The reconnector task exits once it has restored every connection it was asked for. Before #127 it was spawned at most once per pool, so a connection lost after that first pass stayed missing for good. A pool of 5 on a long-lived server lost its connections one at a time to transient connect failures and ended at PoolExhausted, with nothing retrying. The test lets the reconnector finish its first job, then reports another lost connection the way release() does and expects the pool to refill. Removing the `running = false` reset makes it fail with PoolNotRestored. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>