Bring in less hard coded values for ecflow_run

BrianCurtis-NOAA · Mar 22, 2024 · 689a697 · 689a697
1 parent 978a86b
commit 689a697
Show file tree

Hide file tree

Showing 2 changed files with 69 additions and 51 deletions.
diff --git a/tests/rt.sh b/tests/rt.sh
@@ -522,8 +522,8 @@ cleanup() {
 trap '{ echo "rt.sh interrupted"; rt_trap ; }' INT
 trap '{ echo "rt.sh quit"; rt_trap ; }' QUIT
 trap '{ echo "rt.sh terminated"; rt_trap ; }' TERM
-trap '{ echo "rt.sh error on line $LINENO"; cleanup ; }' ERR
-trap '{ echo "rt.sh finished"; cleanup ; }' EXIT
+trap '{ echo "rt.sh error on line $LINENO"; rt_trap ; }' ERR
+trap '{ echo "rt.sh finished"; rt_trap ; }' EXIT
 
 # PATHRT - Path to regression tests directory
 PATHRT="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd -P )"
@@ -544,7 +544,7 @@ else
   exit 1
 fi
 
-source detect_machine.sh # Note: this does not set ACCNR. The "if" block below does.
+source detect_machine.sh
 source rt_utils.sh
 # shellcheck disable=SC1091
 source module-setup.sh
@@ -656,36 +656,12 @@ fi
 # Display the machine and account using the format detect_machine.sh used:
 echo "Machine: ${MACHINE_ID} | Account: ${ACCNR}"
 
-if [[ ${MACHINE_ID} = wcoss2 ]]; then
+if [[ ${MACHINE_ID} = wcoss2 || ${MACHINE_ID} = acorn ]]; then
 
   module load ecflow/5.6.0.13
   module load intel/19.1.3.304
   module load python/3.8.6
 
-  # ECFLOW_START="${ECF_ROOT}/scripts/server_check.sh"
-  # export ECF_OUTPUTDIR="${PATHRT}/ecf_outputdir"
-  # export ECF_COMDIR="${PATHRT}/ecf_comdir"
-  # rm -rf "${ECF_OUTPUTDIR}" "${ECF_COMDIR}"
-  # mkdir -p "${ECF_OUTPUTDIR}"
-  # mkdir -p "${ECF_COMDIR}"
-  export colonifnco=":output"  # hack
-
-  DISKNM=/lfs/h2/emc/nems/noscrub/emc.nems/RT
-  QUEUE=dev
-  COMPILE_QUEUE=dev
-  ROCOTO_SCHEDULER=pbs
-  PARTITION=
-  STMP="/lfs/h2/emc/ptmp"
-  PTMP="/lfs/h2/emc/ptmp"
-  SCHEDULER="pbs"
-
-elif [[ ${MACHINE_ID} = acorn ]]; then
-
-  module load ecflow/5.6.0.13
-  module load intel/19.1.3.304
-  module load python/3.8.6
-  INPUTDATA_ROOT=${INPUTDATA_ROOT:-${DISKNM}/NEMSfv3gfs/input-data-20221101}
-  ECF_ROOT=${ECF_ROOT:-}
   ECFLOW_START="${ECF_ROOT}/scripts/server_check.sh"
   export ECF_OUTPUTDIR="${PATHRT}/ecf_outputdir"
   export ECF_COMDIR="${PATHRT}/ecf_comdir"
@@ -694,7 +670,7 @@ elif [[ ${MACHINE_ID} = acorn ]]; then
   mkdir -p "${ECF_COMDIR}"
   export colonifnco=":output"  # hack
 
-  DISKNM=/lfs/h1/emc/nems/noscrub/emc.nems/RT
+  DISKNM=/lfs/h2/emc/nems/noscrub/emc.nems/RT
   QUEUE=dev
   COMPILE_QUEUE=dev
   ROCOTO_SCHEDULER=pbs
@@ -703,6 +679,30 @@ elif [[ ${MACHINE_ID} = acorn ]]; then
   PTMP="/lfs/h2/emc/ptmp"
   SCHEDULER="pbs"
 
+# elif [[ ${MACHINE_ID} = acorn ]]; then
+
+#   module load ecflow/5.6.0.13
+#   module load intel/19.1.3.304
+#   module load python/3.8.6
+#   INPUTDATA_ROOT=${INPUTDATA_ROOT:-${DISKNM}/NEMSfv3gfs/input-data-20221101}
+#   ECF_ROOT=${ECF_ROOT:-}
+#   ECFLOW_START="${ECF_ROOT}/scripts/server_check.sh"
+#   export ECF_OUTPUTDIR="${PATHRT}/ecf_outputdir"
+#   export ECF_COMDIR="${PATHRT}/ecf_comdir"
+#   rm -rf "${ECF_OUTPUTDIR}" "${ECF_COMDIR}"
+#   mkdir -p "${ECF_OUTPUTDIR}"
+#   mkdir -p "${ECF_COMDIR}"
+#   export colonifnco=":output"  # hack
+
+#   DISKNM=/lfs/h2/emc/nems/noscrub/emc.nems/RT
+#   QUEUE=dev
+#   COMPILE_QUEUE=dev
+#   ROCOTO_SCHEDULER=pbs
+#   PARTITION=
+#   STMP="/lfs/h2/emc/ptmp"
+#   PTMP="/lfs/h2/emc/ptmp"
+  # SCHEDULER="pbs"
+
 elif [[ ${MACHINE_ID} = gaea ]]; then
 
   module use /ncrc/proj/epic/rocoto/modulefiles
@@ -919,7 +919,7 @@ else
 fi
 
 # Does this machine support Rocoto?
-if [[ -n ${ROCOTO} ]]; then
+if [[ ${ROCOTO} == true ]]; then
   if [[ ${MACHINE_ID} != wcoss2 && ${MACHINE_ID} != acorn && ${MACHINE_ID} != expanse && ${MACHINE_ID} != stampede ]]; then
     ROCOTORUN="$(command -v rocotorun)"
     export ROCOTORUN
@@ -933,7 +933,7 @@ if [[ -n ${ROCOTO} ]]; then
 fi
 
 # Does this machine support ecflow?
-if [[ -n ${ECFLOW} ]]; then
+if [[ ${ECFLOW} == true ]]; then
   if [[ ${MACHINE_ID} == wcoss2 && ${MACHINE_ID} == acorn ]]; then
     ECFLOW_START="$(command -v server_check.sh)"
 

diff --git a/tests/rt_utils.sh b/tests/rt_utils.sh
@@ -763,39 +763,57 @@ ecflow_run() {
 
   ECF_HOST="${ECF_HOST:-${HOSTNAME}}"
 
-  set +e
+  #set +e
+  # Make sure ECF_HOST and ECF_PORT are set/ready on systems that have an
+  # explicit ecflow node
+  if [[ ${MACHINE_ID} == wcoss2 || ${MACHINE_ID} == acorn ]]; then
+    readarray -t ECFHOSTLIST < "${ECF_HOSTFILE}"
+    for ECF_HOST in "${ECFHOSTLIST[@]}"
+    do
+      if ssh -t -t "${ECF_HOST}"; then
+        export ECF_HOST
+        break
+      else
+        ECF_HOST=''
+      fi
+    done
+  elif [[ ${MACHINE_ID} == hera || ${MACHINE_ID} == jet ]]; then
+    module load ecflow
+  fi
+  if [[ -z ${ECF_HOST} || -z ${ECF_PORT} ]]; then
+    echo "ERROR: ECF_HOST or ECF_PORT are not set, and rt.sh cannot continue with ECFLOW"
+    exit 1
+  else
+    echo "ECF_HOST: ${ECF_HOST}, ECF_PORT: ${ECF_PORT}"
+  fi
+
+  # Start the ecflow_server
   ecflow_client --ping --host="${ECF_HOST}" --port="${ECF_PORT}"
   not_running=$?
   if [[ ${not_running} -eq 1 ]]; then
     echo "ecflow_server is NOT running on ${ECF_HOST}:${ECF_PORT}"
     if [[ ${MACHINE_ID} == wcoss2 || ${MACHINE_ID} == acorn ]]; then
-      if [[ "${HOST::1}" == "a" ]]; then
-    export ECF_HOST=aecflow01
-      elif [[ "${HOST::1}" == "c" ]]; then
-    export ECF_HOST=cdecflow01
-      elif [[ "${HOST::1}" == "d" ]]; then
-    export ECF_HOST=ddecflow01
-      fi
       #shellcheck disable=SC2029
-      ssh "${ECF_HOST}" "bash -l -c \"module load ecflow && ecflow_start.sh -p ${ECF_PORT}\""
+      ssh "${ECF_HOST}" "bash -l -c \"module load ecflow && ${ECFLOW_START} -p ${ECF_PORT}\""
     elif [[ ${MACHINE_ID} == hera || ${MACHINE_ID} == jet ]]; then
-      module load ecflow
-      echo "On ${MACHINE_ID}, start ecFlow server on dedicated node ${ECF_HOST}"
       #shellcheck disable=SC2029
       ssh "${ECF_HOST}" "bash -l -c \"module load ecflow && ${ECFLOW_START} -d ${RUNDIR_ROOT}/ecflow_server\""
     else
       ${ECFLOW_START} -p "${ECF_PORT}" -d "${RUNDIR_ROOT}/ecflow_server"
     fi
+
+    # Try pinging ecflow server now, and erroring out if not there.
+    ecflow_client --ping --host="${ECF_HOST}" --port="${ECF_PORT}"
+    not_running=$?
+    if [[ ${not_running} -eq 1 ]]; then
+      echo "ERROR: Failure to start ecflow, exiting..."
+      exit 1
+    fi
   else
     echo "ecflow_server is already running on ${ECF_HOST}:${ECF_PORT}"
   fi
-  set -e
-
+  #set -e
   ECFLOW_RUNNING=true
-
-  export ECF_PORT
-  export ECF_HOST
-
   ecflow_client --load="${ECFLOW_RUN}/${ECFLOW_SUITE}.def"
   ecflow_client --begin="${ECFLOW_SUITE}"
   ecflow_client --restart
@@ -819,16 +837,16 @@ ecflow_run() {
 
 ecflow_kill() {
    [[ ${ECFLOW_RUNNING:-false} == true ]] || return
-   set +e
+   #set +e
    ecflow_client --suspend "/${ECFLOW_SUITE}"
    ecflow_client --kill "/${ECFLOW_SUITE}"
    sleep 20
-   ecflow_client --delete=force yes"/${ECFLOW_SUITE}"
+   ecflow_client --delete=force yes "/${ECFLOW_SUITE}"
 }
 
 ecflow_stop() {
    [[ ${ECFLOW_RUNNING:-false} == true ]] || return
-   set +e
+   #set +e
    SUITES=$( ecflow_client --get )
    SUITES=$( grep "^suite" <<< "${SUITES}" )
    #SUITES=$( ecflow_client --get | grep "^suite" )