Repository navigation
query temp table is routed to standby node. #154
Description
Activity
- changed the title
[-]query temp table is routed primary node.[/-][+]query temp table is routed to standby node.[/+]on Mar 12, 2026 Your analysis is correct.
With the current behavior, the cache is not invalidated even if the system catalog changes. As a workaround, you can either disable
enable_shared_relcacheor configurerelcache_expire.We will discuss within the development team whether this behavior can be improved.
Or you can set check_temp_table = trace to avoid the issue.
in addition， unlog table，function maybe have same issue.
That's a known limitation of the system catalog lookup cache system.
thanks for your reply.
set check_temp_table = trace or disable enable_shared_relcache is only available to temp table, but not available to unlog table.
configure relcache_expire is only effective to a certain extent.if you can improve that in future, that's great!
For temp table case, I found a bug. In your scenario iii, a shared relation cache for t1 (t1 is not a temp table") is created. And the query in vi refers to the shared relation cache and believes that t1 is not a temp table. I posted a patch to pgpool-hackes.
https://www.postgresql.org/message-id/20260320.115354.1808938375091443061.ishii%40postgresql.org- added 5 commits that reference this issue
on Mar 25, 2026
Hi,
I met an issue as below:
environment
pgpool 4.6.2 with three backend nodes, one primary and two standby(weight: 0:1:1).
reproduce steps
analyze
concern
so do we have any mechanism to avoid this? I know we can set expire time, but it won't take effect in a timely manner.
in addition， unlog table，function maybe have same issue.
thanks~