[
https://issues.apache.org/jira/browse/FLINK-38710?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Diego Lobba updated FLINK-38710:
--------------------------------
Description:
When changing a property under the {{kubernetes.operator}} namespace from field
{{spec.flinkConfiguration}} of a FlinkDeployment resource version v1beta
changes are apparently not sorting any effect.
The changed manifest is updated, but managed field \{{lastStableSpec }}is not
reflecting changes proposed, despite showing no error or event reporting about
a possible error encountered.
This issue can be also noticed by setting any of such configuration property to
an erroneous value (such as
{{{}kubernetes.operator.deployment.readiness.timeout: 10mingibberish{}}}).
Changes are applied and no error is being reported.
The only way such property are being parsed and further actions considered is
when updating from the same field any property associated to the job manager or
the task manager (es:
{{{}jobmanager.memory.jvm-metaspace.size{}}}). As soon as such property is
updated, the entire flinkConfiguration field is parsed and evaluated, possibly
reporting any error.
*Step to reproduce the issue:*
Let's set an invalid property value:
# Consider FlinkDeployment manifest from attached file flink-deployment.yaml;
# Update field {{{}spec.flinkConfiguration{}}}, changing property
{{{}kubernetes.operator.deployment.readiness.timeout from value {{10min{}}}}}
to value {{10mingib.}}
# apply the manifest. FlinkDeployment spec is expected to be updated,
lastStableSpec is expected not to be updated, no error is expected ({*}bug{*});
# now update field {{spec.flinkConfiguration once again, changing property
jobmanager.memory.jvm-metaspace.size }}from value {{512m}} to value
{{{}510m{}}}.
# apply the manifest. Any managed job manager is expected to be updated, the
operator parses value of spec.flinkConfiguration and reports about the
misconfiguration.
was:
When changing a property under the {{kubernetes.operator}} namespace from field
{{spec.flinkConfiguration}} of a FlinkDeployment resource version v1beta
changes are apparently not sorting any effect.
The changed manifest is updated, but managed field {{lastStableSpec }}is not
reflecting changes proposed, despite showing no error or event reporting about
a possible error encountered.
This issue can be also noticed by setting any of such configuration property to
an erroneous value (such as
{{{}kubernetes.operator.deployment.readiness.timeout: 10mingibberish{}}}).
Changes are applied and no error is being reported.
The only way such property are being parsed and further actions considered is
when updating from the same field any property associated to the job manager or
the task manager (es:
{{{}jobmanager.memory.jvm-metaspace.size{}}}). As soon as such property is
updated, the entire flinkConfiguration field is parsed and evaluated, possibly
reporting any container error.
*Step to reproduce the issue:*
Let's set an invalid property value:
# Consider FlinkDeployment manifest from attached file flink-deployment.yaml;
# Update field {{{}spec.flinkConfiguration{}}}, changing property
{{kubernetes.operator.deployment.readiness.timeout from value {{10min}}}} to
value {{10mingib.}}
# apply the manifest. FlinkDeployment spec is expected to be updated,
lastStableSpec is expected not to be updated, no error is expected ({*}bug{*});
# now update field {{spec.flinkConfiguration once again, changing property
jobmanager.memory.jvm-metaspace.size }}from value {{512m}} to value
{{{}510m{}}}.
# apply the manifest. Any managed job manager is expected to be updated, the
operator parses value of spec.flinkConfiguration and reports about the
misconfiguration.
> operator properties in FlinkDeployment are not taking effect
> ------------------------------------------------------------
>
> Key: FLINK-38710
> URL: https://issues.apache.org/jira/browse/FLINK-38710
> Project: Flink
> Issue Type: Bug
> Components: Kubernetes Operator
> Affects Versions: 1.10.0
> Reporter: Diego Lobba
> Priority: Major
> Attachments: flink-deployment.yaml
>
>
> When changing a property under the {{kubernetes.operator}} namespace from
> field {{spec.flinkConfiguration}} of a FlinkDeployment resource version
> v1beta changes are apparently not sorting any effect.
> The changed manifest is updated, but managed field \{{lastStableSpec }}is not
> reflecting changes proposed, despite showing no error or event reporting
> about a possible error encountered.
> This issue can be also noticed by setting any of such configuration property
> to an erroneous value (such as
> {{{}kubernetes.operator.deployment.readiness.timeout: 10mingibberish{}}}).
> Changes are applied and no error is being reported.
> The only way such property are being parsed and further actions considered is
> when updating from the same field any property associated to the job manager
> or the task manager (es:
> {{{}jobmanager.memory.jvm-metaspace.size{}}}). As soon as such property is
> updated, the entire flinkConfiguration field is parsed and evaluated,
> possibly reporting any error.
>
> *Step to reproduce the issue:*
>
> Let's set an invalid property value:
> # Consider FlinkDeployment manifest from attached file
> flink-deployment.yaml;
> # Update field {{{}spec.flinkConfiguration{}}}, changing property
> {{{}kubernetes.operator.deployment.readiness.timeout from value {{10min{}}}}}
> to value {{10mingib.}}
> # apply the manifest. FlinkDeployment spec is expected to be updated,
> lastStableSpec is expected not to be updated, no error is expected
> ({*}bug{*});
> # now update field {{spec.flinkConfiguration once again, changing property
> jobmanager.memory.jvm-metaspace.size }}from value {{512m}} to value
> {{{}510m{}}}.
> # apply the manifest. Any managed job manager is expected to be updated, the
> operator parses value of spec.flinkConfiguration and reports about the
> misconfiguration.
>
>
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
