On Wed, Jan 22, 2020 at 4:06 AM Nicola Contu <nicola.contu@xxxxxxxxx> wrote: > after a few months, we started having this issue again. > So we revert the work_mem parameter to 600MB instead of 2GB. > But the issue is still there. A query went to segmentation fault, the DB went to recovery mode and our app went to read only for a few minutes. Hi Nicola, Hmm, a segmentation fault sounds like a different problem. Can you please share the exact error messages from PostgreSQL and OS logs? > I understand we can increase max_connections so we can have many more segments. > > My question is : is there a way to understand the number of segments we reached? If you're on Linux, you can probably see them with "ls /dev/shm". > Currently we have 220 max_conn so as your formula is 64 + 2* 220 we have about 500 shared segments. > We would like to increase that number to 300 or 400 but would be great to understand if there is a way to make sure we will solve the issue as it requires a restart of the service. > > I know you were also talking about a redesign this part in PostgreSQL. Do you know if anything has changed in any of the newer versions after 11.5? It's possible that we should increase a couple of constants used the formula -- I'll look into that again. But first I'd like to see if we're even investigating the right problem here.
