test: feature_llmq_signing.py --spork21 flaky on v23.1.x

[thepastaclaw]

feature_llmq_signing.py --spork21 flaky on v23.1.x

Summary

feature_llmq_signing.py --spork21 is still intermittently failing on the
v23.1.x CI lane with the known wait_for_sigs(True, False, True, 15)
timeout. This appears to be the same failure tracked previously in
#7255 and fixed on develop by #7289, but #7289 is not present on
v23.1.x.

Latest occurrence

• PR: ci: tolerate conflict comment permission errors #7322 (ci: tolerate conflict comment permission errors)
• PR head: 6948ad3c273d2f1a4848877bf985347de08c1379
• Base: v23.1.x
• Failing run/job:
  https://github.com/dashpay/dash/actions/runs/25917464462/job/76187839723
• Job: linux64_multiprocess-test / Test source
• Failing test: feature_llmq_signing.py --spork21

Traceback excerpt:

File "test/functional/feature_llmq_signing.py", line 111, in run_test
  wait_for_sigs(True, False, True, 15)
File "test/functional/feature_llmq_signing.py", line 60, in wait_for_sigs
  self.wait_until(lambda: check_sigs(...), timeout = timeout)
AssertionError: Predicate ... not true after 60.0 seconds

Scope check

The triggering PR only changes:

.github/workflows/predict-conflicts.yml

It does not touch source code, test-src.yml, feature_llmq_signing.py, or
the functional test framework. The failure is therefore not caused by #7322.

Existing diagnosis

Issue #7255 tracked this exact intermittent feature_llmq_signing.py --spork21
timeout and was closed as fixed by #7289. That fix merged to develop as:

• 8c42e829b3 — Merge revert: "fix: avoid improper dual way connection attempts" + tests #7289
• 0e33aa2a9d — revert: "fix: avoid improper dual way connection attempts"
• c1cdb751e4 — test: assert QSENDRECSIGS handshake is symmetric under spork21

upstream/v23.1.x does not contain 8c42e829b3 / #7289, so the release
branch can still hit the old flake.

Suggested resolution

Backport #7289 to v23.1.x if the fix is appropriate for the release branch,
or otherwise mark/handle this test as known flaky on v23.1.x CI.

[thepastaclaw]

Additional occurrence from tracker thepastaclaw/tracker#1288 / #7322:

• Run/job: https://github.com/dashpay/dash/actions/runs/25917464462/job/76187839723
• PR head: 6948ad3c273d2f1a4848877bf985347de08c1379
• Base: v23.1.x
• Failing check: linux64_multiprocess-test / Test source
• Failure: feature_llmq_signing.py --spork21 timed out at wait_for_sigs(True, False, True, 15) / check_sigs(...) after 60s.

Scope check: #7322 only changes .github/workflows/predict-conflicts.yml; it does not touch functional tests or LLMQ/source code. This is the same release-branch flake described here; no PR branch change was made.

[thepastaclaw]

Additional occurrence from #7327 / thepastaclaw/tracker#1304:

• Run/job: https://github.com/dashpay/dash/actions/runs/26015843317/job/76468743126
• PR head: fb832662467a241a70c97826d2d3863bd44858f1
• Base: v23.1.x
• Failing check: linux64_sqlite-test / Test source
• Failure: feature_llmq_signing.py --spork21 timed out at wait_for_sigs(True, False, True, 2) after reconnecting the isolated recovery member (feature_llmq_signing.py:201).

Scope check: #7327 only changes feature_dip3_v19.py and the activate_by_name() helper to pass a longer wait_for_sporks_same() timeout for that one DIP3 v19 activation path. It does not modify feature_llmq_signing.py, quorum signing code, or the direct wait_for_sporks_same() use in this test. The modified feature_dip3_v19.py test passed in the same linux64_sqlite job. No PR branch change was made.