Jenkins declarative pipeline job - how to distribute parallel steps across slaves? - jenkins-pipeline

Question

I am running a Declarative Pipeline where one of the steps runs a (very long) integration test. I'm trying to split my test into several smaller ones and run them in parallel over several nodes. I have 8 of these smaller tests and I have 8 nodes (under a label), so I'd like to have each test run on a separate node. Unfortunately, two tests — when run on the same node — interfere with each other, and so both fail.
I need to be able to first get the list of available nodes, and then run the smaller tests in parallel, one of each node; if there are not enough nodes, one of the smaller tests need to wait until the node is finished.
However, what happens is that when asking for a node by label, two of the smaller tests usually get the same node, and so both fail. Nodes are configured to run up to 3 executors, otherwise the whole system halts, so I can't change that.
My current configuration for the smaller test is:
stage('Integration Tests') {
when {
expression {params.TESTS_INTEGRATION}
}
parallel {
stage('Test1') {
agent {node {label 'my_builder'}}
steps {
script {
def shell_script = getShellScript("Test1")
sh "${shell_script}"
}
}
}
I am able to get the list of available slaves from a label like this:
pipeline {
stages {
// ... other stages here ...
stage('NodeList'){
steps {
script {
def nodes = getNodeNames('my_builder')
free_nodes = []
for (def element = 0; element < nodes.size(); element++) {
usenode = nodes[element]
try {
// Give it 5 seconds to run the nodetest function
timeout(time: 5, unit: 'SECONDS') {
node(usenode) {
nodetest()
free_nodes += usenode
}
}
} catch(err) {
}
}
println free_nodes
}
}
}
Where
def getNodeNames (String label) {
def lgroup = Jenkins.instance.getLabel(label)
def nodes = lgroup.getNodes()
def result = []
if (nodes.size() > 0) {
for (def element = 0; element < nodes.size(); element++) {
result += nodes[element].getNodeName()
}
}
return result
}
def nodetest() {
sh('echo alive on \$(hostname)')
}
How can I get the node name programmatically out of the free_nodes array and direct the stage to use that?

Answer 1

I've figured it out, so for the people from the future:
It turns out you can run a Scripted Pipeline inside a Declarative Pipeline, like this:
pipeline {
stage('SomeStage') {
steps {
script {
// ... your scripted pipeline here
}
}
}
The script can do anything, and that includes... running a pipeline!
Here is the script:
script {
def builders = [:]
def nodes = getNodeNames('my_label')
// let's find the free nodes
String[] free_nodes = []
for (def element = 0; element < nodes.size(); element++) {
usenode = nodes[element]
try {
// Give it 5 seconds to run the nodetest function
timeout(time: 5, unit: 'SECONDS') {
node(usenode) {
nodetest()
free_nodes += usenode
}
}
} catch(err) {
// do nothing
}
}
println free_nodes
def tests = params.TESTS_LIST.split(',')
for(int i = 0; i < tests.length; i++) {
// select the test to run
def the_test = tests[i]
// select on which node to run it
def the_node = free_nodes[i % free_nodes.length]
// here comes the scripted pipeline: prepare steps
builders[the_test] = {
// run on the selected node
node(the_node) {
// lock the resource with the name of the node so two tests can't run there at the same time
lock(the_node) {
// name the stage
stage(the_test) {
println "Running on ${NODE_NAME}"
def shell_script = getShellScript("${the_test}")
sh "${shell_script}"
}
}
}
}
}
// run the steps in parallel
parallel builders
}