mon: mdsmap infrastructure for standby states, up:standby-replay

We need a state value for standby mds's.

diff --git a/src/TODO b/src/TODO
index 2f40cef3cadbd9e7367dd9dfb11d3828a49fab99..c01626bd289bf4f0b2cf21417e2a527f135fea8e 100644 (file)
--- a/src/TODO
+++ b/src/TODO
@@ -115,6 +115,10 @@ mds
 - what if a recovery is queued, or in progress, and the inode is then cowed?  can that happen?  
 - proper handling of cache expire messages during rejoin phase?
   -> i think cache expires are fine; the rejoin_ack handler just has to behave if rejoining items go missing
+- add up:standby-replay mode?
+  - tail the mds log as it is written
+  - periodically check head so that we trim, too
+  - a brute-force mode too that will replay from scratch?
 - try_remove_unlinked_dn thing
 - rename: importing inode... also journal imported client map?
 - rerun destro trace against latest, with various journal lengths

diff --git a/src/include/ceph_fs.h b/src/include/ceph_fs.h
index 192fc49fc5e01cc1b78340ab054d1583416c0d22..b9dc338d2f9a5d626971d88aded650910f5151bc 100644 (file)
--- a/src/include/ceph_fs.h
+++ b/src/include/ceph_fs.h
@@ -584,6 +584,7 @@ struct ceph_mds_getmap {
 #define CEPH_MDS_STATE_STANDBY    -5  /* up, idle.  waiting for assignment. */
 #define CEPH_MDS_STATE_CREATING   -6  /* up, creating MDS instance. */
 #define CEPH_MDS_STATE_STARTING   -7  /* up, starting previously stopped mds. */
+#define CEPH_MDS_STATE_STANDBY_REPLAY -8  /* up, tailing active node's journal. */
 
 #define CEPH_MDS_STATE_REPLAY      8  /* up, replaying journal. */
 #define CEPH_MDS_STATE_RESOLVE     9  /* up, disambiguating distributed

diff --git a/src/mds/MDSMap.cc b/src/mds/MDSMap.cc
index 39ea1db3a834986f72f1a522429b8a35a993dd54..cadda5480cd507d5b0b760bff575c9f559122371 100644 (file)
--- a/src/mds/MDSMap.cc
+++ b/src/mds/MDSMap.cc
@@ -32,15 +32,23 @@ void MDSMap::print(ostream& out)
     out << " mds" << *p << "." << mds_inc[*p]
        << " : " << get_state_name(get_state(*p));
     if (have_inst(*p))
-      out << " : " << get_inst(*p)
+      out << " : " << get_inst(*p).addr
          << (is_laggy(get_inst(*p).addr) ? " LAGGY" : "");
-    if (standby_for.count(*p) && !standby_for[*p].empty())
-      out << " : +" << standby_for[*p].size()
-         << " standby " << standby_for[*p]
-         << std::endl;
+    out << "\n";
+    if (standby_for.count(*p) && !standby_for[*p].empty()) {
+      //out << " : +" << standby_for[*p].size() << std::endl;
+      for (set<entity_addr_t>::iterator q = standby_for[*p].begin();
+          q != standby_for[*p].end();
+          q++)
+       out << " mds" << *p << ".? : " << get_state_name(standby[*q].state)
+           << " : " << *q << std::endl;
+    }
   }
   if (!standby_any.empty()) {
-    out << " +" << standby_any.size() << " shared standby " << standby_any << std::endl;
+    for (set<entity_addr_t>::iterator q = standby_any.begin();
+        q != standby_any.end();
+        q++)
+      out << " mds?.? : " << get_state_name(standby[*q].state) << " : " << *q << std::endl;
   }
 }
 

diff --git a/src/mds/MDSMap.h b/src/mds/MDSMap.h
index a42ddccc53c5a2888161d4a8f54a8f8d4d72d0f6..602f11a77bd31c002395573bc1c334f4aba82d22 100644 (file)
--- a/src/mds/MDSMap.h
+++ b/src/mds/MDSMap.h
@@ -61,6 +61,7 @@ class MDSMap {
 
   static const int STATE_BOOT     =   CEPH_MDS_STATE_BOOT;  // up, boot announcement.  destiny unknown.
   static const int STATE_STANDBY  =   CEPH_MDS_STATE_STANDBY;  // up, idle.  waiting for assignment by monitor.
+  static const int STATE_STANDBY_REPLAY = CEPH_MDS_STATE_STANDBY_REPLAY;  // up, replaying active node; ready to take over.
 
   static const int STATE_CREATING  =  CEPH_MDS_STATE_CREATING;  // up, creating MDS instance (new journal, idalloc..).
   static const int STATE_STARTING  =  CEPH_MDS_STATE_STARTING;  // up, starting prior stopped MDS instance.
@@ -83,6 +84,7 @@ class MDSMap {
       // up and out
     case STATE_BOOT:       return "up:boot";
     case STATE_STANDBY:    return "up:standby";
+    case STATE_STANDBY_REPLAY:    return "up:standby-replay";
     case STATE_CREATING:   return "up:creating";
     case STATE_STARTING:   return "up:starting";
       // up and in
@@ -97,6 +99,21 @@ class MDSMap {
     return 0;
   }
 
+  struct standby_t {
+    int32_t mds;
+    int32_t state;
+    void encode(bufferlist& bl) const {
+      ::encode(mds, bl);
+      ::encode(state, bl);
+    }
+    void decode(bufferlist::iterator& bl) {
+      ::decode(mds, bl);
+      ::decode(state, bl);
+    }
+  };
+  WRITE_CLASS_ENCODER(standby_t)
+
+
  protected:
   epoch_t epoch;
   epoch_t client_epoch;       // incremented only when change is significant to client.
@@ -115,7 +132,7 @@ class MDSMap {
   map<int32_t,entity_inst_t> mds_inst;      // up instances
   map<int32_t,int32_t>       mds_inc;       // incarnation count (monotonically increases)
 
-  map<entity_addr_t,int32_t> standby;    // -1 == any
+  map<entity_addr_t,standby_t> standby;    // -1 == any
   map<int32_t, set<entity_addr_t> > standby_for;
   set<entity_addr_t> standby_any;
   
@@ -257,6 +274,17 @@ class MDSMap {
   bool is_stopped(int m)  { return mds_state.count(m) && mds_state[m] == STATE_STOPPED; }
 
   bool is_standby(entity_addr_t a)  { return standby.count(a); }
+  
+  int get_standby_for(entity_addr_t a) {
+    if (standby.count(a))
+      return standby[a].mds;
+    return -1;
+  }
+  int get_standby_state(entity_addr_t a) {
+    if (standby.count(a))
+      return standby[a].state;
+    return STATE_DNE;
+  }
 
   // cluster states
   bool is_full() {
@@ -395,6 +423,7 @@ class MDSMap {
   void print(ostream& out);
   void print_summary(ostream& out);
 };
+WRITE_CLASS_ENCODER(MDSMap::standby_t)
 
 inline ostream& operator<<(ostream& out, MDSMap& m) {
   m.print_summary(out);

diff --git a/src/mon/MDSMonitor.cc b/src/mon/MDSMonitor.cc
index d631e78055b950ad5d2ce4747825f088098076bb..157aa9c3b1166a6438d390032c1493a8c2851254 100644 (file)
--- a/src/mon/MDSMonitor.cc
+++ b/src/mon/MDSMonitor.cc
@@ -350,7 +350,8 @@ bool MDSMonitor::prepare_beacon(MMDSBeacon *m)
        dout(10) << "mds_beacon boot: standby for mds" << standby_for << dendl;
        pending_mdsmap.standby_for[standby_for].insert(addr);
       }
-      pending_mdsmap.standby[addr] = standby_for;
+      pending_mdsmap.standby[addr].mds = standby_for;
+      pending_mdsmap.standby[addr].state = MDSMap::STATE_STANDBY;
       state = MDSMap::STATE_STANDBY;
     } else {
       // join|takeover
@@ -579,7 +580,7 @@ void MDSMonitor::bcast_latest_mds()
   // standby too
   entity_inst_t inst;
   inst.name = entity_name_t::MDS(-1);
-  for (map<entity_addr_t,int32_t>::iterator p = mdsmap.standby.begin();
+  for (map<entity_addr_t,MDSMap::standby_t>::iterator p = mdsmap.standby.begin();
        p != mdsmap.standby.end();
        p++) {
     inst.addr = p->first;
@@ -669,7 +670,7 @@ void MDSMonitor::tick()
        mdsmap.get_state(p->first) != MDSMap::STATE_STOPPED &&
        mdsmap.get_state(p->first) != MDSMap::STATE_FAILED)
       last_beacon[p->second.addr] = g_clock.now();
-  for (map<entity_addr_t,int32_t>::iterator p = mdsmap.standby.begin();
+  for (map<entity_addr_t,MDSMap::standby_t>::iterator p = mdsmap.standby.begin();
        p != mdsmap.standby.end();
        ++p )
     if (last_beacon.count(p->first) == 0)
@@ -742,8 +743,8 @@ void MDSMonitor::tick()
       dout(10) << "no beacon from standby " << addr << " since " << last_beacon[addr]
               << ", removing from standby list"
               << dendl;
-      if (pending_mdsmap.standby[addr] >= 0)
-       pending_mdsmap.standby_for[pending_mdsmap.standby[addr]].erase(addr);
+      if (pending_mdsmap.standby[addr].mds >= 0)
+       pending_mdsmap.standby_for[pending_mdsmap.standby[addr].mds].erase(addr);
       else
        pending_mdsmap.standby_any.erase(addr);
       pending_mdsmap.standby.erase(addr);