janhoy commented on PR #5323: URL: https://github.com/apache/solr/pull/5323#issuecomment-6879745167
🤖 *AI-assisted reply, reviewed by Jan* Thanks Nick, this was a careful read and the follow-up branch made every point easy to check. One framing note first: this PR is a faithful port of the BATS suites. Where a finding would make a test assert more than BATS did, I'd rather keep it out of this PR and take it as its own JIRA + PR, ideally after the port lands, or stacked on this branch if you want to start sooner. Finding by finding: **Authenticated posting and log indexing.** Agreed that `1 files indexed.` and `Committed` do not prove a document landed: `solr post` counts files attempted (SOLR-18653) and `postlogs` prints `Committed` after a failed batch (SOLR-18655), two of the CLI bugs CATS turned up, tracked separately. The BATS original asserted exactly those two strings, so the port is faithful as it stands. Querying for the document and the logs is a good strengthening; please propose it as a follow-up. **TLS rejection causes.** The five bare `assertConnectionFailed()` sites in `startSolrWithSsl` and `testKeystoreReload` mirror BATS's bare `run ! curl` at the same places; tightening them to the certification-path message is a good follow-up too. The "Server refused connection" cases were already handled in 0880d46778f, an ancestor of the commit you reviewed: each CLI assertion there is paired with an in-JVM `assertConnectionFailed()` that requires an SSL handshake failure, which your diff removes. On mTLS: your rewrite's `(bad_certificate|certificate_required)` assertion fails in your own evidence file, while the current assertions pass locally, so the case stays `@AwaitsFix(SOLR-18646)` and the closed-channel behaviour you saw is part of what that JIRA has to sort out. I'd leave the synthetic regex fixtures out; they test constants in the same file against literals rather than Solr. **Example resource-limit warnings.** Good catch, and in scope: the case landed in `test_example.bats` on Nov 11 (SOLR-18527), after the suites were ported, and the main merge missed it. Ported as `ExampleTest.examplesSuppressResourceLimitWarnings`, essentially your hunk. The same commit added `test_secure_example.bats`, which had no counterpart either; ported as `SecureExampleTest`. **Rolling-upgrade execution.** Agreed on the intent, and both requirements are already met by the task: `dockerRollingUpgradeTest` depends on `dockerBuild` and hands the test the built image id, and the test's only skip is an absent Docker daemon, so a missing image or property fails loudly. What remains is that nothing invokes the task in CI yet, a pre-existing gap (the BATS predecessor silently skipped in CI for the same reason), tracked in SOLR-18658. **Filtered process discovery.** You are right that a marker-filtered scan treats a live node outside the marker as stale. That is the design: the marker scopes a PID directory, so whoever sets one must also give its nodes their own `SOLR_PID_DIR`, as the harness does. Its own variables are applied after the inherited environment, so an ambient `SOLR_PID_DIR` cannot override that isolation. I've made the assumption explicit in the `PROCESS_SCAN_MARKER` javadoc rather than change the behaviour. If you think a filtered scan should never delete, that is worth its own discussion. And yes, #5299 can close when this merges. Thanks for the comparison work. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
