Status

Current state: Under discussion  Complete 

Discussion thread: here

JIRA: here 

Vote thread: here

Please keep the discussion on the mailing list rather than commenting on the wiki (wiki discussions get unwieldy fast).

...

We have attempted error handing in accepted KIP: KIP-691: Enhance Transactional Producer Exception Handling - Apache Kafka - Apache Software Foundation which we can perhaps leverage some of the work here.

...

Proposed Changes

We would new error types which will be extended by existing exceptions mentioned in below table.

// Producer-Recoverable
public class AbortableTransactionException extends ApiException {
    public AbortableTransactionException(String message) {
        super(message);
    }
    ...
}

//Producer-Retriable 
public class ProducerRetriableTransactionException extends ApiException {
    public ProducerRetriableTransactionException(String message) {
        super(message);
    }
 	...
}

//Producer-Retriable
public class ProducerImmediateRetriableTransactionException extends ApiException {
    public ProducerImmediateRetriableTransactionException(String message) {
        super(message);
    }
	...
}

//Application-Recoverable
public class ApplicationRecoverableTransactionException extends ApiException {
    public ApplicationRecoverableTransactionException(String message) {
        super(message);
    }
    ...
}

// Invalid-Configuration
public class InvalidConfiguationTransactionException extends ApiException {
    public InvalidConfiguationTransactionException(String message) {
        super(message);
    }
 	...
}

// Extending exception types example
public class InvalidProducerEpochException extends AbortableTransactionException {
    private static final long serialVersionUID = 1L;
    public InvalidProducerEpochException(String message) {
        super(message);
    }
}

Client side code example

are proposing to group exceptions into four types:

1. Retriable: mostly maps onto current retriable errors. Some of these errors may indicate that a metadata update at the producer level is required. Because of that we introduce two subclasses to the retriable error

   1. retry only (just send the request again – after some period of backoff)

   2. refresh metadata and retry (request metadata and maybe modify the request before resending)

2. Abortable: mostly maps onto abortable errors. The error is bubbled to the application layer, and it can choose to roll back any state from the ongoing transaction and abort the transaction. The producer does not need to restart, as after aborting, it can be confident that the state was as it was before the transaction started. The application can also close the producer and react as it does for application-recoverable cases, but it doesn’t need to.

3. Application-Recoverable: maps on to some fatal errors. The error is bubbled to the application layer, and it may need to do a bit more to roll back and clean up state. The producer can not simply abort and know the current state of the partitions. Applications can handle this in different ways – Kafka Streams may rebalance a task and/or close the task and restart it. Another application may read from their own checkpoint to continue. In any case, the producer must restart and will be unusable after encountering this error.

4. Invalid-Configuration: maps to some fatal errors. The error is bubbled up to the application layer. The application can decide what to do. The producer doesn’t need to restart, but the application may chose to close it.

Each error code always represent the same class and rarely rely on client state to determine how to handle. Additionally, while it is good to have a mapping, It is also useful to have a general strategy – ie a typical unknown error (not specified by the client to have a type) should probably be application recoverable.

Note: For Retriable errors, the producer handles retries internally, keeping the failure details hidden from the application. Conversely, other types of exceptions will be surfaced to the application code for handling.

Exception Table 

Below table contains list of exceptions with current and New handing. All exceptions are grouped under a header which represent Exception Name:

Yellow Image Added → Needs modification

Check Image Added → Exception handling match as mentioned in KIP-691

Cross Image Added → Exception handling doesn't match as mentioned in KIP-691

RetriableException

RefreshRetriableException

TransactionAbortableException

ApplicationRecoverableException

InvalidConfigurationException

Public Interfaces

We have below exception classes available in current Kafka code

// Producer Abortable Transaction
public class TransactionAbortableException extends ApiException {
    ...
}

//Producer Retriable  
public abstract class RetriableException extends ApiException {
   ...
}


// Invalid-Configuration
public class InvalidConfigurationException extends ApiException {
 	...
}

We will add new exception types, as listed in the table below, that extend the existing exceptions.

//Producer Refresh and Retriable
public abstract class RefreshRetriableException extends RetriableException {
	...
}

//Application-Recoverable new
public abstract class ApplicationRecoverableException extends ApiException {
    ...
}

We will extend below existing exceptions types to maintain class hierarchy 

// UnknownTopicOrPartitionException and NotLeaderOrFollowerException  extends InvalidMetadataException 
public class InvalidMetadataException extends RefreshRetriableException {
	...
}

// TopicAuthorizationException and GroupAuthorizationException extends AuthorizationException 
public class AuthorizationException extends InvalidConfigurationException {
	...
}

Client side code example

public class TransactionalClientDemo {

    private static final String CONSUMER_GROUP_ID = "my-group-id";
    private static final String OUTPUT_TOPIC = "output";
    private static final String INPUT_TOPIC = "input";
    private static KafkaConsumer<String, String> consumer;
    private static KafkaProducer<String, String> producer;

    public static void main(String[] args) {
        initializeApplication();

        boolean isRunning = true;
        // Continuously poll for records
        while (isRunning) {

public class TransactionalClientDemo {

    private static final String CONSUMER_GROUP_ID = "my-group-id";
    private static final String OUTPUT_TOPIC = "output";
    private static final String INPUT_TOPIC = "input";
    private static KafkaConsumer<String, String> consumer;
    private static KafkaProducer<String, String> producer;

    public static void main(String[] args) {
        initializeApplication();

        boolean isRunning = true;
        // Continuously poll for records
        while (isRunning) {
            try {
                try {
                    // Poll records from Kafka for a timeout of 60 seconds
                    ConsumerRecords<String, String> records = consumer.poll(ofSeconds(60));

                    // Process records to generate word count map
                    Map<String, Integer> wordCountMap = new HashMap<>();

                    for (ConsumerRecord<String, String> record : records) {
                        String[] words = record.value().split(" ");
                        for (String word : words) {
                            wordCountMap.merge(word, 1, Integer::sum);
                        }
                    }

                    // Begin transaction
                    producer.beginTransaction();

                    // Produce word count results to output topic
                    wordCountMap.forEach((key, value) ->
                            producer.send(new ProducerRecord<>(OUTPUT_TOPIC, key, value.toString())));

                    // Determine offsets to commit
                    Map<TopicPartition, OffsetAndMetadata> offsetsToCommit = new HashMap<>();
                    for (TopicPartition partition : records.partitions()) {
                        List<ConsumerRecord<String, String>> partitionedRecords = records.records(partition);
                        long offset = partitionedRecords.get(partitionedRecords.size() - 1).offset();
                        offsetsToCommit.put(partition, new OffsetAndMetadata(offset + 1));
                    }

                    // Send offsets to transaction for atomic commit
                    producer.sendOffsetsToTransaction(offsetsToCommit, CONSUMER_GROUP_ID);

                    // Commit transactiontry {
                    producer.commitTransaction();try {
                } catch (AbortableTransactionException e) {
// Poll records from Kafka for a timeout of 60 seconds
          // Abortable Exception: Handle Kafka exception by aborting transaction. producer.abortTransaction() should not throw abortable exception.
 ConsumerRecords<String, String> records = consumer.poll(ofSeconds(60));

                    // Process  producer.abortTransaction();
    records to generate word count map
                resetToLastCommittedPositions(consumer);
    Map<String, Integer> wordCountMap = new HashMap<>();

       }
            } catchfor (InvalidConfiguationTransactionException eConsumerRecord<String, String> record : records) {
                //   Fatal Error: The error is bubbledString[] upwords to the application layer. The application can decide what to do
= record.value().split(" ");
                      closeAll();
  for (String word : words)     {
						  wordCountMap.merge(word, 1, Integer::sum);
     throw e;
            } catch (KafkaException | ApplicationRecoverableTransactionException e) {}
                  // Application Recoverable: The application must restart
  }

                     closeAll();
// Begin transaction
                 initializeApplication   producer.beginTransaction();

            }
        // Produce word count results to output topic
              }

      }

Full example can be accessed at: https://github.com/apache/kafka/pull/15913/files 

Proposed Changes

We are proposing to group exceptions into four types:

1. Producer-Retriable: mostly maps onto current retriable errors. The producer can handle the retry on its own and the failure is invisible to the application. Some of these errors may indicate that a metadata update at the producer level is required. Because of that we introduce a few subclasses to the retriable error

   1. retry only (just send the request again – after some period of backoff)

   2. refresh metadata and retry (request metadata and maybe modify the request before resending)

2. Producer-Recoverable: mostly maps onto abortable errors. The error is bubbled to the application layer, and it can choose to roll back any state from the ongoing transaction and abort the transaction. The producer does not need to restart, as after aborting, it can be confident that the state was as it was before the transaction started. The application can also close the producer and react as it does for application-recoverable cases, but it doesn’t need to.

3. Application-Recoverable: maps on to some fatal errors. The error is bubbled to the application layer, and it may need to do a bit more to roll back and clean up state. The producer can not simply abort and know the current state of the partitions. Applications can handle this in different ways – streams may rebalance a task and/or close the task and restart it. Another application may read from their own checkpoint to continue. In any case, the producer must restart and will be unusable after encountering this error.

4. Invalid-Configuration: maps to some fatal errors. The error is bubbled up to the application layer. The application can decide what to do. The producer doesn’t need to restart, but the application may chose to close it.

Each error code always represent the same class and rarely rely on client state to determine how to handle.

While it is good to have a mapping, I think it is also useful to have a general strategy – ie a typical unknown error (not specified by the client to have a type) should probably be application recoverable.

Below table contains list of exceptions with current and expected handing:

Yellow Image Removed → Needs modification

Green Image Removed → New exceptions added as part of this KIP

Check Image Removed → Exception handling match as mentioned in KIP-691

Cross Image Removed → Exception handling doesn't match as mentioned in KIP-691

...

...

Thrown scenarios during transaction

...

Current handling

...

Expected Handling

...

Producer API

...

Transaction API

...

Producer API

...

Transaction API

...

ErrorRetry

...

Producer Retriable

...

Producer Retriable

...

Producer Retriable

...

Producer Retriable

...

ErrorRefreshMetadataAndRetry

...

Refresh + Retriable

...

Refresh + Retriable

...

Refresh + Retriable

...

Refresh + Retriable

...

ErrorFenceClient

...

Application Recoverable

...

Application Recoverable

...

Application Recoverable

...

Application Recoverable

...

ErrorBadConfig

...

Invalid Configuration

...

Invalid Configuration

...

Invalid Configuration

...

Invalid Configuration

...

TransactionAbortableException
** note – Added via KIP-890

...

Producer Recoverable

...

Producer Recoverable

...

Producer Recoverable

...

Producer Recoverable

...

IllegalStateException

...

Abortable

...

Sometimes Fatal depending on whether application or Sender caused issue. See: kafka: KAFKA-14831: Illegal state errors should be fatal in transactional producerCLOSED

...

...

...

Au thenticationException

...

Abortable

...

Fatal

...

...

...

InvalidPidMappingException

...

• If the TransactionalId does not exist in the TransactionalId mapping or if the mapped PID is different from that in the request, reply with InvalidPidMapping; otherwise proceed to next step.

...

Sometimes retriable (no longer returned) or abortable

...

Abortable

...

...

...

UnknownProducerIdException

...

Sometimes retriable (no longer returned) or abortable

...

Abortable

...

...

...

ClusterAuthorizationException

TransactionalIdAuthorizationException

UnsupportedVersionException

UnsupportedForMessageFormatException

...

Fatal, except UnsupportedForMessageFormatException which is Abortable

...

Cluster/Transaction Auth → abortable on InitProducerId, other errors abortable

TransactionAuth → fatal, others abortable for AddPartitions, Find Coordinator, EndTxn, AddOffsets

Transaction Auth, UnssupportedForMessageFormat → fatal for OffsetCommit

...

...

...

CorruptRecordException

NotEnoughReplicasAfterAppendException

NotEnoughReplicasException

TimeoutException

...

Retriable if the error is retriable. Otherwise abortable

...

Retriable if the error is retriable otherwise abortable

...

...

...

UnknownTopicOrPartitionException

NotLeaderOrFollowerException

...

Retriable if the error is retriable. Otherwise abortable

...

Retriable if the error is retriable otherwise abortable

...

...

...

InvalidRecordException

InvalidRequiredAcksException

RecordBatchTooLargeException

InvalidTopicException

...

Retriable if the error is retriable. Otherwise abortable

...

Retriable if the error is retriable otherwise abortable

...

...

...

TopicAuthorizationException

GroupAuthorizationException

...

Abortable

...

Abortable

...

...

...

FencedInstanceIdException

CommitFailedException

UnknownMemberIdException

IllegalGenerationExceiption

...

N/A

...

Abortable (TxnOffsetCommit Only)

...

N/A

...

...

InvalidProducerEpochException

...

• If the PID’s epoch number is different from the current TransactionalId PID mapping, reply with the InvalidProducerEpoch error code; otherwise proceed to next step.

...

Abortable

...

Fatal

...

...

...

ProducerFencedException

...

Fatal

...

Fatal

...

...

...

OutOfOrderSequenceException

...

Sometimes retriable or abortable

...

N/A

...

...

N/A

...

InvalidTxnStateException

...

• When client makes a request which involves change of Txn state. However expected state transition is invalid.
  e.g COMPLETE_ABORT to COMMIT

...

Abortable

...

Fatal

...

...

...

KafkaException

...

Abortable (default seems to be abortable)

...

Fatal in most cases, but abortable when there are partition errors

...

...

...

RuntimeException

...

Abortable (default seems to be abortable)

...

Fatal – only thrown as this generic type when correlation ID is wrong. This should be updated as KIP-691 suggests

...

...

...

ConcurrentTransactionsException

...

Retriable

...

Retriable

...

Producer Retriable (KIP-890 may see this)

...

Producer Retriable

...

CoordinatorLoadInProgressException

...

Retriable

...

Retriable

...

Producer Retriable

...

Producer Retriable

...

NotCoordinatorException

CoordinatorNotAvailableException

...

• Check whether there is a previous entry with the same TransactionalId and a higher epoch. If so, throw an exception. In particular, this indicates the log is corrupt. All future transactional RPCs to this coordintaor will result in a `NotCoordinatorForTransactionalId` error code, and this partition of the log will be effectively disabled.

• Check if it is the assigned transaction coordinator for the TransactionalId, if not reply with the NotCoordinatorForTransactionalId error code.

...

Retriable

...

Retriable

...

Refresh + Retriable

...

Refresh + Retriable

wordCountMap.forEach((key, value) ->
                            producer.send(new ProducerRecord<>(OUTPUT_TOPIC, key, value.toString())));

                    // Determine offsets to commit
                    Map<TopicPartition, OffsetAndMetadata> offsetsToCommit = new HashMap<>();
                    for (TopicPartition partition : records.partitions()) {
                        List<ConsumerRecord<String, String>> partitionedRecords = records.records(partition);
                        long offset = partitionedRecords.get(partitionedRecords.size() - 1).offset();
                        offsetsToCommit.put(partition, new OffsetAndMetadata(offset + 1));
                    }

                    // Send offsets to transaction for atomic commit
                    producer.sendOffsetsToTransaction(offsetsToCommit, CONSUMER_GROUP_ID);

                    // Commit transaction
                    producer.commitTransaction();
                } catch (TransactionAbortableException e) {
                    // Abortable Exception: Handle Kafka exception by aborting transaction. producer.abortTransaction() should not throw abortable exception.
                    producer.abortTransaction();
                    resetToLastCommittedPositions(consumer);
                }
            } catch (InvalidConfigurationException e) {
                //  Fatal Error: The error is bubbled up to the application layer. The application can decide what to do
                closeAll();
                throw e;
            } catch (KafkaException | ApplicationRecoverableException e) {
                // Application Recoverable: The application must restart
                closeAll();
                initializeApplication();
            }
        }

    }

Full example can be accessed at: https://github.com/apache/kafka/pull/15913/files 

Note: The producer.abortTransaction() method will not throw an abortable exception to avoid creating a loop.

Compatibility, Deprecation, and Migration Plan

For applications using older code

For existing errors, if the application code already handles the error, it will continue to do so. However, the new TransactionAbortableException  introduced in KIP-890 extends KafkaException, which is treated as fatal by applications. Therefore, we expect TransactionAbortableException  to also be treated as fatal by older code.

Currently, the transactional producer.send returns retriable exception types, such as TimeoutException , which poses a risk of duplicates in Kafka. In this KIP, we will update the transactional producer.send path such that all retriable exceptions will be translated to TransactionAbortableException. Older clients that are using the transactional producer and handling TimeoutException  by retrying the produce operation can update to handle TransactionAbortableException .

For applications using newer code

Applications using newer code, as mentioned in Clientsidecodeexample, have the flexibility to adjust their exception handling logic based on the specific type of exception encountered, enabling them to take appropriate actions.

Handling for OutOfOrderSequenceException  requires additional considerations which is out of scope for this KIP.  OutOfOrderSequenceException  and UnknownProducerIdException  (which extends OutOfOrderSequenceException  ) will not be categorised under any of the defined exception groups in this KIP. This will not impact applications using newer code

...

CorrelationIdMismatchException

...

Application Recoverable

...

Application Recoverable

Compatibility, Deprecation, and Migration Plan

This KIP involves client side changes which only affects the resiliency of new Producer client and Streams. Old clients will continue to be able to use their error handling. Based on type of exception thrown, user needs to change their exception catching logic to take actions against their exception handling.

Test Plan

We will add additional integration and unit tests to check if client acts as expected.

...