Purushottam Sinha created FLINK-41463:
-----------------------------------------
Summary: Connectors / MongoDB: MongoE2ECase fails intermittently
with "pull access denied for flink-configured-taskmanager"
Key: FLINK-41463
URL: https://issues.apache.org/jira/browse/FLINK-41463
Project: Flink
Issue Type: Bug
Components: Connectors / MongoDB, Tests
Reporter: Purushottam Sinha
*Problem*
On {{main}}, {{MongoE2ECase}} sometimes fails while the TaskManager containers
are starting. The CI workflow is fail-fast, so one failure cancels the rest of
the matrix and leaves unrelated PRs red.
*Evidence*
* Failed run:
https://github.com/apache/flink-connector-mongodb/actions/runs/43198030832/job/131113903082
(mongodb7, Flink 2.1.3, JDK 21). It shows {{ContainerLaunchException ...
flink-configured-taskmanager:latest}}, caused by a 404 "pull access denied".
The other 22 jobs were cancelled.
* {{MongoE2ECase.java:88}} sets {{numTaskManagers(2)}}.
{{FlinkTestcontainersConfigurator}} builds one image per TaskManager, and every
one is named {{flink-configured-taskmanager}}. {{FlinkContainers.start()}} then
starts them with {{parallelStream()}}, so the two builds race on the same tag.
* {{v2.0}} already has the fix (FLINK-39098, [PR
#78|https://github.com/apache/flink-connector-mongodb/pull/78]). It never
reached {{main}}.
*Proposed fix*
* In {{MongoE2ECase}}, use {{numTaskManagers(1)}} with
{{numSlotsPerTaskManager(2)}}. This matches {{v2.0}} and the Kafka,
Elasticsearch, Opensearch and Pulsar E2E tests.
*Acceptance*
* {{MongoE2ECase}} passes across the full PR CI matrix on {{main}}.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
