Add support for forecast-only runs on AWS (#2711)

The purpose of this PR it to merge code that allowing global-workflow run on AWS, first focus on ATM forecast only. Resolves: #2709 --------- Co-authored-by: David Huber <[email protected]>
NOAA-EMC · Aug 13, 2024 · eba813f · eba813f
1 parent 5699167
commit eba813f
Show file tree

Hide file tree

Showing 19 changed files with 169 additions and 82 deletions.
diff --git a/env/AWSPW.env b/env/AWSPW.env
@@ -9,8 +9,8 @@ fi
 
 step=$1
 
-export launcher="mpiexec.hydra"
-export mpmd_opt=""
+export launcher="srun -l --export=ALL"
+export mpmd_opt="--multi-prog --output=mpmd.%j.%t.out"
 
 # Configure MPI environment
 export OMP_STACKSIZE=2048000
@@ -35,6 +35,8 @@ fi
 
 if [[ "${step}" = "fcst" ]] || [[ "${step}" = "efcs" ]]; then
 
+    export launcher="srun --mpi=pmi2 -l"
+
     (( nnodes = (ntasks+tasks_per_node-1)/tasks_per_node ))
     (( ufs_ntasks = nnodes*tasks_per_node ))
     # With ESMF threading, the model wants to use the full node

diff --git a/modulefiles/module_base.noaacloud.lua b/modulefiles/module_base.noaacloud.lua
@@ -0,0 +1,49 @@
+help([[
+Load environment to run GFS on noaacloud
+]])
+
+local spack_mod_path=(os.getenv("spack_mod_path") or "None")
+prepend_path("MODULEPATH", spack_mod_path)
+
+load(pathJoin("stack-intel", (os.getenv("stack_intel_ver") or "None")))
+load(pathJoin("stack-intel-oneapi-mpi", (os.getenv("stack_impi_ver") or "None")))
+load(pathJoin("python", (os.getenv("python_ver") or "None")))
+
+load(pathJoin("jasper", (os.getenv("jasper_ver") or "None")))
+load(pathJoin("libpng", (os.getenv("libpng_ver") or "None")))
+load(pathJoin("cdo", (os.getenv("cdo_ver") or "None")))
+--load(pathJoin("R", (os.getenv("R_ver") or "None")))
+
+load(pathJoin("hdf5", (os.getenv("hdf5_ver") or "None")))
+load(pathJoin("netcdf-c", (os.getenv("netcdf_c_ver") or "None")))
+load(pathJoin("netcdf-fortran", (os.getenv("netcdf_fortran_ver") or "None")))
+
+load(pathJoin("nco", (os.getenv("nco_ver") or "None")))
+load(pathJoin("prod_util", (os.getenv("prod_util_ver") or "None")))
+load(pathJoin("grib-util", (os.getenv("grib_util_ver") or "None")))
+load(pathJoin("g2tmpl", (os.getenv("g2tmpl_ver") or "None")))
+load(pathJoin("gsi-ncdiag", (os.getenv("gsi_ncdiag_ver") or "None")))
+load(pathJoin("crtm", (os.getenv("crtm_ver") or "None")))
+load(pathJoin("bufr", (os.getenv("bufr_ver") or "None")))
+load(pathJoin("wgrib2", (os.getenv("wgrib2_ver") or "None")))
+load(pathJoin("py-f90nml", (os.getenv("py_f90nml_ver") or "None")))
+load(pathJoin("py-netcdf4", (os.getenv("py_netcdf4_ver") or "None")))
+load(pathJoin("py-pyyaml", (os.getenv("py_pyyaml_ver") or "None")))
+load(pathJoin("py-jinja2", (os.getenv("py_jinja2_ver") or "None")))
+load(pathJoin("py-pandas", (os.getenv("py_pandas_ver") or "None")))
+load(pathJoin("py-python-dateutil", (os.getenv("py_python_dateutil_ver") or "None")))
+--load(pathJoin("met", (os.getenv("met_ver") or "None")))
+--load(pathJoin("metplus", (os.getenv("metplus_ver") or "None")))
+load(pathJoin("py-xarray", (os.getenv("py_xarray_ver") or "None")))
+
+setenv("WGRIB2","wgrib2")
+setenv("UTILROOT",(os.getenv("prod_util_ROOT") or "None"))
+
+--prepend_path("MODULEPATH", pathJoin("/scratch1/NCEPDEV/global/glopara/git/prepobs/v" .. (os.getenv("prepobs_run_ver") or "None"), "modulefiles"))
+--prepend_path("MODULEPATH", pathJoin("/scratch1/NCEPDEV/global/glopara/git/prepobs/feature-GFSv17_com_reorg_log_update/modulefiles"))
+--load(pathJoin("prepobs", (os.getenv("prepobs_run_ver") or "None")))
+
+--prepend_path("MODULEPATH", pathJoin("/scratch1/NCEPDEV/global/glopara/git/Fit2Obs/v" .. (os.getenv("fit2obs_ver") or "None"), "modulefiles"))
+--load(pathJoin("fit2obs", (os.getenv("fit2obs_ver") or "None")))
+
+whatis("Description: GFS run environment")
diff --git a/modulefiles/module_gwci.noaacloud.lua b/modulefiles/module_gwci.noaacloud.lua
@@ -0,0 +1,15 @@
+help([[
+Load environment to run GFS workflow setup scripts on noaacloud
+]])
+
+prepend_path("MODULEPATH", "/contrib/spack-stack/spack-stack-1.6.0/envs/unified-env/install/modulefiles/Core")
+
+load(pathJoin("stack-intel", os.getenv("2021.3.0")))
+load(pathJoin("stack-intel-oneapi-mpi", os.getenv("2021.3.0")))
+
+load(pathJoin("netcdf-c", os.getenv("4.9.2")))
+load(pathJoin("netcdf-fortran", os.getenv("4.6.1")))
+load(pathJoin("nccmp","1.9.0.1"))
+load(pathJoin("wgrib2", "2.0.8"))
+
+whatis("Description: GFS run setup CI environment")
diff --git a/modulefiles/module_gwsetup.noaacloud.lua b/modulefiles/module_gwsetup.noaacloud.lua
@@ -0,0 +1,20 @@
+help([[
+Load environment to run GFS workflow setup scripts on noaacloud
+]])
+
+load(pathJoin("rocoto"))
+
+prepend_path("MODULEPATH", "/contrib/spack-stack/spack-stack-1.6.0/envs/unified-env/install/modulefiles/Core")
+
+local stack_intel_ver=os.getenv("stack_intel_ver") or "2021.3.0"
+local python_ver=os.getenv("python_ver") or "3.10.3"
+
+load(pathJoin("stack-intel", stack_intel_ver))
+load(pathJoin("python", python_ver))
+load("py-jinja2")
+load("py-pyyaml")
+load("py-numpy")
+local git_ver=os.getenv("git_ver") or "1.8.3.1"
+load(pathJoin("git", git_ver))
+
+whatis("Description: GFS run setup environment")
diff --git a/parm/config/gfs/config.base b/parm/config/gfs/config.base
@@ -483,4 +483,12 @@ export OFFSET_START_HOUR=0
 # Number of regional collectives to create soundings for
 export NUM_SND_COLLECTIVES=${NUM_SND_COLLECTIVES:-9}
 
+# The tracker, genesis, and METplus jobs are not supported on AWS yet
+# TODO: we should place these in workflow/hosts/awspw.yaml as part of AWS setup, not for general.
+if [[ "${machine}" == "AWSPW" ]]; then
+   export DO_TRACKER="NO"
+   export DO_GENESIS="NO"
+   export DO_METP="NO"
+fi
+
 echo "END: config.base"
diff --git a/parm/config/gfs/config.resources b/parm/config/gfs/config.resources
@@ -106,7 +106,8 @@ case ${machine} in
     ;;
   "AWSPW")
     export PARTITION_BATCH="compute"
-    max_tasks_per_node=40
+    npe_node_max=36
+    max_tasks_per_node=36
     # TODO Supply a max mem/node value for AWS
     # shellcheck disable=SC2034
     mem_node_max=""

diff --git a/parm/config/gfs/config.resources.AWSPW b/parm/config/gfs/config.resources.AWSPW
@@ -0,0 +1,10 @@
+#! /usr/bin/env bash
+
+# AWS-specific job resources
+
+export is_exclusive="True"
+
+# shellcheck disable=SC2312
+for mem_var in $(env | grep '^memory_' | cut -d= -f1); do
+  unset "${mem_var}"
+done
diff --git a/sorc/build_all.sh b/sorc/build_all.sh
@@ -145,7 +145,7 @@ build_opts["ww3prepost"]="${_wave_opt} ${_verbose_opt} ${_build_ufs_opt} ${_buil
 
 # Optional DA builds
 if [[ "${_build_ufsda}" == "YES" ]]; then
-   if [[ "${MACHINE_ID}" != "orion" && "${MACHINE_ID}" != "hera" && "${MACHINE_ID}" != "hercules" && "${MACHINE_ID}" != "wcoss2" ]]; then
+   if [[ "${MACHINE_ID}" != "orion" && "${MACHINE_ID}" != "hera" && "${MACHINE_ID}" != "hercules" && "${MACHINE_ID}" != "wcoss2" && "${MACHINE_ID}" != "noaacloud" ]]; then
       echo "NOTE: The GDAS App is not supported on ${MACHINE_ID}.  Disabling build."
    else
       build_jobs["gdas"]=8

diff --git a/sorc/build_ufs.sh b/sorc/build_ufs.sh
@@ -41,30 +41,9 @@ COMPILE_NR=0
 CLEAN_BEFORE=YES
 CLEAN_AFTER=NO
 
-if [[ "${MACHINE_ID}" != "noaacloud" ]]; then
-  BUILD_JOBS=${BUILD_JOBS:-8} ./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
-    set +x
-    # 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
-    set -x
-  fi
-
-  export CMAKE_FLAGS="${MAKE_OPT}"
-  BUILD_JOBS=${BUILD_JOBS:-8} ./build.sh
-  mv "${cwd}/ufs_model.fd/build/ufs_model" "${cwd}/ufs_model.fd/tests/ufs_model.x"
-fi
+BUILD_JOBS=${BUILD_JOBS:-8} ./tests/compile.sh "${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
 
 exit 0
diff --git a/sorc/link_workflow.sh b/sorc/link_workflow.sh
@@ -76,6 +76,7 @@ case "${machine}" in
   "jet")      FIX_DIR="/lfs4/HFIP/hfv3gfs/glopara/git/fv3gfs/fix" ;;
   "s4")       FIX_DIR="/data/prod/glopara/fix" ;;
   "gaea")     FIX_DIR="/gpfs/f5/ufs-ard/world-shared/global/glopara/data/fix" ;;
+  "noaacloud") FIX_DIR="/contrib/global-workflow-shared-data/fix" ;;
   *)
     echo "FATAL: Unknown target machine ${machine}, couldn't set FIX_DIR"
     exit 1

diff --git a/ush/load_fv3gfs_modules.sh b/ush/load_fv3gfs_modules.sh
@@ -20,7 +20,7 @@ source "${HOMEgfs}/versions/run.ver"
 module use "${HOMEgfs}/modulefiles"
 
 case "${MACHINE_ID}" in
-  "wcoss2" | "hera" | "orion" | "hercules" | "gaea" | "jet" | "s4")
+  "wcoss2" | "hera" | "orion" | "hercules" | "gaea" | "jet" | "s4" | "noaacloud")
     module load "module_base.${MACHINE_ID}"
     ;;
   *)

diff --git a/ush/load_ufswm_modules.sh b/ush/load_ufswm_modules.sh
@@ -11,40 +11,21 @@ ulimit_s=$( ulimit -S -s )
 
 source "${HOMEgfs}/ush/detect_machine.sh"
 source "${HOMEgfs}/ush/module-setup.sh"
-if [[ "${MACHINE_ID}" != "noaacloud" ]]; then
-  module use "${HOMEgfs}/sorc/ufs_model.fd/modulefiles"
-  module load "ufs_${MACHINE_ID}.intel"
-  module load prod_util
-  if [[ "${MACHINE_ID}" = "wcoss2" ]]; then
-    module load cray-pals
-    module load cfp
-    module load libjpeg
-    module load craype-network-ucx
-    module load cray-mpich-ucx
-  else
-    export UTILROOT=${prod_util_ROOT}
-  fi
-  module load wgrib2
-  export WGRIB2=wgrib2
-fi
-if [[ "${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="${ndate_path}"
-   fi
+module use "${HOMEgfs}/sorc/ufs_model.fd/modulefiles"
+module load "ufs_${MACHINE_ID}.intel"
+module load prod_util
+if [[ "${MACHINE_ID}" = "wcoss2" ]]; then
+  module load cray-pals
+  module load cfp
+  module load libjpeg
+  module load craype-network-ucx
+  module load cray-mpich-ucx
+else
+  export UTILROOT=${prod_util_ROOT}
 fi
+module load wgrib2
+export WGRIB2=wgrib2
 
 module list
 unset MACHINE_ID

diff --git a/ush/module-setup.sh b/ush/module-setup.sh
@@ -92,10 +92,8 @@ elif [[ ${MACHINE_ID} = discover* ]]; then
 # 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
+    # We are on NOAA Cloud
+    module purge
 
 else
     echo WARNING: UNKNOWN PLATFORM 1>&2

diff --git a/versions/build.noaacloud.ver b/versions/build.noaacloud.ver
@@ -0,0 +1,5 @@
+export stack_intel_ver=2021.3.0
+export stack_impi_ver=2021.3.0
+export spack_env=gsi-addon-env
+source "${HOMEgfs:-}/versions/build.spack.ver"
+export spack_mod_path="/contrib/spack-stack/spack-stack-${spack_stack_ver}/envs/gsi-addon-env/install/modulefiles/Core"
diff --git a/versions/run.noaacloud.ver b/versions/run.noaacloud.ver
@@ -0,0 +1,8 @@
+export stack_intel_ver=2021.3.0
+export stack_impi_ver=2021.3.0
+export spack_env=gsi-addon-env
+
+source "${HOMEgfs:-}/versions/run.spack.ver"
+export spack_mod_path="/contrib/spack-stack/spack-stack-${spack_stack_ver}/envs/gsi-addon-env/install/modulefiles/Core"
+
+export cdo_ver=2.2.0
diff --git a/workflow/hosts.py b/workflow/hosts.py
@@ -52,7 +52,7 @@ def detect(cls):
         elif container is not None:
             machine = 'CONTAINER'
         elif pw_csp is not None:
-            if pw_csp.lower() not in ['azure', 'aws', 'gcp']:
+            if pw_csp.lower() not in ['azure', 'aws', 'google']:
                 raise ValueError(
                     f'NOAA cloud service provider "{pw_csp}" is not supported.')
             machine = f"{pw_csp.upper()}PW"

diff --git a/workflow/hosts/awspw.yaml b/workflow/hosts/awspw.yaml
@@ -1,12 +1,12 @@
-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.
-COMINsyn: '/scratch1/NCEPDEV/global/glopara/com/gfs/prod/syndat' #TODO: This does not yet exist.
+BASE_GIT: '' #TODO: This does not yet exist.
+DMPDIR: '' # TODO: This does not yet exist.
+PACKAGEROOT: '' #TODO: This does not yet exist.
+COMINsyn: '' #TODO: This does not yet exist.
 HOMEDIR: '/contrib/${USER}'
-STMP: '/lustre/${USER}/stmp2/'
-PTMP: '/lustre/${USER}/stmp4/'
-NOSCRUB: ${HOMEDIR}
-ACCOUNT: hwufscpldcld
+STMP: '/lustre/${USER}/stmp/'
+PTMP: '/lustre/${USER}/ptmp/'
+NOSCRUB: '${HOMEDIR}'
+ACCOUNT: '${USER}'
 SCHEDULER: slurm
 QUEUE: batch
 QUEUE_SERVICE: batch
@@ -16,10 +16,11 @@ RESERVATION: ''
 CLUSTERS: ''
 CHGRP_RSTPROD: 'YES'
 CHGRP_CMD: 'chgrp rstprod' # TODO: This is not yet supported.
-HPSSARCH: 'YES'
+HPSSARCH: 'NO'
 HPSS_PROJECT: emc-global #TODO: See `ATARDIR` below.
+BASE_CPLIC: '/bucket/global-workflow-shared-data/ICSDIR/prototype_ICs'
 LOCALARCH: 'NO'
-ATARDIR: '/NCEPDEV/${HPSS_PROJECT}/1year/${USER}/${machine}/scratch/${PSLOT}' # TODO: This will not yet work from AWS.
+ATARDIR: '' # TODO: This will not yet work from AWS.
 MAKE_NSSTBUFR: 'NO'
 MAKE_ACFTBUFR: 'NO'
 SUPPORTED_RESOLUTIONS: ['C48', 'C96'] # TODO: Test and support all cubed-sphere resolutions.
diff --git a/workflow/rocoto/tasks.py b/workflow/rocoto/tasks.py
@@ -209,7 +209,10 @@ def get_resource(self, task_name):
             else:
                 native += ':shared'
         elif scheduler in ['slurm']:
-            native = '--export=NONE'
+            if task_config.get('is_exclusive', False):
+                native = '--exclusive'
+            else:
+                native = '--export=NONE'
             if task_config['RESERVATION'] != "":
                 native += '' if task_name in Tasks.SERVICE_TASKS else ' --reservation=' + task_config['RESERVATION']
             if task_config.get('CLUSTERS', "") not in ["", '@CLUSTERS@']:

diff --git a/workflow/rocoto/workflow_xml.py b/workflow/rocoto/workflow_xml.py
@@ -157,10 +157,16 @@ def _write_crontab(self, crontab_file: str = None, cronint: int = 5) -> None:
 
         strings = ['',
                    f'#################### {pslot} ####################',
-                   f'MAILTO="{replyto}"',
-                   f'{cronintstr} {rocotorunstr}',
-                   '#################################################################',
-                   '']
+                   f'MAILTO="{replyto}"'
+                   ]
+        # AWS need 'SHELL', and 'BASH_ENV' defined, or, the crontab job won't start.
+        if os.environ.get('PW_CSP', None) in ['aws', 'azure', 'google']:
+            strings.extend([f'SHELL="/bin/bash"',
+                            f'BASH_ENV="/etc/bashrc"'
+                            ])
+        strings.extend([f'{cronintstr} {rocotorunstr}',
+                        '#################################################################',
+                        ''])
 
         if crontab_file is None:
             crontab_file = f"{expdir}/{pslot}.crontab"