Skip to content

Instantly share code, notes, and snippets.

@akomakom
Forked from rb2k/gist:8372402
Last active June 3, 2024 15:44
Show Gist options
  • Save akomakom/481507c0dd79ec52a395 to your computer and use it in GitHub Desktop.
Save akomakom/481507c0dd79ec52a395 to your computer and use it in GitHub Desktop.
A jenkins script to clean up workspaces on slaves. On all slaves with less than X free GB, this script removes /slaves/workspace/* if slave is idle. If not idle, it removes /full/job/workspace/path for each job that isn't currently running. There are two versions: a "System Groovy Script" and a Jenkinsfile (Pipeline script)
/**
Jenkins System Groovy script to clean up workspaces on all slaves.
Check if a slave has < X GB of free space, perform cleanup if it's less.
If slave is idle, wipe out everything in the workspace directory as well any extra configured directories.
If slave is busy, wipe out individual job workspace directories for jobs that aren't running.
Either way, remove custom workspaces also if they aren't in use.
**/
import hudson.model.*;
import hudson.util.*;
import jenkins.model.*;
import hudson.FilePath.FileCallable;
import hudson.slaves.OfflineCause;
import hudson.node_monitors.*;
//threshold is in GB and comes from a job parameter
def threshold = Integer.parseInt(build.buildVariableResolver.resolve("CLEAN_THRESHOLD"))
def skippedLabels = [ 'container' ] //don't clean docker slaves
def extraDirectoriesToDelete = [ 'temp' ] //additional paths under slave's root path that should be removed if found
def deleteRemote(def path, boolean deleteContentsOnly) {
boolean result = true
def pathAsString = path.getRemote()
if (path.exists()) {
try {
if (deleteContentsOnly) {
path.deleteContents()
println ".... deleted ALL contents of ${pathAsString}"
} else {
path.deleteRecursive()
println ".... deleted directory ${pathAsString}"
}
} catch (Throwable t) {
println "Failed to delete ${pathAsString}: ${t}"
result = false
}
}
return result
}
def failedNodes = []
for (node in Jenkins.instance.nodes) {
computer = node.toComputer()
if (computer.getChannel() == null) {
continue
}
if (node.assignedLabels.find{ it.expression in skippedLabels }) {
println "Skipping ${node.displayName} based on labels"
continue
}
try {
size = DiskSpaceMonitor.DESCRIPTOR.get(computer).size
roundedSize = size / (1024 * 1024 * 1024) as int
println("node: " + node.getDisplayName() + ", free space: " + roundedSize + "GB. Idle: ${computer.isIdle()}")
if (roundedSize < threshold) {
def prevOffline = computer.isOffline()
if (prevOffline && computer.getOfflineCauseReason().startsWith('disk cleanup from job')) {
prevOffline = false //previous run screwed up, ignore it and clear it at the end
}
if (!prevOffline) {
//don't override any previosly set temporarily offline causes (set by humans possibly)
computer.setTemporarilyOffline(true, new hudson.slaves.OfflineCause.ByCLI("disk cleanup from job ${build.displayName}"))
}
if (computer.isIdle()) {
//It's idle so delete everything under workspace
def workspaceDir = node.rootPath.child('workspace')
if (!deleteRemote(workspaceDir, true)) {
failedNodes << node
}
//delete custom workspaces
Jenkins.instance.getAllItems(TopLevelItem).findAll{item -> item instanceof Job && !("${item.class}".contains('WorkflowJob')) && item.getCustomWorkspace()}.each{ item ->
if (!deleteRemote(node.getRootPath().child(item.customWorkspace), false)) {
failedNodes << node
}
}
extraDirectoriesToDelete.each{
if (!deleteRemote(node.getRootPath().child(it), false)) {
failedNodes << node
}
}
} else {
Jenkins.instance.getAllItems(TopLevelItem).findAll{item -> item instanceof Job && !item.isBuilding() && !("${item.class}".contains('WorkflowJob')) }.each{ item ->
jobName = item.getFullDisplayName()
//println(".. checking workspaces of job " + jobName)
workspacePath = node.getWorkspaceFor(item)
if (!workspacePath) {
println(".... could not get workspace path for ${jobName}")
return
}
//println(".... workspace = " + workspacePath)
customWorkspace = item.getCustomWorkspace()
if (customWorkspace) {
workspacePath = node.getRootPath().child(customWorkspace)
// println(".... custom workspace = " + workspacePath)
}
if (!deleteRemote(workspacePath, false)) {
failedNodes << node
}
}
}
if (!prevOffline) {
computer.setTemporarilyOffline(false, null)
}
}
} catch (Throwable t) {
println "Error with ${node.displayName}: ${t}"
failedNodes << node
}
}
println "\n\nSUMMARY\n\n"
failedNodes.each{node ->
println "\tERRORS with: ${node.displayName}"
}
assert failedNodes.size() == 0
/**
Jenkins Pipeline script to clean up workspaces on all slaves.
Check if a slave has < X GB of free space, perform cleanup if it's less.
If slave is idle, wipe out everything in the workspace directory as well any extra configured directories.
If slave is busy, wipe out individual job workspace directories for jobs that aren't running.
Either way, remove custom workspaces also if they aren't in use.
**/
import hudson.model.*;
import hudson.util.*;
import jenkins.model.*;
import hudson.node_monitors.*;
properties(
[
buildDiscarder(logRotator(artifactDaysToKeepStr: '', artifactNumToKeepStr: '', daysToKeepStr: '', numToKeepStr: '25')),
disableConcurrentBuilds(),
disableResume(),
parameters([
string(defaultValue: '20', description: 'Slave must have less than this # of GB to be cleaned', name: 'CLEAN_THRESHOLD', trim: false),
booleanParam(defaultValue: false, description: 'Enable verbosity', name: 'DEBUG')
]),
pipelineTriggers([cron('@hourly'), pollSCM('')]) //is SCM triggers are set up
]
)
Jenkins.instance.getItemByFullName(env.JOB_NAME).setDescription('''
<pre>
For each slave with low disk space:
* Empties the workspace directory (if slave is idle)
* Deletes individual job workspace directories (if slave is busy)
* Deletes directories matching custom workspace directories configured in any job
* Deletes a hardcoded list of additional directories
Runs at random times to catch recurring jobs
Script from https://gist.github.com/rb2k/8372402
(actually a fork of it with fixes for folders: https://gist.github.com/akomakom/481507c0dd79ec52a395)
</pre>
''')
//threshold is in GB and comes from a job parameter
def skippedLabels = [ 'container' ] //don't clean docker slaves
/** GLOBALS **/
threshold = Integer.parseInt(params.CLEAN_THRESHOLD)
debug = Boolean.valueOf(params.DEBUG)
countFullClean = 0
countPartialClean = 0
failedNodes = []
extraDirectoriesToDelete = [ 'somethingRelative', '/something/absolute' ] //additional paths under slave's root path that should be removed if found (or abs path)
def debug(def string) {
if (debug) {
echo "DEBUG: ${string}"
}
}
def deleteRemote(def path, boolean deleteContentsOnly) {
boolean result = true
def pathAsString = path.getRemote()
debug "Asked to delete ${path} deleteContentsOnly=${deleteContentsOnly ? 'yes' : 'no'}"
if (path.exists()) {
try {
if (deleteContentsOnly) {
path.deleteContents()
echo ".... deleted ALL contents of ${pathAsString}"
} else {
path.deleteRecursive()
echo ".... deleted directory ${pathAsString}"
}
} catch (Throwable t) {
echo "Failed to delete ${pathAsString}: ${t}"
result = false
}
}
return result
}
def cleanNode(node, computer) {
size = DiskSpaceMonitor.DESCRIPTOR.get(computer).size
roundedSize = size / (1024 * 1024 * 1024) as int
echo("node: " + node.getDisplayName() + ", free space: " + roundedSize + "GB. Idle: ${computer.isIdle()}")
if (roundedSize < threshold) {
def prevOffline = computer.isOffline()
if (prevOffline && computer.getOfflineCauseReason().startsWith('disk cleanup from job')) {
prevOffline = false //previous run screwed up, ignore it and clear it at the end
}
if (!prevOffline) {
//don't override any previosly set temporarily offline causes (set by humans possibly)
computer.setTemporarilyOffline(true, new hudson.slaves.OfflineCause.ByCLI("disk cleanup from job ${currentBuild.displayName}"))
}
if (computer.isIdle()) {
//It's idle so delete everything under workspace
countFullClean++
def workspaceDir = node.rootPath.child('workspace')
debug "Deleting workspace ${workspaceDir}"
if (!deleteRemote(workspaceDir, true)) {
failedNodes << node
}
//delete custom workspaces
Jenkins.instance.getAllItems(TopLevelItem).findAll { item -> item instanceof Job && !("${item.class}".contains('WorkflowJob')) && item.getCustomWorkspace() }.each { item ->
debug "Deleting custom workspace ${item.customWorkspace}"
if (!deleteRemote(node.getRootPath().child(item.customWorkspace), false)) {
failedNodes << node
}
}
extraDirectoriesToDelete.each {
def pathToDelete = it.startsWith('/') ? node.createPath(it) : node.getRootPath().child(it)
debug "Deleting extra dir ${pathToDelete}"
if (!deleteRemote(pathToDelete, false)) {
failedNodes << node
}
}
} else {
countPartialClean++
Jenkins.instance.getAllItems(TopLevelItem).findAll { item -> item instanceof Job && !item.isBuilding() && !("${item.class}".contains('WorkflowJob')) }.each { item ->
jobName = item.getFullDisplayName()
//echo(".. checking workspaces of job " + jobName)
workspacePath = node.getWorkspaceFor(item)
if (!workspacePath) {
echo(".... could not get workspace path for ${jobName}")
return
}
//echo(".... workspace = " + workspacePath)
customWorkspace = item.getCustomWorkspace()
if (customWorkspace) {
workspacePath = node.getRootPath().child(customWorkspace)
// echo(".... custom workspace = " + workspacePath)
}
if (!deleteRemote(workspacePath, false)) {
failedNodes << node
}
}
}
if (!prevOffline) {
computer.setTemporarilyOffline(false, null)
}
}
}
timestamps {
for (node in Jenkins.instance.nodes) {
def computer = node.toComputer()
if (computer.getChannel() == null) {
continue
}
if (node.assignedLabels.find { it.expression in skippedLabels }) {
echo "Skipping ${node.displayName} based on labels"
continue
}
try {
cleanNode(node, computer)
} catch (Throwable t) {
echo "Error with ${node.displayName}: ${t}"
failedNodes << node
}
}
echo "\n\n"
failedNodes.each { node ->
echo "\tERRORS with: ${node.displayName}"
}
}
assert failedNodes.size() == 0
currentBuild.setDescription("Full: ${countFullClean}, Partial: ${countPartialClean}")
@akomakom
Copy link
Author

akomakom commented Nov 27, 2019

This will not work on swarms good idea though.
groovy.lang.MissingPropertyException: No such property: build for class: Script1

That's because you were trying to run a "Groovy System Script" as a "Jenkins Pipeline Script". I was in fact using this on swarms.

However, I'm adding a Pipeline version today...

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment