[
https://issues.apache.org/jira/browse/KAFKA-7994?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16845381#comment-16845381
]
Sophie Blee-Goldman commented on KAFKA-7994:
--------------------------------------------
Hi [~Yohan123], I think we need to be careful in assuming a singular view of
streamtime across tasks or even within a single task. Rather than it being an
obstacle that different subtopologies can't "talk" to one another and pass
along a single stream time, I think this actually enforces correctness – each
node has its own sense of time and it doesn't make sense for them to look
upstream for the time as seen by a different node. See
[https://github.com/apache/kafka/pull/6278#|https://github.com/apache/kafka/pull/6278]
> Improve Stream-Time for rebalances and restarts
> -----------------------------------------------
>
> Key: KAFKA-7994
> URL: https://issues.apache.org/jira/browse/KAFKA-7994
> Project: Kafka
> Issue Type: Bug
> Components: streams
> Reporter: Matthias J. Sax
> Assignee: Richard Yu
> Priority: Major
> Attachments: possible-patch.diff
>
>
> We compute a per-partition partition-time as the maximum timestamp over all
> records processed so far. Furthermore, we use partition-time to compute
> stream-time for each task as maximum over all partition-times (for all
> corresponding task partitions). This stream-time is used to make decisions
> about processing out-of-order records or drop them if they are late (ie,
> timestamp < stream-time - grace-period).
> During rebalances and restarts, stream-time is initialized as UNKNOWN (ie,
> -1) for tasks that are newly created (or migrated). In net effect, we forget
> current stream-time for this case what may lead to non-deterministic behavior
> if we stop processing right before a late record, that would be dropped if we
> continue processing, but is not dropped after rebalance/restart. Let's look
> at an examples with a grade period of 5ms for a tumbling windowed of 5ms, and
> the following records (timestamps in parenthesis):
>
> {code:java}
> r1(0) r2(5) r3(11) r4(2){code}
> In the example, stream-time advances as 0, 5, 11, 11 and thus record `r4` is
> dropped as late because 2 < 6 = 11 - 5. However, if we stop processing or
> rebalance after processing `r3` but before processing `r4`, we would
> reinitialize stream-time as -1, and thus would process `r4` on restart/after
> rebalance. The problem is, that stream-time does advance differently from a
> global point of view: 0, 5, 11, 2.
> Note, this is a corner case, because if we would stop processing one record
> earlier, ie, after processing `r2` but before processing `r3`, stream-time
> would be advance correctly from a global point of view.
> A potential fix would be, to store latest observed partition-time in the
> metadata of committed offsets. Thus way, on restart/rebalance we can
> re-initialize time correctly.
> Notice that this particular issue applies for all Stream Tasks in the
> topology. The further down the DAG records flow, the more likely it is that
> the StreamTask will have an incorrect stream time. For instance, if r3 was
> filtered out, the tasks receiving the processed records will compute the
> stream time as 5 instead of the correct timestamp being 11. This entails us
> to also propagate the latest observed partition time as well downstream.
> That means the sources located at the head of the topology must forward the
> partition time to its subtopologies whenever records are sent.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
