[
https://issues.apache.org/jira/browse/HUDI-5159?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
ASF GitHub Bot updated HUDI-5159:
---------------------------------
Labels: pull-request-available (was: )
> Support write a success file to partition when it finished in flink streaming
> append writer
> -------------------------------------------------------------------------------------------
>
> Key: HUDI-5159
> URL: https://issues.apache.org/jira/browse/HUDI-5159
> Project: Apache Hudi
> Issue Type: New Feature
> Components: connectors
> Reporter: KevinyhZou
> Priority: Major
> Labels: pull-request-available
>
> When we use flink streaming job to consume data from mq and wtite to hudi
> partition table, we can not know when a partition is wite finished. And this
> is often necessary to tell the downstream offline task scheduler to run while
> partition is finished.
> I think we can use the flink watermark mechanism to implment this. As
> watermark represents the minimum timestamp in flink streaming job, when the
> watermark is greater than the hudi partition time, it always means the data
> is write finished to hudi parititon in an ordered streaming data, and then
> it is the time to write a success file to the parititon path to represent it
> finished wirte.
> It can be designed as below.
> # Get the field of partitions and values in flink append streaming data,
> this can be implements in AppendWriteFunction, the emit it to downstream;
> # Implement a SuccessFileWriteSink to receive these partition values and
> store them to activePartitions set,
> # Compare the watermark timestamp and the partition timestamp values
> converted from activeParitions set, if the wartermark is greater, set the
> partition to finished partitons set;
> # Iterate the finished partition set, and get the partition path, and write
> success file to it while flink job make checkpoint;
> # Store the active partition set and finished partition set in flink state,
> avoid the data loss while the job failver.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
