-
Bug
-
Resolution: Fixed
-
Minor
-
None
We have an issue with the labels assigned to jenkins slave pods within the jenkins kubernetes plugin. A label in a Kubernetes pod are key/value pairs. Valid label keys have two segments: an optional prefix and name, separated by a slash 
. The issue we are seeing is that the keys are unique for each pod that is spinned up. Following the guidelines here: https://github.com/jenkinsci/kubernetes-plugin means that we end up with a pod label like: `jenkins/worker-a3346222-237b-4882-a8f9-2d1f125e2a44=true`. so instead of putting the unique id for the worker as the value, it is part of the key. This becomes a problem because we get one index per build in elasticsearch. Has anybody thought about this or know a workaround? It seems unnecessary that we have to write a logstash filter to ignore fields starting with `jenkins/`
- duplicates
-
JENKINS-60088 Change pod label - build label from unique key to value.
-
- Resolved
-
- is related to
-
-
- Resolved
-
[JENKINS-56442] Use agent slave pod label value instead of key
I had considered this as another possible option:
```
$ git diff src/main/java/org/csanchez/jenkins/plugins/kubernetes/PodTemplate.java diff --git a/src/main/java/org/csanchez/jenkins/plugins/kubernetes/PodTemplate.java b/src/main/java/org/csanchez/jenkins/plugins/kubernetes/PodTemplate.java index 4cdecb0..74881fe 100644 --- a/src/main/java/org/csanchez/jenkins/plugins/kubernetes/PodTemplate.java +++ b/src/main/java/org/csanchez/jenkins/plugins/kubernetes/PodTemplate.java @@ -360,6 +360,7 @@ public class PodTemplate extends AbstractDescribableImpl<PodTemplate> implements if (!labelSet.isEmpty()) { for (LabelAtom label : labelSet) { // builder.put(label == null ? DEFAULT_ID : "jenkins/" + label.getName(), "true"); + builder.put(label == null ? DEFAULT_ID : "jenkins_build", label.getName()); } } return builder.build();
```
But I felt it was kind of stupid to even consider maintaining my own mods to a jenkins plugin, and instead make the log shipper do the transformation work.
I struggle with the `if (!labelSet.isEmpty())` logic, and the getLabelsMap() in general. I think if we can specify our own label, probably in the config where we define the Kube “cloud”, it would possibly get around all of this. I never explored if there is in fact a way to do that from the config UI (or even if I was properly reading that function).
I too had to work through this in Dec 2018 and worked with Eric in the slack channel discussing it. I couldn't get it to work with logstash and filebeats, in any combination. Instead in the Kube stack that runs our Jenkins builders, I had to install and configure fluentd and use inline ruby scripting to morph the fields. I had to capture everything after the `jenkins/(.+)`, create a new key "jenkins_build=$1". What I had to do to accomplish it is documented here:
https://gist.github.com/mrballcb/c1a8ff4132224e654e85aad80f3a0fec