[
https://issues.apache.org/jira/browse/KAFKA-1124?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13965432#comment-13965432
]
Guozhang Wang commented on KAFKA-1124:
--------------------------------------
Hi Jason, in the new producer the retry would not log as an error, and only
when all retries have exhausted it will report an error. While at the same time
both retry rate and error rate will be reported in DMX. Would this resolve your
problem?
> Sending to a new topic (with auto.create.topics.enable) returns ERROR
> ---------------------------------------------------------------------
>
> Key: KAFKA-1124
> URL: https://issues.apache.org/jira/browse/KAFKA-1124
> Project: Kafka
> Issue Type: Bug
> Affects Versions: 0.8.0, 0.8.1
> Reporter: Jason Rosenberg
> Labels: usability
>
> I had thought this was reported issue, but can't seem to find a previous
> report for it.
> If auto.create.topics.enable is true, a producer still gets an ERROR logged
> on the first attempt to send a message to a new topic, e.g.:
> 2013-11-06 03:00:08,638 ERROR [Thread-1] async.DefaultEventHandler - Failed
> to collate messages by topic, partition due to: Failed to fetch topic
> metadata for topic: mynewtopic
> 2013-11-06 03:00:08,638 INFO [Thread-1] async.DefaultEventHandler - Back off
> for 100 ms before retrying send. Remaining retries = 3
> This usually clears itself up immediately on retry (after 100 ms), as handled
> by the the kafka.producer.async.DefaultEventHandler (with retries enabled).
> However, this is logged to the client as an ERROR, and looks scary, when in
> fact it should have been a normal operation (since we have
> auto.create.topics.enable=true).
> There should be a better interaction here between the producer client and the
> server.
> Perhaps the server can create the topic in flight before returning the
> metadata request.
> Or, if it needs to be asynchronous, it could return a code which indicates
> something like: "The topic doesn't exist yet, it is being created, try again
> shortly".....and have the client automatically retry (even if retries not
> enabled, since it's not an ERROR condition, really).
> The ERROR log level is a problem since apps often have alert systems set up
> to notify when any ERROR happens, etc.....
--
This message was sent by Atlassian JIRA
(v6.2#6252)
