[
https://issues.apache.org/jira/browse/STORM-1549?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15175682#comment-15175682
]
ASF GitHub Bot commented on STORM-1549:
---------------------------------------
Github user revans2 commented on a diff in the pull request:
https://github.com/apache/storm/pull/1174#discussion_r54730661
--- Diff: storm-core/src/jvm/org/apache/storm/task/OutputCollector.java ---
@@ -218,6 +218,16 @@ public void fail(Tuple input) {
_delegate.fail(input);
}
+ /**
+ * Resets the message timeout for any tuple trees to which the given
tuple belongs.
+ * The timeout is reset to Config.TOPOLOGY_MESSAGE_TIMEOUT_SECS.
+ * @param input the tuple to reset timeout for
+ */
--- End diff --
Can we add to the javadoc that this should be used sparingly? This is not
a cheap operation, so having someone do it for each tuple they receive is very
expensive.
> Add support for extending tuple tree timeout
> --------------------------------------------
>
> Key: STORM-1549
> URL: https://issues.apache.org/jira/browse/STORM-1549
> Project: Apache Storm
> Issue Type: New Feature
> Components: storm-core
> Reporter: Stig Rohde Døssing
> Assignee: Stig Rohde Døssing
> Priority: Minor
>
> During the discussion of https://github.com/apache/storm/pull/700 the issue
> of allowing timeout extension in case of unavailable external components
> (such as a web service) came up.
> The current implementation makes tuples fail at a set interval, regardless of
> whether or not replaying them is necessary. This can be irritating in
> topologies that emit to multiple services, since one hanging service will
> cause replays to hit all the working services as well.
> I suggest adding a resetTimeout function to IOutputCollector, which will make
> the relevant ackers and spouts reinsert the tuple tree information in their
> pending maps.
> The intended usage is that a bolt can call this function on an interval if it
> needs to delay expiration, for example if it needs to retry calling a web
> server a few times. It may also be useful for slow topologies that want Storm
> to detect hanging/dropped tuples faster than the max expected complete
> latency of the topology.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
