Idle connections thrash when pool min > 0 (i.e. by default)

I was going to raise this against generic-pool-redux, but I’m not sure if it’s not perhaps an issue with a missing config option here in knex.

The default minimum pool size in knex is 2, so by default 2 connections are kept active at all times. At the same time, there is an idle timeout set to 30 seconds in generic-pool-redux, which cleans away idle connections.

The result is that by default connections are opened and closed every 30 seconds whenever a client is idle. This isn’t much of an issue, until you start scaling the number of clients talking to a database. Then the thrashing LOGN & QUIT requests activating and deactivating idle connections starts blocking real requests from being able to access the database.

Setting min:0 stops the thrashing, but I don’t think this behaviour is correct, especially not as the default behaviour for the library.

Perhaps there is or should be a way to stop idle connections from being cleaned away when there is a min > 0, or perhaps the default min should be 0?

[xxx]

I think this has become an issue again with the switch to generic-pool for pooling as of 0.12.0. We definitely see it with 0.12.9.

edit: We were able to resolve it by add refreshIdle: false to the config

[ErisDS]

Ok cool, so the default changed - that’s a good start. I also went and read the pool2 docs and they say this:

idleTimeout

An integer, in milliseconds, to specify how long a resource must be idle before it may be disposed of. Resources are periodically checked (see syncInterval below) for idleness, and idle resources are disposed of unless they would bring the pool below the configured min value.

This sounds like the logic in pool2 will result in much better behaviour. Thanks for the info!