adixitconfluent commented on code in PR #18093:
URL: https://github.com/apache/kafka/pull/18093#discussion_r1874487856
##########
core/src/main/java/kafka/server/share/SharePartition.java:
##########
@@ -371,52 +370,35 @@ public static RecordState forId(byte id) {
*/
public CompletableFuture<Void> maybeInitialize() {
log.debug("Maybe initialize share partition: {}-{}", groupId,
topicIdPartition);
- CompletableFuture<Void> future = new CompletableFuture<>();
- AtomicReference<Optional<Throwable>> futureException = new
AtomicReference<>(Optional.empty());
// Check if the share partition is already initialized.
- InitializationResult initializationResult =
checkInitializationCompletion();
- if (initializationResult.isComplete()) {
- if (initializationResult.throwable() != null) {
- future.completeExceptionally(initializationResult.throwable());
- } else {
- future.complete(null);
+ try {
+ if (initializedOrThrowException()) {
+ return CompletableFuture.completedFuture(null);
}
- return future;
+ } catch (Exception e) {
+ return CompletableFuture.failedFuture(e);
}
+ // If code reaches here then the share partition is not initialized.
Initialize the share partition.
// All the pending requests should wait to get completed before the
share partition is initialized.
// Attain lock to avoid any concurrent requests to be processed.
lock.writeLock().lock();
- boolean shouldFutureBeCompleted = false;
try {
// Re-check the state to verify if previous requests has already
initialized the share partition.
- initializationResult = checkInitializationCompletion();
- if (initializationResult.isComplete()) {
- if (initializationResult.throwable() != null) {
-
futureException.set(Optional.of(initializationResult.throwable()));
- }
- shouldFutureBeCompleted = true;
- return future;
+ if (initializedOrThrowException()) {
+ return CompletableFuture.completedFuture(null);
Review Comment:
We should not complete the future while holding the write lock here as well.
As mentioned by @AndrewJSchofield in this comment
https://github.com/apache/kafka/pull/18053#discussion_r1871156905, this should
be completed post releasing the lock, else it may cause problems for the
future. Same for the future completion in the `catch` block.
##########
core/src/main/java/kafka/server/share/SharePartition.java:
##########
@@ -1178,32 +1160,20 @@ private boolean stateNotActive() {
return partitionState() != SharePartitionState.ACTIVE;
}
- private void completeInitializationWithException() {
- lock.writeLock().lock();
- try {
- partitionState = SharePartitionState.FAILED;
- } finally {
- lock.writeLock().unlock();
- }
- }
-
- private InitializationResult checkInitializationCompletion() {
+ private boolean initializedOrThrowException() {
SharePartitionState currentState = partitionState();
- switch (currentState) {
- case ACTIVE:
- return new InitializationResult(true);
- case FAILED:
- return new InitializationResult(true, new
IllegalStateException(String.format("Share partition failed to load %s-%s",
groupId, topicIdPartition)));
- case INITIALIZING:
- return new InitializationResult(true, new
LeaderNotAvailableException(String.format("Share partition is already
initializing %s-%s", groupId, topicIdPartition)));
- case FENCED:
- return new InitializationResult(true, new
FencedStateEpochException(String.format("Share partition is fenced %s-%s",
groupId, topicIdPartition)));
- case EMPTY:
- // Do not complete the future as the share partition is not
yet initialized.
- return new InitializationResult(false);
- default:
Review Comment:
Any particular reason why we've removed the default case now?
--
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]
