[
https://issues.apache.org/jira/browse/FLINK-3679?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15880142#comment-15880142
]
ASF GitHub Bot commented on FLINK-3679:
---------------------------------------
Github user tzulitai commented on a diff in the pull request:
https://github.com/apache/flink/pull/3314#discussion_r102665687
--- Diff:
flink-connectors/flink-connector-kafka-0.8/src/main/java/org/apache/flink/streaming/connectors/kafka/internals/SimpleConsumerThread.java
---
@@ -373,16 +370,28 @@ else if (partitionsRemoved) {
keyPayload.get(keyBytes);
}
- final T value =
deserializer.deserialize(keyBytes, valueBytes,
-
currentPartition.getTopic(), currentPartition.getPartition(), offset);
-
- if
(deserializer.isEndOfStream(value)) {
- // remove
partition from subscribed partitions.
-
partitionsIterator.remove();
- continue
partitionsLoop;
- }
-
- owner.emitRecord(value,
currentPartition, offset);
+ final Collector<T>
collector = new Collector<T>() {
--- End diff --
Moving the discussion back a bit:
I don't think this implementation works correctly with exactly-once and how
we checkpoint the consumer's partition offset state.
The problem is that, in `emitRecord`, we will be updating the offset state.
In the changes here, what this means is that we will be considering a record to
have been fully processed as soon as the collector collects something.
For example, lets say the serializer will call `collect` 3 times for
elements deserialized from record R before `deserialize` returns. R has offset
100L. As soon as the first element is collected, the state will be updated to
`finished processing offset 100L`. If now checkpointing is triggered, and we
use that checkpoint to restore, we will be skipping the remaining 2 elements
that were yet to be collected.
Once
> DeserializationSchema should handle zero or more outputs for every input
> ------------------------------------------------------------------------
>
> Key: FLINK-3679
> URL: https://issues.apache.org/jira/browse/FLINK-3679
> Project: Flink
> Issue Type: Bug
> Components: DataStream API, Kafka Connector
> Reporter: Jamie Grier
> Assignee: Haohui Mai
>
> There are a couple of issues with the DeserializationSchema API that I think
> should be improved. This request has come to me via an existing Flink user.
> The main issue is simply that the API assumes that there is a one-to-one
> mapping between input and outputs. In reality there are scenarios where one
> input message (say from Kafka) might actually map to zero or more logical
> elements in the pipeline.
> Particularly important here is the case where you receive a message from a
> source (such as Kafka) and say the raw bytes don't deserialize properly.
> Right now the only recourse is to throw IOException and therefore fail the
> job.
> This is definitely not good since bad data is a reality and failing the job
> is not the right option. If the job fails we'll just end up replaying the
> bad data and the whole thing will start again.
> Instead in this case it would be best if the user could just return the empty
> set.
> The other case is where one input message should logically be multiple output
> messages. This case is probably less important since there are other ways to
> do this but in general it might be good to make the
> DeserializationSchema.deserialize() method return a collection rather than a
> single element.
> Maybe we need to support a DeserializationSchema variant that has semantics
> more like that of FlatMap.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
