[
https://issues.apache.org/jira/browse/CASSANDRA-15543?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17056278#comment-17056278
]
Ekaterina Dimitrova edited comment on CASSANDRA-15543 at 3/10/20, 7:27 PM:
---------------------------------------------------------------------------
Hi Kevin,
"In any case, I believe passing an immutable copy of the
{{failureReasonByEndpoint}} map to the constructor of
Read/WriteFailureException would reduce the chances for the {{number of
failures}} and the failure messages to be inconsistent."
Agree with you. I would personally expect 1 response from node1 where the
schema was altered and two failures with the two messages.
"In addition to that, there's the remaining question of the behavior of
ReadCallback when failures happen (do we fail fast? or do we wait for all
responses to come back/timeout?). Depending on the outcome of that, the test
that is flaky at the moment would need to be adjusted to expect 1 *or* 2
failures in the response."
Or we can say "as soon as it fails more than 0 times" and be safe?
This is my personal interpretation.
I believe [~ifesdjeen] would be the best person to confirm any details but for
test fix this would be enough I think to make it deterministic. I don't see a
bug but confusing responses to the user (same as you, as we discussed on Slack)
was (Author: e.dimitrova):
Hi Kevin,
"In any case, I believe passing an immutable copy of the
{{failureReasonByEndpoint}} map to the constructor of
Read/WriteFailureException would reduce the chances for the {{number of
failures}} and the failure messages to be inconsistent."
Agree with you. I would personally expect 1 response from node1 where the
schema was altered and two failures with the two messages.
"In addition to that, there's the remaining question of the behavior of
ReadCallback when failures happen (do we fail fast? or do we wait for all
responses to come back/timeout?). Depending on the outcome of that, the test
that is flaky at the moment would need to be adjusted to expect 1 *or* 2
failures in the response."
Or we can say "as soon as it fails more than 0 times" and be safe?
This is my personal interpretation.
I believe [~ifesdjeen] would be the best person to confirm any details but for
test fix this would be enough. I don't see a bug but confusing responses to the
user (same as you, as we discussed on Slack)
> flaky test
> org.apache.cassandra.distributed.test.SimpleReadWriteTest.readWithSchemaDisagreement
> -----------------------------------------------------------------------------------------------
>
> Key: CASSANDRA-15543
> URL: https://issues.apache.org/jira/browse/CASSANDRA-15543
> Project: Cassandra
> Issue Type: Bug
> Components: Test/dtest
> Reporter: David Capwell
> Assignee: Kevin Gallardo
> Priority: Normal
> Fix For: 4.0-alpha
>
>
> This fails infrequently, last seen failure was on java 8
> {code}
> junit.framework.AssertionFailedError
> at
> org.apache.cassandra.distributed.test.DistributedReadWritePathTest.readWithSchemaDisagreement(DistributedReadWritePathTest.java:276)
> {code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
