nsivabalan commented on PR #20151: URL: https://github.com/apache/hudi/pull/20151#issuecomment-6186595861
One more false positive, found after the review round: **the cleaner check flagged every pure-append table.** A table that only receives inserts never produces a clean instant, and that is correct — cleaning retires superseded file slices, and an append-only workload never supersedes anything. Every file group has one slice, nothing is replaced, so the cleaner runs and finds no work. The check read that empty cleaner timeline as a stalled service, so such a table came out UNHEALTHY every day with nothing the operator could do about it. Fixed in `f786277318f1`. Append-only is read from the commit metadata rather than from `hoodie.datasource.write.operation`, because what the table actually received is the question: a writer configured for upsert that has only ever inserted still has nothing to clean, and a property file can be stale or absent. One `upsert`, `delete` or `insert_overwrite` anywhere on the timeline and the normal cadence judgement applies again. It is conservative on failure — an unreadable commit, or an operation type the metadata does not carry, means the exemption does not apply and the check judges as before. Suppressing a real finding is the worse error, so the exemption has to be earned by every completed write on the timeline. Three tests: append-only is HEALTHY with the reason in the summary, an upsert table with no clean is still UNHEALTHY, and a timeline that is 59 inserts plus a single upsert is still judged. -- 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]
