canvas-lms/Jenkinsfile.selenium.flakey_spec_catcher

#!/usr/bin/env groovy

/*
 * Copyright (C) 2019 - present Instructure, Inc.
 *
 * This file is part of Canvas.
 *
 * Canvas is free software: you can redistribute it and/or modify it under
 * the terms of the GNU Affero General Public License as published by the Free
 * Software Foundation, version 3 of the License.
 *
 * Canvas is distributed in the hope that it will be useful, but WITHOUT ANY
 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
 * A PARTICULAR PURPOSE. See the GNU Affero General Public License for more
 * details.
 *
 * You should have received a copy of the GNU Affero General Public License along
 * with this program. If not, see <http://www.gnu.org/licenses/>.
 */

library "canvas-builds-library@${env.CANVAS_BUILDS_REFSPEC}"

@groovy.transform.Field
def partitions = []
@groovy.transform.Field
def changed_tests = ''
@groovy.transform.Field
def fsc_status = null
def FSC_TIMEOUT = configuration.getInteger('fsc-timeout', 20)

def isPlugin() {
  return env.GERRIT_PROJECT == "canvas-lms" ? "0" : "1"
}

def getDockerWorkDir() {
  return env.GERRIT_PROJECT == "canvas-lms" ? "/usr/src/app" : "/usr/src/app/gems/plugins/${env.GERRIT_PROJECT}"
}

def getLocalWorkDir() {
  return env.GERRIT_PROJECT == "canvas-lms" ? "." : "gems/plugins/${env.GERRIT_PROJECT}"
}

def setupNode() {
  cleanAndSetup()
  sh 'rm -vrf ./tmp'
  def refspecToCheckout = env.GERRIT_PROJECT == "canvas-lms" ? env.GERRIT_REFSPEC : env.CANVAS_LMS_REFSPEC

  checkoutRepo("canvas-lms", refspecToCheckout, 2)

  if (env.IS_PLUGIN == "1") {
    dir(env.LOCAL_WORKDIR) {
      checkoutRepo(GERRIT_PROJECT, env.GERRIT_REFSPEC, 2)
    }
  }

  sh 'build/new-jenkins/docker-compose-pull.sh'
  sh 'build/new-jenkins/docker-compose-build-up.sh'
}

def computeTestCount() {
  // oops, probably should have added an easier way to _count_ tests...
  sh 'rm -vrf $LOCAL_WORKDIR/tmp'
  sh 'mkdir -v $LOCAL_WORKDIR/tmp'
  sh 'chmod -vv 777 $LOCAL_WORKDIR/tmp'

  if (FORCE_FAILURE == '1') {
    changed_tests = 'spec/force_failure_spec.rb'
  } else {
    sh '''
      docker run --volume $(pwd)/$LOCAL_WORKDIR/.git:$DOCKER_WORKDIR/.git \
                 --volume $(pwd)/$LOCAL_WORKDIR/tmp:$DOCKER_WORKDIR/tmp \
                 --env FSC_IGNORE_FILES \
                 -w=$DOCKER_WORKDIR \
                 $PATCHSET_TAG \
          bash -c "flakey_spec_catcher --use-parent --dry-run-quiet > $DOCKER_WORKDIR/tmp/test_list"
    '''

    changed_tests = readFile("$env.LOCAL_WORKDIR/tmp/test_list").trim()
  }

  echo "raw result from catcher: \n====\n$changed_tests\n===="

  if(changed_tests.length() == 0) {
    echo "no tests found to execute"

    return
  }

  def changed_tests_arr = changed_tests.split('\n').collect {
    "$DOCKER_WORKDIR/$it"
  }

  sh """
    docker-compose --project-name canvas-lms0 run --rm -v \$(pwd)/\$LOCAL_WORKDIR/tmp:\$DOCKER_WORKDIR/tmp \
      canvas \
      bash -c "bundle exec rspec --dry-run ${changed_tests_arr.join(' ')} --require ./build/new-jenkins/rspec_location.rb --format RSpecLocationFormatter --out ${env.DOCKER_WORKDIR}/tmp/test_list_split"
  """

  changed_tests = readFile("$env.LOCAL_WORKDIR/tmp/test_list_split").trim()
  changed_tests_arr = changed_tests.split('\n')

  echo "raw result from splitter: \n====\n$changed_tests\n===="

  if(changed_tests.length() == 0) {
    echo "no tests found to execute"

    return
  }

  def test_count = changed_tests_arr.size()
  def weighted_values = changed_tests_arr.collect {
    it.contains('selenium') ? env.SELENIUM_RATIO.toInteger() : 1
  }

  def distributed_factor = env.TESTS_PER_NODE.toInteger()

  def i = 0
  def cur_partition = []
  def cur_weight = 0

  while(i < test_count) {
    cur_partition.add(changed_tests_arr[i])
    cur_weight += weighted_values[i]

    if(cur_weight >= distributed_factor) {
      partitions.add(cur_partition)

      cur_partition = []
      cur_weight = 0
    }

    i++
  }

  if(cur_partition.size() > 0) {
    partitions.add(cur_partition)
  }

  echo "expected nodes to run on for $test_count tests: ${partitions.size()}"
}

def executeFlakeySpecCatcher(prefix = 'main') {
  sh 'rm -vrf tmp'
  try {
    sh 'seq 0 $((DOCKER_PROCESSES-1)) | parallel "build/new-jenkins/rspec-flakey-spec-catcher.sh {}"'
  } catch (org.jenkinsci.plugins.workflow.steps.FlowInterruptedException e) {
    if (e.causes[0] instanceof org.jenkinsci.plugins.workflow.steps.TimeoutStepExecution.ExceededTimeout) {
      echo "Failing the build due to timeouts"
      fsc_status = 'timeout'
      throw e
    } else if(e.causes[0] instanceof jenkins.model.CauseOfInterruption.UserInterruption) {
      echo "Build aborted"
      fsc_status = 'aborted'
    } else {
      throw e
    }
  } finally {
    sh "build/new-jenkins/docker-copy-files.sh /usr/src/app/tmp/fsc.out tmp canvas_ --allow-error --clean-dir"
    archiveArtifacts artifacts: "tmp/**/fsc.out", allowEmptyArchive: true
  }
}

def sendSlack(success) {
  def color = fsc_status == 'timeout' ? 'warning' : (success ? 'good' : 'danger')
  def jobInfo = "<https://gerrit.instructure.com/$env.GERRIT_CHANGE_NUMBER|Gerrit> | <$env.BUILD_URL|Jenkins>"
  def message = "$jobInfo\n$changed_tests"
  if (fsc_status == 'timeout') {
    message = "Timeout Occurred!\n$message"
  }
  slackSend channel: '#flakey_spec_catcher_noisy', color: color, message: message
}

def cleanupFn(status) {
  try {
    if(fsc_status != 'aborted') {
      sendSlack(status == 'SUCCESS')
    }
  } finally {
    sh 'rm -vrf ./tmp/'
    libraryScript.execute 'bash/docker-cleanup.sh --allow-failure'
  }
}

pipeline {
  agent none
  options {
    ansiColor('xterm')
    timeout(time: FSC_TIMEOUT)
    timestamps()
  }

  environment {
    GERRIT_PORT = '29418'
    GERRIT_URL = "$GERRIT_HOST:$GERRIT_PORT"
    BUILD_REGISTRY_FQDN = configuration.buildRegistryFQDN()
    COMPOSE_FILE = 'docker-compose.new-jenkins.yml:docker-compose.new-jenkins-selenium.yml:docker-compose.new-jenkins-flakey-spec-catcher.yml'
    IS_PLUGIN = isPlugin()
    DOCKER_WORKDIR = getDockerWorkDir()
    LOCAL_WORKDIR = getLocalWorkDir()
    FORCE_FAILURE = configuration.forceFailureFSC()
    // fsc errors when running specs from gems.
    // until we figure out how to run them, we should ignore them
    FSC_IGNORE_FILES = "gems/.*/spec/,spec/contracts/"
    POSTGRES_PASSWORD = 'sekret'
    SELENIUM_VERSION = "3.141.59-20201119"

    // Targeting 10 minutes / node, each node runs DOCKER_PROCESSES threads and
    // repeats each test FSC_REPEAT_FACTOR times.
    // Assumption: non-selenium tests take 500ms / test
    // Assumption: selenium tests take 500ms * SELENIUM_RATIO / test
    DOCKER_PROCESSES = 3
    FSC_REPEAT_FACTOR = 10
    MAX_NODES = configuration.getString('fsc-max-nodes', '10')
    SELENIUM_RATIO = 20
    TESTS_PER_NODE = 200
  }

  stages {
    stage('Environment') {
      steps {
        script {
          protectedNode('canvas-docker', { status -> cleanupFn(status) }) {
            stage('Setup') {
              setupNode()
            }

            stage("Compute Build Distribution") {
              computeTestCount()
            }

            stage("Run Flakey Spec Catcher") {
              if(partitions.size() == 0) {
                return
              } else if(partitions.size() >= env.MAX_NODES.toInteger()) {
                error "Refusing to use more than ${env.MAX_NODES} nodes to catch flakey specs. Consider breaking this change into smaller pieces."

                return
              }

              def stages = [:]

              stages["flakey set 00"] = {
                withEnv(["FSC_TESTS=${partitions[0].join(',')}"]) {
                  executeFlakeySpecCatcher()
                }
              }

              for(int i = 1; i < partitions.size(); i++) {
                // make sure to create a new index variable so this value gets
                // captured by the lambda
                def index = i
                def node_number = (index).toString().padLeft(2, '0')

                stages["flakey set $node_number"] = {
                  protectedNode('canvas-docker') {
                    withEnv(["FSC_TESTS=${partitions[index].join(',')}"]) {
                      try {
                        setupNode()
                        executeFlakeySpecCatcher("node$node_number")
                      }
                      finally {
                        sh 'rm -vrf ./tmp'
                        libraryScript.execute 'bash/docker-cleanup.sh --allow-failure'
                      }
                    }
                  }
                }
              }
              parallel(stages)
            }
          }
        }
      }
    }
  }
}