Make MDS evaluates the overload situation with the same criterion

Currently, the prep_rebalancer requires the mds_load of itself
being greater than average level mds_bal_min_rebalance percent
to be counted as exporter, however, for the other mds, as long
as their mdsload is greater than average level, they could be
consider as exporter. This inconsistent standard would cause
gray area in the rebalance activity: one may expect the other
do some work, but it didn't. With the increasing amount of mds,
this inconsistent decision would get worse and affects the
optimization effect of rebalancer.

Signed-off-by: Jianyu Li <joannyli@tencent.com>

diff --git a/src/mds/MDBalancer.cc b/src/mds/MDBalancer.cc
index ae3d9a9553a883751847dd79731edefca7977a48..85aeab8645e3278bb2a8988d0f255824e7fdf45a 100644 (file)
--- a/src/mds/MDBalancer.cc
+++ b/src/mds/MDBalancer.cc
@@ -338,7 +338,7 @@ void MDBalancer::send_heartbeat()
   for (set<mds_rank_t>::iterator p = up.begin(); p != up.end(); ++p) {
     if (*p == mds->get_nodeid())
       continue;
-    MHeartbeat *hb = new MHeartbeat(load, beat_epoch);
+    MHeartbeat *hb = new MHeartbeat(load, beat_epoch, last_epoch_under);
     hb->get_import_map() = import_map;
     messenger->send_message(hb,
                             mds->mdsmap->get_inst(*p));
@@ -384,6 +384,7 @@ void MDBalancer::handle_heartbeat(MHeartbeat *m)
     }
   }
   mds_import_map[ who ] = m->get_import_map();
+  mds_last_epoch_under_info[who] = m->get_last_epoch_under();
 
   //dout(0) << "  load is " << load << " have " << mds_load.size() << dendl;
 
@@ -663,7 +664,8 @@ void MDBalancer::prep_rebalance(int beat)
        dout(15) << "   mds." << it->second << " is importer" << dendl;
        importers.insert(pair<double,mds_rank_t>(it->first,it->second));
        importer_set.insert(it->second);
-      } else {
+      } else if ((it->first > target_load * (1.0 + g_conf->mds_bal_min_rebalance)) && 
+        (it->second == whoami || !mds_last_epoch_under_info[it->second] || beat_epoch - mds_last_epoch_under_info[it->second] >= 2)){
        dout(15) << "   mds." << it->second << " is exporter" << dendl;
        exporters.insert(pair<double,mds_rank_t>(it->first,it->second));
        exporter_set.insert(it->second);

diff --git a/src/mds/MDBalancer.h b/src/mds/MDBalancer.h
index d23185b22f898e1ffe278178813bfc62b5683991..ae7aa21ccc69bc5be8ecae38a39088d1b97eaae5 100644 (file)
--- a/src/mds/MDBalancer.h
+++ b/src/mds/MDBalancer.h
@@ -142,6 +142,7 @@ private:
   map<mds_rank_t, mds_load_t>  mds_load;
   map<mds_rank_t, double>       mds_meta_load;
   map<mds_rank_t, map<mds_rank_t, float> > mds_import_map;
+  map<mds_rank_t, int> mds_last_epoch_under_info;
 
   // per-epoch state
   double          my_load, target_load;

diff --git a/src/messages/MHeartbeat.h b/src/messages/MHeartbeat.h
index ad0d8f8e4818a915b3d3986a10a1ca587d12a7b8..ff3d178d5b02564860349954633d60f41f97860e 100644 (file)
--- a/src/messages/MHeartbeat.h
+++ b/src/messages/MHeartbeat.h
@@ -20,24 +20,29 @@
 #include "msg/Message.h"
 
 class MHeartbeat : public Message {
+  static const int HEAD_VERSION = 2;
+  static const int COMPAT_VERSION = 1;
   mds_load_t load;
   __s32        beat = 0;
+  __s32        last_epoch_under = 0;
   map<mds_rank_t, float> import_map;
 
  public:
   mds_load_t& get_load() { return load; }
   int get_beat() { return beat; }
+  int get_last_epoch_under() { return last_epoch_under; }
 
   map<mds_rank_t, float>& get_import_map() {
     return import_map;
   }
 
   MHeartbeat()
-    : Message(MSG_MDS_HEARTBEAT), load(utime_t()) { }
-  MHeartbeat(mds_load_t& load, int beat)
-    : Message(MSG_MDS_HEARTBEAT),
+    : Message(MSG_MDS_HEARTBEAT, HEAD_VERSION, COMPAT_VERSION), load(utime_t()) { }
+  MHeartbeat(mds_load_t& load, int beat, int last_epoch_under)
+    : Message(MSG_MDS_HEARTBEAT, HEAD_VERSION, COMPAT_VERSION),
       load(load) {
     this->beat = beat;
+    this->last_epoch_under = last_epoch_under;
   }
 private:
   ~MHeartbeat() override {}
@@ -49,6 +54,7 @@ public:
     ::encode(load, payload);
     ::encode(beat, payload);
     ::encode(import_map, payload);
+    ::encode(last_epoch_under, payload);
   }
   void decode_payload() override {
     bufferlist::iterator p = payload.begin();
@@ -56,6 +62,11 @@ public:
     ::decode(load, now, p);
     ::decode(beat, p);
     ::decode(import_map, p);
+    if (header.version >= 2) {
+      ::decode(last_epoch_under, p);
+    } else {
+      last_epoch_under = 0;
+    }
   }
 
 };