[
https://issues.apache.org/jira/browse/THRIFT-4469?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
James E. King III updated THRIFT-4469:
--------------------------------------
Affects Version/s: (was: 1.0)
> isServing is not thread safe
> ----------------------------
>
> Key: THRIFT-4469
> URL: https://issues.apache.org/jira/browse/THRIFT-4469
> Project: Thrift
> Issue Type: Bug
> Components: Java - Library
> Affects Versions: 0.9.3, 0.10.0, 0.11.0, 0.12.0
> Reporter: Robert Joseph Evans
> Priority: Major
> Time Spent: 10m
> Remaining Estimate: 0h
>
> We have some integration tests in storm were we create a TServer and call
> serve on it in a background thread, because it is a blocking call.
> Then in the main thread we wait for up to 10 seconds for isServing to return
> success with a 100ms sleep in between each call.
> I don't know if this is a java bug or what, but in some not so rare cases
> isServing does not return true during the entire 10 seconds and the test
> fails. The test prints out a stack trace when this happens and the
> background thread running the server is always at
> https://github.com/apache/thrift/blob/b636ffb613ab49e0f037fbe696d28a4b17a72c5f/lib/java/src/org/apache/thrift/server/AbstractNonblockingServer.java#L95
> happily serving.
> So either we got very unlucky many times and it takes just a few ms longer
> than 10 seconds to come up some times, and < 1 sec other times, or java is
> doing something strange some of the time, and because Server.isServing is not
> volatile the checking thread caches a copy of it in a register and never
> updates it during the 10 seconds of the test run.
> Even if it is a java bug, which I could argue probably is, it would be good
> to make isServing volatile.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
