Re: Heavy LWLockTranche buffer_mapping in Postgres 9.6.10 server - Mailing list pgsql-admin

believe, Transparent Huge Pages is disabled and Huge Pages are enabled to the tune of shared_buffers.

HTH

On Sun, 2 Feb 2020 at 04:46, Justin Lu <justin.lu5432@gmail.com> wrote:

Experts,

We are seeing very heavy LWLockTranche buffer_mapping in db recently. 

There server had 24 core, 128GB of RAM, SSD data file system, on Unbuntu 16.04.6.

The shared_buffers was at 32GB. 1/4 of over RAM size. No issue on checkpoints (avg time 29 min apart).

After seeing the heavy wait, we added 64GB more RAM and increased shared_buffers to 48GB, effective_cache_size to 90GB. But it seems there is no impact on the buffer mapping waits at all. 

There doesn't seem to be big changes in the db load pattern, but since the heavy buffer mapping appeared, the cpu load jumped 3 - 4 times.

file system I/O times/timing has been increasing as the load on the db increased over the last few months. But not by any means jumped.

temp files generated by user sorting activities also have been increasing, and increase of the work_mem after adding RAM doesn't seem to help much on this.

Any suggestions on what to look for?

thanks,

Justin