Fix NODEFAIL test on cheyenne.

config/cesm/machines/config_machines.xml

@@ -219,8 +219,10 @@
     <mpirun mpilib="default">
       <executable>mpiexec_mpt</executable>
       <arguments>
+	<arg name="anum_tasks"> -np $TOTALPES</arg>
 	<arg name="labelstdout">-p "%g:"</arg>
-	<arg name="threadplacement"> omplace </arg>
+	<!-- the omplace argument needs to be last -->
+	<arg name="zthreadplacement"> omplace </arg>
       </arguments>
     </mpirun>
     <mpirun mpilib="mpi-serial">

config/config_tests.xml

@@ -166,6 +166,45 @@ PRE    pause-resume test: by default a BFB test of pause-resume cycling
 
 LII    CLM initial condition interpolation test
 
+======================================================================
+    Infrastructural tests for CIME. These are used by scripts_regression_tests.
+    Users won't generally run these.
+======================================================================
+
+
+TESTBUILDFAIL     Insta-fail build step. Used to confirm that failed
+                  builds are caught and reported correctly.
+
+TESTBUILDFAILEXC  Insta-fail build step by failing to init. Used to test
+                  correct behavior when exceptions are generated.
+
+TESTRUNFAIL       Insta-fail run step. Used to confirm that model run
+                  failures are caught and reported correctly.
+
+TESTRUNFAILEXC    Insta-fail run step via exception. Used to test correct
+                  correct behavior when exceptions are generated.
+
+TESTRUNPASS       Insta-pass run step. Used to test that run that work
+                  are reported correctly.
+
+TESTMEMLEAKFAIL   Insta-fail memleak step. Used to test that memleaks are
+                  detected and reported correctly.
+
+TESTMEMLEAKPASS   Insta-pass memleak step. Used to test that non-memleaks are
+                  reported correctly.
+
+TESTRUNDIFF       Produces a canned hist file. Env var TESTRUNDIFF_ALTERNATE can
+                  be used to cause a DIFF. Used to check that baseline diffs are
+                  detected and reported correctly.
+
+TESTTESTDIFF      Simulates internal test diff (non baseline). Used to check that
+                  internal comparison failures are detected and reported correctly.
+
+TESTRUNSLOWPASS   After 5 minutes of sleep, pass run step. Used to test timeouts
+                  and kills.
+
+NODEFAIL          Tests restart upon detected node failure. Generates fake failures,
+                  the number of which is controlled by NODEFAIL_NUM_FAILS.
 
 -->
 
@@ -366,7 +405,8 @@ LII    CLM initial condition interpolation test
   <test NAME="NODEFAIL">
     <DESC>For testing infra only. Tests restart upon detected node failure</DESC>
     <INFO_DBUG>1</INFO_DBUG>
-    <STOP_OPTION>ndays</STOP_OPTION>
+    <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>

scripts/lib/CIME/SystemTests/nodefail.py

@@ -56,8 +56,16 @@ def _restart_fake_phase(self):
         env_mach_specific.set_value("run_exe", fake_exe_file)
         self._case.flush(flushall=True)
 
+        # This flag is needed by mpt to run a script under mpiexec
+        mpilib = self._case.get_value("MPILIB")
+        if mpilib == "mpt":
+            os.environ["MPI_SHEPHERD"] = "true"
+
         self.run_indv(suffix=None)
 
+        if mpilib == "mpt":
+            del os.environ["MPI_SHEPHERD"]
+
         env_mach_specific = self._case.get_env("mach_specific")
         env_mach_specific.set_value("run_exe", prev_run_exe)
         self._case.flush(flushall=True)