Skip to content

[KAFKA-6730] Simplify State Store Recovery #5013

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
mjsax merged 11 commits into from
Jun 5, 2018
Merged

[KAFKA-6730] Simplify State Store Recovery #5013

Show file tree
Hide file tree
Changes from 9 commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
9b41afa
[KAFKA-6730] Simplify State Store Recovery
May 12, 2018
a129b4e
Making change
May 12, 2018
4883a43
ensuring updatedEndOffsets is called once
May 16, 2018
804a850
Fixing post initialization
May 19, 2018
3963257
Merge branch 'trunk' into KAFKA-6730
ConcurrencyPractitioner May 21, 2018
78d81c3
Adding removal of partitions
May 22, 2018
8207dbe
Removing conditional
May 26, 2018
303fbab
Patching
May 29, 2018
4a33d9c
Strengthening conditional
May 29, 2018
dcb5f0c
Reverting previous unforeseen error
Jun 1, 2018
e742d31
Moving partition addition
Jun 4, 2018
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
64 changes: 20 additions & 44 deletions streams/src/main/java/org/apache/kafka/streams/processor/internals/StoreChangelogReader.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,7 +26,6 @@
import org.apache.kafka.common.utils.LogContext;
import org.apache.kafka.streams.KeyValue;
import org.apache.kafka.streams.errors.StreamsException;
import org.apache.kafka.streams.errors.TaskMigratedException;
import org.apache.kafka.streams.processor.StateRestoreListener;
import org.slf4j.Logger;

Expand All @@ -50,6 +49,7 @@ public class StoreChangelogReader implements ChangelogReader {
private final Map<TopicPartition, StateRestorer> stateRestorers = new HashMap<>();
private final Map<TopicPartition, StateRestorer> needsRestoring = new HashMap<>();
private final Map<TopicPartition, StateRestorer> needsInitializing = new HashMap<>();
private Map<TopicPartition, Long> updatedEndOffsets = new HashMap<>();

public StoreChangelogReader(final Consumer<byte[], byte[]> restoreConsumer,
final StateRestoreListener userStateRestoreListener,
Expand All @@ -66,11 +66,11 @@ public void register(final StateRestorer restorer) {
needsInitializing.put(restorer.partition(), restorer);
}

/**
* @throws TaskMigratedException if another thread wrote to the changelog topic that is currently restored
*/
public Collection<TopicPartition> restore(final RestoringTasks active) {
if (!needsInitializing.isEmpty()) {
final Set<TopicPartition> remainingPartitions = new HashSet<>(needsRestoring.keySet());
remainingPartitions.removeAll(updatedEndOffsets.keySet());
updatedEndOffsets.putAll(restoreConsumer.endOffsets(remainingPartitions));
initialize();
}

Expand All @@ -79,11 +79,23 @@ public Collection<TopicPartition> restore(final RestoringTasks active) {
return completed();
}

final Set<TopicPartition> restoringPartitions = new HashSet<>(needsRestoring.keySet());
try {
final ConsumerRecords<byte[], byte[]> allRecords = restoreConsumer.poll(10);
for (final TopicPartition partition : restoringPartitions) {
restorePartition(allRecords, partition, active.restoringTaskFor(partition));
final ConsumerRecords<byte[], byte[]> records = restoreConsumer.poll(10);
final Iterator<TopicPartition> iterator = needsRestoring.keySet().iterator();
final HashSet<TopicPartition> completedPartitions = new HashSet<>();
while (iterator.hasNext()) {
final TopicPartition partition = iterator.next();
final StateRestorer restorer = stateRestorers.get(partition);
final long pos = processNext(records.records(partition), restorer, updatedEndOffsets.get(partition));
restorer.setRestoredOffset(pos);
if (restorer.hasCompleted(pos, updatedEndOffsets.get(partition))) {

@mjsax mjsax May 22, 2018

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

we can also remove the entry in updatedEndOffsets if this is true.

restorer.restoreDone();
updatedEndOffsets.remove(partition);
completedPartitions.add(partition);

@mjsax mjsax May 29, 2018

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

One more thing -- missed this before: can't we remove completedPartitions and call iterator.remove() instead?

Maybe we need to change the iterator to iterate over the HashMap instead of the "key-set" of the HashMap though.

}
}
for (TopicPartition partition : completedPartitions) {
needsRestoring.remove(partition);
}
} catch (final InvalidOffsetException recoverableException) {
log.warn("Restoring StreamTasks failed. Deleting StreamTasks stores to recreate from scratch.", recoverableException);
Expand Down Expand Up @@ -240,41 +252,6 @@ public void reset() {
needsInitializing.clear();
}

/**
* @throws TaskMigratedException if another thread wrote to the changelog topic that is currently restored
*/
private void restorePartition(final ConsumerRecords<byte[], byte[]> allRecords,
final TopicPartition topicPartition,
final Task task) {
final StateRestorer restorer = stateRestorers.get(topicPartition);
final Long endOffset = endOffsets.get(topicPartition);
final long pos = processNext(allRecords.records(topicPartition), restorer, endOffset);
restorer.setRestoredOffset(pos);
if (restorer.hasCompleted(pos, endOffset)) {
if (pos > endOffset) {
throw new TaskMigratedException(task, topicPartition, endOffset, pos);
}

// need to check for changelog topic
if (restorer.offsetLimit() == Long.MAX_VALUE) {
final Long updatedEndOffset = restoreConsumer.endOffsets(Collections.singletonList(topicPartition)).get(topicPartition);
if (!restorer.hasCompleted(pos, updatedEndOffset)) {
throw new TaskMigratedException(task, topicPartition, updatedEndOffset, pos);
}
}


log.debug("Completed restoring state from changelog {} with {} records ranging from offset {} to {}",
topicPartition,
restorer.restoredNumRecords(),
restorer.startingOffset(),
restorer.restoredOffset());

restorer.restoreDone();
needsRestoring.remove(topicPartition);
}
}

private long processNext(final List<ConsumerRecord<byte[], byte[]>> records,
final StateRestorer restorer,
final Long endOffset) {
Expand Down Expand Up @@ -326,7 +303,6 @@ private boolean hasPartition(final TopicPartition topicPartition) {
return true;
}
}

return false;
}
}
1 change: 0 additions & 1 deletion streams/src/main/java/org/apache/kafka/streams/processor/internals/TaskManager.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -309,7 +309,6 @@ void setConsumer(final Consumer<byte[], byte[]> consumer) {
/**
* @throws IllegalStateException If store gets registered after initialized is already finished
* @throws StreamsException if the store's change log does not contain the partition
* @throws TaskMigratedException if the task producer got fenced or consumer discovered changelog offset changes (EOS only)
*/
boolean updateNewAndRestoringTasks() {
active.initializeNewTasks();
Expand Down
65 changes: 0 additions & 65 deletions .../src/test/java/org/apache/kafka/streams/processor/internals/StoreChangelogReaderTest.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,6 @@
import org.apache.kafka.common.utils.Utils;
import org.apache.kafka.streams.KeyValue;
import org.apache.kafka.streams.errors.StreamsException;
import org.apache.kafka.streams.errors.TaskMigratedException;
import org.apache.kafka.streams.processor.StateRestoreListener;
import org.apache.kafka.test.MockRestoreCallback;
import org.apache.kafka.test.MockStateRestoreListener;
Expand Down Expand Up @@ -377,46 +376,6 @@ public void shouldRestorePartitionsRegisteredPostInitialization() {
assertThat(callbackTwo.restored.size(), equalTo(3));
}

@Test
public void shouldThrowTaskMigratedExceptionIfEndOffsetGetsExceededDuringRestoreForChangelogTopic() {
final int messages = 10;
setupConsumer(messages, topicPartition);
consumer.updateEndOffsets(Collections.singletonMap(topicPartition, 5L));
changelogReader.register(new StateRestorer(topicPartition, restoreListener, null, Long.MAX_VALUE, true, "storeName"));

expect(active.restoringTaskFor(topicPartition)).andReturn(task);
replay(active);

try {
changelogReader.restore(active);
fail("Should have thrown TaskMigratedException");
} catch (final TaskMigratedException expected) {
/* ignore */
}
}


@Test
public void shouldThrowTaskMigratedExceptionIfChangelogTopicUpdatedDuringRestoreProcessFoundInSecondCheck() {
final int messages = 10;
setupConsumer(messages, topicPartition);
// in this case first call to endOffsets returns correct value, but a second thread has updated the changelog topic
// so a subsequent call to endOffsets returns a value exceeding the expected end value
consumer.addEndOffsets(Collections.singletonMap(topicPartition, 15L));
changelogReader.register(new StateRestorer(topicPartition, restoreListener, null, Long.MAX_VALUE, true, "storeName"));

expect(active.restoringTaskFor(topicPartition)).andReturn(task);
replay(active);

try {
changelogReader.restore(active);
fail("Should have thrown TaskMigratedException");
} catch (final TaskMigratedException expected) {
// verifies second block threw exception with updated end offset
assertTrue(expected.getMessage().contains("end offset 15, current offset 10"));
}
}


@Test
public void shouldNotThrowTaskMigratedExceptionIfSourceTopicUpdatedDuringRestoreProcess() {
Expand All @@ -434,30 +393,6 @@ public void shouldNotThrowTaskMigratedExceptionIfSourceTopicUpdatedDuringRestore
}


@Test
public void shouldThrowTaskMigratedExceptionIfEndOffsetGetsExceededDuringRestoreForChangelogTopicEOSEnabled() {
final int totalMessages = 10;
assignPartition(totalMessages, topicPartition);
// records 0..4
addRecords(5, topicPartition, 0);
//EOS enabled commit marker at offset 5 so rest of records 6..10
addRecords(5, topicPartition, 6);
consumer.assign(Collections.<TopicPartition>emptyList());

// end offsets should start after commit marker of 5 from above
consumer.updateEndOffsets(Collections.singletonMap(topicPartition, 6L));
changelogReader.register(new StateRestorer(topicPartition, restoreListener, null, Long.MAX_VALUE, true, "storeName"));

expect(active.restoringTaskFor(topicPartition)).andReturn(task);
replay(active);
try {
changelogReader.restore(active);
fail("Should have thrown task migrated exception");
} catch (final TaskMigratedException expected) {
/* ignore */
}
}

@Test
public void shouldNotThrowTaskMigratedExceptionDuringRestoreForChangelogTopicWhenEndOffsetNotExceededEOSEnabled() {
final int totalMessages = 10;
Expand Down