[
https://issues.apache.org/jira/browse/THRIFT-3891?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15837187#comment-15837187
]
ASF GitHub Bot commented on THRIFT-3891:
----------------------------------------
Github user jeking3 commented on a diff in the pull request:
https://github.com/apache/thrift/pull/1080#discussion_r97708735
--- Diff: lib/cpp/src/thrift/server/TNonblockingServer.cpp ---
@@ -1524,7 +1532,7 @@ void TNonblockingIOThread::breakLoop(bool error) {
}
// sets a flag so that the loop exits on the next event
- event_base_loopbreak(eventBase_);
+ event_base_loopexit(eventBase_, 0);
--- End diff --
I am curious to understand why this was changed. event_base_loopexit will
cause the lop to continue processing until drained. This function is called
breakLoop and I assume that means stop as soon as possible, which is what it
was doing.
> TNonblockingServer configured with more than one IO threads does not always
> return from serve() upon stop()
> -----------------------------------------------------------------------------------------------------------
>
> Key: THRIFT-3891
> URL: https://issues.apache.org/jira/browse/THRIFT-3891
> Project: Thrift
> Issue Type: Bug
> Components: C++ - Library
> Affects Versions: 0.9.3
> Reporter: Buğra Gedik
> Priority: Minor
> Attachments: patch.diff
>
>
> Using {{TNonblockingServer}}, when the number of IO threads is > 1, there is
> race condition in which {{stop()}} does not properly unblock {{serve()}}.
> The problem manifests itself when {{stop()}} is called (obviously from a
> different thread) soon after {{serve()}}.
> The core issue is that, {{event_base_loopbreak()}} is called within the
> {{stop()}} sequence without checking whether the IO thread has actually
> entered its event loop. The documentation of {{event_base_loopbreak()}} says
> (http://www.wangafu.net/~nickm/libevent-book/Ref3_eventloop.html)
> {quote}
> Note also that event_base_loopexit(base,NULL) and event_base_loopbreak(base)
> act differently when no event loop is running: loopexit schedules the next
> instance of the event loop to stop right after the next round of callbacks
> are run (as if it had been invoked with EVLOOP_ONCE) whereas loopbreak only
> stops a currently running loop, and has no effect if the event loop isn’t
> running.
> {quote}
> Attached is a patch (against the released 0.9.3 version of the codebase).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
