[BEAM-8582] Python SDK emits duplicate records for Default and AfterWatermark triggers

• Type: Bug
• Status: Triage Needed
• Priority: P2
• Resolution: Fixed
• Fix Version/s: 2.19.0
• Component/s: sdk-py-core
• Assignee: Sam Rohde
• Reporter: Sam Rohde
• Created: 07/Nov/19 21:05
• Updated: 13/Apr/23 11:00
• Resolved: 10/Dec/19 22:46

Description

This was found after fixing https://issues.apache.org/jira/browse/BEAM-8581. The fix for 8581 was to pass in the input watermark. Previously, it was using MIN_TIMESTAMP for all of its EOW calculations. By giving it a proper input watermark, this bug started to manifest.

The DefaultTrigger and AfterWatermark do not clear their timers after the watermark passed the end of the endow, leading to duplicate records being emitted.

Fix: Clear the watermark timer when the watermark reaches the end of the window.

Issue Links

• causes: BEAM-10054 Direct Runner execution stalls with test pipeline