Fix the FileBasedDeadLetterQueueReconsumer dup issue #2476
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #2476 +/- ##
==========================================
Coverage 49.62% 49.62%
- Complexity 4808 5157 +349
==========================================
Files 941 941
Lines 57661 57675 +14
Branches 6233 6233
==========================================
+ Hits 28614 28623 +9
- Misses 27007 27010 +3
- Partials 2040 2042 +2
🚀 New features to boost your workflow:
|
damccorm
reviewed
Jul 14, 2025
|
|
||
| PCollection<ResourceId> movedFiles = | ||
| input | ||
| .apply("MoveFiles", ParDo.of(new MoveFiles())) |
There was a problem hiding this comment.
I think this change introduces the possibility of data loss which is worse. Specifically, we could run into the following scenario:
- Bundle x contains
file-1,file-2,file-3 file-1is successfully renamed totmp-file-1, but the bundle fails on renamingfile-2
file-1 is now orphaned as tmp-file-1. I guess the way it is written, we would actually get recurrent errors which is a little better, but still not ideal.
Instead, I'd propose the following tweak:
- We get a timestamp
FOOassociated with each generate sequence firing. We already get this for free, we'd just need to propogate it through - Instead of renaming the file
tmp-<file name>, we rename ittmp-FOO-<file name> - Instead of only looking for
<file name>when doing the rename operation, we look for<file name>and rename it OR if<file name>is not present, we look fortmp-FOO-<file name>and use that instead. Iftmp-FOO-<file name>and<file name>are both not present, we log and move on, assuming that something else has claimed the file.
So the algorithm would be:
TriggerConsumeDLQ, AsFilePattern, MatchFiles, <new extract_ts function> -> (ts, file) tuples
in MoveFiles:
renamed_file_name = "tmp-${ts}-${original_file_name}"
if exists(renamed_file_name):
return
if !exists(original_file_name):
log.warning('skipping, handled by different pass')
mv(original_file_name, renamed_file_name)
Then the rest of the logic would stay the same.
| ParDo.of(new MoveAndConsumeFn(fileContents, fileMetadata)) | ||
| .withOutputTags(fileContents, TupleTagList.of(fileMetadata))); | ||
|
|
||
| results | ||
| .get(fileMetadata) | ||
| .setCoder(MetadataCoder.of()) | ||
| .setCoder(ResourceIdCoder.of()) |
There was a problem hiding this comment.
I think this is inevitable, but we are breaking update compatibility. Maybe worth calling out in the PR title so that it gets into the resource notes.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fixes #2236
This PR changes the process from a "match -> read -> delete" sequence to a "match -> atomic rename -> read -> delete" sequence.
MoveFiles:
A new MoveFiles step is introduced, which renames each matched dead-letter file by adding a tmp- prefix to its name. his atomic "move" operation effectively claims the file, ensuring that subsequent scans won't find the original file and attempt to process it again.
After the files are moved, a Reshuffle transform is applied to guarantee that the rename operation is completed.