[BEAM-8543] Dataflow streaming timers are not strictly time ordered when set earlier mid-bundle

• Type: Bug
• Status: Triage Needed
• Priority: P3
• Resolution: Unresolved
• Affects Version/s: 2.13.0
• Component/s: runner-dataflow
• Assignee: Unassigned
• Reporter: Jan Lukavský
• Created: 01/Nov/19 17:11
• Updated: 13/Apr/23 11:02

Description

Let's suppose we have the following situation:

• statful ParDo with two timers - timerA and timerB
• timerA is set for window.maxTimestamp() + 1
• timerB is set anywhere between <windowStart, windowEnd), let's denote that timerB.timestamp
• input watermark moves to BoundedWindow.TIMESTAMP_MAX_VALUE

Then the order of timers is as follows (correct):

• timerB
• timerA

But, if timerB sets another timer (say for timerB.timestamp + 1), then the order of timers will be:

• timerB (timerB.timestamp)
• timerA (BoundedWindow.TIMESTAMP_MAX_VALUE)
• timerB (timerB.timestamp + 1)

Which is not ordered by timestamp. The reason for this is that when the input watermark update is evaluated, the WatermarkManager,extractFiredTimers() will produce both timerA and timerB. That would be correct, but when timerB sets another timer, that breaks this.

Issue Links

• causes: BEAM-10991 Timers don't release watermark holds in dataflow on 2.24
• is a clone of: BEAM-7520 DirectRunner timers are not strictly time ordered
• links to: GitHub Pull Request #11924
• links to: GitHub Pull Request #12424
• links to: GitHub Pull Request #15794