[
https://issues.apache.org/jira/browse/FLINK-32531?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Ivan Stoiev updated FLINK-32531:
--------------------------------
Description:
Maybe it's not a bug, but some documentation issue.
[Docs|https://nightlies.apache.org/flink/flink-kubernetes-operator-docs-main/docs/custom-resource/job-management/#recovery-of-missing-job-deployments]
says:
??When HA is enabled, the operator can recover the Flink cluster deployments in
cases when it was accidentally deleted by the user or some external process.??
My use cases are stadalone jobs with savepoint as {{{}upgradeMode{}}}, and we
want to ensure it's state, even when explicity deleted by the user. Starting a
job withouth its previous state is something that we need to avoid at all cost.
In our tests (using Flink 1.16, kubernetes HA and operator 1.4.0), when we
delete the FlinkDeployment object, all HA metadata is removed. And when
applying the same FlinkDeployment object, flink always start withouth state.
Looking at the code and related issues it seems the normal behaviour, but the
docs lead us to wrong conclusions. Is there something that we are missing here?
Is there some configuration/situation that could replicate the behaviour
described on docs.
was:
Maybe it's not a bug, but some documentation issue.
[Docs|https://nightlies.apache.org/flink/flink-kubernetes-operator-docs-main/docs/custom-resource/job-management/#recovery-of-missing-job-deployments]
says:
??When HA is enabled, the operator can recover the Flink cluster deployments in
cases when it was accidentally deleted by the user or some external process.??
My use cases are stadalone jobs with savepoint as {{{}upgradeMode{}}}, and we
want to ensure it's state, even when explicity deleted by the user. Starting a
job withouth its previous state is something that we need to avoid at all costs.
In our tests (using Flink 1.16, kubernetes HA and operator 1.4.0), when we
delete the FlinkDeployment object, all HA metadata is removed. And when
applying the same FlinkDeployment object, flink always start withouth state.
Looking at the code and related issues it seems the normal behaviour, but the
docs lead us to wrong conclusions. Is there something that we are missing here?
Is there some configuration/situation that could replicate the behaviour
described on docs.
> "Recovery of missing job deployments" feature not working as described
> ----------------------------------------------------------------------
>
> Key: FLINK-32531
> URL: https://issues.apache.org/jira/browse/FLINK-32531
> Project: Flink
> Issue Type: Bug
> Components: Kubernetes Operator
> Reporter: Ivan Stoiev
> Priority: Minor
>
> Maybe it's not a bug, but some documentation issue.
> [Docs|https://nightlies.apache.org/flink/flink-kubernetes-operator-docs-main/docs/custom-resource/job-management/#recovery-of-missing-job-deployments]
> says:
> ??When HA is enabled, the operator can recover the Flink cluster deployments
> in cases when it was accidentally deleted by the user or some external
> process.??
> My use cases are stadalone jobs with savepoint as {{{}upgradeMode{}}}, and we
> want to ensure it's state, even when explicity deleted by the user. Starting
> a job withouth its previous state is something that we need to avoid at all
> cost.
> In our tests (using Flink 1.16, kubernetes HA and operator 1.4.0), when we
> delete the FlinkDeployment object, all HA metadata is removed. And when
> applying the same FlinkDeployment object, flink always start withouth state.
> Looking at the code and related issues it seems the normal behaviour, but the
> docs lead us to wrong conclusions. Is there something that we are missing
> here? Is there some configuration/situation that could replicate the
> behaviour described on docs.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
