[FLINK-16047] Blink planner produces wrong aggregate results with state clean up

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 1.9.0
• Fix Version/s: 1.9.3, 1.10.1, 1.11.0
• Component/s: Table SQL / Planner
• Labels: pull-request-available
• Assignee: Jark Wu
• Reporter: Timo Walther
• Created: 13/Feb/20 14:16
• Updated: 17/Mar/20 10:27
• Resolved: 17/Mar/20 09:16

Description

It seems that FLINK-10674 has not been ported to the Blink planner.

Because state clean up happens in processing time, it might be the case that retractions are arriving after the state has been cleaned up. Before these changes, a new accumulator was created and invalid retraction messages were emitted. This change drops retraction messages for which no accumulator exists.

These lines are missing in org.apache.flink.table.runtime.operators.aggregate.GroupAggFunction:

if (null == accumulators) {
      // Don't create a new accumulator for a retraction message. This
      // might happen if the retraction message is the first message for the
      // key or after a state clean up.
      if (!inputC.change) {
        return
      }
      // first accumulate message
      firstRow = true
      accumulators = function.createAccumulators()
    } else {
      firstRow = false
    }

The bug has not been verified. I spotted it only by looking at the code.

Issue Links

• links to: GitHub Pull Request #11419