Hi All,
I have a PostgreSQL 13 server that is currently crashing every few of days while spinning on a LWLock and timing out. The server has 88 cores (that includes hyper threading) with an average of 900 connections and during the crash there were around 130 active sessions (a DSS system with a mix load of queries and data loads/updates). We get a spike in system CPU time as the processes are spinning on the lock trying to get XidGen before crashing.
Is there any tunable that can be used or is there any performance improvements in version 14 that may alleviate the issue?
2022-08-01 14:31:32.388237-04 | active | LWLock | XidGen | 11
Aug 1 14:32:49 pgdb02 postgres[910]: [124-1] dbc1-PANIC: stuck spinlock detected at LWLockWaitListLock, lwlock.c:918
Aug 1 14:32:54 pgdb02 postgres[43475]: [8818-1] dbc1-PANIC: stuck spinlock detected at LWLockWaitListLock, lwlock.c:918
Aug 1 14:32:58 pgdb02 postgres[98902]: [125-1] dbc1-PANIC: stuck spinlock detected at LWLockWaitListLock, lwlock.c:918
Thanks,
Rui