutils-devel
+ BuildRequires: libiscsi-devel
+ BuildRequires: ncurses-devel
+ BuildRequires: libattr-devel
+-BuildRequires: libusbx-devel >= 1.0.22
++BuildRequires: gperftools-devel
++BuildRequires: libusbx-devel >= 1.0.21
+ %if %{have_usbredir}
+ BuildRequires: usbredir-devel >= 0.7.1
+ %endif
+@@ -856,12 +859,12 @@ BuildRequires: virglrenderer-devel
+ # For smartcard NSS support
+ BuildRequires: nss-devel
+ %endif
+-BuildRequires: libseccomp-devel >= 2.4.0
++#Requires: libseccomp >= 2.4.0
+ # For network block driver
+ BuildRequires: libcurl-devel
+ BuildRequires: libssh-devel
+-BuildRequires: librados-devel
+-BuildRequires: librbd-devel
++#BuildRequires: librados-devel
++#BuildRequires: librbd-devel
+ %if %{have_gluster}
+ # For gluster block driver
+ BuildRequires: glusterfs-api-devel
+@@ -955,25 +958,25 @@ hardware for a full system such as a PC
+ 
+ %package -n qemu-kvm-core
+ Summary: qemu-kvm core components
++Requires: gperftools-libs
+ Requires: qemu-img = %{epoch}:%{version}-%{release}
+ %ifarch %{ix86} x86_64
+ Requires: seabios-bin >= 1.10.2-1
+ Requires: sgabios-bin
+-Requires: edk2-ovmf
+ %endif
+ %ifarch aarch64
+ Requires: edk2-aarch64
+ %endif
+ 
+ %ifnarch aarch64 s390x
+-Requires: seavgabios-bin >= 1.12.0-3
+-Requires: ipxe-roms-qemu >= 20170123-1
++Requires: seavgabios-bin >= 1.11.0-1
++Requires: ipxe-roms-qemu >= 20181214-1
++Requires: /usr/share/ipxe.efi
+ %endif
+ %ifarch %{power64}
+ Requires: SLOF >= %{SLOF_gittagdate}-1.git%{SLOF_gittagcommit}
+ %endif
+ Requires: %{name}-common = %{epoch}:%{version}-%{release}
+-Requires: libseccomp >= 2.4.0
+ # For compressed guest memory dumps
+ Requires: lzo snappy
+ %if %{have_kvm_setup}
+@@ -1085,15 +1088,15 @@ This package provides the additional iSC
+ Install this package if you want to access iSCSI volumes.
+ 
+ 
+-%package  block-rbd
+-Summary: QEMU Ceph/RBD block driver
+-Requires: %{name}-common%{?_isa} = %{epoch}:%{version}-%{release}
+-
+-%description block-rbd
+-This package provides the additional Ceph/RBD block driver for QEMU.
+-
+-Install this package if you want to access remote Ceph volumes
+-using the rbd protocol.
++#%package  block-rbd
++#Summary: QEMU Ceph/RBD block driver
++#Requires: %{name}-common%{?_isa} = %{epoch}:%{version}-%{release}
++#
++#%description block-rbd
++#This package provides the additional Ceph/RBD block driver for QEMU.
++#
++#Install this package if you want to access remote Ceph volumes
++#using the rbd protocol.
+ 
+ 
+ %package  block-ssh
+@@ -1117,12 +1120,14 @@ the Secure Shell (SSH) protocol.
+ # --build-id option is used for giving info to the debug packages.
+ buildldflags="VL_LDFLAGS=-Wl,--build-id"
+ 
+-%global block_drivers_list qcow2,raw,file,host_device,nbd,iscsi,rbd,blkdebug,luks,null-co,nvme,copy-on-read,throttle
++#%global block_drivers_list qcow2,raw,file,host_device,nbd,iscsi,rbd,blkdebug,luks,null-co,nvme,copy-on-read,throttle
++%global block_drivers_list qcow2,raw,file,host_device,nbd,iscsi,blkdebug,luks,null-co,nvme,copy-on-read,throttle
+ 
+ %if 0%{have_gluster}
+     %global block_drivers_list %{block_drivers_list},gluster
+ %endif
+ 
++[ -e /usr/bin/sphinx-build ] || ln -s sphinx-build-3 /usr/bin/sphinx-build
+ ./configure  \
+  --prefix="%{_prefix}" \
+  --libdir="%{_libdir}" \
+@@ -1152,15 +1157,15 @@ buildldflags="VL_LDFLAGS=-Wl,--build-id"
+ %else
+   --disable-numa \
+ %endif
+-  --enable-rbd \
++  --disable-rbd \
+ %if 0%{have_librdma}
+   --enable-rdma \
+ %else
+   --disable-rdma \
+ %endif
+   --disable-pvrdma \
+-  --enable-seccomp \
+-%if 0%{have_spice}
++  --disable-seccomp \
++%if %{have_spice}
+   --enable-spice \
+   --enable-smartcard \
+   --enable-virglrenderer \
+@@ -1179,7 +1184,7 @@ buildldflags="VL_LDFLAGS=-Wl,--build-id"
+ %else
+   --disable-usb-redir \
+ %endif
+-  --disable-tcmalloc \
++  --enable-tcmalloc \
+ %ifarch x86_64
+   --enable-libpmem \
+ %else
+@@ -1193,9 +1198,7 @@ buildldflags="VL_LDFLAGS=-Wl,--build-id"
+ %endif
+   --python=%{__python3} \
+   --target-list="%{buildarch}" \
+-  --block-drv-rw-whitelist=%{block_drivers_list} \
+   --audio-drv-list= \
+-  --block-drv-ro-whitelist=vmdk,vhdx,vpc,https,ssh \
+   --with-coroutine=ucontext \
+   --tls-priority=NORMAL \
+   --disable-bluez \
+@@ -1262,7 +1265,7 @@ buildldflags="VL_LDFLAGS=-Wl,--build-id"
+   --disable-sanitizers \
+   --disable-hvf \
+   --disable-whpx \
+-  --enable-malloc-trim \
++  --disable-malloc-trim \
+   --disable-membarrier \
+   --disable-vhost-crypto \
+   --disable-libxml2 \
+@@ -1308,7 +1311,7 @@ make V=1 %{?_smp_mflags} $buildldflags
+ cp -a %{kvm_target}-softmmu/qemu-system-%{kvm_target} qemu-kvm
+ 
+ gcc %{SOURCE6} $RPM_OPT_FLAGS $RPM_LD_FLAGS -o ksmctl
+-gcc %{SOURCE35} $RPM_OPT_FLAGS $RPM_LD_FLAGS -o udev-kvm-check
++#gcc %{SOURCE35} $RPM_OPT_FLAGS $RPM_LD_FLAGS -o udev-kvm-check
+ 
+ %install
+ %define _udevdir %(pkg-config --variable=udevdir udev)
+@@ -1343,8 +1346,8 @@ mkdir -p $RPM_BUILD_ROOT%{testsdir}/test
+ mkdir -p $RPM_BUILD_ROOT%{testsdir}/tests/qemu-iotests
+ mkdir -p $RPM_BUILD_ROOT%{testsdir}/scripts/qmp
+ 
+-install -p -m 0755 udev-kvm-check $RPM_BUILD_ROOT%{_udevdir}
+-install -p -m 0644 %{SOURCE34} $RPM_BUILD_ROOT%{_udevrulesdir}
++#install -p -m 0755 udev-kvm-check $RPM_BUILD_ROOT%{_udevdir}
++#install -p -m 0644 %{SOURCE34} $RPM_BUILD_ROOT%{_udevrulesdir}
+ 
+ install -m 0644 scripts/dump-guest-memory.py \
+                 $RPM_BUILD_ROOT%{_datadir}/%{name}
+@@ -1562,6 +1565,8 @@ rm -rf $RPM_BUILD_ROOT%{qemudocdir}/inte
+ # Remove spec
+ rm -rf $RPM_BUILD_ROOT%{qemudocdir}/specs
+ 
++%global __os_install_post %(echo '%{__os_install_post}' | sed -e 's!/usr/lib[^[:space:]]*/brp-python-bytecompile[[:space:]].*$!!g')
++
+ %check
+ export DIFF=diff; make check V=1
+ 
+@@ -1645,8 +1650,8 @@ useradd -r -u 107 -g qemu -G kvm -d / -s
+ %config(noreplace) %{_sysconfdir}/sysconfig/ksm
+ %{_unitdir}/ksmtuned.service
+ %{_sbindir}/ksmtuned
+-%{_udevdir}/udev-kvm-check
+-%{_udevrulesdir}/81-kvm-rhel.rules
++#%{_udevdir}/udev-kvm-check
++#%{_udevrulesdir}/81-kvm-rhel.rules
+ %ghost %{_sysconfdir}/kvm
+ %config(noreplace) %{_sysconfdir}/ksmtuned.conf
+ %dir %{_sysconfdir}/%{name}
+@@ -1711,8 +1716,8 @@ useradd -r -u 107 -g qemu -G kvm -d / -s
+ %{_libexecdir}/vhost-user-gpu
+ %{_datadir}/%{name}/vhost-user/50-qemu-gpu.json
+ %endif
+-%{_libexecdir}/virtiofsd
+-%{_datadir}/%{name}/vhost-user/50-qemu-virtiofsd.json
++#%{_libexecdir}/virtiofsd
++#%{_datadir}/%{name}/vhost-user/50-qemu-virtiofsd.json
+ 
+ %files -n qemu-img
+ %defattr(-,root,root)
+@@ -1748,8 +1753,8 @@ useradd -r -u 107 -g qemu -G kvm -d / -s
+ %files block-iscsi
+ %{_libdir}/qemu-kvm/block-iscsi.so
+ 
+-%files block-rbd
+-%{_libdir}/qemu-kvm/block-rbd.so
++#%files block-rbd
++#%{_libdir}/qemu-kvm/block-rbd.so
+ 
+ %files block-ssh
+ %{_libdir}/qemu-kvm/block-ssh.so

rpm/vitastor-el7.spec

@@ -20,6 +20,7 @@ Requires:       rh-nodejs12
 Requires:       rh-nodejs12-npm
 Requires:       liburing >= 0.6
 Requires:       libJerasure2
+Requires:       lpsolve
 
 %description
 Vitastor is a small, simple and fast clustered block storage (storage for VM drives),

rpm/vitastor-el8.spec

@@ -18,6 +18,7 @@ Requires:       qemu-kvm = 4.2.0-29.el8.6
 Requires:       nodejs >= 10
 Requires:       liburing >= 0.6
 Requires:       libJerasure2
+Requires:       lpsolve
 
 %description
 Vitastor is a small, simple and fast clustered block storage (storage for VM drives),