[
https://issues.apache.org/jira/browse/BEAM-14484?focusedWorklogId=772929&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-772929
]
ASF GitHub Bot logged work on BEAM-14484:
-----------------------------------------
Author: ASF GitHub Bot
Created on: 20/May/22 17:16
Start Date: 20/May/22 17:16
Worklog Time Spent: 10m
Work Description: jrmccluskey commented on code in PR #17724:
URL: https://github.com/apache/beam/pull/17724#discussion_r878372580
##########
sdks/go/pkg/beam/core/sdf/sdf.go:
##########
@@ -73,8 +73,8 @@ type RTracker interface {
// reason), then this function returns nil as the residual.
//
// If the split fraction is 0 (e.g. a self-checkpointing split)
TrySplit() should return either
- // a nil primary or an RTracker that is both bounded and has size 0.
This ensures that there is
- // no data that is lost by not being rescheduled for execution later.
+ // a nil primary or a restriction that represents no remaining work.
This will ensure that there
+ // is not data loss.
Review Comment:
So the returned primary restriction is supposed to be the same as the
restriction the RTracker holds as part of a typical runner-side split, so in
general we do care that the returns are configured correctly.
We do fail the pipeline to prevent the data loss, but I do prefer calling
out the why of failing the pipeline.
Issue Time Tracking
-------------------
Worklog Id: (was: 772929)
Time Spent: 6h 50m (was: 6h 40m)
> Improve error message surrounding primary returns in the self-checkpointing
> code
> --------------------------------------------------------------------------------
>
> Key: BEAM-14484
> URL: https://issues.apache.org/jira/browse/BEAM-14484
> Project: Beam
> Issue Type: Improvement
> Components: sdk-go
> Reporter: Jack McCluskey
> Assignee: Jack McCluskey
> Priority: P1
> Time Spent: 6h 50m
> Remaining Estimate: 0h
>
> The error message in the Go SDK harness around returned primaries in the
> self-checkpointing code
> ([https://github.com/apache/beam/blob/ea1f292e9cf31fc8c4803b10d811f0d3ee184ae7/sdks/go/pkg/beam/core/runtime/exec/datasource.go#L375)]
> is unclear and should be made more explicit. It should also guide the user
> towards making sure that the restriction behaves properly in the
> self-checkpointing case.
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
