[nick-boss-tech]

🤖 AI text below 🤖 (posted on behalf of Nick Shanin)

Ready for review, but held as a draft due to the project's open-PR cap; I'll mark it open on request.

https://issues.apache.org/jira/browse/SOLR-18576

What happens today

BasicDistributedZk2Test can index into a node whose ZooKeeper session has just expired, and the update fails with "Updates are disabled".

After ChaosMonkey.expireSession returns, the test calls blockUntilConnected and then indexes (test lines 152-164 at the merge base). expireSession returns once the client reports disconnected or its session id changes, but the expired-session event reaches Curator later. The wait can return before the event arrives, so the update races the expiry. In a CI run on #5194, the update was rejected with 503 "Cannot talk to ZooKeeper - Updates are disabled" 4 ms after the session-expired event was logged and 7 ms after ChaosMonkey started the expiry.

What this change does

The test waits for the expired-session event, then for reconnection, before it indexes.

A connection-state listener is registered before the session is expired. The test waits on a latch that counts down on the expired-session callback, then calls blockUntilConnected, and removes the listener before indexing. Only BasicDistributedZk2Test.java changes. Production code is unchanged.

Proof

The proof is inconclusive by construction: the race did not reproduce before or after the change, so these runs show the test still passes, not that the CI failure is gone.

• BasicDistributedZk2Test, seed 7E5494314774A841 (the seed from the failing run): 1 test, 0 failures, 0 errors. Verified 2026-10-23 at head 4c8f12f526f.
• BasicDistributedZk2Test, 4 random seeds: 4 tests, 0 failures, 0 errors. Verified 2026-10-23 at head 4c8f12f526f.
• Module check :solr:core:check -x test passes at that head.

Limits

The race is timing-dependent, and the cause is inferred from the CI log, not confirmed by a failing run.

• The test already allows up to 30 seconds for the expired-session event and then for reconnection. If either wait times out, the test fails at that wait, not at the update.
• Other tests that call expireSession are not changed.
• Changelog: none. The change is test-only.

AI-generated description. Nick Shanin directed the work and takes responsibility for this contribution.