lnbest0707 commented on code in PR #278:
URL:
https://github.com/apache/flink-connector-kafka/pull/278#discussion_r3483705226
##########
flink-connector-kafka/src/main/java/org/apache/flink/connector/kafka/dynamic/source/enumerator/DynamicKafkaSourceEnumerator.java:
##########
@@ -728,27 +747,208 @@ public void addSplitsBack(List<DynamicKafkaSourceSplit>
splits, int subtaskId) {
clusterEnumeratorMap
.get(kafkaClusterId)
.addSplitsBack(kafkaPartitionSplits.get(kafkaClusterId), subtaskId);
+ } else if (failOnInactiveCluster) {
+ throw new IllegalStateException(
+ String.format(
+ "Cannot reassign split for active cluster %s
because its"
+ + " enumerator is unavailable",
+ kafkaClusterId));
} else {
logger.warn(
"Split refers to inactive cluster {} with current
clusters being {}",
kafkaClusterId,
clusterEnumeratorMap.keySet());
}
}
-
- handleNoMoreSplits();
}
/** NOTE: this happens at startup and failover. */
@Override
public void addReader(int subtaskId) {
logger.debug("Adding reader {}", subtaskId);
- splitAssignmentStrategy.onReaderAdded(subtaskId);
+ ReaderInfo readerInfo = enumContext.registeredReaders().get(subtaskId);
+ if (readerInfo != null) {
+ List<DynamicKafkaSourceSplit> reportedSplits =
+ readerInfo.getReportedSplitsOnRegistration();
+ if (!reportedSplits.isEmpty()) {
+ pendingReportedSplitsByReader.put(subtaskId, new
ArrayList<>(reportedSplits));
+ }
+ }
+
+ if (tryCompletePendingReaderRegistration()) {
+ return;
+ }
+
+ addReaderToClusterEnumerators(subtaskId);
+ handleNoMoreSplits();
+ }
+
+ private boolean tryCompletePendingReaderRegistration() {
+ boolean hasPendingRecovery =
+ initialReaderRegistrationPending ||
!pendingReportedSplitsByReader.isEmpty();
+ if (!hasPendingRecovery) {
+ return false;
+ }
+ if (!firstDiscoveryComplete || !allReadersRegistered()) {
+ return true;
+ }
+
+ if (initialReaderRegistrationPending) {
+ initialReaderRegistrationPending = false;
+ }
+ if (!pendingReportedSplitsByReader.isEmpty()) {
+ reassignReportedSplits();
+ } else {
+ flushPendingSplitAssignmentsForRegisteredReaders();
+ }
+ handleNoMoreSplits();
+ flushPendingMetadataUpdateEvents();
+ return true;
+ }
+
+ private boolean allReadersRegistered() {
+ return enumContext.registeredReaders().size() ==
enumContext.currentParallelism();
+ }
- // assign pending splits from the sub enumerator
+ private void addReaderToClusterEnumerators(int subtaskId) {
+ splitAssignmentStrategy.onReaderAdded(subtaskId);
clusterEnumeratorMap.forEach(
(cluster, subEnumerator) ->
subEnumerator.addReader(subtaskId));
- handleNoMoreSplits();
+ }
+
+ private void flushPendingSplitAssignmentsForRegisteredReaders() {
+ List<Integer> registeredReaders = new
ArrayList<>(enumContext.registeredReaders().keySet());
+ Collections.sort(registeredReaders);
+ for (int readerId : registeredReaders) {
+ addReaderToClusterEnumerators(readerId);
+ }
+ }
+
+ private void reassignReportedSplits() {
+ Map<String, ReportedSplit> activeReportedSplits = new TreeMap<>();
+ Map<Integer, List<DynamicKafkaSourceSplit>> retainedSplitsByReader =
new TreeMap<>();
+ long currentTimeMillis = System.currentTimeMillis();
+
+ for (Entry<Integer, List<DynamicKafkaSourceSplit>> readerSplits :
+ new TreeMap<>(pendingReportedSplitsByReader).entrySet()) {
+ int readerId = readerSplits.getKey();
+ for (DynamicKafkaSourceSplit split : readerSplits.getValue()) {
+ if (isSplitActive(split)) {
+ DynamicKafkaSourceSplit activeSplit =
split.clearRetention();
+ ReportedSplit previous =
+ activeReportedSplits.putIfAbsent(
+ activeSplit.splitId(),
+ new ReportedSplit(activeSplit, readerId));
+ if (previous != null) {
Review Comment:
This part may have some conflict with FLINK-39837/#279 as that one is
applying the max-offset merge on retained offsets.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
