Merge branch 'NOAA-EMC:develop' into aws-forecast-only

ci/scripts/utils/launch_java_agent.sh

@@ -1,8 +1,70 @@
 #!/bin/env bash
+
 set -e
 
+# ==============================================================================
+# Script Name: launch_java_agent.sh
+#
+# Description:
+#        This script automates the process of launching a Jenkins agent
+#        on a specified machine. It ensures that the necessary
+#        prerequisites are met, such as the availability of JAVA_HOME,
+#        the Jenkins agent launch directory, and proper authentication
+#        with GitHub.
+#
+#       It then proceeds to check if the Jenkins node is online and
+#       decides whether to launch the Jenkins agent based on the node's
+#       status. The agent is launched in the background,
+#       and its PID is logged for reference.
+#
+# Prerequisites:
+#         JAVA_HOME must be set to a valid JDK installation.
+#         Jenkins agent launch directory must exist and be specified.
+#         GitHub CLI (gh) must be installed and authenticated for messeging
+#           from the Jenkins controller to GitHub PR via shell commands.
+#           Jenkins agent launch directory must exist and be specified.
+#           TODO: Must use GitHub CLI v2.25.1 (newer versoins have issues)
+#           https://github.com/cli/cli/releases/download/v2.25.1/gh_2.25.1_linux_amd64.tar.gz
+#         Jenkins controller URL and authentication token must be provided.
+#         jenkins-secret-file:
+#           Must be present in the Jenkins agent launch directory.
+#           This file contains the secret key for the Jenkins agent
+#           established by the Jenkins administrator for each Node.
+#         jenkins_token:
+#           Must be present in the Jenkins agent launch directory.
+#           This file contains the user authentication token for the Jenkins controller
+#           to use the Remote API. This token can be generated by the user
+#           on the Jenkins controller.
+#         controller_user:
+#           Must be set to the Jenkins controller username corresponing to the jenkins_token.
+#
+# Usage: ./launch_java_agent.sh [now] [-f]
+#        The optional 'now' argument forces the script to launch the Jenkins
+#        agent without waiting before trying again.
+#        The optional 'force' argument forces the script to launch the Jenkins regarless of the node status.
+#
+# ==============================================================================
+
+force_launch="False"
+skip_wait="False"
+while getopts ":fnh" flag; do
+  case "${flag}" in
+    f) force_launch="True";;
+    n) skip_wait="True";;
+    h) echo "Usage: ./launch_java_agent.sh [now] [force]
+Two mutually exclusive optional arguments:
+    -n (now) causes the script to launch the Jenkins agent without waiting before trying again.
+    -f (force) forces the script to launch the Jenkins regarless of its connection status."
+       exit 0 ;;
+    *) echo "Unknown flag: ${flag}"
+       exit 1;;
+  esac
+done
+
 controller_url="https://jenkins.epic.oarcloud.noaa.gov"
-controller_user="terry.mcguinness"
+controller_user=${controller_user:-"terry.mcguinness"}
+controller_user_auth_token="jenkins_token"
+
 HOMEgfs="$(cd "$(dirname  "${BASH_SOURCE[0]}")/../../.." >/dev/null 2>&1 && pwd )"
 host=$(hostname)
 
@@ -13,12 +75,10 @@ host=$(hostname)
 source "${HOMEgfs}/ush/detect_machine.sh"
 case ${MACHINE_ID} in
   hera | orion | hercules | wcoss2)
-    echo "Launch Jenkins Java Controler on ${MACHINE_ID}"
-    ;;
+    echo "Launch Jenkins Java Controler on ${MACHINE_ID}";;
   *)
     echo "Unsupported platform. Exiting with error."
-    exit 1
-    ;;
+    exit 1;;
 esac
 
 LOG=lanuched_agent-$(date +%Y%m%d%M).log
@@ -43,9 +103,16 @@ echo "JAVA VERSION: "
 ${JAVA} -version
 
 export GH="${HOME}/bin/gh"
-command -v "${GH}"
+[[ -f "${GH}" ]] || echo "gh is not installed in ${HOME}/bin"
 ${GH} --version
 
+check_mark=$(gh auth status -t 2>&1 | grep "Token:" | awk '{print $1}') || true
+if [[ "${check_mark}" != "✓" ]]; then
+  echo "gh not authenticating with emcbot token"
+  exit 1
+fi
+echo "gh authenticating with emcbot TOKEN ok"
+
 if [[ -d "${JENKINS_AGENT_LANUCH_DIR}" ]]; then
   echo "Jenkins Agent Lanuch Directory: ${JENKINS_AGENT_LANUCH_DIR}"
 else
@@ -56,22 +123,62 @@ cd "${JENKINS_AGENT_LANUCH_DIR}"
 
 if ! [[ -f agent.jar ]]; then
   curl -sO "${controller_url}/jnlpJars/agent.jar"
+  echo "Updated agent.jar downloaded"
+fi
+
+if [[ ! -f "${controller_user_auth_token}" ]]; then
+   echo "User Jenkins authetication TOKEN to the controller for using the Remote API does not exist"
+   exit 1
 fi
+JENKINS_TOKEN=$(cat "${controller_user_auth_token}")
+
+cat << EOF > parse.py
+#!/usr/bin/env python3
+import json,sys
+with open(sys.argv[1], 'r') as file:
+    data = json.load(file)
+print(data.get('offline','True'))
+EOF
+chmod u+x parse.py
 
-JENKINS_TOKEN=$(cat jenkins_token)
+check_node_online() {
+    rm -f curl_response
+    curl_response=$(curl --silent -u "${controller_user}:${JENKINS_TOKEN}" "${controller_url}/computer/${MACHINE_ID^}-EMC/api/json?pretty=true") || true
+    if [[ "${curl_response}" == "" ]]; then
+       echo "ERROR: Jenkins controller not reachable. Exiting with error."
+       exit 1
+    fi
+    echo -n "${curl_response}" > curl_response
+    ./parse.py curl_response
+}
+
+lauch_agent () {
+    echo "Launching Jenkins Agent on ${host}"
+    command="nohup ${JAVA} -jar agent.jar -jnlpUrl ${controller_url}/computer/${MACHINE_ID^}-EMC/jenkins-agent.jnlp  -secret @jenkins-secret-file -workDir ${JENKINS_WORK_DIR}"
+    echo -e "Launching Jenkins Agent on ${host} with the command:\n${command}" >& "${LOG}"
+    ${command} >> "${LOG}" 2>&1 &
+    nohup_PID=$!
+    echo "Java agent running on PID: ${nohup_PID}" >> "${LOG}" 2>&1
+}
+
+if [[ "${force_launch}" == "True" ]]; then
+  lauch_agent
+  exit
+fi
 
-# 
-offline=$(curl --silent -u "${controller_user}:${JENKINS_TOKEN}" "${controller_url}/computer/${MACHINE_ID^}-EMC/api/json?pretty=true" | grep '\"offline\"' | awk '{gsub(/,/,"");print $3}') || true
-echo "Jenkins Agent offline setting: ${offline}"
+offline=$(set -e; check_node_online)
 
-if [[ "${offline}" == "true" ]]; then
-  echo "Jenkins Agent is offline. Lanuching Jenkins Agent on ${host}"
-  command="nohup ${JAVA} -jar agent.jar -jnlpUrl ${controller_url}/computer/${MACHINE_ID^}-EMC/jenkins-agent.jnlp  -secret @jenkins-secret-file -workDir ${JENKINS_WORK_DIR}"
-  echo -e "Lanuching Jenkins Agent on ${host} with the command:\n${command}" >& "${LOG}"
-  ${command} >> "${LOG}" 2>&1 &
-  nohup_PID=$!
-  echo "Java agent running on PID: ${nohup_PID}" >> "${LOG}" 2>&1
-  echo "Java agent running on PID: ${nohup_PID}"
+if [[ "${offline}" != "False" ]]; then
+   if [[ "${skip_wait}" != "True" ]]; then
+      echo "Jenkins Agent is offline. Waiting 5 more minutes to check again in the event it is a temp network issue"
+      sleep 300
+      offline=$(set -e; check_node_online)
+   fi
+   if [[ "${offline}" != "False" ]]; then
+     lauch_agent
+    else
+      echo "Jenkins Agent is online (nothing done)"
+    fi
 else
   echo "Jenkins Agent is online (nothing done)"
 fi

parm/config/gfs/config.resources

@@ -890,8 +890,8 @@ case ${step} in
     threads_per_task=1
     walltime_gdas="03:00:00"
     walltime_gfs="06:00:00"
-    ntasks=4
-    tasks_per_node=4
+    ntasks=1
+    tasks_per_node=1
     export memory="80G"
     ;;
 

ush/parsing_namelists_FV3.sh

@@ -389,7 +389,7 @@ cat >> input.nml <<EOF
   iopt_tbot    = ${iopt_tbot:-"2"}
   iopt_stc     = ${iopt_stc:-"1"}
   iopt_trs     = ${iopt_trs:-"2"}
-  iopt_diag    = ${iopt_diag:-"1"}
+  iopt_diag    = ${iopt_diag:-"2"}
   debug        = ${gfs_phys_debug:-".false."}
   nstf_name    = ${nstf_name}
   nst_anl      = ${nst_anl}

ush/parsing_namelists_FV3_nest.sh

@@ -446,7 +446,7 @@ cat >> "${nml_file}" <<EOF
   iopt_tbot    = ${iopt_tbot:-"2"}
   iopt_stc     = ${iopt_stc:-"1"}
   iopt_trs     = ${iopt_trs:-"2"}
-  iopt_diag    = ${iopt_diag:-"1"}
+  iopt_diag    = ${iopt_diag:-"2"}
   debug        = ${gfs_phys_debug:-".false."}
   nstf_name    = ${nstf_name}
   nst_anl      = ${nst_anl}

workflow/rocoto/tasks.py

@@ -24,7 +24,7 @@ class Tasks:
                    'prepsnowobs', 'snowanl',
                    'fcst',
                    'atmanlupp', 'atmanlprod', 'atmupp', 'goesupp',
-                   'atmosprod', 'oceanprod', 'iceprod',
+                   'atmos_prod', 'ocean_prod', 'ice_prod',
                    'verfozn', 'verfrad', 'vminmon',
                    'metp',
                    'tracker', 'genesis', 'genesis_fsu',