MOD-15862 / MOD-15914 Fix Linux SSL CI hang + testUnevenWork timeout

[gabsow]

Summary

Cumulative fix for the LibMR Linux SSL CI hangs. Combines all MOD-15862 work (already discussed in #93) plus the targeted MOD-15914 follow-up fix for testUnevenWork:

1. MOD-15862 — original 6h hang in testSendRetriesMechanizm under single-shard --tls. Was hardcoded for 3 INNERCOMM sends but under TLS libmr only sends 2 (the slower handshake means the initial send goes through the resend loop and burns retry clang setup fixes #1). Rewritten to tolerate [1, MSG_MAX_RETRIES] attempts with bounded waits.
2. MOD-15862 (defensive) — added timeout-minutes: 45 / timeout-minutes: 25 SSL-step caps on both Linux and macOS workflows. Linux SSL step also gets PYTHONUNBUFFERED=1 + --test-timeout 300 so future hung tests are named, not silent.
3. MOD-15914 — testUnevenWork on Redis 7.2 + oss-cluster shards-count 3 + TLS was timing out at 200+s. Root cause: the with TimeLimit(2) SIGALRM was being starved by ssl.create_default_context() -> load_default_certs() -> set_default_verify_paths() running inside the ssl C extension on every loop iteration (because env.getConnection() was called inside the hot loop, opening fresh TCP sockets each time). Fix: lift getConnection outside the loop so redis-py reuses the pooled TLS connection — .ping() becomes pure socket I/O and yields to Python often enough for SIGALRM to fire on schedule.

Stacking history

This branch was originally a follow-up stacked on #93. Retargeted to master so it can get the full Linux/macOS workflow CI (which is gated on branches: [master]). PR #93 will be subsumed by this PR; if you want only one of the two merged, merge this.

Test plan

• Linux all 3 Redis versions pass Default + SSL
• macOS all 3 Redis versions pass Default + SSL (no 6h hangs)
• Linux Valgrind failure on testUnevenWork is unchanged (pre-existing — same MOD-15914 root cause shows under valgrind too, but the per-test cap will fire quickly now instead of running for the whole 6h job cap)

🤖 Generated with Claude Code

---

Note

Low Risk
Changes are limited to GitHub workflows, test harness flags, Valgrind suppressions, and Python integration tests; no production libmr send/retry logic is modified in this diff.

Overview
This PR hardens Linux and macOS CI against long SSL test hangs and makes failures easier to diagnose. Jobs get a 45-minute cap; the SSL test step is limited to 25 minutes on both platforms. On Linux, SSL runs also set PYTHONUNBUFFERED, pass RLTEST_EXTRA_ARGS: --test-timeout 300 through run_tests.sh, and upload test logs as artifacts on failure.

testUnevenWork now opens one Redis client per shard before the loop instead of calling env.getConnection() every iteration, so TLS runs reuse pooled connections and TimeLimit(2) can fire instead of starving inside repeated cert loading.

testSendRetriesMechanizm no longer assumes exactly three retry sends: it counts attempts in a bounded loop (tolerating 2 vs 3 under TLS vs plain), documents the MR_HelloResponseArrived retry off-by-one, and fixes the “no reconnect after give up” check so assertion failures are not swallowed.

Valgrind gets a new suppression for a pre-existing shutdown race on MR_CreateExecution / deferred dispose.

^{Reviewed by Cursor Bugbot for commit b9143a8. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

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

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit cfcec04. Configure here.}