Hrm, I'll need to look into this - I could have sworn I'd specifically made sure that when is evaluated before the agent is handled.

Hi there, the same thing should apply to docker agent.

The docker agent container should not be started before the when condition is evaluated.

Since there are some cases where users may need to have the agent allocated before when is evaluated, we have to do it in this order. Sorry.

This is a really huge pain for our use of Jenkins - it can extend the time for checkin processing by minutes.  If you can't do this for the standard "when" clause, could you generate a new "quick when" clause that behaves like the old version?  I'm happy to use a restricted version that can only look at flags set for the overall run.

Yeah this is pretty disappointing. It would be interesting to know how often 'when' clauses require node processing. It seems like this would be the exception.

How does this get processed? Could there be something like a "requireNodeAllocation: true" in the when {} block, or has the ship sailed by the time that would get evaluated?

So I think I may be able to make it so that when types are marked internally as to whether they require an agent. Off the top of my head, the only one that I'd mark as requiring an agent is expression, since I can't tell what you might be doing in there. So if you used any when condition but expression, it wouldn't go into an agent first, but if you used expression (including inside a nested anyOf, allOf, or not), it would go into an agent first. How's that sound?

Actually a common use case of ours is to have expression inside the when; we have stages which toggle based on an environment variable set on the pipeline job itself, so that we can have a single Jenkinsfile handling both pre- and post-merge workflows.

So for us this wouldn't work, but we can deal with it by having separate Jenkinsfiles and/or using shared libraries instead.

Example pipeline:

ciBuild = env.CI == null ? true : env.CI.toLowerCase() == 'true'

pipeline {
  agent none
  stages {
    stage('Build') {
      when { expression { return !ciBuild } }
      agent { label 'someLabel' }
      steps {
        sh "echo Build"
      }
    }
    stage('Test') {
      when { expression { return ciBuild } }
      agent { label 'someOtherLabel' }
      steps {
        sh "echo Test"
      }
    }
  }
}

What specifically brought me here is that someOtherLabel points to a huge Docker image which takes a while to spin up; if it's not already cached it can take 20 minutes to pull down. Once I noticed this I got a little scared because our pipelines actually have several more stages, most of which are skipped in the ciBuild scenario. The extra time spent allocating nodes would generate complaints.

Subsequent to commenting I noticed that I can put a when inside a parallel, and if I have {{ agent none }} as above, no node will be allocated. Can you have parallel with a single substage? Even if you have to have a dummy placeholder stage, people might be fine trading the extra logging and UI noise for reduced node allocations.

So it turns out that it's easier to just add a beforeAgent field for when - the default will be to require the agent first, but if beforeAgent true is specified, it'll enter the agent after the when evaluation.

PR incoming.

PR up at https://github.com/jenkinsci/pipeline-model-definition-plugin/pull/232, with docs PR up at https://github.com/jenkins-infra/jenkins.io/pull/1290

Code changed in jenkins
User: Andrew Bayer
Path:
pipeline-model-api/src/main/java/org/jenkinsci/plugins/pipeline/modeldefinition/ast/ModelASTWhen.java
pipeline-model-api/src/main/resources/ast-schema.json
pipeline-model-definition/src/main/groovy/org/jenkinsci/plugins/pipeline/modeldefinition/model/StageConditionals.groovy
pipeline-model-definition/src/main/groovy/org/jenkinsci/plugins/pipeline/modeldefinition/parser/JSONParser.groovy
pipeline-model-definition/src/main/groovy/org/jenkinsci/plugins/pipeline/modeldefinition/parser/ModelParser.groovy
pipeline-model-definition/src/main/groovy/org/jenkinsci/plugins/pipeline/modeldefinition/parser/RuntimeASTTransformer.groovy
pipeline-model-definition/src/main/resources/org/jenkinsci/plugins/pipeline/modeldefinition/Messages.properties
pipeline-model-definition/src/main/resources/org/jenkinsci/plugins/pipeline/modeldefinition/ModelInterpreter.groovy
pipeline-model-definition/src/test/java/org/jenkinsci/plugins/pipeline/modeldefinition/AbstractModelDefTest.java
pipeline-model-definition/src/test/java/org/jenkinsci/plugins/pipeline/modeldefinition/ValidatorTest.java
pipeline-model-definition/src/test/java/org/jenkinsci/plugins/pipeline/modeldefinition/WhenStageTest.java
pipeline-model-definition/src/test/resources/json/whenBeforeAgentTrue.json
pipeline-model-definition/src/test/resources/whenBeforeAgentFalse.groovy
pipeline-model-definition/src/test/resources/whenBeforeAgentTrue.groovy
pipeline-model-definition/src/test/resources/whenBeforeAgentUnspecified.groovy
http://jenkins-ci.org/commit/pipeline-model-definition-plugin/ae2406d833ccb9b199e9c4865d302ef882ed1dae
Log:
[FIXED JENKINS-44461] Add beforeAgent flag for when

If true, the when condition will be evaluated before entering the
stage's agent (if defined). The default is false - better safe than
sorry there.

Merged, will be in Declarative 1.2.6.

Code changed in jenkins
User: Andrew Bayer
Path:
content/doc/book/pipeline/syntax.adoc
http://jenkins-ci.org/commit/jenkins.io/09ee78185f68c7212613202ef20d7523504acf1c
Log:
JENKINS-44461 Add documentation for new when->beforeAgent field

Code changed in jenkins
User: Liam Newman
Path:
content/doc/book/pipeline/syntax.adoc
http://jenkins-ci.org/commit/jenkins.io/003b9ead4337beba1b4a835a2496f2d7ce1ad365
Log:
Merge pull request #1290 from abayer/jenkins-44461

JENKINS-44461 Add documentation for new when->beforeAgent field

Compare: https://github.com/jenkins-infra/jenkins.io/compare/ff40d07092a3...003b9ead4337

Bulk closing resolved issues.