karmazilla / nanopool

The attributes owner and currentLease are set to null after the Connecter has entered the AVAILABLE state whereby the gates are opened for the Connector to be re-leased.

The data race essentially means that we are at risk of returning null from Connector.getConnection.

This uncritical set is bad behavior since it might overwrite an OUTDATED or SHUTDOWN state.

Throwing a RuntimeException in a pre-release hook will prevent the hook from returning to the pool and risk an unrecoverable Connector leak.

Resize is signaled with an exception that is thrown from the FsmMixin.getConnection method, which is then re-executed. This will cause the pre-connect hooks to re-run. This is bad because it is reasonable to expect the pre-connect and post-connect hooks to have symmetrical invocation patterns.

The Connector.invalidate method is called during the pool shutdown procedure, and that in turn always closes the associated PooledConnection instance - this may be too early if the connection is leased out and therefor in use.

In this case, we should instead (probably) wait with closing the physical connection until the connection returns to the pool.

A pool that is resized and closed simultaneously might loose the reference to the new Connectors array. With unlucky timing this could cause connection leaks, although the risk is low since it is a very narrow window.

Create a ContentionHandler that resizes the pool by some predetermined factor, until it reaches a max.