Summary

• Adds maxUseTime configuration to Pool that enables detection of leaked connections (connections checked out longer than the threshold)
• When the pool is exhausted and stale connections are detected, their slots are recovered allowing new connections to be created
• Tracks checkout timestamps on Connection objects via markCheckedOut()/getCheckedOutAt()
• Feature is opt-in (disabled by default with maxUseTime=0) for full backwards compatibility

Problem

The Sentry error "Pool 'console' is empty (size 75)" (CLOUD-3K4N, 3267 events) occurs when all pool connections are checked out and none are returned. This happens when connections are leaked — popped but never reclaimed due to consumer code errors, timeouts, or edge cases. Without recovery, leaked connections permanently reduce pool capacity until full exhaustion.

How it works

1. Each connection records a timestamp when checked out (markCheckedOut())
2. When pop() fails to get a connection and maxUseTime > 0, it scans active connections for any exceeding the threshold
3. Stale connections are removed from tracking and their slots freed (connectionsCreated decremented)
4. The retry loop then creates a fresh connection in the freed slot

Test plan

• testPoolRecoversStaleConnections — verifies stale connections are detected and new connections created in their place
• testPoolDoesNotRecoverNonStaleConnections — verifies non-stale connections are not evicted
• testPoolStaleRecoveryDisabledByDefault — verifies the feature is off when maxUseTime=0
• All 90 existing tests pass
• Lint (pint --preset psr12) passes
• Static analysis (phpstan) passes

🤖 Generated with Claude Code