[jira] [Comment Edited] (KAFKA-4675) Subsequent CreateTopic command could be lost after a DeleteTopic command

["huxi (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/KAFKA-4675?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15833336#comment-15833336 ] 

huxi edited comment on KAFKA-4675 at 1/22/17 7:26 AM:
------------------------------------------------------

Failed to reproduce this issue with snippet below:

{code}
val zkUtils = ZkUtils("localhost:2181", 30000, 30000, false)
AdminUtils.deleteTopic(zkUtils, "old-topic")
AdminUtils.createTopic(zkUtils, "new-topic", 1, 1)
{code}

After running the code above, topic "new-topic" will be created. Seems subsequent CreateTopic command has still been invoked. [~guozhang] Does the code reflect the way you run into this problem?


was (Author: huxi_2b):
Failed to reproduce this issue with snippet below:

{code}
val zkUtils = ZkUtils("localhost:2181", 30000, 30000, false)
AdminUtils.deleteTopic(zkUtils, "old-topic")
AdminUtils.createTopic(zkUtils, "new-topic", 1, 1)
{code}

After running the code above, topic "new-topic" will be created. Seems subsequent CreateTopic command has still been invoked. Does the code reflect the way you run into this problem?

> Subsequent CreateTopic command could be lost after a DeleteTopic command
> ------------------------------------------------------------------------
>
>                 Key: KAFKA-4675
>                 URL: https://issues.apache.org/jira/browse/KAFKA-4675
>             Project: Kafka
>          Issue Type: Bug
>            Reporter: Guozhang Wang
>              Labels: admin
>
> This is discovered while investigating KAFKA-3896: If an admin client sends a delete topic command and a create topic command consecutively, even if it wait for the response of the previous command before issuing the second, there is still a race condition that the create topic command could be "lost".
> This is because currently these commands are all asynchronous as defined in KIP-4, and controller will return the response once it has written the corresponding data to ZK path, which can be handled by different listener threads at different paces, and if the thread handling create is faster than the other, the executions could be effectively re-ordered.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)