On 2018-08-22 11:14:48 -0300, Avi Vallarapu wrote:
> Seams like you are with the default or fairly less checkpoint_* settings.
>
> You need to look at several patterns to understand the best values for
> these settings.
>
> For now, i can say 300 seconds for checkpoint_timeout could be fairly very
> less. Try increasing that with a reload.
>
> See this =>
> https://www.percona.com/blog/2018/06/15/tuning-postgresql-for-sysbench-tpcc/
>
> Please do note that delaying checkpointing could also increase the time for
> crash recovery/shutdown, etc.........
This seems to have not that much to do with the problem the OP has, no?
I suggest just using pmap -d $process_id_of_checkpointer
that'll likely show that it doesn't actually use much memory, but that
the RSS originates from the shared memory it has touched.
- Andres