[
https://issues.apache.org/jira/browse/FLINK-9674?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16531034#comment-16531034
]
ASF GitHub Bot commented on FLINK-9674:
---------------------------------------
Github user florianschmidt1994 commented on a diff in the pull request:
https://github.com/apache/flink/pull/6216#discussion_r199719951
--- Diff:
flink-end-to-end-tests/test-scripts/test_queryable_state_restart_tm.sh ---
@@ -85,20 +90,23 @@ function run_test() {
exit 1
fi
- local
current_num_checkpoints=current_num_checkpoints$(get_completed_number_of_checkpoints
${JOB_ID})
-
kill_random_taskmanager
latest_snapshot_count=$(cat $FLINK_DIR/log/*out* | grep "on snapshot"
| tail -n 1 | awk '{print $4}')
echo "Latest snapshot count was ${latest_snapshot_count}"
- sleep 65 # this is a little longer than the heartbeat timeout so that
the TM is gone
+ # wait until the TM loss was detected
+ wait_for_job_state_transition ${JOB_ID} "RESTARTING" "CREATED"
start_and_wait_for_tm
+ wait_job_running ${JOB_ID}
+
+ local current_num_checkpoints="$(get_completed_number_of_checkpoints
${JOB_ID})"
--- End diff --
Why did you move this from `before killing the TM` to `after having the TM
restarted again` ?
> Remove 65s sleep in QueryableState E2E test
> -------------------------------------------
>
> Key: FLINK-9674
> URL: https://issues.apache.org/jira/browse/FLINK-9674
> Project: Flink
> Issue Type: Improvement
> Components: Queryable State, Tests
> Affects Versions: 1.5.0, 1.6.0
> Reporter: Chesnay Schepler
> Assignee: Chesnay Schepler
> Priority: Major
> Labels: pull-request-available
>
> The {{test_queryable_state_restart_tm.sh}} kills a taskmanager, waits for the
> loss to be noticed, starts a new tm and waits for the job to continue.
> {code}
> kill_random_taskmanager
> [...]
> sleep 65 # this is a little longer than the heartbeat timeout so that the TM
> is gone
> start_and_wait_for_tm
> {code}
> Instead of waiting for a fixed amount of time that is tied to some config
> value we should wait for a specific event, like the job being canceled.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
