fix(streams): transaction probe must not lease an AR connection

[mhenrixon]

Summary

Stream#current_open_transaction probes for an open transaction via ActiveRecord::Base.connection. On Rails 7.2+ that call takes a sticky, executor-scoped lease, which leaks pool connections in two ways:

1. Non-executor threads (the Coalescer's Concurrent::ScheduledTask flush thread, any app worker thread not wrapped by the Rails executor): the lease is never released — one AR pool connection pinned per thread for its lifetime.
2. Inside connection_pool.with_connection: the sticky flag defeats the block-exit release, so the caller's connection leaks when its thread dies. The pool then reports it dead-but-in-use (stat → dead: N), and those slots are gone.

Both variants only fire on the durable path — ephemeral broadcasts return broadcast_ephemeral(wrapped) unless use_durable before the probe. That's why flipping streams_default_broadcast_mode = :durable surfaced it downstream: Zazu's bulk-payment fan-out spec sizes its worker threads to the exact AR pool, and one run left the pool at {connections: 6, busy: 1, dead: 5, waiting: 6} — 5 dead leases from the workers' with_connection blocks plus one live lease pinned to the coalescer flush thread. Deterministic ActiveRecord::ConnectionTimeoutError on every run.

Production impact isn't limited to specs: every coalesced/durable broadcast flushed on the coalescer thread pins an AR connection in web/worker processes.

Fix

Probe connection_pool.active_connection? instead — returns the calling thread/fiber's existing lease or nil, never a checkout. Semantics are unchanged: a transaction is per-lease, so a thread holding no connection has no open transaction to defer on. The old code's fresh checkout always answered nil in that situation anyway — at the price of the leak.

Tests

spec/pgbus/streams/transaction_probe_spec.rb (real AR via the dummy app):

• probe from a lease-less thread leases nothing (red on main: active_connection? returns the leaked lease)
• probe from a lease-less thread returns nil
• broadcasting inside with_connection doesn't defeat its release (red on main: dead-owner in-use connection)
• probe still finds the open transaction when the caller holds one

Existing AR stubs in streams_spec.rb updated to the new connection_pool.active_connection? contract. Full streams suite: 383 green. (spec/rubocop/cop/pgbus/no_ruby_timeout_spec.rb fails to load on a clean main checkout locally — pre-existing, unrelated.)

Summary by CodeRabbit

• Bug Fixes

  • Improved transaction detection during broadcasts to avoid unnecessarily acquiring or leaking database connections.
  • Preserved correct connection cleanup on worker threads.
  • Broadcasts now proceed immediately when no transaction is open.

• Tests

  • Added regression coverage for connection handling and open transaction detection.
  • Updated transaction test coverage to reflect active connection behavior.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository UI (base), Organization UI (inherited)

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 36608e9b-0e21-4f6e-9eb8-81a3fdd434d0

📥 Commits

Reviewing files that changed from the base of the PR and between b3b5c61 and 99b1691.

📒 Files selected for processing (3)

• lib/pgbus/streams.rb
• spec/pgbus/streams/transaction_probe_spec.rb
• spec/pgbus/streams_spec.rb

---

📝 Walkthrough

Walkthrough

Stream#current_open_transaction now inspects only already-leased ActiveRecord connections. Tests cover connection cleanup, no-lease behavior, immediate broadcasts, and open transaction detection.

Changes

ActiveRecord transaction probing

Layer / File(s)	Summary
Use active connection leases for transaction probing lib/pgbus/streams.rb	current_open_transaction uses active_connection? and avoids acquiring an unused connection.
Validate connection lifecycle and transaction detection spec/pgbus/streams/transaction_probe_spec.rb, spec/pgbus/streams_spec.rb	Tests cover unleased threads, worker-thread cleanup, immediate broadcasts, and existing open transactions. Test doubles now use the connection pool interface.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related PRs

• mhenrixon/pgbus#81: Introduced the transaction helper refined by this change.
• mhenrixon/pgbus#158: Addresses related ActiveRecord connection lifecycle behavior in a different code path.

Suggested labels: bug, testing, connections

Poem

A rabbit checks the leased connection,
No checkout springs from inspection.
Open transactions still appear,
Worker threads leave their leases clear.
Tests hop softly through the flow.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely describes the main change: preventing the transaction probe from leasing an ActiveRecord connection.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/transaction-probe-connection-leak

---

_{Comment @coderabbitai help to get the list of available commands.}