Fail to get new connections

[yaroot]

This is a follow up for #110

Client would throw this error and not be able to connect through pgagroal

pg8000.exceptions.ProgrammingError: {'S': 'FATAL', 'V': 'FATAL', 'C': '53300', 'M': 'connection pool is full'}

[yaroot]

pgagroal.log.gz

Metrics are in UTC, logs are in UTC+9:00

[jesperpedersen]

Thanks for the trace !

However, it shows that it is maxes out on 70 connections so the error is correct if you used the configuration in #110. Could you do another run and keep an eye on /methics ?

Also, if there is a problem check the output from ps aux | grep pgagroal ? The pgagroal_active_connections should be in line with the actual connections...

Do you have a test case that I can use locally ?

[yaroot]

Sorry the problem is that I've only opened 60 connections (30 from each program) and after I ^ced the clients it stayed at 70.

I put the code in this repo

[jesperpedersen]

This is very helpful, thank you !

Although you are missing a conn.rollback() and a conn.close() I can replicate the issue you are seeing when doing the ^C testing.

[jesperpedersen]

I have pushed a fix to master and 0.9.x - if would be great if you could do a rerun of all your test cases.

The main issue is that the clients doesn't know about the cached connections since there is no prefill in play. I'll think some more about that scenario...

[yaroot]

I rerun the test on 5e01f52 for a few hours, nothing happened so far, still trying to trigger connection pool is full, usually the scala code would do the trick.

I've added .rollback() on exception (the code didn't even reach this branch since the last few commits), the connection will be closed by the with construction.

[jesperpedersen]

ok, sounds good. My tests with your clients are successful now.

[yaroot]

Closing as there's nothing more I could find, it's a pleasure to work with you.