[jira] [Updated] (KAFKA-7994) Improve Stream-Time for rebalances and restarts

["Sophie Blee-Goldman (Jira)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/KAFKA-7994?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Sophie Blee-Goldman updated KAFKA-7994:
---------------------------------------
    Description: 
We compute a per-partition partition-time as the maximum timestamp over all records processed so far. Before 2.3 this was used to determine the logical stream-time used to make decisions about processing out-of-order records or drop them if they are late (ie, timestamp < stream-time - grace-period). Preserving the stream-time is necessary to ensure deterministic results (see KAFKA-9368), and although the processor-time is now used instead of partition-time, preserving the partition-time is a first step towards improving the overall stream-time semantics.

The partition-time is also used by the TimestampExtractor. It gets passed in to #extract and can be used to determine a rough timestamp estimate if the actual timestamp is missing, corrupt, etc. This means in the corner case where the next record to be processed after a rebalance/restart cannot have its actual timestamp determined, we have no idea way of coming up with a reasonable guess and the record will likely have to be dropped.

 

A potential fix would be, to store latest observed partition-time in the metadata of committed offsets. This way, on restart/rebalance we can re-initialize partition-time correctly.

  was:
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.


> 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
>             Fix For: 2.4.0
>
>         Attachments: possible-patch.diff
>
>
> We compute a per-partition partition-time as the maximum timestamp over all records processed so far. Before 2.3 this was used to determine the logical stream-time used to make decisions about processing out-of-order records or drop them if they are late (ie, timestamp < stream-time - grace-period). Preserving the stream-time is necessary to ensure deterministic results (see KAFKA-9368), and although the processor-time is now used instead of partition-time, preserving the partition-time is a first step towards improving the overall stream-time semantics.
> The partition-time is also used by the TimestampExtractor. It gets passed in to #extract and can be used to determine a rough timestamp estimate if the actual timestamp is missing, corrupt, etc. This means in the corner case where the next record to be processed after a rebalance/restart cannot have its actual timestamp determined, we have no idea way of coming up with a reasonable guess and the record will likely have to be dropped.
>  
> A potential fix would be, to store latest observed partition-time in the metadata of committed offsets. This way, on restart/rebalance we can re-initialize partition-time correctly.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)