Exception due to db connection problem about rapns

[ileitch]

The issue is in fact the exact same as #69. The connection hasn't been lost, it is just unable to create one due to the pool being full.

[tpoland]

Actually, it looks like the error message may be mixing terms here, or perhaps there are two separate exceptions. There is a connect_error with MySQL because the database authentication does not complete within MySQL's connect_timeout. Increasing the timeout from 5 to 10 seconds should help with this set global connect_timeout=10;.

[ileitch]

@tpoland I don't think that will help. The ActiveRecord connection pool is on a per-thread basis. rapns worker threads are long lived and therefore never release their claim on a database connection. Increasing the timeout will not help because no threads are going to release their connection.

This does however make me think that the solution to this is to simply have workers explicitly release connections like a Rails request lifecycle does.

[tpoland]

@ileitch I was thinking that this should prevent the ActiveRecord ConnectTimeout perhaps abandoning less connections in the pool. Maybe the pool change could be combined with restarts of the RAPNS worker threads (via monit) at regular intervals so that you don't end up with too massive a pool.

[ileitch]

Please try 2.0.5.rc1