[jira] [Updated] (STORM-73) TridentTopology with a RichSpout and tuples that fail

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

     [ https://issues.apache.org/jira/browse/STORM-73?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Rick Kellogg updated STORM-73:
------------------------------
    Component/s: storm-core

> TridentTopology with a RichSpout and tuples that fail
> -----------------------------------------------------
>
>                 Key: STORM-73
>                 URL: https://issues.apache.org/jira/browse/STORM-73
>             Project: Apache Storm
>          Issue Type: Bug
>          Components: storm-core
>            Reporter: James Xu
>            Priority: Minor
>
> https://github.com/nathanmarz/storm/issues/323
> Please see the following gist for a testcase: https://gist.github.com/3679478
> Tested this with 0.8.1
> What this Spout does is keep a counter for the tuples it emits and splits them into sequences. The the ack() and fail() methods expect that they receive exactly the number of ack or fail calls you would expect as emited for each sequence.
> But watching the console, after around 13 sequences:
> Acking sequence: 0
> Acking sequence: 1
> Acking sequence: 2
> Acking sequence: 3
> Acking sequence: 4
> Acking sequence: 5
> Acking sequence: 6
> Acking sequence: 7
> Acking sequence: 8
> Acking sequence: 9
> Acking sequence: 10
> Acking sequence: 11
> Acking sequence: 12
> Acking sequence: 13
> Failing sequence: 15
> Some sequence didn't fail completly!! {14=396}
> Failing sequence: 16
> Some sequence didn't fail completly!! {14=396}
> Failing sequence: 17
> etc... etc.. Keeps failing from now on
> -----------
> barrywhart: This sounds very similar to a problem we were having last week on 0.8.0 on a conventional topology (non-Trident) written in Python. The spout would hang forever after about 100 messages.
> In our case we were able to downgrade to 0.7.4, and it has been working fine since.



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