[
https://issues.apache.org/jira/browse/KAFKA-2880?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15023729#comment-15023729
]
Jason Gustafson commented on KAFKA-2880:
----------------------------------------
[~ewencp] Looks like we're missing a disconnect check on the client response.
Looking over the code, error handling for the listTopics() and partitionsFor()
API clearly hasn't been given enough attention. How about we expand the scope
of this ticket to include the following cases?
1. Unauthorized topics: raise TopicAuthorizationException
2. Miscellaneous topic errors (e.g. leader/replica not available): retry until
request timeout expires
3. Disconnect errors: retry until request timeout expires
4. Request timeout: raise TimeoutException
> Fetcher.getTopicMetadata NullPointerException when broker cannot be reached
> ---------------------------------------------------------------------------
>
> Key: KAFKA-2880
> URL: https://issues.apache.org/jira/browse/KAFKA-2880
> Project: Kafka
> Issue Type: Bug
> Components: clients
> Affects Versions: 0.9.0.0
> Reporter: Ewen Cheslack-Postava
> Assignee: Jason Gustafson
>
> The Fetcher class will throw a NullPointerException if a broker cannot be
> reached:
> {quote}
> Exception in thread "main" java.lang.NullPointerException
> at
> org.apache.kafka.common.requests.MetadataResponse.<init>(MetadataResponse.java:130)
> at
> org.apache.kafka.clients.consumer.internals.Fetcher.getTopicMetadata(Fetcher.java:203)
> at
> org.apache.kafka.clients.consumer.KafkaConsumer.partitionsFor(KafkaConsumer.java:1143)
> at
> org.apache.kafka.connect.util.KafkaBasedLog.start(KafkaBasedLog.java:126)
> at
> org.apache.kafka.connect.storage.KafkaOffsetBackingStore.start(KafkaOffsetBackingStore.java:85)
> at org.apache.kafka.connect.runtime.Worker.start(Worker.java:108)
> at org.apache.kafka.connect.runtime.Connect.start(Connect.java:56)
> at
> org.apache.kafka.connect.cli.ConnectDistributed.main(ConnectDistributed.java:62)
> {quote}
> This is trivially reproduced by trying to start Kafka Connect in distributed
> mode (i.e. connect-distributed.sh config/connect-distributed.properties) with
> no broker running. However, it's not specific to Kafka Connect, it just
> happens to use the consumer in a way that triggers it reliably.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
