[
https://issues.apache.org/jira/browse/FLINK-21000?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17466756#comment-17466756
]
Hang Ruan commented on FLINK-21000:
-----------------------------------
In my solution, I will propose some changes as following:
# We need to add a new scope for the operator coordinator in the job manager.
This operator coordinator scope would be
{{<host>.jobmanager.<job_name>.<operator_name>.coordinator as default.}}
# Add the method {{OperatorCoordinator.Context#metricGroup()}} to get the
{{OperatorCoordinatorMetricGroup}} in the operator coordinator context, and
modify its implementation class
# Add two implementation class {{InternalOperatorCoordinatorMetricGroup}} and
{{InternalSplitEnumeratorMetricGroup}} for the operator coordinator and split
enumerator
# {{Modify the JobManagerJobMetricGroup, let the
}}{{OperatorCoordinatorMetricGroup}} be the subcomponent of it
# Modify other classes to pass the metric group from {{SchedulerBase}} to
{{{}SourceCoordinatorContext{}}}, and let the {{SourceCoordinatorContext}}
return a actual split enumerator metric group instead of null
Because of changing the public interface {{MetricOptions}} in configuration, we
may need to raise a FLIP.
> set MetricGroup in SourceCoordinatorContext
> -------------------------------------------
>
> Key: FLINK-21000
> URL: https://issues.apache.org/jira/browse/FLINK-21000
> Project: Flink
> Issue Type: Bug
> Components: Connectors / Common
> Affects Versions: 1.12.0
> Reporter: Steven Zhen Wu
> Assignee: Hang Ruan
> Priority: Not a Priority
> Labels: auto-deprioritized-major, auto-deprioritized-minor
>
> Right now, SourceCoordinatorContext returns null. We need a valid metric
> group to publish source metrics (e.g. number of pending splits) in source
> enumerator.
> cc [~sewen] [~jqin]
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
