[
https://issues.apache.org/jira/browse/FLINK-12683?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16852600#comment-16852600
]
vinoyang commented on FLINK-12683:
----------------------------------
Hi [~klion26] Here is my plan:
Changing these two methods' parameter list to :
{code:java}
public void receiveDeclineMessage(DeclineCheckpoint message) -> public void
receiveDeclineMessage(DeclineCheckpoint message, TaskManagerLocation
taskManagerLocation);
public boolean receiveAcknowledgeMessage(AcknowledgeCheckpoint message) ->
public boolean receiveAcknowledgeMessage(AcknowledgeCheckpoint message,
TaskManagerLocation taskManagerLocation);
{code}
So that we do not need to introduce the {{ExecutionGraph}} or the mapping
relationship of {{ExecutionAttemptID}} and {{Execution}} to the
{{CheckpointCoordinator}}.
Currently, the caller of both methods are {{LegacyScheduler}} , in this class
we can access the instance of {{ExecutionGraph}} and get the Execution by
{{ExecutionAttemptID}}, then we can call
{{Execution#getAssignedResourceLocation}}. WDYT?
> Provide task manager's location information for checkpoint coordinator
> specific log messages
> --------------------------------------------------------------------------------------------
>
> Key: FLINK-12683
> URL: https://issues.apache.org/jira/browse/FLINK-12683
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Checkpointing
> Reporter: vinoyang
> Assignee: vinoyang
> Priority: Major
>
> Currently, the {{AcknowledgeCheckpoint}} does not contain the task manager's
> location information. When a task's snapshot task sends an ack message to the
> coordinator, we can only log this message:
> {code:java}
> Received late message for now expired checkpoint attempt 6035 from
> ccd88d08bf82245f3466c9480fb5687a of job 775ef8ff0159b071da7804925bbd362f.
> {code}
> Sometimes we need to get this sub task's location information to do the
> further debug work, e.g. stack trace dump. But, without the location
> information, It will not help to quickly locate the problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
