Bg: storage executes ingest, rocksdb triggers write stall, storage will always report leader election failure, query storage, it will report rpc timeout
reason：GraphStorageServiceHandler, RaftexService,RaftPart shares the cpu threads pool
The first step of RaftPart's heartbeat will write rocksdb, if now rocksdb write stall
, a host has a large number of part leaders, which will cause all cpu thread pools to be blocked in write rocksdb (assuming the write stall exceeds 60s，ex：ingest), at this time storage raft will fail, and storage query will also fail
I think raftPart should be created separately without affecting the call of rpc
HarrisChu
commented
1 year ago
Bg: storage executes ingest, rocksdb triggers write stall, storage will always report leader election failure, query storage, it will report rpc timeout reason：GraphStorageServiceHandler, RaftexService,RaftPart shares the cpu threads pool
The first step of RaftPart's heartbeat will write rocksdb, if now rocksdb write stall , a host has a large number of part leaders, which will cause all cpu thread pools to be blocked in write rocksdb (assuming the write stall exceeds 60s，ex：ingest), at this time storage raft will fail, and storage query will also fail I think raftPart should be created separately without affecting the call of rpc