[FLINK-6004] [kinesis] Allow FlinkKinesisConsumer to skip non-deserializable records #5269
Conversation
This file contains 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
zentol
reviewed
Jan 31, 2018
| @@ -484,7 +484,10 @@ protected Properties getConsumerConfiguration() { | |||
| */ | |||
| protected final void emitRecordAndUpdateState(T record, long recordTimestamp, int shardStateIndex, SequenceNumber lastSequenceNumber) { | |||
| synchronized (checkpointLock) { | |||
| sourceContext.collectWithTimestamp(record, recordTimestamp); | |||
| if (record != null) { | |||
| sourceContext.collectWithTimestamp(record, recordTimestamp); | |||
There was a problem hiding this comment.
Are we silently skipping the record or do we log somewhere that a record was invalid?
There was a problem hiding this comment.
We currently do not have a log for that.
I'll add a warning log if the record is null.
zentol
reviewed
Jan 31, 2018
| for (int i = 0; i < numShards; i++) { | ||
| fetcher.emitRecordAndUpdateState("record-" + i, 10L, i, new SequenceNumber("seq-num-1")); | ||
| assertEquals(new SequenceNumber("seq-num-1"), testShardStates.get(i).getLastProcessedSequenceNum()); | ||
| assertEquals(new StreamRecord<>("record-" + i, 10L), sourceContext.removeLatestOutput()); |
zentol
reviewed
Jan 31, 2018
|
@zentol |
…nesisDataFetcherTest and related classes The previous implementation of the TestableKinesisDataFetcher was confusing in various ways, causing it hard to be re-used for other tests. This commit contains the following various cleaups: - Remove confusing mocks of source context and checkpoint lock. We now allow users of the TestableKinesisDataFetcher to provide a source context, which should provide the checkpoint lock. - Remove override of emitRecordAndUpdateState(). Strictly speaking, that method should be final. It was previously overriden to allow verifying how many records were output by the fetcher. That verification would be better implemented within a mock source context. - Properly parameterize the output type for the TestableKinesisDataFetcher. - Remove use of PowerMockito in KinesisDataFetcherTest. - Use CheckedThreads to properly capture any exceptions in fetcher / consumer threads in unit tests. - Use assertEquals / assertNull instead of assertTrue where-ever appropriate.
Prior to this commit, several unit tests in KinesisDataFetcherTest relied on sleeps to wait until a certain operation happens, in order for the test to pass. This commit removes those sleeps and replaces the test behaviours with OneShotLatches.
This commit acknowledges that null can be returned from the deserialization schema, if the message cannot be deserialized. If null is returned for a Kinesis record, no output is produced for that record, while the sequence number in the shard state is still advanced so that the record is effectively accounted as processed.
|
@zentol I've rebased, and added a warning when skipping records. Could you have another quick look? |
|
looks good, +1 |
|
Thanks, merging .. |
tzulitai
added a commit
to tzulitai/flink
that referenced
this pull request
Feb 1, 2018
This commit acknowledges that null can be returned from the deserialization schema, if the message cannot be deserialized. If null is returned for a Kinesis record, no output is produced for that record, while the sequence number in the shard state is still advanced so that the record is effectively accounted as processed. This closes apache#5269.
tzulitai
added a commit
to tzulitai/flink
that referenced
this pull request
Feb 2, 2018
This commit acknowledges that null can be returned from the deserialization schema, if the message cannot be deserialized. If null is returned for a Kinesis record, no output is produced for that record, while the sequence number in the shard state is still advanced so that the record is effectively accounted as processed. This closes apache#5269.
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.
What is the purpose of the change
This PR is based on #5268, which includes fixes to harden Kinesis unit tests. Only the last commit is relevant.
In the past, we allowed the Flink Kafka Consumer to skip corrupted Kafka records which cannot be deserialized. In reality pipelines, it is entirely normal that this could happen.
This PR adds this functionality to the Flink Kinesis Consumer also.
Brief change log
KinesisDeserializationSchemathatnullcan be returned if a message cannot be deserialized.recordisnullinKinesisDataFetcher::emitRecordAndUpdateState(...), do not collect any output for the record.KinesisDataFetcherTest::testSkipCorruptedRecord()to verify feature.Verifying this change
Additional
KinesisDataFetcherTest::testSkipCorruptedRecord()test verifies this change.Does this pull request potentially affect one of the following parts:
@Public(Evolving): (yes / no)Documentation