[
https://issues.apache.org/jira/browse/HUDI-1007?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17275658#comment-17275658
]
sivabalan narayanan commented on HUDI-1007:
-------------------------------------------
[~liujinhui]: So, you mean to say with spark2.4, and w/ config "
spark.streaming.kafka.allowNonConsecutiveOffsets" set, you are still hitting
the exception? in other words, if offsets sent to Kafka to consume have been
compacted, fetching from Kafka fails? Can you try using a vanilla Kafka
consumer with some older offset (which was compacted) and see what happens. I
don't think we need to make changes in Hudi/deltastreamer for this. Kafka
should the one moving the offset to some valid offset and start serving msgs
from there.
> When earliestOffsets is greater than checkpoint, Hudi will not be able to
> successfully consume data
> ---------------------------------------------------------------------------------------------------
>
> Key: HUDI-1007
> URL: https://issues.apache.org/jira/browse/HUDI-1007
> Project: Apache Hudi
> Issue Type: Bug
> Components: DeltaStreamer
> Reporter: liujinhui
> Assignee: liujinhui
> Priority: Major
> Labels: user-support-issues
> Fix For: 0.8.0
>
> Original Estimate: 24h
> Remaining Estimate: 24h
>
> Use deltastreamer to consume kafka,
> When earliestOffsets is greater than checkpoint, Hudi will not be able to
> successfully consume data
> org.apache.hudi.utilities.sources.helpers.KafkaOffsetGen#checkupValidOffsets
> boolean checkpointOffsetReseter = checkpointOffsets.entrySet().stream()
> .anyMatch(offset -> offset.getValue() <
> earliestOffsets.get(offset.getKey()));
> return checkpointOffsetReseter ? earliestOffsets : checkpointOffsets;
> Kafka data is continuously generated, which means that some data will
> continue to expire.
> When earliestOffsets is greater than checkpoint, earliestOffsets will be
> taken. But at this moment, some data expired. In the end, consumption fails.
> This process is an endless cycle. I can understand that this design may be to
> avoid the loss of data, but it will lead to such a situation, I want to fix
> this problem, I want to hear your opinion
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
