Merge pull request #4667 from jedwards4b/fix/rest_n_in_tests

Fix/rest n in tests

CIME/Servers/__init__.py

@@ -1,9 +1,9 @@
 # pylint: disable=import-error
-from distutils.spawn import find_executable
+from shutil import which
 
-has_gftp = find_executable("globus-url-copy")
-has_svn = find_executable("svn")
-has_wget = find_executable("wget")
+has_gftp = which("globus-url-copy")
+has_svn = which("svn")
+has_wget = which("wget")
 has_ftp = True
 try:
     from ftplib import FTP

CIME/SystemTests/ers.py

@@ -16,29 +16,18 @@ def __init__(self, case, **kwargs):
         SystemTestsCommon.__init__(self, case, **kwargs)
 
     def _ers_first_phase(self):
-        stop_n = self._case.get_value("STOP_N")
-        stop_option = self._case.get_value("STOP_OPTION")
-        rest_n = self._case.get_value("REST_N")
-        expect(stop_n > 0, "Bad STOP_N: {:d}".format(stop_n))
-
-        expect(stop_n > 2, "ERROR: stop_n value {:d} too short".format(stop_n))
-        logger.info(
-            "doing an {0} {1} initial test with restart file at {2} {1}".format(
-                str(stop_n), stop_option, str(rest_n)
-            )
-        )
+        self._rest_n = self._set_restart_interval()
         self.run_indv()
 
     def _ers_second_phase(self):
         stop_n = self._case.get_value("STOP_N")
         stop_option = self._case.get_value("STOP_OPTION")
 
-        rest_n = int(stop_n / 2 + 1)
-        stop_new = stop_n - rest_n
+        stop_new = stop_n - self._rest_n
         expect(
             stop_new > 0,
             "ERROR: stop_n value {:d} too short {:d} {:d}".format(
-                stop_new, stop_n, rest_n
+                stop_new, stop_n, self._rest_n
             ),
         )
         rundir = self._case.get_value("RUNDIR")

CIME/SystemTests/restart_tests.py

@@ -1,6 +1,5 @@
 """
 Abstract class for restart tests
-
 """
 
 from CIME.SystemTests.system_tests_compare_two import SystemTestsCompareTwo
@@ -34,6 +33,7 @@ def __init__(
     def _case_one_setup(self):
         stop_n = self._case1.get_value("STOP_N")
         expect(stop_n >= 3, "STOP_N must be at least 3, STOP_N = {}".format(stop_n))
+        self._set_restart_interval()
 
     def _case_two_setup(self):
         rest_n = self._case1.get_value("REST_N")

CIME/SystemTests/system_tests_common.py

@@ -1,6 +1,7 @@
 """
 Base class for CIME system tests
 """
+
 from CIME.XML.standard_module_setup import *
 from CIME.XML.env_run import EnvRun
 from CIME.XML.env_test import EnvTest
@@ -117,6 +118,74 @@ def __init__(
         self._dry_run = False
         self._user_separate_builds = False
         self._expected_num_cmp = None
+        self._rest_n = None
+
+    def _set_restart_interval(self):
+        stop_n = self._case.get_value("STOP_N")
+        stop_option = self._case.get_value("STOP_OPTION")
+        self._case.set_value("REST_OPTION", stop_option)
+        # We need to make sure the run is long enough and to set REST_N to a
+        # value that makes sense for all components
+        maxncpl = 10000
+        minncpl = 0
+        for comp in self._case.get_values("COMP_CLASSES"):
+            if comp == "CPL":
+                continue
+            compname = self._case.get_value("COMP_{}".format(comp))
+
+            # ignore stub components in this test.
+            if compname == "s{}".format(comp.lower()):
+                ncpl = None
+            else:
+                ncpl = self._case.get_value("{}_NCPL".format(comp))
+
+            if ncpl and maxncpl > ncpl:
+                maxncpl = ncpl
+            if ncpl and minncpl < ncpl:
+                minncpl = ncpl
+
+        ncpl_base_period = self._case.get_value("NCPL_BASE_PERIOD")
+        if ncpl_base_period == "hour":
+            coupling_secs = 3600 / maxncpl
+            timestep = 3600 / minncpl
+        elif ncpl_base_period == "day":
+            coupling_secs = 86400 / maxncpl
+            timestep = 86400 / minncpl
+        elif ncpl_base_period == "year":
+            coupling_secs = 31536000 / maxncpl
+            timestep = 31536000 / minncpl
+        elif ncpl_base_period == "decade":
+            coupling_secs = 315360000 / maxncpl
+            timestep = 315360000 / minncpl
+
+        # Convert stop_n to units of coupling intervals
+        factor = 1
+        if stop_option == "nsteps":
+            factor = timestep
+        elif stop_option == "nminutes":
+            factor = 60
+        elif stop_option == "nhours":
+            factor = 3600
+        elif stop_option == "ndays":
+            factor = 86400
+        elif stop_option == "nyears":
+            factor = 315360000
+        else:
+            expect(False, f"stop_option {stop_option} not available for this test")
+
+        stop_n = int(stop_n * factor // coupling_secs)
+        rest_n = int((stop_n // 2 + 1) * coupling_secs / factor)
+
+        expect(stop_n > 0, "Bad STOP_N: {:d}".format(stop_n))
+
+        expect(stop_n > 2, "ERROR: stop_n value {:d} too short".format(stop_n))
+        logger.info(
+            "doing an {0} {1} initial test with restart file at {2} {1}".format(
+                str(stop_n), stop_option, str(rest_n)
+            )
+        )
+        self._case.set_value("REST_N", rest_n)
+        return rest_n
 
     def _init_environment(self, caseroot):
         """

CIME/case/case.py

@@ -270,10 +270,10 @@ def initialize_derived_attributes(self):
             self.tasks_per_node = env_mach_pes.get_tasks_per_node(
                 self.total_tasks, self.thread_count
             )
-
             self.num_nodes, self.spare_nodes = env_mach_pes.get_total_nodes(
                 self.total_tasks, self.thread_count
             )
+
             self.num_nodes += self.spare_nodes
 
         logger.debug(

CIME/case/case_submit.py

@@ -234,6 +234,8 @@ def submit(
     caseroot = self.get_value("CASEROOT")
     if self.get_value("TEST"):
         casebaseid = self.get_value("CASEBASEID")
+        if os.path.exists(os.path.join(caseroot, "env_test.xml")):
+            self.set_initial_test_values()
         # This should take care of the race condition where the submitted job
         # begins immediately and tries to set RUN phase. We proactively assume
         # a passed SUBMIT phase. If this state is already PASS, don't set it again

CIME/code_checker.py

@@ -24,6 +24,7 @@
 
 logger = logging.getLogger(__name__)
 
+
 ###############################################################################
 def _run_pylint(all_files, interactive):
     ###############################################################################
@@ -80,16 +81,6 @@ def _run_pylint(all_files, interactive):
 
     return result
 
-    # if stat != 0:
-    #     if interactive:
-    #         logger.info("File %s has pylint problems, please fix\n    Use command: %s" % (on_file, cmd))
-    #         logger.info(out + "\n" + err)
-    #     return (on_file, out + "\n" + err)
-    # else:
-    #     if interactive:
-    #         logger.info("File %s has no pylint problems" % on_file)
-    #     return (on_file, "")
-
 
 ###############################################################################
 def _matches(file_path, file_ends):

CIME/data/config/config_tests.xml

@@ -320,7 +320,6 @@ NODEFAIL          Tests restart upon detected node failure. Generates fake failu
     <INFO_DBUG>1</INFO_DBUG>
     <STOP_OPTION>ndays</STOP_OPTION>
     <STOP_N>7</STOP_N>
-    <REST_N>$STOP_N / 2 + 1</REST_N>
     <REST_OPTION>$STOP_OPTION</REST_OPTION>
     <HIST_N>$STOP_N</HIST_N>
     <HIST_OPTION>$STOP_OPTION</HIST_OPTION>
@@ -333,7 +332,6 @@ NODEFAIL          Tests restart upon detected node failure. Generates fake failu
     <INFO_DBUG>1</INFO_DBUG>
     <STOP_OPTION>ndays</STOP_OPTION>
     <STOP_N>7</STOP_N>
-    <REST_N>$STOP_N / 2 + 1</REST_N>
     <REST_OPTION>$STOP_OPTION</REST_OPTION>
     <HIST_N>$STOP_N</HIST_N>
     <HIST_OPTION>$STOP_OPTION</HIST_OPTION>
@@ -540,7 +538,6 @@ NODEFAIL          Tests restart upon detected node failure. Generates fake failu
     <STOP_OPTION>nsteps</STOP_OPTION>
     <OCN_NCPL>$ATM_NCPL</OCN_NCPL>
     <STOP_N>11</STOP_N>
-    <REST_N>$STOP_N / 2 + 1</REST_N>
     <REST_OPTION>$STOP_OPTION</REST_OPTION>
     <HIST_N>$STOP_N</HIST_N>
     <HIST_OPTION>$STOP_OPTION</HIST_OPTION>