[ https://issues.apache.org/jira/browse/STORM-391?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14236581#comment-14236581 ]
ASF GitHub Bot commented on STORM-391: -------------------------------------- Github user harshach commented on the pull request: https://github.com/apache/storm/pull/338#issuecomment-65884958 @Lewuathe its just one kafka request right? . KafkaSpout makes one request if that request fails than you'll get a exception and if that exception happens to be UnknownTopic than you can do the sleep like you have it now. IMO its better to rely on Kafka api and accept their exception than have a work around using zookeeper. Also current code assumes that users always going to use kafka zookeeper as part KafkaSpout config. This is not a right assumption users can also give broker hosts instead kafka hosts. In that case KafkaSpout uses storm zookeepers to store the offset info. > KafkaSpout to await for the topic > --------------------------------- > > Key: STORM-391 > URL: https://issues.apache.org/jira/browse/STORM-391 > Project: Apache Storm > Issue Type: Improvement > Affects Versions: 0.9.2-incubating > Reporter: Alexey Raga > Assignee: Kai Sasaki > Labels: features > > When topic does not yet exist and the consumer is asked to consume from it, > the default behaviour for Kafka heigh-level consumer is to "await" for the > topic without a failure. > KafkaSpout currently fails trying to get the partition information about the > topic that does not exist. > It may be a good idea to have the same common behaviour in KafkaSpout and it > can probably be implemented through the zookeeper watchers: if topic does not > exist, then set up a watcher and don't do anything until it yields. -- This message was sent by Atlassian JIRA (v6.3.4#6332)