Merge pull request #13367 from tchaikov/wip-qa-jewel-x-singleton

test: Thrasher: update pgp_num of all expanded pools if not yet

Reviewed-by: Sage Weil <sage@redhat.com>

qa/tasks/ceph_manager.py

@@ -112,6 +112,7 @@ class Thrasher:
         self.logger = logger
         self.config = config
         self.revive_timeout = self.config.get("revive_timeout", 150)
+        self.pools_to_fix_pgp_num = set()
         if self.config.get('powercycle'):
             self.revive_timeout += 120
         self.clean_wait = self.config.get('clean_wait', 0)
@@ -137,12 +138,13 @@ class Thrasher:
             self.config = dict()
         # prevent monitor from auto-marking things out while thrasher runs
         # try both old and new tell syntax, in case we are testing old code
-        try:
-            manager.raw_cluster_cmd('--', 'tell', 'mon.*', 'injectargs',
-                                    '--mon-osd-down-out-interval 0')
-        except Exception:
-            manager.raw_cluster_cmd('--', 'mon', 'tell', '*', 'injectargs',
-                                    '--mon-osd-down-out-interval 0')
+        self.saved_options = {}
+        first_mon = teuthology.get_first_mon(manager.ctx, self.config).split('.')
+        opt_name = 'mon_osd_down_out_interval'
+        self.saved_options[opt_name] = manager.get_config(first_mon[0],
+                                                          first_mon[1],
+                                                          opt_name)
+        self._set_config('mon', '*', opt_name, 0)
         # 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
@@ -172,6 +174,18 @@ class Thrasher:
         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)
+        try:
+            whom = '.'.join([service_type, service_id])
+            self.ceph_manager.raw_cluster_cmd('--', 'tell', whom,
+                                              'injectargs', opt_arg)
+        except Exception:
+            self.ceph_manager.raw_cluster_cmd('--', service_type,
+                                              'tell', service_id,
+                                              '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()
@@ -515,18 +529,24 @@ class Thrasher:
         Increase the size of the pool
         """
         pool = self.ceph_manager.get_pool()
+        orig_pg_num = self.ceph_manager.get_pool_pg_num(pool)
         self.log("Growing pool %s" % (pool,))
         self.ceph_manager.expand_pool(pool,
                                       self.config.get('pool_grow_by', 10),
                                       self.max_pgs)
+        if orig_pg_num < self.ceph_manager.get_pool_pg_num(pool):
+            self.pools_to_fix_pgp_num.add(pool)
 
-    def fix_pgp_num(self):
+    def fix_pgp_num(self, pool=None):
         """
         Fix number of pgs in pool.
         """
-        pool = self.ceph_manager.get_pool()
+        if pool is None:
+            pool = self.ceph_manager.get_pool()
         self.log("fixing pg num pool %s" % (pool,))
         self.ceph_manager.set_pool_pgpnum(pool)
+        if pool in self.pools_to_fix_pgp_num:
+            self.pools_to_fix_pgp_num.remove(pool)
 
     def test_pool_min_size(self):
         """
@@ -839,6 +859,13 @@ class Thrasher:
                         Scrubber(self.ceph_manager, self.config)
             self.choose_action()()
             time.sleep(delay)
+        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 opt, value in self.saved_options.iteritems():
+            self._set_config('mon', '*', opt, value)
+        self.saved_options.clear()
         self.all_up()
 
 
@@ -1235,7 +1262,7 @@ class CephManager:
             proc = self.admin_socket(service_type, service_id,
                                      args, check_status=False, stdout=stdout)
             if proc.exitstatus is 0:
-                break
+                return proc
             else:
                 tries += 1
                 if (tries * 5) > timeout:
@@ -1259,6 +1286,16 @@ class CephManager:
                 return i
         assert False
 
+    def get_config(self, service_type, service_id, name):
+        """
+        :param node: like 'mon.a'
+        :param name: the option name
+        """
+        proc = self.wait_run_admin_socket(service_type, service_id,
+                                          ['config', 'show'])
+        j = json.loads(proc.stdout.getvalue())
+        return j[name]
+
     def set_config(self, osdnum, **argdict):
         """
         :param osdnum: osd number