postgres: ECONNRESET errors after a while

Hi @porsager !

Similar to #43, I’m also encountering similar ECONNRESET errors with a PostgreSQL service on Render.com, but with a very low number of connections:

Apr 27 10:22:53 PM  node:internal/process/promises:246
Apr 27 10:22:53 PM            triggerUncaughtException(err, true /* fromPromise */);
Apr 27 10:22:53 PM            ^
Apr 27 10:22:53 PM  
Apr 27 10:22:53 PM  Error: read ECONNRESET    at TCP.onStreamRead (node:internal/stream_base_commons:211:20) {
Apr 27 10:22:53 PM    errno: -104,
Apr 27 10:22:53 PM    code: 'ECONNRESET',
Apr 27 10:22:53 PM    syscall: 'read'
Apr 27 10:22:53 PM  }
Apr 27 10:22:54 PM  error Command failed with exit code 1.

It seems like after a certain amount of time, the connection with PostgreSQL goes into this state where the ECONNRESET errors start happening. Once it’s in that state, it fails all queries for a while (a few minutes?). But then it eventually “wakes up” again (I guess Postgres.js eventually figured it out and reconnected at that point).

I’m using postgres@2.0.0-beta.5 - I’m assuming this beta version is not super unstable like this…? I never had any issues with this with Heroku’s PostgreSQL service.

I’ve also asked for assistance from the Render.com support team over here: https://community.render.com/t/econnreset-with-node-js-and-postgresql/1498

About this issue

Original URL
State: closed
Created 3 years ago
Reactions: 1
Comments: 17 (16 by maintainers)

Most upvoted comments

Maybe I’ll wait for this feature to land in non-beta v2 and then we can close this.

karlhorky on Jul 30, 2021

Ok, some updates:

connect_timeout: 10 had no effect
idle_timeout: 2 has the service working again now 🤯

So I guess Render.com terminates idle PostgreSQL connections on the server side… I’ll ask about this on the community page.

Maybe a good thing to add to the Postgres.js docs too, in case this is a more common configuration…

karlhorky on Apr 28, 2021

I’ve just released beta.6 which simply sets keepalive from the client to 60 seconds (a postgres db connection is fairly high bandwidth already, so I feel 60 seconds is a good start). I’m curious if something as simple as this will do? Would you mind giving it a go?

If not we could look into doing like the go library, which appears to retry if this connection failure happens.

porsager on Jul 7, 2021

Hi Karl.

Yeah, this is probably a common scenario.

Could be included with the part you already contributed about lambdas etc. 🙂 https://github.com/porsager/postgres#idle-timeout

porsager on Apr 28, 2021