With Jenkins declarative syntax, it's possible to run parallel stages with no top-level agent. This ends up consuming two executors, since the top level agent is marked 'none':
pipeline {
agent none
stages {
stage('Run on parallel nodes') {
parallel {
stage('Do one thing') {
agent any
steps {
...
}
stage('Do another thing') {
agent any
steps {
...
}
}
}
}
}
}
With scripted pipelines, which requires a top-level 'node' element, this is seemingly not possible. This ends up consuming three executors, even though only two are doing real work:
node {
stage('Run on parallel nodes') {
parallel ([
'Do one thing': {
node() {
...
}
},
'Do another thing': {
node() {
...
}
}
])
}
}
Is a 'lightweight' top level executor possible with scripted pipelines?
Scripted pipelines don't require a top-level node allocation. This is just wrong and can be left out.
Related
I have a parallel stage setup, and would like to know if it's possible to run a script prior to the nested stages, so something like this:
stage('E2E-PR-CYPRESS') {
when {
allOf {
expression {
return fileExists("cypress.json")
}
branch "PR-*"
}
}
steps {
script {
stash name: 'cypress-dir', includes: 'cypress/**/*'
}
}
parallel {
stage('Cypress Tests 1') {
agent { label 'aws_micro_slave_e2e' }
options { skipDefaultCheckout() }
steps {
runE2eTests()
}
}
stage('Cypress Tests 2') {
agent { label 'aws_micro_slave_e2e' }
options { skipDefaultCheckout() }
steps {
runE2eTests()
}
}
}
post {
always {
e2eAfterCypressRun(this, true)
}
}
}
I know the above is wrong, I get the error Only one of "matrix", "parallel", "stages", or "steps" allowed for stage "E2E-PR-CYPRESS"
I already have the stash script in a setup stage at the beginning of my pipeline, but I'd like to be able to restart from this stage above on Jenkins, and so need the stash part in this stage as the parallel stages need to unstash the contents.
Updated Answer:
After playing a bit with the Restart from a Stage option there is seems to be a nice feature designed exactly for your needs called Preserving stashes for Use with Restarted Stages:
Normally, when you run the stash step in your Pipeline, the resulting
stash of artifacts is cleared when the Pipeline completes, regardless
of the result of the Pipeline. Since stash artifacts aren’t accessible
outside of the Pipeline run that created them, this has not created
any limitations on usage. But with Declarative stage restarting, you
may want to be able to unstash artifacts from a stage which ran before
the stage you’re restarting from.
To enable this, there is a job property that allows you to configure a
maximum number of completed runs whose stash artifacts should be
preserved for reuse in a restarted run. You can specify anywhere from
1 to 50 as the number of runs to preserve.
This job property can be configured in your Declarative Pipeline’s options section, as below:
options {
preserveStashes()
// or
preserveStashes(buildCount: 5)
}
This built in feature is exactly what you need to solve your issue without any special modifications to your code, as it will allow you to rerun the pipeline from any stage and still use the existing file that were previously stashed.
Original Answer:
You can actually achieve this quite simply using the scripted syntax for the parallel command, and it will also allow you to avoid the duplicate code in the parallel stages.
parallel: Execute in parallel
Takes a map from branch names to closures and an optional argument failFast which will terminate all branches upon a failure in any other branch:
parallel firstBranch: {
// do something
}, secondBranch: {
// do something else
},
failFast: true|false
In your case it can look like:
stage('E2E-PR-CYPRESS') {
when {
allOf {
expression {
return fileExists("cypress.json")
}
branch "PR-*"
}
}
steps {
script {
stash name: 'cypress-dir', includes: 'cypress/**/*'
// Define the parallel execution stages
def stages = ['Cypress Tests 1', 'Cypress Tests 2']
// Create the parallel executions and run them
parallel stages.collectEntries {
["Running ${it}": {
node('aws_micro_slave_e2e') {
skipDefaultCheckout()
runE2eTests()
}
}]
}
}
}
post {
always {
e2eAfterCypressRun(this, true)
}
}
}
This way you can easily add more parallel steps by updating the stages list, or even receive it as an input parameter. In addition you can create the parallel executions by different labels or tests suits, instead of the stage name.
You can add a Prepare stage at the top like this:
stages{
stage('Preperation'){
when {
allOf {
expression {
return fileExists("cypress.json")
}
branch "PR-*"
}
}
steps {
script {
stash name: 'cypress-dir', includes: 'cypress/**/*'
}
}
}
stage('E2E-PR-CYPRESS') {
parallel {
stage('Cypress Tests 1') {
agent { label 'aws_micro_slave_e2e' }
options { skipDefaultCheckout() }
steps {
runE2eTests()
}
}
stage('Cypress Tests 2') {
agent { label 'aws_micro_slave_e2e' }
options { skipDefaultCheckout() }
steps {
runE2eTests()
}
}
}
}
}
post {
always {
e2eAfterCypressRun(this, true)
}
}
An out of the box concept
Propose splitting the job into 2 parts taking the following into consideration:
Currently use an EC2 plugin, as the current agents are EC2
Running the parallel stages with the same stashed content ready to unstash
Create jenkins pipeline job 1:
This job will checkout the workspace with any type of agent
Create a packer json to create a customised AMI for the EC2
The customised AMI will stash the contents and move to a directory that will appear on the EC2 when the agent is built
Output the AMI ID, run a groovy job to update the EC2 plugin AMI ID with the customised AMI ID to temporarily set the AMI in memory on Jenkins
pipeline {
agent {
docker {
test-container
}
}
options {
buildDiscarder(
logRotator(
numToKeepStr: '10',
artifactNumToKeepStr: '10'
)
)
ansiColor('xterm')
gitConnection("git")
}
stages {
stage('Run Stash Cypress Functional Test') {
steps {
dir('functional-test') {
// develop branch is canary build, all other branches are stable builds
script {
sh """
# script to stash cypress tests
"""
}
}
}
}
stage('Functional Test AMI Build') {
steps {
dir('functional-test/packer') {
withAWS(role: 'PackerBuild', roleAccount: '123456789012', roleSessionName: 'Jenkins-Workflow-FunctionalTest-Packer') {
script {
sh """
# packer json script will require to copy contents from workspace, run the script to stash content
# packer json script will require to capture new AMI ID
# https://discuss.devopscube.com/t/how-to-get-the-ami-id-after-a-packer-build/36
# https://www.packer.io/docs/post-processors/manifest
packer validate FunctionalTestPacker.json
packer build -debug FunctionalTestPacker.json
# grab AMI ID and export as jenkins env variable
"""
}
}
}
}
}
stage('run groovy script to update AMI ID on EC2 plugin') {
steps {
dir(groovy job dir) {
script {
sh """
# run groovy job to update AMI on Jenkins EC2 plugin
# https://gist.github.com/vrivellino/97954495938e38421ba4504049fd44ea
"""
}
}
}
}
stage('Kickoff Functional Test Deploy') {
// pipeline checkbox parameter, when ticked it will automatically kick off the functional test pipeline
when {
expression {params.RUN_TESTS.toBoolean()}
}
steps {
script{
env.branch = params.BRANCH
sh """
echo "Branch is ${branch}"
"""
}
build job: 'workflow/CypressFunctionaTestDeployAndRun',
parameters: [
string(name: 'BRANCH', value: env.branch)
],
wait : false
}
}
}
post {
always {
cleanWs()
}
}
}
Create jenkins pipeline job 2:
This job will create the EC2 agents via the plugin from the customised AMI from pipeline job 1
This means your agents will have the same workspace ready to unstash - so you can execute a parallel run
Also you could move a lot of your user data script that is in the EC2 plugin as part of the customised AMI build, thus cut down the time for each EC2 agent to get ready to carry out execution
pipeline {
stages {
stage('E2E-PR-CYPRESS') {
when {
allOf {
expression {
return fileExists("cypress.json")
}
branch "PR-*"
}
}
}
parallel {
stage('Cypress Tests 1') {
agent { label 'aws_micro_slave_e2e' }
options { skipDefaultCheckout() }
steps {
runE2eTests()
}
}
stage('Cypress Tests 2') {
agent { label 'aws_micro_slave_e2e' }
options { skipDefaultCheckout() }
steps {
runE2eTests()
}
}
}
}
post {
always {
e2eAfterCypressRun(this, true)
}
}
}
Is there a way to execute step outside of the Jenkins agent?
Suppose that I have following structure of Jenkinsfile:
pipeline {
agent none
stages {
stage('Example Stage') {
agent { someAgent }
steps {
run something ...
input ...
}
}
}
}
I would like to execute input outside of an agent to not block it for hours (timeout is not the answer ;))
One of the possible solutions is to execute the logic in separate stages but i'm trying to avoid creating additional ones.
You could use node instead of agent:
pipeline {
agent none
stages {
stage('Example Stage') {
steps {
node( someAgent ) {
run something ...
}
// outside of any agent
input ...
}
}
}
}
I'm using Declarative Pipelines 1.3.2 plugin and I want to use the same agent (as in only specifying the agent directive once) in multiple parallel stages:
stage('Parallel Deployment')
{
agent { dockerfile { label 'docker'; filename 'Dockerfile'; } }
parallel
{
stage('A') { steps { ... } }
stage('B') { steps { ... } }
}
}
However, Jenkins complains:
"agent" is not allowed in stage "Parallel Deployment" as it contains parallel stages
A solution is to duplicate the agent directive for each parallel stage, but this is tedious and leads to lot of duplicated code with many parallel stages:
stage('Parallel Deployment')
{
parallel
{
stage('A') {
agent { dockerfile { label 'docker'; filename 'Dockerfile'; } }
steps { ... }
}
stage('B') {
agent { dockerfile { label 'docker'; filename 'Dockerfile'; } }
steps { ... }
}
}
}
Is there a more idiomatic solution, or is duplicating agent directive necessary for each of the parallel stages?
Specifying the agent at pipeline level can be a solution, but has the potential downside that the agent is up & running for the whole duration of the build.
Also note that this means each stage (that doesn't define its own agent) is run on the same agent instance, not agent type. If the parallel processes are CPU / resource intensive, this might not be what you want.
Still, if you want to run parallel stages on one instance, and can't or want not to define the agent at pipeline level, here's a workaround for the declarative syntax:
stage('Parallel Deployment') {
agent { dockerfile { label 'docker'; filename 'Dockerfile'; } }
stages {
stage('A & B') {
parallel {
stage('A') { steps { ... } }
stage('B') { steps { ... } }
}
}
}
}
Or you go for a scripted pipeline, which doesn't have this limitation.
Declare the agent at Pipeline level so all stages run on the same agent.
I have a list of long running Gradle tasks on different sub projects in my project. I would like to run these in parallel using Jenkins declarative pipeline.
I was hoping something like this might work:
projects = [":a", ":b", ":c"]
pipeline {
stage("Deploy"){
parallel {
for(project in projects){
stage(project ) {
when {
expression {
someConditionalFunction(project)
}
}
steps {
sh "./gradlew ${project}:someLongrunningGradleTask"
}
}
}
}
}
}
Needless to say that gives a compile error since it was expecting stage instead of for. Any ideas on how to overcome this? Thanks
I was trying to reduce duplicated code in my existing Jenkinsfile using declarative pipeline syntax. Finally I was able to wrap my head around the difference between scripted and declarative syntax.
It is possible to use scripted pipeline syntax in a declarative pipeline by wrapping it with a script {} block.
Check out my example below: you will see that all three parallel stages finish at the same time after waking up from the sleep command.
def jobs = ["JobA", "JobB", "JobC"]
def parallelStagesMap = jobs.collectEntries {
["${it}" : generateStage(it)]
}
def generateStage(job) {
return {
stage("stage: ${job}") {
echo "This is ${job}."
sh script: "sleep 15"
}
}
}
pipeline {
agent any
stages {
stage('non-parallel stage') {
steps {
echo 'This stage will be executed first.'
}
}
stage('parallel stage') {
steps {
script {
parallel parallelStagesMap
}
}
}
}
}
Parallel wants a map structure. You are doing this a little inside-out. Build your map and then just pass it to parallel, rather than trying to iterate inside parallel.
Option 2 on this page shows you a way to do something similar to what you are trying.
At this link you can find a complex way I did this similar to a matrix/multi-config job:
I have a declarative pipeline.
In this pipeline I want various stages not executed by only one but multiple nodes (later stages, which are node specific, depend on these). Is this somehow possible?
sure, you can select different nodes in different stages based on label:
pipeline {
agent none
stages {
stage('build') {
steps {
node('docker') {
sh 'echo $HOSTNAME'
}
}
}
stage('test') {
steps {
node('rbenv') {
sh 'echo $HOSTNAME'
}
}
}
}
}
does that make sense?
You can follow the following format in your pipeline job to perform specific tasks on specific nodes:
node('master') {
.....................
<some task to perform>
.....................
}
node('slave1 && slave2') {
.....................
<some task to perform>
.....................
}