mon/PGMonitor: add max pgs per osd warning

This is just like the min warning, but a max.  It should help warn users
away from having too many PGs in their cluster, which can be very
dangerous.

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

src/common/config_opts.h

@@ -178,6 +178,7 @@ OPTION(mon_accept_timeout, OPT_FLOAT, 10.0)    // on leader, if paxos update isn
 OPTION(mon_pg_create_interval, OPT_FLOAT, 30.0) // no more than every 30s
 OPTION(mon_pg_stuck_threshold, OPT_INT, 300) // number of seconds after which pgs can be considered inactive, unclean, or stale (see doc/control.rst under dump_stuck for more info)
 OPTION(mon_pg_warn_min_per_osd, OPT_INT, 20)  // min # pgs per (in) osd before we warn the admin
+OPTION(mon_pg_warn_max_per_osd, OPT_INT, 500)  // max # pgs per (in) osd before we warn the admin
 OPTION(mon_pg_warn_max_object_skew, OPT_FLOAT, 10.0) // max skew few average in objects per pg
 OPTION(mon_pg_warn_min_objects, OPT_INT, 10000)  // do not warn below this object #
 OPTION(mon_pg_warn_min_pool_objects, OPT_INT, 1000)  // do not warn on pools below this object #

src/mon/PGMonitor.cc

@@ -2097,6 +2097,16 @@ void PGMonitor::get_health(list<pair<health_status_t,string> >& summary,
 	detail->push_back(make_pair(HEALTH_WARN, ss.str()));
     }
   }
+  if (num_in && g_conf->mon_pg_warn_max_per_osd > 0) {
+    int per = sum_pg_up / num_in;
+    if (per > g_conf->mon_pg_warn_max_per_osd) {
+      ostringstream ss;
+      ss << "too many PGs per OSD (" << per << " > max " << g_conf->mon_pg_warn_max_per_osd << ")";
+      summary.push_back(make_pair(HEALTH_WARN, ss.str()));
+      if (detail)
+	detail->push_back(make_pair(HEALTH_WARN, ss.str()));
+    }
+  }
   if (!pg_map.pg_stat.empty()) {
     for (ceph::unordered_map<int,pool_stat_t>::const_iterator p = pg_map.pg_pool_sum.begin();
 	 p != pg_map.pg_pool_sum.end();