[
https://issues.apache.org/jira/browse/CASSANDRA-15550?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17036338#comment-17036338
]
Brandon Williams edited comment on CASSANDRA-15550 at 2/13/20 4:37 PM:
-----------------------------------------------------------------------
This one has been around quite some time: ~4 years. In defense of that, it did
sometimes take nearly 300 runs to reproduce it. The cause of this very tight
race is that aborting the session is done asynchronously in nonPeriodicTasks,
so we need to make sure that isn't pending when we call finishTasks since it
checks the {code}completed{code} variable which won't be true until the task is
aborted. Patch here: https://github.com/driftx/cassandra/tree/CASSANDRA-15550
to do this, but it does raise the question to me of should finishTasks actually
_not_ finish if the abort hasn't set completed to true? Why would we need to
call it in idempotent fashion? As evidenced by the test it seems like this is
at least _potential_ for leaking a reference.
With the patch the test passed a couple thousand times.
was (Author: brandon.williams):
This one has been around quite some time: ~4 years. In defense of that, it did
sometimes take nearly 300 runs to reproduce it. The cause of this very tight
race is that aborting the session is done asynchronously in nonPeriodicTasks,
so we need to make sure that isn't pending when we call finishTasks since it
checks the `completed` variable which won't be true until the task is aborted.
Patch here: https://github.com/driftx/cassandra/tree/CASSANDRA-15550 to do
this, but it does raise the question to me of should finishTasks actually _not_
finish if the abort hasn't set completed to true? Why would we need to call it
in idempotent fashion? As evidenced by the test it seems like this is at least
_potential_ for leaking a reference.
With the patch the test passed a couple thousand times.
> Fix flaky test org.apache.cassandra.streaming.StreamTransferTaskTest
> testFailSessionDuringTransferShouldNotReleaseReferences
> ----------------------------------------------------------------------------------------------------------------------------
>
> Key: CASSANDRA-15550
> URL: https://issues.apache.org/jira/browse/CASSANDRA-15550
> Project: Cassandra
> Issue Type: Bug
> Components: Test/unit
> Reporter: David Capwell
> Assignee: Brandon Williams
> Priority: Normal
> Fix For: 4.0-alpha
>
>
> {code}
> junit.framework.AssertionFailedError: expected:<0> but was:<1>
> at
> org.apache.cassandra.streaming.StreamTransferTaskTest.testFailSessionDuringTransferShouldNotReleaseReferences(StreamTransferTaskTest.java:181)
> at
> java.base/jdk.internal.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
> at
> java.base/jdk.internal.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
> at
> java.base/jdk.internal.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
> {code}
> Failure was on java 11
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
