fixed osd failure reporting; monitorstore mkfs; monitorstore cleanup

git-svn-id: https://ceph.svn.sf.net/svnroot/ceph@1123 29311d96-e01e-0410-9327-a35deaab8ce9

diff --git a/branches/riccardo/monitor2/client/Client.cc b/branches/riccardo/monitor2/client/Client.cc
index 2b85f1a7ad97d7461b86e618008c36eb4806bd43..928096ace466769fb25bc8ae199b3a4131c3707d 100644 (file)
--- a/branches/riccardo/monitor2/client/Client.cc
+++ b/branches/riccardo/monitor2/client/Client.cc
@@ -2617,8 +2617,10 @@ int Client::lazyio_synchronize(int fd, off_t offset, size_t count)
 }
 
 
-void Client::ms_handle_failure(Message *m, entity_name_t dest, const entity_inst_t& inst)
+void Client::ms_handle_failure(Message *m, const entity_inst_t& inst)
 {
+  msg_addr_t dest = inst.name;
+
   if (dest.is_mon()) {
     // resend to a different monitor.
     int mon = monmap->pick_mon(true);

diff --git a/branches/riccardo/monitor2/client/Client.h b/branches/riccardo/monitor2/client/Client.h
index de16639c89e642affe398c30bbba6fbb64265b4a..e00c6eae3ca4814e7a6ae82fa560eb6176bc0821 100644 (file)
--- a/branches/riccardo/monitor2/client/Client.h
+++ b/branches/riccardo/monitor2/client/Client.h
@@ -582,7 +582,7 @@ protected:
 
   int describe_layout(char *fn, list<ObjectExtent>& result);
 
-  void ms_handle_failure(Message*, entity_name_t dest, const entity_inst_t& inst);
+  void ms_handle_failure(Message*, const entity_inst_t& inst);
 };
 
 #endif

diff --git a/branches/riccardo/monitor2/mon/MonitorStore.cc b/branches/riccardo/monitor2/mon/MonitorStore.cc
index cce0d759393a703829a5fa43ac245a3d6cf4cdd3..a56dcb229e4f01e2fbe7563c56bb94f6af7d7f76 100644 (file)
--- a/branches/riccardo/monitor2/mon/MonitorStore.cc
+++ b/branches/riccardo/monitor2/mon/MonitorStore.cc
@@ -43,9 +43,9 @@ void MonitorStore::mkfs()
   dout(1) << "mkfs" << endl;
 
   char cmd[200];
-  sprintf(cmd, "/bin/rm -rf %s/*", dir.c_str());
+  sprintf(cmd, "/bin/rm -r %s ; mkdir %s", dir.c_str(), dir.c_str());
   dout(1) << cmd << endl;
-  //system(cmd);
+  system(cmd);
 }
 
 
@@ -68,9 +68,9 @@ version_t MonitorStore::get_int(const char *a, const char *b)
   version_t val = atoi(buf);
   
   if (b) {
-    dout(10) << "get_int " << a << "/" << b << " = " << val << endl;
+    dout(15) << "get_int " << a << "/" << b << " = " << val << endl;
   } else {
-    dout(10) << "get_int " << a << " = " << val << endl;
+    dout(15) << "get_int " << a << " = " << val << endl;
   }
   return val;
 }
@@ -82,18 +82,23 @@ void MonitorStore::put_int(version_t val, const char *a, const char *b)
   sprintf(fn, "%s/%s", dir.c_str(), a);
   if (b) {
     ::mkdir(fn, 0755);
-    dout(10) << "set_int " << a << "/" << b << " = " << val << endl;
+    dout(15) << "set_int " << a << "/" << b << " = " << val << endl;
     sprintf(fn, "%s/%s/%s", dir.c_str(), a, b);
   } else {
-    dout(10) << "set_int " << a << " = " << val << endl;
+    dout(15) << "set_int " << a << " = " << val << endl;
   }
   
+  char vs[30];
+  sprintf(vs, "%lld\n", val);
+
   char tfn[200];
   sprintf(tfn, "%s.new", fn);
-  FILE *f = ::fopen(tfn, "w");
-  assert(f);
-  ::fprintf(f, "%lld\n", val);
-  ::fclose(f);
+
+  int fd = ::open(tfn, O_WRONLY|O_CREAT);
+  assert(fd > 0);
+  ::fchmod(fd, 0644);
+  ::write(fd, vs, strlen(vs));
+  ::close(fd);
   ::rename(tfn, fn);
 }
 
@@ -105,10 +110,10 @@ bool MonitorStore::exists_bl(const char *a, const char *b)
 {
   char fn[200];
   if (b) {
-    dout(10) << "exists_bl " << a << "/" << b << endl;
+    dout(15) << "exists_bl " << a << "/" << b << endl;
     sprintf(fn, "%s/%s/%s", dir.c_str(), a, b);
   } else {
-    dout(10) << "exists_bl " << a << endl;
+    dout(15) << "exists_bl " << a << endl;
     sprintf(fn, "%s/%s", dir.c_str(), a);
   }
   
@@ -130,9 +135,9 @@ int MonitorStore::get_bl(bufferlist& bl, const char *a, const char *b)
   int fd = ::open(fn, O_RDONLY);
   if (!fd) {
     if (b) {
-      dout(10) << "get_bl " << a << "/" << b << " DNE" << endl;
+      dout(15) << "get_bl " << a << "/" << b << " DNE" << endl;
     } else {
-      dout(10) << "get_bl " << a << " DNE" << endl;
+      dout(15) << "get_bl " << a << " DNE" << endl;
     }
     return 0;
   }
@@ -149,9 +154,9 @@ int MonitorStore::get_bl(bufferlist& bl, const char *a, const char *b)
   ::close(fd);
 
   if (b) {
-    dout(10) << "get_bl " << a << "/" << b << " = " << bl.length() << " bytes" << endl;
+    dout(15) << "get_bl " << a << "/" << b << " = " << bl.length() << " bytes" << endl;
   } else {
-    dout(10) << "get_bl " << a << " = " << bl.length() << " bytes" << endl;
+    dout(15) << "get_bl " << a << " = " << bl.length() << " bytes" << endl;
   }
 
   return len;
@@ -163,10 +168,10 @@ int MonitorStore::put_bl(bufferlist& bl, const char *a, const char *b)
   sprintf(fn, "%s/%s", dir.c_str(), a);
   if (b) {
     ::mkdir(fn, 0755);
-    dout(10) << "put_bl " << a << "/" << b << " = " << bl.length() << " bytes" << endl;
+    dout(15) << "put_bl " << a << "/" << b << " = " << bl.length() << " bytes" << endl;
     sprintf(fn, "%s/%s/%s", dir.c_str(), a, b);
   } else {
-    dout(10) << "put_bl " << a << " = " << bl.length() << " bytes" << endl;
+    dout(15) << "put_bl " << a << " = " << bl.length() << " bytes" << endl;
   }
   
   char tfn[200];
@@ -183,7 +188,7 @@ int MonitorStore::put_bl(bufferlist& bl, const char *a, const char *b)
        it != bl.buffers().end();
        it++) 
     ::write(fd, it->c_str(), it->length());
-  
+  ::fchmod(fd, 0644);
   ::fsync(fd);
   ::close(fd);
   ::rename(tfn, fn);

diff --git a/branches/riccardo/monitor2/msg/Dispatcher.h b/branches/riccardo/monitor2/msg/Dispatcher.h
index 97e59bbb7819d6f66630ef79290a0cdf0d73fb4d..8b6fe923814276671561fcb3d017b20f1d4129f1 100644 (file)
--- a/branches/riccardo/monitor2/msg/Dispatcher.h
+++ b/branches/riccardo/monitor2/msg/Dispatcher.h
@@ -27,7 +27,7 @@ class Dispatcher {
   virtual void dispatch(Message *m) = 0;
 
   // how i deal with transmission failures.
-  virtual void ms_handle_failure(Message *m, entity_name_t dest, const entity_addr_t& addr) { delete m; }
+  virtual void ms_handle_failure(Message *m, const entity_inst_t& inst) { delete m; }
 };
 
 #endif

diff --git a/branches/riccardo/monitor2/msg/SimpleMessenger.cc b/branches/riccardo/monitor2/msg/SimpleMessenger.cc
index 99a04a82918989bbbcc2e1b56b713fee0ce7d983..f31f8378b648720b51f83fe99c40b1cdfb0ed123 100644 (file)
--- a/branches/riccardo/monitor2/msg/SimpleMessenger.cc
+++ b/branches/riccardo/monitor2/msg/SimpleMessenger.cc
@@ -683,7 +683,7 @@ void Rank::Pipe::fail(list<Message*>& out)
            k != j->second.end();
            ++k) {
        derr(1) << "pipe(" << peer_addr << ' ' << this << ").fail on " << **k << " to " << j->first << " inst " << peer_addr << endl;
-        i->first->ms_handle_failure(*k, j->first, peer_addr);
+        i->first->ms_handle_failure(*k, (*k)->get_dest_inst());
       }
 }
 

diff --git a/branches/riccardo/monitor2/osd/OSD.cc b/branches/riccardo/monitor2/osd/OSD.cc
index 7e3bcc5568eed2b52e5ceae445583f795c45cb37..1799a4109cb7559bf1d5a76c9272b109803f2819 100644 (file)
--- a/branches/riccardo/monitor2/osd/OSD.cc
+++ b/branches/riccardo/monitor2/osd/OSD.cc
@@ -731,6 +731,7 @@ void OSD::ms_handle_failure(Message *m, const entity_inst_t& inst)
   entity_name_t dest = inst.name;
 
   if (g_conf.ms_die_on_failure) {
+    dout(0) << "ms_handle_failure " << inst << " on " << *m << endl;
     exit(0);
   }
 
@@ -1018,6 +1019,8 @@ void OSD::advance_map(ObjectStore::Transaction& t)
     //cerr << "osdmap " << osdmap->get_ctime() << " logger start " << logger->get_start() << endl;
     logger->set_start( osdmap->get_ctime() );
 
+    assert(g_conf.osd_mkfs);  // make sure we did a mkfs!
+
     // create PGs
     for (int nrep = 1; 
          nrep <= MIN(g_conf.num_osd, g_conf.osd_max_rep);    // for low osd counts..  hackish bleh