Hi Hackers,
My company (NTT Comware) and NTT OSS Center did verification of
partitioned table on PG14dev, and we faced a problem that consumed
huge memory when we created a Foreign key constraint on a partitioned
table including 500 partitioning tables and inserted some data.
We investigated it a little to use the "pg_backend_memory_contextes"
command and realized a "CachedPlan" of backends increased dramatically.
See below:
Without FKs
==============================
CachedPlan 0kB
With FKs (the problem is here)
==============================
CachedPlan 710MB
Please find the attached file to reproduce the problem.
We know two things as following:
- Each backend uses the size of CachedPlan
- The more increasing partitioning tables, the more CachedPlan
consuming
If there are many backends, it consumes about the size of CachedPlan x
the number of backends. It may occur a shortage of memory and OOM killer.
We think the affected version are PG12 or later. I believe it would be
better to fix the problem. Any thoughts?
Regards,
Tatsuro Yamada