[
https://issues.apache.org/jira/browse/BEAM-14536?focusedWorklogId=776503&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-776503
]
ASF GitHub Bot logged work on BEAM-14536:
-----------------------------------------
Author: ASF GitHub Bot
Created on: 31/May/22 18:36
Start Date: 31/May/22 18:36
Worklog Time Spent: 10m
Work Description: damccorm commented on PR #17782:
URL: https://github.com/apache/beam/pull/17782#issuecomment-1142516057
I caught up with Jack briefly, and the takeaways were:
1) If you call TryClaim, you are saying that you are going to attempt to
process the data (and it is your restriction's responsibility to handle that).
This is consistent with the Java guidance (see
https://beam.apache.org/documentation/programming-guide/#user-initiated-checkpoint).
2) Because of this, the residual restriction should _never_ contain a point
that the primary restriction has already tried to claim.
3) Currently, for any 0.0 split point, the new primary is `[original.Start,
original.Claimed)` and the residual is `[original.Claimed, original.End)`. This
is wrong since the residual now contains `original.Claimed`.
I've updated my PR to fix this in a general way
Issue Time Tracking
-------------------
Worklog Id: (was: 776503)
Time Spent: 2h (was: 1h 50m)
> Offsetrange tracker panics when splitting at 0.0 without claiming work
> ----------------------------------------------------------------------
>
> Key: BEAM-14536
> URL: https://issues.apache.org/jira/browse/BEAM-14536
> Project: Beam
> Issue Type: Bug
> Components: sdk-go
> Reporter: Danny McCormick
> Assignee: Danny McCormick
> Priority: P2
> Time Spent: 2h
> Remaining Estimate: 0h
>
> Right now, if you try to call TrySplit on an offsetrange restriction with a
> fraction of 0.0 and without first claiming work, it sets the primary
> restriction to \{Start, Start-1}. This causes newSplitResult to panic -
> https://github.com/apache/beam/blob/ff39fcb5229b15140e41a61bd09f7d590730e93a/sdks/go/pkg/beam/core/runtime/exec/sdf.go#L859
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
