[ 
https://issues.apache.org/jira/browse/IGNITE-25599?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17956597#comment-17956597
 ] 

Sergey Chugunov edited comment on IGNITE-25599 at 6/6/25 1:54 PM:
------------------------------------------------------------------

My findings when fixing this test.
Test name itself suggests that it is about proper handling of discard messages 
that are responsible for removing pending messages from internal queue in 
ServerImpl. But the test implementation is rather vague and doesn't make it 
obvious what is checked and how these checks are done.

It looks more like a smoke test verifying that pending messages are not resent 
after being discarded in situation when a node in cluster is restarted. I 
examined an unmodified version of the test and found out that when it fails 
locally, there are always some not discarded pending messages. It means that 
there is a race within the test. But at the same time closing the race makes 
the test look weird, like it verifies nothing. Well, if we allow pending 
messages to be discarded (basically what we need to do to close the race), we 
don't expect these messages to be resent again.

One possible explanation for the existence of the test is that at some point in 
time pending messages functionality wasn't stable enough, and already discarded 
messages could indeed be resent after another node restart. Anyway, I see this 
test as a useful one as pending messages functionality is pretty complicated, 
and even smoke tests don't hurt, especially a lightweight ones.




was (Author: sergeychugunov):
My findings when fixing this test.
Test name itself suggests that it is about proper handling of discard messages 
that enable to remove pending messages from internal queue in ServerImpl. But 
the test implementation is rather vague and doesn't make it obvious what is 
checked and how these checks are done.

It looks more like a smoke test verifying that pending messages are not resent 
after being discarded in situation when a node in cluster is restarted. I 
examined an unmodified version of the test and found out that when it fails 
locally, there are always some not discarded pending messages. It means that 
there is a race within the test. But at the same time closing the race makes 
the test look weird, like it verifies nothing. Well, if we allow pending 
messages to be discarded (basically what we need to do to close the race), we 
don't expect these messages to be resent again.

One possible explanation for the existence of the test is that at some point in 
time pending messages functionality wasn't stable enough, and already discarded 
messages could indeed be resent after another node restart. Anyway, I see this 
test as a useful one as pending messages functionality is pretty complicated, 
and even smoke tests don't hurt, especially a lightweight ones.



> TcpDiscoverySslTrustedSelfTest is flaky on TC
> ---------------------------------------------
>
>                 Key: IGNITE-25599
>                 URL: https://issues.apache.org/jira/browse/IGNITE-25599
>             Project: Ignite
>          Issue Type: Bug
>            Reporter: Sergey Chugunov
>            Assignee: Sergey Chugunov
>            Priority: Major
>             Fix For: 2.18
>
>          Time Spent: 20m
>  Remaining Estimate: 0h
>
> Test testDiscoveryEventsDiscard is [flaky on master 
> branch|https://ci2.ignite.apache.org/project.html?projectId=IgniteTests24Java8&testNameId=3386094243324005637&tab=testDetails&branch_IgniteTests24Java8=%3Cdefault%3E]
>  with failure rate ~ 30%.
> Failures are reproducible locally.
> Test fails with an assertion in test code as test TcpDiscoverySpi declares it 
> sees a duplicated messages being sent:
> {code:java}
> [ERROR][tcp-disco-msg-worker-[crd]-#49%tcp.TcpDiscoverySelfTest0%-#813%tcp.TcpDiscoverySelfTest0%][TcpDiscoverySelfTest$TestEventDiscardSpi]
>  Send duplicated message: TcpDiscoveryCustomEventMessage [msg=null, 
> super=TcpDiscoveryAbstractMessage 
> [sndNodeId=ad7215e8-5eb2-4ccc-aa7c-2bf12d500000, 
> id=a9ad4304791-ad7215e8-5eb2-4ccc-aa7c-2bf12d500000, 
> verifierNodeId=ad7215e8-5eb2-4ccc-aa7c-2bf12d500000, topVer=2, pendingIdx=0, 
> failedNodes=null, isClient=false]]
> {code}



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

Reply via email to