NXBT-2300: cleanup windb job

[alexistimic]

Cleanup windb job (pipeline way)

https://qa.nuxeo.org/jenkins/job/atimic/job/atimic_windows_slaves_cleanup/

[jcarsique]

See https://jenkins.io/doc/pipeline/examples/#jobs-in-parallel ?

I'd recommend to use the @NonCPS for the loops.

[jcarsique]

LGTM after recommended changes.

---

Next improvement is likely to parallelize the tasks and handle busy nodes.

Because of the planed deletion, it makes sense that the node must be fully idle (even if or especially if there are multiple executors).

But the node() step reserves an executor and waits for it being assigned. So it would also work fine for the busy nodes while they have only one executor: the task will wait for being accepted.

Then it would be required to parallelize the cleanup tasks. Anyway, that is already whished if not required.
See https://support.cloudbees.com/hc/en-us/articles/230922168-Pipeline-Parallel-execution-of-tasks

Finally, in case of a node with multiple executors, you would have to find a strategy to forbid simultaneous execution of other tasks, or skip the cleanup.

[jcarsique]

candidate for lib?

[jcarsique]

multiline to ease maintenance?

[jcarsique]

provide output and clearer steps about the (ignored) nodes, for instance:

import jenkins.model.Jenkins

def winNodes = [];
for (slave in jenkins.model.Jenkins.instance.getNodes()) {
    def nodeName = slave.getNodeName()
    if (!nodeName.contains('windb')) {
        continue
    }
    def computer = slave.toComputer()
    if (!computer.isOnline()) {
        println "Node '$nodeName' is offline - skip cleanup"
    } else if (computer.isIdle()) {
        winNodes.add(nodeName)
    } else {
        println "Node '$nodeName' is busy - skip cleanup"
    }
}

Use computer.isIdle() instead of computer.countBusy() == 0 (although equivalent since countBusy() = countExecutors()-countIdle())

[jcarsique]

node(winNode) does not work?

[jcarsique]

👍

[tmartins]

👍