22.08.2018 16:43, Raghavendra Rao J S V wrote:
We have a database cluster as "db1_data". Under this cluster we have two databases. one is db1 and other is qovr. I surprised to see as "checkpointer process" is consuming 8.73GB of memory(RSS value as 9158892). Why "checkpointer process" is consuming this much amount of memory and how to limit the usage of the "checkpointer process" memory.
RSS value is not reasonable to determine memory leaks because it takes into account shared segments (e.g. from shared buffer cache). As a long-lived process checkpointer process tries to flush and as a consequence to touch each buffer cell therefore its RSS approaches to local allocated memory plus shared_buffers.
If you want to know the real local memory consumption you may to use python utility *smem* to see unshared local memory size.
--
Regards,
Maksim Milyutin