Dennis-Mircea Ciupitu created FLINK-40583:
---------------------------------------------

             Summary: Filter operator keys on the remaining runtime config paths
                 Key: FLINK-40583
                 URL: https://issues.apache.org/jira/browse/FLINK-40583
             Project: Flink
          Issue Type: Improvement
          Components: Kubernetes Operator
            Reporter: Dennis-Mircea Ciupitu


As of now, the operator and autoscaler properties ({{{}kubernetes.operator.{}}} 
and {{{}job.autoscaler.{}}}) can be exchanged between the operator and the 
Flink cluster in both directions. FLINK-40380 closed one path, but yet 2 are 
still left.
 - On the reading path, the operator reads a running job's config from three 
REST endpoints and merges them. FLINK-40380 added a filter that drops 
{{kubernetes.operator.}} and {{job.autoscaler.}} keys, but only on one of them, 
leaving the JobManager endpoint path uncovered.
 - On the writing path, when a session job is submitted, the jar run request is 
built from the raw config, and the operator sends its own keys to the 
JobManager on every submit. Currently, Flink ignores them and never hands them 
back. Those properties can also hold sensitive information, such as 
{{{}kubernetes.operator.user.artifacts.http.header{}}}, which holds the header 
used to download job artifacts, so it is better not to send any of them.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

Reply via email to