Jenkins pipeline

Jenkinsfile

@@ -0,0 +1,183 @@
+def MACHINE = 'none'
+def machine = 'none'
+def HOME = 'none'
+
+pipeline {
+    agent { label 'built-in' }
+
+    options {
+        disableConcurrentBuilds(abortPrevious: true)
+        skipDefaultCheckout(true)
+        buildDiscarder(logRotator(numToKeepStr: '3'))
+    }
+
+    stages {
+
+        stage('Get Machine') {
+            agent { label 'built-in' }
+            steps {
+                script {
+                    MACHINE = 'none'
+                    for (label in pullRequest.labels) {
+                        echo "Label: ${label}"
+                        if ((label.matches("CI-Hera-Ready"))) {
+                            MACHINE = 'hera'
+                        } else if ((label.matches("CI-Orion-Ready"))) {
+                            MACHINE = 'orion'
+                        } else if ((label.matches("CI-Hercules-Ready"))) {
+                            MACHINE = 'hercules'
+                        }
+                    }
+                    machine = MACHINE[0].toUpperCase() + MACHINE.substring(1)
+                }
+            }
+        }
+
+        stage('Get Common Workspace') {
+            agent { label "${MACHINE}-emc" }
+            steps ( timeout(time: 1, unit: 'HOURS') ) {
+                script {
+                    HOME = "${WORKSPACE}/TESTDIR"
+                    pullRequest.addLabel("CI-${machine}-Building")
+                    if (pullRequest.labels.contains("CI-${machine}-Ready")) {
+                        pullRequest.removeLabel("CI-${machine}-Ready")
+                    }
+                }
+            }
+        }
+
+        stage('Build') {
+            matrix {
+                agent { label "${MACHINE}-emc" }
+                axes {
+                    axis { 
+                        name "system"
+                        values "gfs", "gefs"
+                    }
+                }
+                stages {
+                    stage("build system") {
+                        steps {
+                            script {
+                                def HOMEgfs = "${HOME}/${system}"
+                                properties([parameters([[$class: 'NodeParameterDefinition', allowedSlaves: ['built-in','Hera-EMC','Orion-EMC'], defaultSlaves: ['built-in'], name: '', nodeEligibility: [$class: 'AllNodeEligibility'], triggerIfResult: 'allCases']])])
+                                sh( script: "mkdir -p ${HOMEgfs}", returnStatus: true)
+                                dir(HOMEgfs) {
+                                    checkout scm
+                                    env.MACHINE_ID = MACHINE
+                                    if (fileExists("sorc/BUILT_semaphor")) {
+                                        sh( script: "cat sorc/BUILT_semaphor", returnStdout: true).trim()
+                                        pullRequest.comment("Cloned PR already built (or build skipped) on ${machine} in directory ${HOMEgfs}")
+                                    } else {
+                                        sh( script: "git submodule update --init --recursive", returnStatus: true) 
+                                        if (system == "gfs") {
+                                            dir("${HOMEgfs}/sorc") {
+                                                sh( script: "./build_all.sh -gu -j 4", returnStatus: false)
+                                                sh( script: "./link_workflow.sh", returnStatus: false)
+                                                sh( script: "echo ${HOMEgfs} > BUILT_semaphor", returnStatus: true)
+                                            }
+                                        } else if (system == "gefs") {
+                                            // TODO: need to add gefs build arguments from a yaml file
+                                            dir("${HOMEgfs}/sorc") {
+                                                sh( script: "./build_all.sh -gu -j 4", returnStatus: false)
+                                                sh( script: "./link_workflow.sh", returnStatus: false)
+                                                sh( script: "echo ${HOMEgfs} > BUILT_semaphor", returnStatus: true)
+                                            }
+                                        }
+                                    }
+                                }
+                            }
+                        }
+                    }
+                }
+            }
+        }
+
+        stage('Setup RUNTESTS') {
+            agent { label "${MACHINE}-emc" }
+            steps {
+                script {
+                    sh( script: "mkdir -p ${HOME}/RUNTESTS", returnStatus: true)
+                    //TODO cannot get pullRequest.labels.contains("CI-${machine}-Building") to work
+                    pullRequest.removeLabel("CI-${machine}-Building")
+                    pullRequest.addLabel("CI-${machine}-Running")
+                }
+            }
+
+        }
+
+        stage('Run Tests') {
+            matrix {
+                agent { label "${MACHINE}-emc" }
+                axes {
+                    axis {
+                        name "Case"
+                        values "C48_ATM", "C48_S2SWA_gefs", "C48_S2SW", "C96_atm3DVar"
+                    }
+                }
+                stages {
+                    stage('Create Experiment') {
+                        steps {
+                                script {
+                                    env.RUNTESTS = "${HOME}/RUNTESTS"
+                                    def HOMEgfs = "${HOME}/gfs"
+                                    env.HOME = HOMEgfs
+                                    sh( script: "rm -Rf ${RUNTESTS}/EXPDIR/${Case}_*" )
+                                    sh( script: "rm -Rf ${RUNTESTS}/COMROOT/${Case}_*" )
+                                    sh( script: "${HOMEgfs}/ci/scripts/utils/ci_utils_wrapper.sh create_experiment ${HOMEgfs}/ci/cases/pr/${Case}.yaml", returnStatus: true)
+                                }
+                        }
+                    }
+                    stage('Run Experiments') {
+                        steps {
+                            script {
+                                def HOMEgfs = "${HOME}/gfs"
+                                pslot = sh( script: "${HOMEgfs}/ci/scripts/utils/ci_utils_wrapper.sh get_pslot ${HOME}/RUNTESTS ${Case}", returnStdout: true ).trim()
+                                pullRequest.comment("Running experiments: ${Case} with pslot ${pslot} on ${machine}")
+                            try {
+                                sh( script: "${HOMEgfs}/ci/scripts/run-check_ci.sh ${HOME} ${pslot}", returnStatus: false)
+                                pullRequest.comment("SUCCESS running experiments: ${Case} on ${machine}")
+                            } catch (Exception e) {
+                                pullRequest.comment("FAILURE running experiments: ${Case} on ${machine}")
+                                error("Failed to run experiments ${Case} on ${machine}")
+                                }
+                            }
+                        }
+                    }
+                }
+            }
+        }
+
+    }
+
+    post {
+        success {
+            script {
+                //if ( pullRequest.labels.contains( "CI-${machine}-Running" ) ) {
+                //   pullRequest.removeLabel("CI-${machine}-Running")
+                //}
+                // TODO: contains mehthod does not work
+                pullRequest.removeLabel("CI-${machine}-Running")
+                pullRequest.addLabel("CI-${machine}-Passed")
+                def timestamp = new Date().format("MM dd HH:mm:ss", TimeZone.getTimeZone('America/New_York'))
+                pullRequest.comment("SUCCESSFULLY ran all CI Cases on ${machine} at ${timestamp}")
+            }
+            cleanWs()
+        }
+        failure {
+            script {
+                pullRequest.removeLabel("CI-${machine}-Running")
+                pullRequest.addLabel("CI-${machine}-Failed")
+                def timestamp = new Date().format("MM dd HH:mm:ss", TimeZone.getTimeZone('America/New_York'))
+                pullRequest.comment("CI FAILED ${machine} at ${timestamp}\n\nBuilt and ran in directory ${HOME}")
+                if (fileExists('${HOME}/RUNTESTS/ci.log')) {
+                    def fileContent = readFile '${HOME}/RUNTESTS/ci.log'
+                        fileContent.eachLine { line ->
+                        archiveArtifacts artifacts: "${line}", fingerprint: true
+                    }
+                }
+            }
+        }
+    }
+
+}

ci/scripts/run-check_ci.sh

@@ -21,7 +21,9 @@ pslot=${2:-${pslot:-?}}        # Name of the experiment being tested by this scr
 #     │   └── ${pslot}
 #     └── EXPDIR
 #         └── ${pslot}
-HOMEgfs="${TEST_DIR}/HOMEgfs"
+# Two system build directories created at build time gfs, and gdas
+# TODO: Make this configurable (for now all scripts run from gfs for CI at runtime)
+HOMEgfs="${TEST_DIR}/gfs"
 RUNTESTS="${TEST_DIR}/RUNTESTS"
 
 # Source modules and setup logging

ci/scripts/run-check_ci_stub.sh

@@ -0,0 +1,11 @@
+#!/bin/bash
+
+set -eu
+
+HOMEgfs=$1
+pslot=$2
+
+echo -e "HOMEgfs: ${HOMEgfs}\npslpot: ${pslot}\n in ${PWD}" >> "${HOMEgfs}/ci/scripts/run-check_ci_stub.log"
+if [[ ${pslot} =~ "C48_S2SW" ]]; then
+     exit 1
+fi

ci/scripts/utils/ci_utils.sh

@@ -1,24 +1,102 @@
 #!/bin/env bash
 
-function cancel_slurm_jobs() {
-
-  # Usage: cancel_slurm_jobs <substring>
-  # Example: cancel_slurm_jobs "C48_ATM_3c4e7f74"
+function cancel_batch_jobs() {
+  # Usage: cancel_batch_jobs <substring>
+  # Example: cancel_batch_jobs "C48_ATM_3c4e7f74"
   #
-  # Cancel all Slurm jobs that have the given substring in their name
+  # Cancel all batch jobs that have the given substring in their name
   # So like in the example all jobs with "C48_ATM_3c4e7f74"
   # in their name will be canceled
 
   local substring=$1
   local job_ids
-  job_ids=$(squeue -u "${USER}" -h -o "%i")
-
-  for job_id in ${job_ids}; do
-    job_name=$(sacct -j "${job_id}" --format=JobName%100 | head -3 | tail -1 | sed -r 's/\s+//g') || true
-    if [[ "${job_name}" =~ ${substring} ]]; then
-      echo "Canceling Slurm Job ${job_name} with: scancel ${job_id}"
-      scancel "${job_id}"
-      continue
-    fi
-  done
+
+  # cancel pbs jobs <substring>
+  if [[ ${MACHINE_ID} == "wcoss2" ]]; then
+    job_ids=$(qstat -u "${USER}" | awk '{print $1}') || true
+
+    for job_id in ${job_ids}; do
+      job_name=$(qstat -f "${job_id}" | grep Job_Name | awk '{print $3}') || true
+      if [[ "${job_name}" =~ ${substring} ]]; then
+        echo "Canceling PBS Job ${job_name} with: qdel ${job_id}"
+        qdel "${job_id}"
+        continue
+      fi
+    done
+  # cancel slurm jobs <substring>
+  else 
+    job_ids=$(squeue -u "${USER}" -h -o "%i")
+
+    for job_id in ${job_ids}; do
+      job_name=$(sacct -j "${job_id}" --format=JobName%100 | head -3 | tail -1 | sed -r 's/\s+//g') || true
+      if [[ "${job_name}" =~ ${substring} ]]; then
+        echo "Canceling Slurm Job ${job_name} with: scancel ${job_id}"
+        scancel "${job_id}"
+        continue
+      fi
+    done
+  fi
+}
+
+
+function get_pr_case_list () {
+
+    #############################################################
+    # loop over every yaml file in the PR's ci/cases
+    # and create an run directory for each one for this PR loop
+    #############################################################
+    for yaml_config in "${HOMEgfs}/ci/cases/pr/"*.yaml; do
+      case=$(basename "${yaml_config}" .yaml) || true
+      echo "${case}"
+    done
+}
+
+function get_pslot_list () {
+
+    local RUNTESTS="${1}"
+
+    #############################################################
+    # loop over expdir directories in RUNTESTS
+    # and create list of the directory names (pslot) with the hash tag
+    #############################################################
+    for pslot_dir in "${RUNTESTS}/EXPDIR/"*; do
+      pslot=$(basename "${pslot_dir}") || true
+      echo "${pslot}"
+    done
+
+}
+
+function get_pslot () {
+
+    local RUNTESTS="${1}"
+    local case="${2}"
+
+    #############################################################
+    # loop over expdir directories in RUNTESTS
+    # and return the name of the pslot with its tag that matches the case
+    #############################################################
+    for pslot_dir in "${RUNTESTS}/EXPDIR/"*; do
+      pslot=$(basename "${pslot_dir}")
+      check_case=$(echo "${pslot}" | rev | cut -d"_" -f2- | rev) || true
+      if [[ "${check_case}" == "${case}" ]]; then
+        echo "${pslot}"
+        break
+      fi
+    done
+
+}
+
+function create_experiment () {
+
+  local yaml_config="${1}"
+  cd "${HOMEgfs}" || exit 1
+  pr_sha=$(git rev-parse --short HEAD)
+  case=$(basename "${yaml_config}" .yaml) || true
+  export pslot=${case}_${pr_sha}
+
+  source "${HOMEgfs}/ci/platforms/config.${MACHINE_ID}"
+  source "${HOMEgfs}/workflow/gw_setup.sh"
+
+  "${HOME}/workflow/create_experiment.py" --yaml "${yaml_config}"
+
 }

ci/scripts/utils/ci_utils_wrapper.sh

@@ -0,0 +1,9 @@
+#!/usr/bin/env bash
+
+HOMEgfs="$(cd "$(dirname  "${BASH_SOURCE[0]}")/../../.." >/dev/null 2>&1 && pwd )"
+source "${HOMEgfs}/ush/detect_machine.sh"
+
+utitilty_function="${1}"
+
+source "${HOMEgfs}/ci/scripts/utils/ci_utils.sh"
+${utitilty_function} "${@:2}"

sorc/build_all_stub.sh

@@ -0,0 +1,5 @@
+#!/bin/bash
+
+set -eu
+
+echo "Ran build all stub in ${PWD}" > "BUILD_STUB_RAN_$(date +%d%H%M)"