Merge pull request #3973 from clumens/ctslab-fixes

Fix various problems in ctslab tests

Author: nrwahl2

lib/pengine/unpack.c

@@ -897,6 +897,9 @@ unpack_resources(const xmlNode *xml_resources, pcmk_scheduler_t *scheduler)
         /* pcs's CI tests look for this specific error message. Confer with the
          * pcs team before changing it. If the dependency still exists, bump the
          * CRM_FEATURE_SET and inform the pcs maintainers.
+         *
+         * Also, ResyncCIB.errors_to_ignore() looks for this specific error
+         * message as well.
          */
         pcmk__config_err("Resource start-up disabled since no fencing "
                          "resources have been defined. Either configure some "

python/pacemaker/_cts/patterns.py

@@ -76,7 +76,7 @@ def __init__(self):
             "Pat:Fencing_start": r"Requesting peer fencing .* targeting %s",
             "Pat:Fencing_ok": r"pacemaker-fenced.*:\s*Operation .* targeting %s by .* for .*@.*: OK",
             "Pat:Fencing_recover": r"pacemaker-schedulerd.*: Recover\s+%s",
-            "Pat:Fencing_active": r"stonith resource .* is active on 2 nodes (attempting recovery)",
+            "Pat:Resource_active": r"resource .* might be active on \d+ nodes \(attempting recovery\)",
             "Pat:Fencing_probe": r"pacemaker-controld.* Result of probe operation for %s on .*: Error",
 
             "Pat:RscOpOK": r"pacemaker-controld.*:\s+Result of %s operation for %s.*: (0 \()?OK",

python/pacemaker/_cts/scenarios.py

@@ -302,19 +302,20 @@ def audit(self, local_ignore=None):
             if self._bad_news:
                 match = self._bad_news.look(0)
 
-            if match:
-                add_err = True
+            if not match:
+                break
 
-                for ignore in ignorelist:
-                    if add_err and re.search(ignore, match):
-                        add_err = False
+            add_err = True
 
-                if add_err:
-                    self._cm.log(f"BadNews: {match}")
-                    self.incr("BadNews")
-                    errcount += 1
-            else:
-                break
+            for ignore in ignorelist:
+                if add_err and re.search(ignore, match):
+                    add_err = False
+                    break
+
+            if add_err:
+                self._cm.log(f"BadNews: {match}")
+                self.incr("BadNews")
+                errcount += 1
         else:
             print("Big problems")
             if not should_continue(self._cm.env):

python/pacemaker/_cts/tests/cibsecret.py

@@ -66,6 +66,25 @@ def _insert_dummy(self, node):
 
         return ""
 
+    def _remove_dummy(self, node):
+        """Remove the previously created dummy resource on the given node."""
+        pats = [
+            self._cm.templates["Pat:RscOpOK"] % ("stop", self._rid)
+        ]
+
+        watch = self.create_watch(pats, 60)
+        watch.set_watch()
+        self._cm.remove_dummy_rsc(node, self._rid)
+
+        with Timer(self._logger, self.name, "removeDummy"):
+            watch.look_for_all()
+
+        if watch.unmatched:
+            self.debug("Failed to find patterns when removing dummy resource")
+            return repr(watch.unmatched)
+
+        return ""
+
     def _check_cib_value(self, node, expected):
         """Check that the secret has the expected value."""
         (rc, lines) = self._rsh(node, f"crm_resource -r {self._rid} -g {self._secret}",
@@ -224,6 +243,7 @@ def __call__(self, node):
             return False
 
         self._test_secrets_removed()
+        self._remove_dummy(node)
 
         return self.success()
 

python/pacemaker/_cts/tests/componentfail.py

@@ -79,7 +79,7 @@ def __call__(self, node):
             # Ignore actions for fence devices if fencer will respawn
             # (their registration will be lost, and probes will fail)
             self._okerrpatterns = [
-                self._cm.templates["Pat:Fencing_active"],
+                self._cm.templates["Pat:Resource_active"],
             ]
             (_, lines) = self._rsh(node, "crm_resource -c", verbose=1)
 

python/pacemaker/_cts/tests/resynccib.py

@@ -64,7 +64,7 @@ def errors_to_ignore(self):
         # Errors that occur as a result of the CIB being wiped
         return [
             r"error.*: v1 patchset error, patch failed to apply: Application of an update diff failed",
-            r"error.*: Resource start-up disabled since no STONITH resources have been defined",
-            r"error.*: Either configure some or disable STONITH with the stonith-enabled option",
-            r"error.*: NOTE: Clusters with shared data need STONITH to ensure data integrity"
+            r"error.*: Resource start-up disabled since no fencing resources have been defined. "
+                       "Either configure some or disable fencing with the fencing-enabled option. "
+                       "NOTE: Clusters with shared data need fencing to ensure data integrity."
         ]

python/pacemaker/_cts/tests/splitbraintest.py

@@ -200,7 +200,8 @@ def errors_to_ignore(self):
             r"Another DC detected:",
             r"(ERROR|error).*: .*Application of an update diff failed",
             r"pacemaker-controld.*:.*not in our membership list",
-            r"CRIT:.*node.*returning after partition"
+            r"CRIT:.*node.*returning after partition",
+            self._cm.templates["Pat:Resource_active"],
         ]
 
     def is_applicable(self):

python/pacemaker/_cts/tests/stonithdtest.py

@@ -125,7 +125,7 @@ def errors_to_ignore(self):
         return [
             self._cm.templates["Pat:Fencing_start"] % ".*",
             self._cm.templates["Pat:Fencing_ok"] % ".*",
-            self._cm.templates["Pat:Fencing_active"],
+            self._cm.templates["Pat:Resource_active"],
             r"error.*: Operation 'reboot' targeting .* by .* for stonith_admin.*: Timer expired"
         ]