[
https://issues.apache.org/jira/browse/KAFKA-1602?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Joel Koshy updated KAFKA-1602:
------------------------------
Description:
As discussed with Joel offline, today the offset commit response contains an
error code for each partition, but the error code will almost always be the
same across all partitions. Offset commits for a given consumer instance in a
consumer group all go to the same partition of the offsets topic. So there are
only two known cases when we may want it to be different: (i) some partitions
may have very large metadata leading to a metadata-size-too-large error (ii) if
we don't use compression (which will be the case until we have KAFKA-1374)
there could be partial appends. Given that: item (ii) is a corner case and if
we use compression (which we will turn on by default once KAFKA-1374 is done)
item (i) will be eliminated; and that item (i) will be easier to handle by just
failing the whole request and sending back a single error code (of metadata too
large); we can just have a single error-code in the response.
Note that this will be a protocol change and hence we need to bump up the
response version number.
was:
As discussed with Joel offline, today the offset commit response contains an
error code for each partition, but the error code will be actually the same
across all partitions except it is failed due to metadata-size-too-large. This
is because only one partition-append is involved for committing the offsets.
Hence we do not need to use a per-partition error code in the response.
Note that this will be a protocol change and hence we need to bump up the
response version number.
> Use single error code for offset commit response
> ------------------------------------------------
>
> Key: KAFKA-1602
> URL: https://issues.apache.org/jira/browse/KAFKA-1602
> Project: Kafka
> Issue Type: Bug
> Reporter: Guozhang Wang
> Labels: newbie++
>
> As discussed with Joel offline, today the offset commit response contains an
> error code for each partition, but the error code will almost always be the
> same across all partitions. Offset commits for a given consumer instance in a
> consumer group all go to the same partition of the offsets topic. So there
> are only two known cases when we may want it to be different: (i) some
> partitions may have very large metadata leading to a metadata-size-too-large
> error (ii) if we don't use compression (which will be the case until we have
> KAFKA-1374) there could be partial appends. Given that: item (ii) is a corner
> case and if we use compression (which we will turn on by default once
> KAFKA-1374 is done) item (i) will be eliminated; and that item (i) will be
> easier to handle by just failing the whole request and sending back a single
> error code (of metadata too large); we can just have a single error-code in
> the response.
> Note that this will be a protocol change and hence we need to bump up the
> response version number.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
