[
https://issues.apache.org/jira/browse/FLINK-10531?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16677018#comment-16677018
]
ASF GitHub Bot commented on FLINK-10531:
----------------------------------------
kl0u opened a new pull request #7036: [FLINK-10531][e2e] Fix unstable TTL
end-to-end test.
URL: https://github.com/apache/flink/pull/7036
As describe on the JIRA, the problem is that the clock on the machine on
Travis seems to have jumped backwards.
This resulted in a mismatch between the elements discarded as expired by
Flink's internal TTL mechanism, and the "user-code" in the test that computes
the expired elements based on the timestamp of the latest update.
I repeat the explanation from the JIRA here for reference:
-------------------------------------------------------------------------------------------
The way the test works is the following. For a given state:
1) it keeps the state with TTL
2) and keeps all the updates to that state irrespective of their timestamp
(e.g. all the elements added in a ListState)
To verify correctness:
1) it fetches the state from Flink, as cleaned up by the internal TTL
mechanism
2) it reconstructs the expected state from the stored updates by taking the
timestamp
of the latest update (ts) discarding elements with timestamp ts-ttl
As you can see from the stacktrace in the error from Travis, the latest
update has timestamp ts=1538918066021
while there are elements in the list with timestamps greater than ts (e.g.
1538918066136). This means that the internal
clock on that machine went backwards, so Flink's TTL may have removed
elements that appear in the expected state
of the test, as it takes as current timestamp the ts=1538918066021.
The fix is simply to assume that (for the test), processing time increases
monotonically and ignore "updates from the past".
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
> State TTL RocksDb backend end-to-end test failed on Travis
> ----------------------------------------------------------
>
> Key: FLINK-10531
> URL: https://issues.apache.org/jira/browse/FLINK-10531
> Project: Flink
> Issue Type: Bug
> Components: Tests
> Affects Versions: 1.6.1
> Reporter: Till Rohrmann
> Assignee: Kostas Kloudas
> Priority: Critical
> Labels: pull-request-available, test-stability
>
> The {{State TTL RocksDb backend end-to-end test}} end-to-end test failed on
> Travis.
> https://travis-ci.org/apache/flink/jobs/438226190
> https://api.travis-ci.org/v3/job/438226190/log.txt
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
