When performing a backup or database migration, database connections are not drained

Issue Summary

When a backup is kicked off (whether using the backup client or DIY scripts), and when in-app database migration is performed, the system starts out by draining in-use database connections and waiting until the pool is idle.

A change made in Bitbucket Server/Data Center 4.14 introduced a wrapper around the DataSource which the draining code was not updated to handle. As a result, database connections are not drained and backup/migration are allowed to continue with some number of connections potentially still in use.

Steps to Reproduce

1. Put the system into maintenance mode
2. Trigger a backup

Expected Results

Database connections are drained.

Actual Results

Database connections are not drained.

The following warning appears in the logs:

2021-03-09 22:32:44,789 WARN  [hz.hazelcast.cached.thread-33]  c.a.s.i.db.DefaultDatabaseManager The DataSource for the current database does not implement ForcedDrainable. Existing connections will not be closed.

Workaround

There is no workaround for this issue; it requires an in-code fix.