Constant disk activity from database

[jenda69]

Hello!
I didn't find anyone mentioning this and I want to ask...
I got myself Raspberry Pi 5 for christmas and I installed Immich v1.91.4.

I moved all volumes (pgdata, model-cache and of course library), to an external hard drive to save space on the sdcard and noticed that the disk never spins down even though all the library scanning and other jobs are finished.
After watching the drive more closely, there's actually activity every 2-3 seconds.

I used htop to see what's going on, I found that the activity I see are writes by postgreSQL.
I set log_statement = 'all' in {pgdata}/postgresql.conf and in docker logs immich_postgres I found this over and over again:

(...)
2023-12-26 11:26:25.745 UTC [45] 172.30.0.6(39054): LOG:  execute <unnamed>: SELECT pg_notify($1, $2)
2023-12-26 11:26:25.745 UTC [45] 172.30.0.6(39054): DETAIL:  parameters: $1 = 'socket.io#/', $2 = '{"type":2,"uid":"6aa222483d0868e1"}'
2023-12-26 11:26:26.500 UTC [48] 172.30.0.5(34152): LOG:  execute <unnamed>: SELECT pg_notify($1, $2)
2023-12-26 11:26:26.500 UTC [48] 172.30.0.5(34152): DETAIL:  parameters: $1 = 'socket.io#/', $2 = '{"type":2,"uid":"097f5a2b4b34a9cf"}'
2023-12-26 11:26:30.747 UTC [45] 172.30.0.6(39054): LOG:  execute <unnamed>: SELECT pg_notify($1, $2)
2023-12-26 11:26:30.747 UTC [45] 172.30.0.6(39054): DETAIL:  parameters: $1 = 'socket.io#/', $2 = '{"type":2,"uid":"6aa222483d0868e1"}'
2023-12-26 11:26:31.502 UTC [48] 172.30.0.5(34152): LOG:  execute <unnamed>: SELECT pg_notify($1, $2)
2023-12-26 11:26:31.502 UTC [48] 172.30.0.5(34152): DETAIL:  parameters: $1 = 'socket.io#/', $2 = '{"type":2,"uid":"097f5a2b4b34a9cf"}'
(...)

I don't really understand programming, node.js, postgreSQL nor docker, so this communication between immich_server and immich_microservices is where I failed to continue following the track...

I ultimately plan to move everything to a SSD so drive not spinning down won't be an issue (useless constant writes could be though!), but I'm curious.
What is this activity for? Is possible to reduce it?

Thanks!

[danieldietzler]

We use the database for websocket notifications, so this is what you're seeing. It should stop if there are no active clients (e.g. no open browser windows, not using the app, ...)

[jenda69]

How long does it take to stop? I closed all the browser windows about 30 minutes ago and force stopped phone app a couple of minutes ago in case it was running in the background. HDD still crunches every few seconds.

[danieldietzler]

Normally a few seconds lol

[danieldietzler]

Ah wait, I think there have been some recent changes and now the different services (microservices, server, ml) also communicate with this pg notify thing. (#5669)

[danieldietzler]

If there is no activity at all it shouldn't send any events though...?

[jenda69]

If I let the browser window open, there's load of other db activity like...
LOG: execute <unnamed>: SELECT DISTINCT "distinctAlias"."UserTokenEntity_id" AS "ids_UserTokenEntity_id" FROM (...)
But if I close everything and I'd expect it to just be idle, these pg_notifys from IP addresses of immich_server and immich_microservices remain appearing every couple of seconds.

[bo0tzz]

We use the socket.io library for notifications between the components of Immich and to connected clients over websockets. The backend connection between components runs via Postgres, and by default has a heartbeat message every 5 seconds (see https://socket.io/docs/v4/postgres-adapter/#options). That's what you're seeing here.

[jenda69]

Aha! I actually stumbled upon this webpage but didn't notice there's heartbeat. After some brief reading about postgreSQL, I guess there always will be writes because postgreSQL has to write everything to that thing called WAL.

[SolidBrave]

could that be changed without any issues to like every 60s?