Fix db.client.connection.count metric drifting over time

[verdie-g]

Problem

The db.client.connection.count UpDownCounter metric drifts in long-running applications, producing nonsensical values like +5M idle / -5M used connections.

Root Cause

Three bugs in the sync connection pool metric recording:

Bug 1 — Phantom IDLE transitions

ConnectionPool.make_connection() and BlockingConnectionPool.make_connection() recorded IDLE +1 for new connections, then get_connection() unconditionally recorded IDLE -1. While balanced in isolation, this creates unnecessary IDLE transitions that drift if any step is disrupted (e.g. OTel collector not yet initialized). The async pool already avoids this with an is_created check.

Bug 2 — Wrong pool name on unowned connections

release() recorded USED -1 against the literal string "unknown_pool" when owns_connection() returned False (e.g. after a fork). The real pool's USED counter was never decremented.

Bug 3 — Full queue silent drop

BlockingConnectionPool.release() swallowed both the USED -1 and IDLE +1 recordings when put_nowait() raised Full. The USED counter leaked +1 permanently and the dropped connection was never disconnected.

Fix

Bug	Fix
Phantom IDLE	Remove IDLE +1 from make_connection(), add is_created check in get_connection() — new connections record only USED +1, reused connections record IDLE -1, USED +1. Matches the async pool.
Wrong pool name	Use get_pool_name(self) instead of "unknown_pool"
Full queue drop	In except Full, disconnect the connection and record USED -1

Tests

9 new tests in tests/test_observability/test_connection_count_bugs.py covering all three bugs across both ConnectionPool and BlockingConnectionPool.

---

Note

Medium Risk
Changes only observability counter recording in hot pool paths (get_connection/release), but incorrect metrics previously misled production monitoring; behavior of Redis I/O is unchanged.

Overview
Fixes drift in the db.client.connection.count observability UpDownCounter for sync ConnectionPool and BlockingConnectionPool.

Metric accounting: Stops recording IDLE +1 when make_connection() creates a socket. On get_connection(), new connections now emit only USED +1; reused pool connections still record IDLE -1 and USED +1 (aligned with the async pools).

Release edge cases: When the pool does not own a connection (e.g. after fork), USED -1 is attributed with get_pool_name(self) instead of the hard-coded "unknown_pool". For BlockingConnectionPool, if put_nowait() raises Full, the connection is disconnected, removed from _connections, and USED -1 is recorded so counters do not leak and reset() does not double-decrement.

Tests: Adds mocked pool lifecycle tests in tests/test_connection.py for new vs reused connections, full lifecycle netting to zero, unowned release pool name, full-queue release, and reset after a dropped connection.

^{Reviewed by Cursor Bugbot for commit dccf545. Bugbot is set up for automated code reviews on this repo. Configure here.}

[jit-ci]

Hi, I’m Jit, a friendly security platform designed to help developers build secure applications from day zero with an MVS (Minimal viable security) mindset.

In case there are security findings, they will be communicated to you as a comment inside the PR.

Hope you’ll enjoy using Jit.

Questions? Comments? Want to learn more? Get in touch with us.

[petyaslavova]

Hi @verdie-g, thank you for your contribution! Please ping me when the PR is ready for review :)

[verdie-g]

Hi @petyaslavova, it should be ready to review :)

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Reviewed by Cursor Bugbot for commit 3738fea. Configure here.}