Merge pull request #4572 from jedwards4b/jedwards/fix_dry_run_and_res…

…olve_subgroup

fix issue getting correct parameters for job subgroups
Merge without fixing e3sm test - seems to be a container issue.

CIME/XML/env_batch.py

@@ -813,6 +813,7 @@ def submit_jobs(
                     dry_run=dry_run,
                     workflow=workflow,
                 )
+
                 batch_job_id = str(alljobs.index(job)) if dry_run else result
                 depid[job] = batch_job_id
                 jobcmds.append((job, result))
@@ -921,35 +922,33 @@ def _submit_single_job(
             logger.info("Starting job script {}".format(job))
             function_name = job.replace(".", "_")
             job_name = "." + job
-            if not dry_run:
-                args = self._build_run_args(
-                    job,
-                    True,
-                    skip_pnl=skip_pnl,
-                    set_continue_run=resubmit_immediate,
-                    submit_resubmits=workflow and not resubmit_immediate,
-                )
-                try:
-                    if hasattr(case, function_name):
-                        getattr(case, function_name)(
-                            **{k: v for k, (v, _) in args.items()}
-                        )
+            args = self._build_run_args(
+                job,
+                True,
+                skip_pnl=skip_pnl,
+                set_continue_run=resubmit_immediate,
+                submit_resubmits=workflow and not resubmit_immediate,
+            )
+            try:
+                if hasattr(case, function_name):
+                    getattr(case, function_name)(**{k: v for k, (v, _) in args.items()})
+                else:
+                    expect(
+                        os.path.isfile(job_name),
+                        "Could not find file {}".format(job_name),
+                    )
+                    if dry_run:
+                        return os.path.join(self._caseroot, job_name)
                     else:
-                        expect(
-                            os.path.isfile(job_name),
-                            "Could not find file {}".format(job_name),
-                        )
                         run_cmd_no_fail(
                             os.path.join(self._caseroot, job_name),
                             combine_output=True,
                             verbose=True,
                             from_dir=self._caseroot,
                         )
-                except Exception as e:
-                    # We don't want exception from the run phases getting into submit phase
-                    logger.warning(
-                        "Exception from {}: {}".format(function_name, str(e))
-                    )
+            except Exception as e:
+                # We don't want exception from the run phases getting into submit phase
+                logger.warning("Exception from {}: {}".format(function_name, str(e)))
 
             return
 
@@ -1088,10 +1087,10 @@ def _submit_single_job(
             # add ` before cd $CASEROOT and at end of command
             submitcmd = submitcmd.replace("cd $CASEROOT", "'cd $CASEROOT") + "'"
 
+        submitcmd = case.get_resolved_value(submitcmd, subgroup=job)
         if dry_run:
             return submitcmd
         else:
-            submitcmd = case.get_resolved_value(submitcmd)
             logger.info("Submitting job script {}".format(submitcmd))
             output = run_cmd_no_fail(submitcmd, combine_output=True)
             jobid = self.get_job_id(output)

CIME/XML/env_workflow.py

@@ -112,7 +112,13 @@ def get_job_specs(self, case, job):
         if ngpus_per_node > max_gpus_per_node:
             ngpus_per_node = max_gpus_per_node
 
-        return task_count, num_nodes, tasks_per_node, thread_count, ngpus_per_node
+        return (
+            task_count,
+            num_nodes,
+            tasks_per_node,
+            thread_count,
+            ngpus_per_node,
+        )
 
     # pylint: disable=arguments-differ
     def get_value(self, item, attribute=None, resolved=True, subgroup="PRIMARY"):

CIME/XML/generic_xml.py

@@ -612,7 +612,9 @@ def set_value(
 
         return value if valnodes else None
 
-    def get_resolved_value(self, raw_value, allow_unresolved_envvars=False):
+    def get_resolved_value(
+        self, raw_value, allow_unresolved_envvars=False, subgroup=None
+    ):
         """
         A value in the xml file may contain references to other xml
         variables or to environment variables. These are refered to in
@@ -662,7 +664,8 @@ def get_resolved_value(self, raw_value, allow_unresolved_envvars=False):
             logger.debug("find: {}".format(var))
             # The overridden versions of this method do not simply return None
             # so the pylint should not be flagging this
-            ref = self.get_value(var)  # pylint: disable=assignment-from-none
+            # pylint: disable=assignment-from-none
+            ref = self.get_value(var, subgroup=subgroup)
 
             if ref is not None:
                 logger.debug("resolve: " + str(ref))

CIME/case/case.py

@@ -481,7 +481,7 @@ def get_value(self, item, attribute=None, resolved=True, subgroup=None):
 
             if result is not None:
                 if resolved and isinstance(result, str):
-                    result = self.get_resolved_value(result)
+                    result = self.get_resolved_value(result, subgroup=subgroup)
                     vtype = env_file.get_type_info(item)
                     if vtype is not None and vtype != "char":
                         result = convert_to_type(result, vtype, item)
@@ -548,13 +548,17 @@ def get_type_info(self, item):
 
         return result
 
-    def get_resolved_value(self, item, recurse=0, allow_unresolved_envvars=False):
+    def get_resolved_value(
+        self, item, recurse=0, allow_unresolved_envvars=False, subgroup=None
+    ):
         num_unresolved = item.count("$") if item else 0
         recurse_limit = 10
         if num_unresolved > 0 and recurse < recurse_limit:
             for env_file in self._env_entryid_files:
                 item = env_file.get_resolved_value(
-                    item, allow_unresolved_envvars=allow_unresolved_envvars
+                    item,
+                    allow_unresolved_envvars=allow_unresolved_envvars,
+                    subgroup=subgroup,
                 )
             if "$" not in item:
                 return item
@@ -563,6 +567,7 @@ def get_resolved_value(self, item, recurse=0, allow_unresolved_envvars=False):
                     item,
                     recurse=recurse + 1,
                     allow_unresolved_envvars=allow_unresolved_envvars,
+                    subgroup=subgroup,
                 )
 
         return item

CIME/case/case_submit.py

@@ -12,8 +12,6 @@
 from CIME.locked_files import unlock_file, lock_file
 from CIME.test_status import *
 
-import socket
-
 logger = logging.getLogger(__name__)
 
 
@@ -39,6 +37,7 @@ def _submit(
     batch_args=None,
     workflow=True,
     chksum=False,
+    dryrun=False,
 ):
     if job is None:
         job = case.get_first_job()
@@ -164,9 +163,6 @@ def _submit(
         case.check_case(skip_pnl=skip_pnl, chksum=chksum)
         if job == case.get_primary_job():
             case.check_DA_settings()
-            if case.get_value("MACH") == "mira":
-                with open(".original_host", "w") as fd:
-                    fd.write(socket.gethostname())
 
     # Load Modules
     case.load_env()
@@ -185,16 +181,20 @@ def _submit(
         mail_type=mail_type,
         batch_args=batch_args,
         workflow=workflow,
+        dry_run=dryrun,
     )
-
     xml_jobids = []
-    for jobname, jobid in job_ids.items():
-        logger.info("Submitted job {} with id {}".format(jobname, jobid))
-        if jobid:
-            xml_jobids.append("{}:{}".format(jobname, jobid))
+    if dryrun:
+        for job in job_ids:
+            xml_jobids.append("{}:{}".format(job[0], job[1]))
+    else:
+        for jobname, jobid in job_ids.items():
+            logger.info("Submitted job {} with id {}".format(jobname, jobid))
+            if jobid:
+                xml_jobids.append("{}:{}".format(jobname, jobid))
 
     xml_jobid_text = ", ".join(xml_jobids)
-    if xml_jobid_text:
+    if xml_jobid_text and not dryrun:
         case.set_value("JOB_IDS", xml_jobid_text)
 
     return xml_jobid_text
@@ -214,6 +214,7 @@ def submit(
     batch_args=None,
     workflow=True,
     chksum=False,
+    dryrun=False,
 ):
     if resubmit_immediate and self.get_value("MACH") in ["mira", "cetus"]:
         logger.warning(
@@ -266,6 +267,7 @@ def submit(
             batch_args=batch_args,
             workflow=workflow,
             chksum=chksum,
+            dryrun=dryrun,
         )
         run_and_log_case_status(
             functor,
@@ -353,7 +355,7 @@ def check_case(self, skip_pnl=False, chksum=False):
 
     expect(
         self.get_value("BUILD_COMPLETE"),
-        "Build complete is " "not True please rebuild the model by calling case.build",
+        "Build complete is not True please rebuild the model by calling case.build",
     )
     logger.info("Check case OK")
 

CIME/tests/test_unit_case.py

@@ -86,6 +86,7 @@ def test_submit(
                 batch_args=None,
                 workflow=True,
                 chksum=True,
+                dryrun=False,
             )
 
 

docker/Dockerfile

@@ -84,7 +84,7 @@ RUN curl -L -k -o "${PWD}/pnetcdf.tar.gz" \
             make install && \
             rm -rf "${PWD}/pnetcdf"
 
-RUN mkdir /root/.cime
+RUN mkdir -p /root/.cime /storage/timings
 
 COPY config_machines.xml /root/.cime/
 COPY docker.cmake /root/.cime/