Move analysis post to new stand-alone task (#1979)

This PR addresses issue #1130. A new task `postanl.sh` has been added such that the
`anl` post-processing is not it's own task within the Rocoto workflow. This change
prevents the GDAS experiment from breaking due to failures caused by the `anl` not
being available for the first cycle.

Resolves #1130

jobs/JGLOBAL_ATMOS_POST

@@ -17,18 +17,6 @@ export g2tmpl_ver=${g2tmpl_ver:-v1.5.0}
 ##############################################
 export CDUMP=${RUN/enkf}
 
-
-##############################################
-# TODO: Remove this egregious HACK
-##############################################
-if [[ "${SDATE:-}" = "${PDY}${cyc}" ]]; then
-  if [[ ${post_times} = "anl" ]]; then
-    echo "No offline post-processing in the first half cycle for analysis"
-    exit 0
-  fi
-fi
-
-
 ##############################################
 # Begin JOB SPECIFIC work
 ##############################################

jobs/rocoto/post.sh

@@ -4,7 +4,6 @@ source "${HOMEgfs}/ush/preamble.sh"
 
 ###############################################################
 ## NCEP post driver script
-## FHRGRP : forecast hour group to post-process (e.g. 0, 1, 2 ...)
 ## FHRLST : forecast hourlist to be post-process (e.g. anl, f000, f000_f001_f002, ...)
 ###############################################################
 
@@ -16,11 +15,7 @@ status=$?
 export job="post"
 export jobid="${job}.$$"
 
-if [ ${FHRGRP} = 'anl' ]; then
-    fhrlst="anl"
-else
-    fhrlst=$(echo ${FHRLST} | sed -e 's/_/ /g; s/f/ /g; s/,/ /g')
-fi
+fhrlst=$(echo ${FHRLST} | sed -e 's/_/ /g; s/f/ /g; s/,/ /g')
 
 #---------------------------------------------------------------
 for fhr in ${fhrlst}; do

jobs/rocoto/postanl.sh

@@ -0,0 +1 @@
+post.sh

workflow/applications/gfs_cycled.py

@@ -101,7 +101,7 @@ def get_task_names(self):
         """
 
         gdas_gfs_common_tasks_before_fcst = ['prep']
-        gdas_gfs_common_tasks_after_fcst = ['post']
+        gdas_gfs_common_tasks_after_fcst = ['postanl', 'post']
         # if self.do_ocean:  # TODO: uncomment when ocnpost is fixed in cycled mode
         #    gdas_gfs_common_tasks_after_fcst += ['ocnpost']
         gdas_gfs_common_tasks_after_fcst += ['vrfy']

workflow/hosts/awspw.yaml

@@ -21,4 +21,4 @@ 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.
+SUPPORTED_RESOLUTIONS: ['C48', 'C96'] # TODO: Test and support all cubed-sphere resolutions.

workflow/rocoto/gfs_tasks.py

@@ -549,24 +549,37 @@ def _fcst_cycled(self):
         return task
 
     def post(self):
-        add_anl_to_post = False
-        if self.app_config.mode in ['cycled']:
-            add_anl_to_post = True
+        return self._post_task('post')
+
+    def postanl(self):
+        postenvars = self.envars.copy()
+        postenvar_dict = {'FHRLST': 'anl',
+                          'ROTDIR': self._base.get('ROTDIR')}
+
+        for key, value in postenvar_dict.items():
+            postenvars.append(rocoto.create_envar(name=key, value=str(value)))
 
-        return self._post_task('post', add_anl_to_post=add_anl_to_post)
+        deps = []
+        atm_anl_path = self._template_to_rocoto_cycstring(self._base["COM_ATMOS_ANALYSIS_TMPL"])
+        data = f'{atm_anl_path}/{self.cdump}.t@Hz.loganl.txt'
+        dep_dict = {'type': 'data', 'data': data}
+        deps.append(rocoto.add_dependency(dep_dict))
+        dependencies = rocoto.create_dependency(dep=deps)
+        resources = self.get_resource('post')
+        task = create_wf_task('postanl', resources, cdump=self.cdump, envar=postenvars, dependency=dependencies,
+                              cycledef=self.cdump)
+
+        return task
 
     def ocnpost(self):
         if self.app_config.mode in ['forecast-only']:  # TODO: fix ocnpost in cycled mode
-            return self._post_task('ocnpost', add_anl_to_post=False)
+            return self._post_task('ocnpost')
 
-    def _post_task(self, task_name, add_anl_to_post=False):
+    def _post_task(self, task_name):
         if task_name not in ['post', 'ocnpost']:
             raise KeyError(f'Invalid post-processing task: {task_name}')
 
-        if task_name in ['ocnpost']:
-            add_anl_to_post = False
-
-        def _get_postgroups(cdump, config, add_anl=False):
+        def _get_postgroups(cdump, config):
 
             fhmin = config['FHMIN']
             fhmax = config['FHMAX']
@@ -591,8 +604,6 @@ def _get_postgroups(cdump, config, add_anl=False):
             fhrs = [f'f{fhr:03d}' for fhr in fhrs]
             fhrs = np.array_split(fhrs, ngrps)
             fhrs = [fhr.tolist() for fhr in fhrs]
-            if add_anl:
-                fhrs.insert(0, ['anl'])
 
             grp = ' '.join(f'_{fhr[0]}-{fhr[-1]}' if len(fhr) > 1 else f'_{fhr[0]}' for fhr in fhrs)
             dep = ' '.join([fhr[-1] for fhr in fhrs])
@@ -610,14 +621,13 @@ def _get_postgroups(cdump, config, add_anl=False):
         dependencies = rocoto.create_dependency(dep_condition='or', dep=deps)
 
         postenvars = self.envars.copy()
-        postenvar_dict = {'FHRGRP': '#grp#',
-                          'FHRLST': '#lst#',
+        postenvar_dict = {'FHRLST': '#lst#',
                           'ROTDIR': self._base.get('ROTDIR')}
         for key, value in postenvar_dict.items():
             postenvars.append(rocoto.create_envar(name=key, value=str(value)))
 
         varname1, varname2, varname3 = 'grp', 'dep', 'lst'
-        varval1, varval2, varval3 = _get_postgroups(self.cdump, self._configs[task_name], add_anl=add_anl_to_post)
+        varval1, varval2, varval3 = _get_postgroups(self.cdump, self._configs[task_name])
         vardict = {varname2: varval2, varname3: varval3}
 
         cycledef = 'gdas_half,gdas' if self.cdump in ['gdas'] else self.cdump