qa: disable mon-health-to-clog in upgrade test

qa/suites/rados/upgrade/luminous-x-singleton/2-partial-upgrade/firsthalf.yaml

@@ -10,6 +10,7 @@ tasks:
 - ceph.restart:
     daemons: [mon.a, mon.b, mon.c]
     wait-for-healthy: false
+    mon-health-to-clog: false
 - ceph.restart:
     daemons: [osd.0, osd.1, osd.2]
     wait-for-healthy: false

qa/suites/upgrade/luminous-x/parallel/3-upgrade-sequence/upgrade-all.yaml

@@ -5,6 +5,7 @@ upgrade-sequence:
    sequential:
    - ceph.restart:
        daemons: [mon.a, mon.b, mon.c, mgr.x]
+       mon-health-to-clog: false
    - ceph.restart:
        daemons: [osd.0, osd.1, osd.2, osd.3]
        wait-for-healthy: false

qa/suites/upgrade/luminous-x/parallel/3-upgrade-sequence/upgrade-mon-osd-mds.yaml

@@ -15,6 +15,7 @@ upgrade-sequence:
    - ceph.restart:
        daemons: [mon.b, mon.c, mgr.x]
        wait-for-healthy: true
+       mon-health-to-clog: false
    - sleep:
        duration: 60
    - ceph.restart:

qa/suites/upgrade/luminous-x/stress-split/2-partial-upgrade/firsthalf.yaml

@@ -9,4 +9,5 @@ tasks:
 - print: "**** done install.upgrade osd.0"
 - ceph.restart:
     daemons: [mon.a,mon.b,mon.c,mgr.x,osd.0,osd.1,osd.2]
+    mon-health-to-clog: false
 - print: "**** done ceph.restart 1st half"

qa/tasks/ceph.py

@@ -1344,6 +1344,39 @@ def created_pool(ctx, config):
                 new_pool, 'pg_num')
 
 
+@contextlib.contextmanager
+def tweaked_option(ctx, config):
+    """
+    set an option, and then restore it with its original value
+
+    Note, due to the way how tasks are executed/nested, it's not suggested to
+    use this method as a standalone task. otherwise, it's likely that it will
+    restore the tweaked option at the /end/ of 'tasks' block.
+    """
+    saved_options = {}
+    # we can complicate this when necessary
+    options = ['mon-health-to-clog']
+    type_, id_ = 'mon', '*'
+    cluster = config.get('cluster', 'ceph')
+    manager = ctx.managers[cluster]
+    if id_ == '*':
+        get_from = next(teuthology.all_roles_of_type(ctx.cluster, type_))
+    else:
+        get_from = id_
+    for option in options:
+        if option not in config:
+            continue
+        value = 'true' if config[option] else 'false'
+        option = option.replace('-', '_')
+        old_value = manager.get_config(type_, get_from, option)
+        if value != old_value:
+            saved_options[option] = old_value
+            manager.inject_args(type_, id_, option, value)
+    yield
+    for option, value in saved_options.items():
+        manager.inject_args(type_, id_, option, value)
+
+
 @contextlib.contextmanager
 def restart(ctx, config):
     """
@@ -1375,12 +1408,14 @@ def restart(ctx, config):
 
     daemons = ctx.daemons.resolve_role_list(config.get('daemons', None), CEPH_ROLE_TYPES, True)
     clusters = set()
-    for role in daemons:
-        cluster, type_, id_ = teuthology.split_role(role)
-        ctx.daemons.get_daemon(type_, id_, cluster).restart()
-        clusters.add(cluster)
-
     manager = ctx.managers['ceph']
+
+    with tweaked_option(ctx, config):
+        for role in daemons:
+            cluster, type_, id_ = teuthology.split_role(role)
+            ctx.daemons.get_daemon(type_, id_, cluster).restart()
+            clusters.add(cluster)
+
     for dmon in daemons:
         if '.' in dmon:
             dm_parts = dmon.split('.')

qa/tasks/ceph_manager.py

@@ -153,7 +153,7 @@ def tmp(x):
                                            first_mon[1],
                                            opt)
             self.saved_options.append((service, opt, old_value))
-            self._set_config(service, '*', opt, new_value)
+            manager.inject_args(service, '*', opt, new_value)
         # initialize ceph_objectstore_tool property - must be done before
         # do_thrash is spawned - http://tracker.ceph.com/issues/18799
         if (self.config.get('powercycle') or
@@ -180,13 +180,6 @@ def tmp(x):
         if self.noscrub_toggle_delay:
             self.noscrub_toggle_thread = gevent.spawn(self.do_noscrub_toggle)
 
-    def _set_config(self, service_type, service_id, name, value):
-        opt_arg = '--{name} {value}'.format(name=name, value=value)
-        whom = '.'.join([service_type, service_id])
-        self.ceph_manager.raw_cluster_cmd('--', 'tell', whom,
-                                          'injectargs', opt_arg)
-
-
     def cmd_exists_on_osds(self, cmd):
         allremotes = self.ceph_manager.ctx.cluster.only(\
             teuthology.is_type('osd', self.cluster)).remotes.keys()
@@ -392,10 +385,12 @@ def revive_osd(self, osd=None, skip_admin_check=False):
         self.dead_osds.remove(osd)
         self.live_osds.append(osd)
         if self.random_eio > 0 and osd is self.rerrosd:
-            self.ceph_manager.raw_cluster_cmd('tell', 'osd.'+str(self.rerrosd),
-                          'injectargs', '--', '--filestore_debug_random_read_err='+str(self.random_eio))
-            self.ceph_manager.raw_cluster_cmd('tell', 'osd.'+str(self.rerrosd),
-                          'injectargs', '--', '--bluestore_debug_random_read_err='+str(self.random_eio))
+            self.ceph_manager.inject_args('osd', self.rerrosd,
+                                          'filestore_debug_random_read_err',
+                                          self.random_eio)
+            self.ceph_manager.inject_args('osd', self.rerrosd,
+                                          'bluestore_debug_random_read_err',
+                                          self.random_eio)
 
 
     def out_osd(self, osd=None):
@@ -893,8 +888,9 @@ def do_optrack_toggle(self):
                 osd_state = "false"
             else:
                 osd_state = "true"
-            self.ceph_manager.raw_cluster_cmd_result('tell', 'osd.*',
-                             'injectargs', '--osd_enable_op_tracker=%s' % osd_state)
+            self.ceph_manager.inject_args('osd', '*',
+                                          'osd_enable_op_tracker',
+                                          osd_state)
             gevent.sleep(delay)
 
     @log_exc
@@ -952,10 +948,12 @@ def do_thrash(self):
         delay = self.config.get("op_delay", 5)
         self.rerrosd = self.live_osds[0]
         if self.random_eio > 0:
-            self.ceph_manager.raw_cluster_cmd('tell', 'osd.'+str(self.rerrosd),
-                          'injectargs', '--', '--filestore_debug_random_read_err='+str(self.random_eio))
-            self.ceph_manager.raw_cluster_cmd('tell', 'osd.'+str(self.rerrosd),
-                          'injectargs', '--', '--bluestore_debug_random_read_err='+str(self.random_eio))
+            self.ceph_manager.inject_args('osd', self.rerrosd,
+                                          'filestore_debug_random_read_err',
+                                          self.random_eio)
+            self.ceph_manager.inject_args('osd', self.rerrosd,
+                                          'bluestore_debug_random_read_err',
+                                          self.random_eio)
         self.log("starting do_thrash")
         while not self.stopping:
             to_log = [str(x) for x in ["in_osds: ", self.in_osds,
@@ -985,16 +983,16 @@ def do_thrash(self):
             time.sleep(delay)
         self.all_up()
         if self.random_eio > 0:
-            self.ceph_manager.raw_cluster_cmd('tell', 'osd.'+str(self.rerrosd),
-                          'injectargs', '--', '--filestore_debug_random_read_err=0.0')
-            self.ceph_manager.raw_cluster_cmd('tell', 'osd.'+str(self.rerrosd),
-                          'injectargs', '--', '--bluestore_debug_random_read_err=0.0')
+            self.ceph_manager.inject_args('osd', self.rerrosd,
+                                          'filestore_debug_random_read_err', '0.0')
+            self.ceph_manager.inject_args('osd', self.rerrosd,
+                                          'bluestore_debug_random_read_err', '0.0')
         for pool in list(self.pools_to_fix_pgp_num):
             if self.ceph_manager.get_pool_pg_num(pool) > 0:
                 self.fix_pgp_num(pool)
         self.pools_to_fix_pgp_num.clear()
         for service, opt, saved_value in self.saved_options:
-            self._set_config(service, '*', opt, saved_value)
+            self.ceph_manager.inject_args(service, '*', opt, saved_value)
         self.saved_options = []
         self.all_up_in()
 
@@ -1481,6 +1479,13 @@ def get_config(self, service_type, service_id, name):
         j = json.loads(proc.stdout.getvalue())
         return j[name]
 
+    def inject_args(self, service_type, service_id, name, value):
+        whom = '{0}.{1}'.format(service_type, service_id)
+        if isinstance(value, bool):
+            value = 'true' if value else 'false'
+        opt_arg = '--{name}={value}'.format(name=name, value=value)
+        self.raw_cluster_cmd('--', 'tell', whom, 'injectargs', opt_arg)
+
     def set_config(self, osdnum, **argdict):
         """
         :param osdnum: osd number
@@ -2314,11 +2319,9 @@ def kill_osd(self, osd):
             remote.console.power_off()
         elif self.config.get('bdev_inject_crash') and self.config.get('bdev_inject_crash_probability'):
             if random.uniform(0, 1) < self.config.get('bdev_inject_crash_probability', .5):
-                self.raw_cluster_cmd(
-                    '--', 'tell', 'osd.%d' % osd,
-                    'injectargs',
-                    '--bdev-inject-crash %d' % self.config.get('bdev_inject_crash'),
-                )
+                self.inject_args(
+                    'osd', osd,
+                    'bdev-inject-crash', self.config.get('bdev_inject_crash'))
                 try:
                     self.ctx.daemons.get_daemon('osd', osd, self.cluster).wait()
                 except:
@@ -2340,9 +2343,8 @@ def blackhole_kill_osd(self, osd):
         """
         Stop osd if nothing else works.
         """
-        self.raw_cluster_cmd('--', 'tell', 'osd.%d' % osd,
-                             'injectargs',
-                             '--objectstore-blackhole')
+        self.inject_args('osd', osd,
+                         'objectstore-blackhole', True)
         time.sleep(2)
         self.ctx.daemons.get_daemon('osd', osd, self.cluster).stop()