[
https://issues.apache.org/jira/browse/FLINK-21377?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Chesnay Schepler updated FLINK-21377:
-------------------------------------
Description:
The {{JobMasterTest}} contains tests for queryable state.
These tests assume that any KV operation can be executed from the moment the
JobMaster is created, when in practice the only requirement is for them to work
once the job is in a deploying state. As a result these tests now fail with the
{{DeclarativeScheduler}} because it is still waiting for resources, and hasn't
even created an {{ExecutionGraph}} yet.
We should make sure to provide enough slots for the job to be deployed.
was:
The {{JobMasterTest}} contains 2 tests for queryable state
({{testRequestKvStateWithIrrelevantRegistration}},
{{testRegisterAndUnregisterKvState}}) that work against the JobMaster but
effectively test behavior of the backing scheduler.
We should refactor these tests to work against the schedulers. This will make
it easier for us to accommodate the broken behavior in these tests; they assume
QS operations to be possible right after the JM was started, but there's no
guarantee that this is the case. For the DeclarativeScheduler this is only
possible after we have received enough slots to schedule the job.
> Adjust queryable-state JobMaster tests
> --------------------------------------
>
> Key: FLINK-21377
> URL: https://issues.apache.org/jira/browse/FLINK-21377
> Project: Flink
> Issue Type: Sub-task
> Components: Runtime / Coordination, Tests
> Reporter: Chesnay Schepler
> Assignee: Chesnay Schepler
> Priority: Major
> Fix For: 1.13.0
>
>
> The {{JobMasterTest}} contains tests for queryable state.
> These tests assume that any KV operation can be executed from the moment the
> JobMaster is created, when in practice the only requirement is for them to
> work once the job is in a deploying state. As a result these tests now fail
> with the {{DeclarativeScheduler}} because it is still waiting for resources,
> and hasn't even created an {{ExecutionGraph}} yet.
> We should make sure to provide enough slots for the job to be deployed.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
