Merge pull request #9394 from xiexingguo/xxg-wip-batchpgstates

mon/PGMonitor: batch filter pg states; add sanity check

Reviewed-by: Kefu Chai <kchai@redhat.com>

doc/dev/osd_internals/recovery_reservation.rst

@@ -62,7 +62,7 @@ to the monitor. The state chart can set:
 
  - recovery_wait: waiting for local/remote reservations
  - recovering: recovering
- - wait_backfill: waiting for remote backfill reservations
+ - backfill_wait: waiting for remote backfill reservations
  - backfilling: backfilling
  - backfill_toofull: backfill reservation rejected, OSD too full
 

src/mon/PGMap.cc

@@ -1733,16 +1733,9 @@ void PGMap::generate_test_instances(list<PGMap*>& o)
   }
 }
 
-void PGMap::get_filtered_pg_stats(const string& state, int64_t poolid, int64_t osdid,
+void PGMap::get_filtered_pg_stats(uint32_t state, int64_t poolid, int64_t osdid,
                                   bool primary, set<pg_t>& pgs)
 {
-  int type = 0;
-  if (state != "all") {
-    type = pg_string_state(state);
-    if (type == -1)
-      assert(0 == "invalid type");
-  }
-
   for (ceph::unordered_map<pg_t, pg_stat_t>::const_iterator i = pg_stat.begin();
        i != pg_stat.end();
        ++i) {
@@ -1750,7 +1743,7 @@ void PGMap::get_filtered_pg_stats(const string& state, int64_t poolid, int64_t o
       continue;
     if ((osdid >= 0) && !(i->second.is_acting_osd(osdid,primary)))
       continue;
-    if ((state != "all") && !(i->second.state & type))
+    if (!(i->second.state & state))
       continue;
     pgs.insert(i->first);
   }

src/mon/PGMap.h

@@ -300,7 +300,7 @@ public:
   void dump_osd_blocked_by_stats(Formatter *f) const;
   void print_osd_blocked_by_stats(std::ostream *ss) const;
 
-  void get_filtered_pg_stats(const string& state, int64_t poolid, int64_t osdid,
+  void get_filtered_pg_stats(uint32_t state, int64_t poolid, int64_t osdid,
                              bool primary, set<pg_t>& pgs);
   void recovery_summary(Formatter *f, list<string> *psl,
                         const pool_stat_t& delta_sum) const;

src/mon/PGMonitor.cc

@@ -621,17 +621,20 @@ void PGMonitor::handle_statfs(MonOpRequestRef op)
             << session->caps << dendl;
     return;
   }
-  MStatfsReply *reply;
-
-  dout(10) << "handle_statfs " << *statfs << " from " << statfs->get_orig_source() << dendl;
 
   if (statfs->fsid != mon->monmap->fsid) {
-    dout(0) << "handle_statfs on fsid " << statfs->fsid << " != " << mon->monmap->fsid << dendl;
+    dout(0) << "handle_statfs on fsid " << statfs->fsid
+            << " != " << mon->monmap->fsid << dendl;
     return;
   }
 
+
+  dout(10) << "handle_statfs " << *statfs
+           << " from " << statfs->get_orig_source() << dendl;
+
   // fill out stfs
-  reply = new MStatfsReply(mon->monmap->fsid, statfs->get_tid(), get_last_committed());
+  MStatfsReply *reply = new MStatfsReply(mon->monmap->fsid, statfs->get_tid(),
+    get_last_committed());
 
   // these are in KB.
   reply->h.st.kb = pg_map.osd_sum.kb;
@@ -932,23 +935,15 @@ void PGMonitor::check_osd_map(epoch_t epoch)
     }
   }
 
-  bool propose = false;
-  if (pg_map.last_osdmap_epoch < epoch) {
-    pending_inc.osdmap_epoch = epoch;
-    propose = true;
-  }
+  assert(pg_map.last_osdmap_epoch < epoch);
+  pending_inc.osdmap_epoch = epoch;
+  map_pg_creates();
+  register_new_pgs();
 
-  if (map_pg_creates())
-    propose = true;
-  if (register_new_pgs())
-    propose = true;
+  if (need_check_down_pgs || !need_check_down_pg_osds.empty())
+    check_down_pgs();
 
-  if ((need_check_down_pgs || !need_check_down_pg_osds.empty()) &&
-      check_down_pgs())
-    propose = true;
-
-  if (propose)
-    propose_pending();
+  propose_pending();
 }
 
 void PGMonitor::register_pg(OSDMap *osdmap,
@@ -1034,7 +1029,7 @@ void PGMonitor::register_pg(OSDMap *osdmap,
   }
 }
 
-bool PGMonitor::register_new_pgs()
+void PGMonitor::register_new_pgs()
 {
   // iterate over crush mapspace
   OSDMap *osdmap = &mon->osdmon()->osdmap;
@@ -1116,10 +1111,9 @@ bool PGMonitor::register_new_pgs()
 
   dout(10) << "register_new_pgs registered " << created << " new pgs, removed "
            << removed << " uncreated pgs" << dendl;
-  return (created || removed);
 }
 
-bool PGMonitor::map_pg_creates()
+void PGMonitor::map_pg_creates()
 {
   OSDMap *osdmap = &mon->osdmon()->osdmap;
 
@@ -1187,9 +1181,7 @@ bool PGMonitor::map_pg_creates()
   }
   if (changed) {
     dout(10) << __func__ << " " << changed << " pgs changed primary" << dendl;
-    return true;
   }
-  return false;
 }
 
 void PGMonitor::send_pg_creates()
@@ -1300,12 +1292,12 @@ void PGMonitor::_try_mark_pg_stale(
   }
 }
 
-bool PGMonitor::check_down_pgs()
+void PGMonitor::check_down_pgs()
 {
   dout(10) << "check_down_pgs last_osdmap_epoch "
 	   << pg_map.last_osdmap_epoch << dendl;
   if (pg_map.last_osdmap_epoch == 0)
-    return false;
+    return;
 
   // use the OSDMap that matches the one pg_map has consumed.
   std::unique_ptr<OSDMap> osdmap;
@@ -1315,8 +1307,6 @@ bool PGMonitor::check_down_pgs()
   osdmap.reset(new OSDMap);
   osdmap->decode(bl);
 
-  bool ret = false;
-
   // if a large number of osds changed state, just iterate over the whole
   // pg map.
   if (need_check_down_pg_osds.size() > (unsigned)osdmap->get_num_osds() *
@@ -1329,7 +1319,6 @@ bool PGMonitor::check_down_pgs()
           p.second.acting_primary != -1 &&
           osdmap->is_down(p.second.acting_primary)) {
 	_try_mark_pg_stale(osdmap.get(), p.first, p.second);
-	ret = true;
       }
     }
   } else {
@@ -1340,7 +1329,6 @@ bool PGMonitor::check_down_pgs()
 	  assert(stat.acting_primary == osd);
 	  if ((stat.state & PG_STATE_STALE) == 0) {
 	    _try_mark_pg_stale(osdmap.get(), pgid, stat);
-	    ret = true;
 	  }
 	}
       }
@@ -1348,8 +1336,6 @@ bool PGMonitor::check_down_pgs()
   }
   need_check_down_pgs = false;
   need_check_down_pg_osds.clear();
-
-  return ret;
 }
 
 inline string percentify(const float& a) {
@@ -1756,7 +1742,6 @@ bool PGMonitor::preprocess_command(MonOpRequestRef op)
     int64_t pool = -1;
     vector<string>states;
     set<pg_t> pgs;
-    set<string> what;
     cmd_getval(g_ceph_context, cmdmap, "pool", pool);
     cmd_getval(g_ceph_context, cmdmap, "osd", osd);
     cmd_getval(g_ceph_context, cmdmap, "states", states);
@@ -1772,17 +1757,31 @@ bool PGMonitor::preprocess_command(MonOpRequestRef op)
     }
     if (states.empty())
       states.push_back("all");
+
+    uint32_t state = 0;
+
     while (!states.empty()) {
-      string state = states.back();
-      what.insert(state);
-      pg_map.get_filtered_pg_stats(state,pool,osd,primary,pgs);
+      string state_str = states.back();
+
+      if (state_str == "all") {
+        state = -1;
+        break;
+      } else {
+        int filter = pg_string_state(state_str);
+        assert(filter != -1);
+        state |= filter;
+      }
+
       states.pop_back();
     }
+
+    pg_map.get_filtered_pg_stats(state, pool, osd, primary, pgs);
+
     if (f && !pgs.empty()) {
-      pg_map.dump_filtered_pg_stats(f.get(),pgs);
+      pg_map.dump_filtered_pg_stats(f.get(), pgs);
       f->flush(ds);
     } else if (!pgs.empty()) {
-      pg_map.dump_filtered_pg_stats(ds,pgs);
+      pg_map.dump_filtered_pg_stats(ds, pgs);
     }
     r = 0;
   } else if (prefix == "pg dump_stuck") {

src/mon/PGMonitor.h

@@ -115,17 +115,13 @@ private:
 
   /**
    * check latest osdmap for new pgs to register
-   *
-   * @return true if we updated pending_inc (and should propose)
    */
-  bool register_new_pgs();
+  void register_new_pgs();
 
   /**
    * recalculate creating pg mappings
-   *
-   * @return true if we updated pending_inc
    */
-  bool map_pg_creates();
+  void map_pg_creates();
 
   void send_pg_creates();
   epoch_t send_pg_creates(int osd, Connection *con, epoch_t next);
@@ -136,9 +132,8 @@ private:
    * clears need_check_down_pgs
    * clears need_check_down_pg_osds
    *
-   * @return true if we updated pending_inc (and should propose)
    */
-  bool check_down_pgs();
+  void check_down_pgs();
   void _try_mark_pg_stale(const OSDMap *osdmap, pg_t pgid,
 			  const pg_stat_t& cur_stat);
 

src/osd/osd_types.cc

@@ -839,7 +839,7 @@ std::string pg_state_string(int state)
     oss << "repair+";
   if ((state & PG_STATE_BACKFILL_WAIT) &&
       !(state &PG_STATE_BACKFILL))
-    oss << "wait_backfill+";
+    oss << "backfill_wait+";
   if (state & PG_STATE_BACKFILL)
     oss << "backfilling+";
   if (state & PG_STATE_BACKFILL_TOOFULL)