kennknowles commented on a change in pull request #15603:
URL: https://github.com/apache/beam/pull/15603#discussion_r718808531
##########
File path: sdks/python/apache_beam/transforms/trigger.py
##########
@@ -161,12 +159,42 @@ def with_prefix(self, prefix):
class DataLossReason(Flag):
- """Enum defining potential reasons that a trigger may cause data loss."""
+ """Enum defining potential reasons that a trigger may cause data loss.
+
+ These flags should only cover when the trigger is the cause, though windowing
+ can be taken into account. For instance, AfterWatermark may not flag itself
+ as finishing if the windowing doesn't allow lateness.
+ """
+
+ # Trigger will never be the source of data loss.
NO_POTENTIAL_LOSS = 0
+
+ # Trigger may finish. In this case, data that comes in after the trigger may
+ # be lost. Example: AfterCount(1) will stop firing after the first element.
MAY_FINISH = auto()
+
+ # Trigger has a condition that is not guaranteed to ever be met. In this
+ # case, data that comes in may be lost. Example: AfterCount(42) will lose
+ # 20 records if only 20 come in, since the condition to fire was never met.
CONDITION_NOT_GUARANTEED = auto()
Review comment:
Yea that comment is outdated. In the current implementation, all
buffered elements must be emitted at GC time. Any runner that does not produce
those 32 elements as output is broken.
Incidentally, the data loss that unsafe triggers has to do with is not this
- it is about if the trigger fires and then further data is just ignored
because it "finished".
--
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]
