Commit Graph
26 Commits
Author SHA1 Message Date
Vitaliy Filippov 84919a10a9 Extract msgr_iothread to separate .{cpp,h} 2026-05-18 01:17:56 +03:00
Vitaliy Filippov f78d7d4efc Do not capture client pointer in ping lambda 2026-05-11 16:30:01 +03:00
Vitaliy Filippov 2a5028d17f Use robin_hood_map in messenger 2026-03-30 02:06:10 +03:00
Vitaliy Filippov 07915c2881 Use surrogate peer IDs instead of FDs 2026-03-30 02:06:10 +03:00
Vitaliy Filippov 44c895dc30 Fix clients possibly not destroyed after being switched to RDMA
It could in theory lead to hung operations on RDMA client disconnections.

Workflow (probable):
* OSD switches an outbound connection to PEER_RDMA and removes the FD from epoll
* But a receive request in io_uring is still active
* Receive response is never handled, refs remains > 0
* The connection is dropped, but osd_client_t is not destroyed because refs > 0
* Outbound ops are not canceled, suboperations hang, primary operations hang too
2026-03-17 01:04:17 +03:00
Vitaliy Filippov 2ed6760447 Remove ODP support 2026-02-17 02:35:46 +03:00
Vitaliy Filippov aacfdf0dec Rename try_recv_rdma to init_recv_rdma 2025-12-07 21:06:36 +03:00
Vitaliy Filippov 65872f5d0e Remove RDMA context FD handlers during msgr destroy 2025-11-08 14:11:03 +03:00
Vitaliy Filippov 0be86a306d Remove old liburing version support as it's now included 2025-08-24 16:55:58 +03:00
Vitaliy Filippov 6022a61329 Decouple break_pg_locks from outbound OSD disconnections 2025-06-05 02:48:54 +03:00
Vitaliy Filippov cd028612c8 Use a separate osd_client_t::in_osd_num for inbound OSD connections 2025-06-05 02:09:41 +03:00
Vitaliy Filippov b2416afb28 Lock PGs on secondary OSDs to allow local reads and guarantee splitbrain prevention 2025-05-10 15:18:00 +03:00
Vitaliy Filippov 5692630005 Move check_sequencing indication into config response features subkey 2025-05-05 02:07:50 +03:00
Vitaliy Filippov 2ab0ae3bc9 Check operation sequencing and stop clients when it breaks 2025-05-02 17:01:50 +03:00
Vitaliy Filippov 9556eeae45 Implement io_uring zero-copy send support 2025-05-01 18:47:10 +03:00
Vitaliy Filippov 80e5f8ba76 Add missing WITH_RDMACM defines 2025-04-12 19:06:27 +03:00
Vitaliy Filippov 013f688ffe Run check_peer_config on RDMA-CM connections too 2025-04-02 01:32:28 +03:00
Vitaliy Filippov 3e5f0be52c Use separate port numbers for RDMA-CM 2025-04-01 16:16:03 +03:00
Vitaliy Filippov 8508e78288 Add an alternative RDMA implementation via RDMA-CM
Required for non-RoCE cards: iWARP and, possibly, Infiniband
2025-03-31 21:01:25 +03:00
Vitaliy Filippov f32dea02bf Support multiple RDMA networks 2025-03-31 21:01:25 +03:00
Vitaliy Filippov a103065d12 Support multiple OSD networks and separate OSD cluster network 2025-03-31 21:01:15 +03:00
Vitaliy Filippov ccc32b9e68 Use TCP on RDMA connection failure 2025-03-23 12:04:23 +03:00
Vitaliy Filippov 8b0389b4e8 Log RDMA ibv_modify_qp() errors 2025-03-22 15:58:13 +03:00
Vitaliy Filippov 7de38250ad Auto-select RDMA device based on osd_network 2024-11-16 18:38:57 +03:00
Vitaliy Filippov abbba6ade4 Support handling TCP I/O in simple separate io_uring-based I/O threads
Required mainly for clients, allows to scale parallel client I/O with TCP
from 100-150k iops to ~400k iops and from 2-3 GB/s to at least 7-8 GB/s
with 4 I/O threads, at the same time increasing Q=1 latency by 2x thread
switching delay, which is ~10 us when CPU powersaving is disabled and may
be as high as 200 us when it's enabled.
2024-07-04 13:29:20 +03:00
Vitaliy Filippov c79b38bd26 Move all sources to subdirs 2024-05-15 11:06:01 +03:00