Xeli commented on a change in pull request #6594: [FLINK-9311] [pubsub] Added
PubSub source connector with support for checkpointing (ATLEAST_ONCE)
URL: https://github.com/apache/flink/pull/6594#discussion_r257503663
##########
File path:
flink-streaming-java/src/main/java/org/apache/flink/streaming/api/functions/source/MessageAcknowledgingSourceBase.java
##########
@@ -148,16 +149,12 @@ public void
initializeState(FunctionInitializationContext context) throws Except
if (context.isRestored()) {
LOG.info("Restoring state for the {}.",
getClass().getSimpleName());
- List<SerializedCheckpointData[]> retrievedStates = new
ArrayList<>();
+ List<Map<Long, Set<UId>>> retrievedStates = new
ArrayList<>();
for (SerializedCheckpointData[] entry :
this.checkpointedState.get()) {
- retrievedStates.add(entry);
+
retrievedStates.add(SerializedCheckpointData.toDeque(entry, idSerializer));
}
+ pendingCheckpoints =
SerializedCheckpointData.combine(retrievedStates);
- // given that the parallelism of the function is 1, we
can only have at most 1 state
Review comment:
To give some more context for these changes:
The MessageAcknowledingSourceBase can deduplicate messages it has already
seen because it keeps a list of messageIds. The PubSubSource actually uses this
too:
https://github.com/apache/flink/pull/6594/files#diff-5d1317f139d297f4e67ff13defa4d66cR119
The thing is because the list isn't shared and pubsub (nor RabbitMQ afaik)
won't guarantee the message will be sent to the same taskmanager this
deduplication will only work with a parallelism of one.
In the case of this pubsub connector we only guarantee `at least once` so we
like to be able to use a higher parallelism. It still tries to deduplicate but
it is more of a best effort.
Though this actually does mean that when you're using the pubsub connector
with a parallelism of 1 it becomes exactly-once.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
