[Iceberg CDC sink] Split late data and create commit windows by ahmedabu98 · Pull Request #40006 · apache/beam

ahmedabu98 · 2026-09-03T23:11:40Z

Adds the CDC sink's windowing stage:

CommitWindows groups each destination's keyed changes into event-time commit windows and sorts them by primary key and sequence number via SortValues transform
SplitLateData diverts every late pane to a replayable dead-letter output instead of letting it reach a window the committer may already have sealed.

Part of #39979

Thank you for your contribution! Follow this checklist to help us incorporate your contribution quickly and easily:

Mention the appropriate issue in your description (for example: addresses #123), if applicable. This will automatically add a link to the pull request in the issue. If you would like the issue to automatically close on merging the pull request, comment fixes #<ISSUE NUMBER> instead.
Update CHANGES.md with noteworthy changes.
If this contribution is large, please file an Apache Individual Contributor License Agreement.

See the Contributor Guide for more tips on how to make review process smoother.

To check the build health, please visit https://github.com/apache/beam/blob/master/.test-infra/BUILD_STATUS.md

GitHub Actions Tests Status (on master branch)

See CI.md for more information about GitHub Actions CI or the workflows README to see a list of phrases to trigger workflows.

github-actions · 2026-09-04T00:41:51Z

Checks are failing. Will not request review until checks are succeeding. If you'd like to override that behavior, comment assign set of reviewers

…commit-windows

github-actions · 2026-09-05T06:45:28Z

Assigning reviewers:

R: @chamikaramj for label java.

Note: If you would like to opt out of this review, comment assign to next reviewer.

Available commands:

stop reviewer notifications - opt out of the automated review tooling
remind me after tests pass - tag the comment author after tests pass
waiting on author - shift the attention set back to the author (any comment or push by the author will return the attention set to the reviewers)

The PR bot will only process comments in the main thread (not review comments).

chamikaramj

Thanks!

chamikaramj · 2026-09-09T05:22:10Z

+ */
+final class CommitWindows
+    extends PTransform<
+        PCollection<KV<KV<String, Integer>, KV<byte[], CdcRecord>>>, CommitWindows.Result> {


PCollection<KV<KV<String, Integer>, KV<byte[], CdcRecord>>> is pretty hard to follow.

Shall we introduce new types to simplify this ? For example,

ShardIdentifier -> KV<String, Integer>

SortKeyAndRecord -> KV<byte[], CdcRecord>

Alternatively we can use Beam Rows or AutoValue classes with schema here.

Added DestinationShard instead of KV<String, Integer>.

Doing SortKeyAndRecord would be a little more complicated since the SortValues transform requires the value part to be of type KV<byte[], Object>. We can technically add adapters before/after to convert from SortKeyAndRecord to KV<byte[], CdcRecord> and back, but it would need an additional custom coder and I don't think is necessary

chamikaramj · 2026-09-09T05:27:20Z

+    return new Result(input.getPipeline(), sorted, deadLetter, deadLetterSchema);
+  }
+
+  /** Applies the commit-window assignment for the input's boundedness; see the class Javadoc. */


for the input's boundedness -> based on whether input is bounded or not ?

Commit-window assignment is applied in both cases. Added some more language to clarify

chamikaramj · 2026-09-09T05:45:39Z

+
+/**
+ * Splits late data in {@link CommitWindows}: every {@linkplain PaneInfo.Timing#LATE late} pane is
+ * diverted to a DLQ side output; on-time and early panes pass through unchanged.


We can't actually commit late data instead of sending to DLQ ? (probably reasonable but just to double check).

Yeah, if we let late data through to the commit stage, it'll just get skipped since its window would have already been committed. Better to capture it here and provide it in the DLQ

chamikaramj · 2026-09-09T06:07:03Z

+    Coder<?> valueCoder = ((KvCoder<?, ?>) inputCoder).getValueCoder();
+    checkArgument(
+        valueCoder instanceof KvCoder, "expected a KvCoder input value coder, got %s", valueCoder);
+    Coder<?> recordCoder = ((KvCoder<?, ?>) valueCoder).getValueCoder();


This kind of casts can be brittel and can be avoided by introducing new types as mentioned in the other comment I believe.

Introducing a new SortKeyAndRecord type will also need its own custom coder, which we'll still have to extract here using casting.

chamikaramj · 2026-09-09T06:11:32Z

+        onTimeUnsorted.apply(
+            "SortBySeqKind",
+            SortValues.create(
+                BufferedExternalSorter.options().withMemoryMB(config.getSorterMemoryMB())));


This could lead to OOMs since we are sorting entire windows per shard. Let's make sure there's enough memory for the default case.

In-memory usage of the sorter is controlled by config.getSorterMemoryMB(). Anything over that is spilled over into on-disk files. I used 100MB because that's already the default in BufferedExternalSorter. We're just re-using it here and allowing the user to override.

chamikaramj · 2026-09-09T06:12:21Z

+        "EventTimeWindows",
+        Window.<KV<KV<String, Integer>, KV<byte[], CdcRecord>>>into(
+                FixedWindows.of(
+                    checkStateNotNull(


Do this check early ?

Yes it will be checked earlier in the top-level class (coming in a later PR)

ahmedabu98 · 2026-09-14T15:47:46Z

Thanks @chamikaramj, can you PTAL?

split late data and create commit windows

9cc7089

github-actions Bot added java io labels Sep 3, 2026

ahmedabu98 changed the title ~~[Iceberg] Split late data and create commit windows~~ [Iceberg CDC sink] Split late data and create commit windows Sep 3, 2026

Merge branch 'master' of https://github.com/ahmedabu98/beam into cdc-…

934a80c

…commit-windows

github-actions Bot added the Next Action: Reviewers label Sep 5, 2026

chamikaramj reviewed Sep 9, 2026

View reviewed changes

address comments

42bf43c

github-actions Bot added iceberg Iceberg IO connector and removed iceberg Iceberg IO connector labels Sep 14, 2026

chamikaramj approved these changes Sep 14, 2026

View reviewed changes

ahmedabu98 merged commit 544e40f into apache:master Sep 14, 2026
23 checks passed

Conversation

ahmedabu98 commented Sep 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

GitHub Actions Tests Status (on master branch)

Uh oh!

github-actions Bot commented Sep 4, 2026

Uh oh!

github-actions Bot commented Sep 5, 2026

Uh oh!

chamikaramj left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ahmedabu98 commented Sep 14, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

ahmedabu98 commented Sep 3, 2026 •

edited

Loading