[
https://issues.apache.org/jira/browse/HDDS-347?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16580110#comment-16580110
]
Xiaoyu Yao commented on HDDS-347:
---------------------------------
Thanks [~GeLiXin] for the details. If we consider the order of the state change
and the log output, should we use GenericTestUtils.waitFor the
logCapturer.getOutput().contains the expected message first and then validate
the isContainerClosed state? This way, the wait behavior will be deterministic
with minimal unnecessary sleep.
> Fix : testCloseContainerViaStandaAlone fails sometimes
> ------------------------------------------------------
>
> Key: HDDS-347
> URL: https://issues.apache.org/jira/browse/HDDS-347
> Project: Hadoop Distributed Data Store
> Issue Type: Bug
> Reporter: LiXin Ge
> Assignee: LiXin Ge
> Priority: Major
> Fix For: 0.2.1
>
> Attachments: HDDS-347.000.patch
>
>
> This issue was finded in the automatic JenKins unit test of HDDS-265.
> The container life cycle state is : Open -> Closing -> closed, this test
> submit the container close command and wait for container state change to
> *not equal to open*, actually even when the state condition(not equal to
> open) is satisfied, the container may still in process of closing, so the LOG
> which will printf after the container closed can't be find sometimes and the
> test fails.
> {code:java|title=KeyValueContainer.java|borderStyle=solid}
> try {
> writeLock();
> containerData.closeContainer();
> File containerFile = getContainerFile();
> // update the new container data to .container File
> updateContainerFile(containerFile);
> } catch (StorageContainerException ex) {
> {code}
> Looking at the code above, the container state changes from CLOSING to CLOSED
> in the first step, the remaining *updateContainerFile* may take hundreds of
> milliseconds, so even we modify the test logic to wait for the *CLOSED* state
> will not guarantee the test success, too.
> These are two way to fix this:
> 1, Remove one of the double check which depends on the LOG.
> 2, If we have to preserve the double check, we should wait for the *CLOSED*
> state and sleep for a while to wait for the LOG appears.
> patch 000 is based on the second way.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
