kafka-connect-hdfs failure due to corrupt WAL file

[Prabhu V <vp...@gmail.com>]

Hi,

I am using kafka-connect-hdfs in a 2 nodes and one of the nodes had to be
rebooted when the process was running.

Upon restart the process fails with

16/08/02 21:43:30 ERROR hdfs.TopicPartitionWriter: Recovery failed at state
RECOVERY_PARTITION_PAUSED
org.apache.kafka.connect.errors.ConnectException: java.io.IOException:
Filesystem closed
        at io.confluent.connect.hdfs.wal.FSWAL.apply(FSWAL.java:131)

I looked into the code and looks like when the WAL log file may be
corrupt/unreadeable, it causes this issue.

Can I delete the WAL files to fix this issue ? I dont mind duplicate events.

Thanks,
Prabhu

Re: kafka-connect-hdfs failure due to corrupt WAL file

[Ewen Cheslack-Postava <ew...@confluent.io>]

If you're not worried about duplicates, then yes, you can delete the WAL to
recover.

By the way, we're aware of some issues in that code that we're addressing
here: https://github.com/confluentinc/kafka-connect-hdfs/pull/96

-Ewen

On Tue, Aug 2, 2016 at 3:12 PM, Prabhu V <vp...@gmail.com> wrote:

> Hi,
>
> I am using kafka-connect-hdfs in a 2 nodes and one of the nodes had to be
> rebooted when the process was running.
>
> Upon restart the process fails with
>
> 16/08/02 21:43:30 ERROR hdfs.TopicPartitionWriter: Recovery failed at state
> RECOVERY_PARTITION_PAUSED
> org.apache.kafka.connect.errors.ConnectException: java.io.IOException:
> Filesystem closed
>         at io.confluent.connect.hdfs.wal.FSWAL.apply(FSWAL.java:131)
>
> I looked into the code and looks like when the WAL log file may be
> corrupt/unreadeable, it causes this issue.
>
> Can I delete the WAL files to fix this issue ? I dont mind duplicate
> events.
>
> Thanks,
> Prabhu
>



-- 
Thanks,
Ewen