weiqingy commented on PR #29045: URL: https://github.com/apache/flink/pull/29045#issuecomment-6051962058
Hi @RocMarshal, #28953 is merged so this one is rebased onto master and out of draft. The diff is standalone now: one commit, 7 files, test-only. It adds restore coverage for the early-fire interval join, one program per time mode because the two keep different operator state. The processing-time one closes a real gap. The schedule state PR-5 added only registers when the time mode is proctime, so the existing row-time program never touched it. That program uses an `INFINITE` after-restore source with no after-restore input, following `GroupWindowAggregateProcTimeRestoreTest`. With no further input there is no `MAX_WATERMARK` and so no cleanup path, which is what keeps it deterministic: the single emitted row can only come from the restored schedule. Renaming the state descriptor makes it fail with an empty result, so the assertion really does depend on restored state. PTAL when you have time. Thanks! -- 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]
