Add AWS functionality to workflow setup and forecast job

env/AWSPW.env

@@ -0,0 +1,137 @@
+#! /usr/bin/env bash
+
+if [[ $# -ne 1 ]]; then
+
+    echo "Must specify an input argument to set runtime environment variables!"
+    echo "argument can be any one of the following:"
+    echo "atmanlrun atmensanlrun aeroanlrun landanlrun"
+    echo "anal sfcanl fcst post vrfy metp"
+    echo "eobs eupd ecen efcs epos"
+    echo "postsnd awips gempak"
+    exit 1
+
+fi
+
+step=$1
+
+export npe_node_max=36
+export launcher="mpiexec.hydra"
+export mpmd_opt=""
+
+# Configure MPI environment
+export OMP_STACKSIZE=2048000
+export NTHSTACK=1024000000
+
+ulimit -s unlimited
+ulimit -a
+
+if [[ "${step}" = "fcst" ]] || [[ "${step}" = "efcs" ]]; then
+
+    if [[ "${CDUMP}" =~ "gfs" ]]; then
+        nprocs="npe_${step}_gfs"
+        ppn="npe_node_${step}_gfs" || ppn="npe_node_${step}"
+    else
+        nprocs="npe_${step}"
+        ppn="npe_node_${step}"
+    fi
+    (( nnodes = (${!nprocs}+${!ppn}-1)/${!ppn} ))
+    (( ntasks = nnodes*${!ppn} ))
+    # With ESMF threading, the model wants to use the full node
+    export APRUN_UFS="${launcher} -n ${ntasks}"
+    unset nprocs ppn nnodes ntasks
+
+elif [[ "${step}" = "post" ]]; then
+
+    nth_max=$((npe_node_max / npe_node_post))
+
+    export NTHREADS_NP=${nth_np:-1}
+    [[ ${NTHREADS_NP} -gt ${nth_max} ]] && export NTHREADS_NP=${nth_max}
+    export APRUN_NP="${launcher} -n ${npe_post}"
+
+    export NTHREADS_DWN=${nth_dwn:-1}
+    [[ ${NTHREADS_DWN} -gt ${nth_max} ]] && export NTHREADS_DWN=${nth_max}
+    export APRUN_DWN="${launcher} -n ${npe_dwn}"
+
+elif [[ "${step}" = "ecen" ]]; then
+
+    nth_max=$((npe_node_max / npe_node_ecen))
+
+    export NTHREADS_ECEN=${nth_ecen:-${nth_max}}
+    [[ ${NTHREADS_ECEN} -gt ${nth_max} ]] && export NTHREADS_ECEN=${nth_max}
+    export APRUN_ECEN="${launcher} -n ${npe_ecen}"
+
+    export NTHREADS_CHGRES=${nth_chgres:-12}
+    [[ ${NTHREADS_CHGRES} -gt ${npe_node_max} ]] && export NTHREADS_CHGRES=${npe_node_max}
+    export APRUN_CHGRES="time"
+
+    export NTHREADS_CALCINC=${nth_calcinc:-1}
+    [[ ${NTHREADS_CALCINC} -gt ${nth_max} ]] && export NTHREADS_CALCINC=${nth_max}
+    export APRUN_CALCINC="${launcher} -n ${npe_ecen}"
+
+elif [[ "${step}" = "esfc" ]]; then
+
+    nth_max=$((npe_node_max / npe_node_esfc))
+
+    export NTHREADS_ESFC=${nth_esfc:-${nth_max}}
+    [[ ${NTHREADS_ESFC} -gt ${nth_max} ]] && export NTHREADS_ESFC=${nth_max}
+    export APRUN_ESFC="${launcher} -n ${npe_esfc}"
+
+    export NTHREADS_CYCLE=${nth_cycle:-14}
+    [[ ${NTHREADS_CYCLE} -gt ${npe_node_max} ]] && export NTHREADS_CYCLE=${npe_node_max}
+    export APRUN_CYCLE="${launcher} -n ${npe_esfc}"
+
+elif [[ "${step}" = "epos" ]]; then
+
+    nth_max=$((npe_node_max / npe_node_epos))
+
+    export NTHREADS_EPOS=${nth_epos:-${nth_max}}
+    [[ ${NTHREADS_EPOS} -gt ${nth_max} ]] && export NTHREADS_EPOS=${nth_max}
+    export APRUN_EPOS="${launcher} -n ${npe_epos}"
+
+elif [[ "${step}" = "postsnd" ]]; then
+
+    export CFP_MP="YES"
+
+    nth_max=$((npe_node_max / npe_node_postsnd))
+
+    export NTHREADS_POSTSND=${nth_postsnd:-1}
+    [[ ${NTHREADS_POSTSND} -gt ${nth_max} ]] && export NTHREADS_POSTSND=${nth_max}
+    export APRUN_POSTSND="${launcher} -n ${npe_postsnd}"
+
+    export NTHREADS_POSTSNDCFP=${nth_postsndcfp:-1}
+    [[ ${NTHREADS_POSTSNDCFP} -gt ${nth_max} ]] && export NTHREADS_POSTSNDCFP=${nth_max}
+    export APRUN_POSTSNDCFP="${launcher} -n ${npe_postsndcfp} ${mpmd_opt}"
+
+elif [[ "${step}" = "awips" ]]; then
+
+    nth_max=$((npe_node_max / npe_node_awips))
+
+    export NTHREADS_AWIPS=${nth_awips:-2}
+    [[ ${NTHREADS_AWIPS} -gt ${nth_max} ]] && export NTHREADS_AWIPS=${nth_max}
+    export APRUN_AWIPSCFP="${launcher} -n ${npe_awips} ${mpmd_opt}"
+
+elif [[ "${step}" = "gempak" ]]; then
+
+    export CFP_MP="YES"
+
+    if [[ ${CDUMP} == "gfs" ]]; then
+        npe_gempak=${npe_gempak_gfs}
+        npe_node_gempak=${npe_node_gempak_gfs}
+    fi
+
+    nth_max=$((npe_node_max / npe_node_gempak))
+
+    export NTHREADS_GEMPAK=${nth_gempak:-1}
+    [[ ${NTHREADS_GEMPAK} -gt ${nth_max} ]] && export NTHREADS_GEMPAK=${nth_max}
+    export APRUN="${launcher} -n ${npe_gempak} ${mpmd_opt}"
+
+
+elif [[ "${step}" = "fit2obs" ]]; then
+
+    nth_max=$((npe_node_max / npe_node_fit2obs))
+
+    export NTHREADS_FIT2OBS=${nth_fit2obs:-1}
+    [[ ${NTHREADS_FIT2OBS} -gt ${nth_max} ]] && export NTHREADS_FIT2OBS=${nth_max}
+    export MPIRUN="${launcher} -n ${npe_fit2obs}"
+
+fi

jobs/rocoto/fcst.sh

@@ -12,9 +12,12 @@ source "${HOMEgfs}/ush/preamble.sh"
 source "${HOMEgfs}/ush/detect_machine.sh"
 set +x
 source "${HOMEgfs}/ush/module-setup.sh"
-module use "${HOMEgfs}/sorc/ufs_model.fd/tests"
-module load modules.ufs_model.lua
-module load prod_util
+if [[ "${MACHINE_ID}" != "noaacloud" ]]; then
+  module use "${HOMEgfs}/sorc/ufs_model.fd/tests"
+  module load modules.ufs_model.lua
+  module load prod_util
+fi
+
 if [[ "${MACHINE_ID}" = "wcoss2" ]]; then
   module load cray-pals
 fi
@@ -30,6 +33,25 @@ if [[ "${MACHINE_ID}" = "hera" ]]; then
 #elif [[ "${MACHINE_ID}" = "wcoss2" ]]; then
 #  module load "python/3.7.5"
 fi
+if [[ "${MACHINE_ID}" == "noaacloud" ]]; then
+   if [[ "${PW_CSP:-}" = "aws" ]]; then
+
+      # TODO: This can be cleaned-up; most of this is a hack for now.
+      module use "/contrib/spack-stack/envs/ufswm/install/modulefiles/Core"
+      module load "stack-intel"
+      module load "stack-intel-oneapi-mpi"
+      module use -a "/contrib/spack-stack/miniconda/modulefiles/miniconda/"
+      module load "py39_4.12.0"
+      module load "ufs-weather-model-env/1.0.0"
+      export NETCDF="/contrib/spack-stack/miniconda/apps/miniconda/py39_4.12.0"
+      # TODO: Are there plans for EPIC to maintain this package or should GW provide support?
+      export UTILROOT="/contrib/global-workflow/NCEPLIBS-prod_util" 
+      export PATH="${PATH}:/contrib/global-workflow/bin"
+      ndate_path="$(command -v ndate)"
+      export NDATE="${ndate_path}"
+   fi
+fi
+
 module list
 unset MACHINE_ID
 set_trace

sorc/build_ufs.sh

@@ -32,10 +32,28 @@ COMPILE_NR=0
 CLEAN_BEFORE=YES
 CLEAN_AFTER=NO
 
-./tests/compile.sh "${MACHINE_ID}" "${MAKE_OPT}" "${COMPILE_NR}" "intel" "${CLEAN_BEFORE}" "${CLEAN_AFTER}"
-mv "./tests/fv3_${COMPILE_NR}.exe" ./tests/ufs_model.x
-mv "./tests/modules.fv3_${COMPILE_NR}.lua" ./tests/modules.ufs_model.lua
-cp "./modulefiles/ufs_common.lua" ./tests/ufs_common.lua
-cp "./modulefiles/ufs_common_spack.lua" ./tests/ufs_common_spack.lua
+if [[ "${MACHINE_ID}" != "noaacloud" ]]; then
+  ./tests/compile.sh "${MACHINE_ID}" "${MAKE_OPT}" "${COMPILE_NR}" "intel" "${CLEAN_BEFORE}" "${CLEAN_AFTER}"
+  mv "./tests/fv3_${COMPILE_NR}.exe" ./tests/ufs_model.x
+  mv "./tests/modules.fv3_${COMPILE_NR}.lua" ./tests/modules.ufs_model.lua
+  cp "./modulefiles/ufs_common.lua" ./tests/ufs_common.lua
+else
+
+  if [[ "${PW_CSP:-}" == "aws" ]]; then
+    # TODO: This will need to be addressed further when the EPIC stacks are available/supported.
+    module use /contrib/spack-stack/envs/ufswm/install/modulefiles/Core
+    module load stack-intel
+    module load stack-intel-oneapi-mpi
+    module load ufs-weather-model-env/1.0.0
+    # TODO: It is still uncertain why this is the only module that is 
+    # missing; check the spack build as this needed to be added manually.
+    module load w3emc/2.9.2 # TODO: This has similar issues for the EPIC stack.
+    module list
+  fi
+
+  export CMAKE_FLAGS="${MAKE_OPT}"
+  ./build.sh 
+  mv "${cwd}/ufs_model.fd/build/ufs_model" "${cwd}/ufs_model.fd/tests/ufs_model.x"
+fi
 
 exit 0

ush/detect_machine.sh

@@ -38,6 +38,13 @@ case $(hostname -f) in
   *) MACHINE_ID=UNKNOWN ;;  # Unknown platform
 esac
 
+if [[ ${MACHINE_ID} == "UNKNOWN" ]]; then 
+   case ${PW_CSP:-} in
+      "aws" | "google" | "azure") MACHINE_ID=noaacloud ;;
+      *) PW_CSP="UNKNOWN"
+   esac
+fi
+
 # Overwrite auto-detect with MACHINE if set
 MACHINE_ID=${MACHINE:-${MACHINE_ID}}
 

ush/module-setup.sh

@@ -102,6 +102,14 @@ elif [[ ${MACHINE_ID} = discover* ]]; then
     export PATH=${PATH}:${SPACK_ROOT}/bin
     . "${SPACK_ROOT}"/share/spack/setup-env.sh
 
+# TODO: This can likely be made more general once other cloud
+# platforms come online.
+elif [[ ${MACHINE_ID} = "noaacloud" ]]; then
+
+    export SPACK_ROOT=/contrib/global-workflow/spack-stack/spack
+    export PATH=${PATH}:${SPACK_ROOT}/bin
+    . "${SPACK_ROOT}"/share/spack/setup-env.sh
+
 else
     echo WARNING: UNKNOWN PLATFORM 1>&2
 fi

workflow/hosts.py

@@ -15,14 +15,15 @@ class Host:
     """
 
     SUPPORTED_HOSTS = ['HERA', 'ORION', 'JET',
-                       'WCOSS2', 'S4', 'CONTAINER']
+                       'WCOSS2', 'S4', 'CONTAINER', 'AWSPW']
 
     def __init__(self, host=None):
 
         detected_host = self.detect()
 
         if host is not None and host != detected_host:
-            raise ValueError(f'detected host: "{detected_host}" does not match host: "{host}"')
+            raise ValueError(
+                f'detected host: "{detected_host}" does not match host: "{host}"')
 
         self.machine = detected_host
         self.info = self._get_info
@@ -57,7 +58,8 @@ def detect(cls):
     @property
     def _get_info(self) -> dict:
 
-        hostfile = Path(os.path.join(os.path.dirname(__file__), f'hosts/{self.machine.lower()}.yaml'))
+        hostfile = Path(os.path.join(os.path.dirname(__file__),
+                        f'hosts/{self.machine.lower()}.yaml'))
         try:
             info = YAMLFile(path=hostfile)
         except FileNotFoundError:

workflow/hosts/awspw.yaml

@@ -0,0 +1,24 @@
+BASE_GIT: '/scratch1/NCEPDEV/global/glopara/git' #TODO: This does not yet exist.
+DMPDIR: '/scratch1/NCEPDEV/global/glopara/dump' # TODO: This does not yet exist.
+PACKAGEROOT: '/scratch1/NCEPDEV/global/glopara/nwpara' #TODO: This does not yet exist.
+COMROOT: '/scratch1/NCEPDEV/global/glopara/com' #TODO: This does not yet exist.
+COMINsyn: '${COMROOT}/gfs/prod/syndat' #TODO: This does not yet exist.
+HOMEDIR: '/contrib/${USER}'
+STMP: '/lustre/${USER}/stmp2/'
+PTMP: '/lustre/${USER}/stmp4/'
+NOSCRUB: $HOMEDIR
+ACCOUNT: hwufscpldcld
+SCHEDULER: slurm
+QUEUE: batch
+QUEUE_SERVICE: batch
+PARTITION_BATCH: compute
+PARTITION_SERVICE: compute
+CHGRP_RSTPROD: 'YES'
+CHGRP_CMD: 'chgrp rstprod' # TODO: This is not yet supported. 
+HPSSARCH: 'YES'
+HPSS_PROJECT: emc-global #TODO: See `ATARDIR` below.
+LOCALARCH: 'NO'
+ATARDIR: '/NCEPDEV/${HPSS_PROJECT}/1year/${USER}/${machine}/scratch/${PSLOT}' # TODO: This will not yet work from AWS.
+MAKE_NSSTBUFR: 'NO'
+MAKE_ACFTBUFR: 'NO'
+SUPPORTED_RESOLUTIONS: ['C48'] # TODO: Test and support all cubed-sphere resolutions.