From: Colin Patrick McCabe <cmccabe@alumni.cmu.edu>
Date: Mon, 11 Oct 2010 23:42:35 +0000 (-0700)
Subject: mon: add 'ceph health' command
X-Git-Tag: v0.23~171
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=fbb5a457bacc656cd31cb5cfeff7f461f829fa79;p=ceph.git

mon: add 'ceph health' command

Create MDSMonitor::get_health and OSDMonitor::get_health to check the
health of the MDSes and OSDes, respectively.

Signed-off-by: Colin McCabe <colinm@hq.newdream.net>
---

diff --git a/src/include/types.h b/src/include/types.h
index e439924867d8..b747ed579e27 100644
--- a/src/include/types.h
+++ b/src/include/types.h
@@ -477,4 +477,27 @@ inline ostream& operator<<(ostream& out, const ceph_mon_subscribe_item& i)
 	     << ((i.flags & CEPH_SUBSCRIBE_ONETIME) ? "" : "+");
 }
 
+enum health_status_t {
+  HEALTH_ERR = 0,
+  HEALTH_WARN = 1,
+  HEALTH_OK = 2,
+};
+
+#ifdef __cplusplus
+inline ostream& operator<<(ostream &oss, health_status_t status) {
+  switch (status) {
+    case HEALTH_ERR:
+      oss << "HEALTH_ERR";
+      break;
+    case HEALTH_WARN:
+      oss << "HEALTH_WARN";
+      break;
+    case HEALTH_OK:
+      oss << "HEALTH_OK";
+      break;
+  }
+  return oss;
+};
+#endif
+
 #endif
diff --git a/src/mon/MDSMonitor.cc b/src/mon/MDSMonitor.cc
index 642d9f170764..4c87a150990c 100644
--- a/src/mon/MDSMonitor.cc
+++ b/src/mon/MDSMonitor.cc
@@ -414,6 +414,24 @@ void MDSMonitor::committed()
   tick();
 }
 
+enum health_status_t MDSMonitor::get_health(std::ostream &ss)
+{
+  health_status_t ret(HEALTH_OK);
+
+  bool ok = true;
+  if (mdsmap.is_stopped()) {
+    ss << "mdsmap: STATUS_ERROR. Mdsmap is stopped! ";
+    ret = HEALTH_ERR;
+  }
+  else if (!mdsmap.is_full()) {
+    ss << "mdsmap: STATUS_WARN. Mdsmap is not full. ";
+    ret = HEALTH_WARN;
+  }
+  if (ret > HEALTH_OK) {
+    ss << "mdsmap: " << "'" << mdsmap << "'" << std::endl;
+  }
+  return ret;
+}
 
 bool MDSMonitor::preprocess_command(MMonCommand *m)
 {
diff --git a/src/mon/MDSMonitor.h b/src/mon/MDSMonitor.h
index 3bb6e724bc02..ba05d048c87d 100644
--- a/src/mon/MDSMonitor.h
+++ b/src/mon/MDSMonitor.h
@@ -83,6 +83,7 @@ class MDSMonitor : public PaxosService {
   bool preprocess_offload_targets(MMDSLoadTargets *m);
   bool prepare_offload_targets(MMDSLoadTargets *m);
 
+  enum health_status_t get_health(std::ostream &ss);
   bool preprocess_command(MMonCommand *m);
   bool prepare_command(MMonCommand *m);
 
diff --git a/src/mon/Monitor.cc b/src/mon/Monitor.cc
index 37ee6db4f6a4..54a387edba03 100644
--- a/src/mon/Monitor.cc
+++ b/src/mon/Monitor.cc
@@ -298,6 +298,7 @@ void Monitor::handle_command(MMonCommand *m)
   bufferlist rdata;
   string rs;
   int r = -EINVAL;
+  rs = "unrecognized subsystem";
   if (!m->cmd.empty()) {
     if (m->cmd[0] == "mds") {
       mdsmon()->dispatch(m);
@@ -339,7 +340,10 @@ void Monitor::handle_command(MMonCommand *m)
       authmon()->dispatch(m);
       return;
     }
-    rs = "unrecognized subsystem";
+    if (m->cmd[0] == "health") {
+      monmon()->dispatch(m);
+      return;
+    }
   } else 
     rs = "no command";
 
diff --git a/src/mon/MonmapMonitor.cc b/src/mon/MonmapMonitor.cc
index fab922be690d..4027449c0067 100644
--- a/src/mon/MonmapMonitor.cc
+++ b/src/mon/MonmapMonitor.cc
@@ -18,6 +18,8 @@
 
 #include "messages/MMonCommand.h"
 #include "common/Timer.h"
+#include "mon/MDSMonitor.h"
+#include "mon/OSDMonitor.h"
 
 #include <sstream>
 #include "config.h"
@@ -150,10 +152,30 @@ bool MonmapMonitor::preprocess_command(MMonCommand *m)
     else if (m->cmd[1] == "remove")
       return false;
   }
+  else if (m->cmd[0] == "health") {
+    ostringstream oss;
+    health_status_t overall = HEALTH_OK;
+    try {
+      health_status_t ret;
+      ret = mon->mdsmon()->get_health(oss);
+      if (ret < overall)
+	overall = ret;
+      ret = mon->osdmon()->get_health(oss);
+      if (ret < overall)
+	overall = ret;
+    }
+    catch (const std::exception &e) {
+      oss << " monmapmonitor: caught exception while "
+	  << "checking health: '" << e.what() << "'";
+    }
+    ss << overall << oss.str();
+    r = 0;
+  }
 
   if (r != -1) {
     string rs;
     getline(ss, rs);
+
     mon->reply_command(m, r, rs, rdata, paxos->get_version());
     return true;
   } else
diff --git a/src/mon/OSDMonitor.cc b/src/mon/OSDMonitor.cc
index ca57c96f475f..5ab2d9555eeb 100644
--- a/src/mon/OSDMonitor.cc
+++ b/src/mon/OSDMonitor.cc
@@ -1016,7 +1016,27 @@ void OSDMonitor::mark_all_down()
   propose_pending();
 }
 
+enum health_status_t OSDMonitor::get_health(std::ostream &ss)
+{
+  enum health_status_t ret(HEALTH_OK);
+
+  int num_osds = osdmap.get_num_osds();
+  int num_up_osds = osdmap.get_num_up_osds();
+  int num_in_osds = osdmap.get_num_in_osds();
 
+  if (num_osds == 0) {
+    ss << " osdmonitor: no OSDS in osdmap!";
+    ret = HEALTH_ERR;
+  }
+  else if ((num_up_osds != num_osds) ||
+	   (num_in_osds != num_osds)) {
+    ss << " osdmonitor: num_osds = " << num_osds << ", ";
+    ss << "num_up_osds = " << num_up_osds << ", num_in_osds = ";
+    ss << num_in_osds;
+    ret = HEALTH_WARN;
+  }
+  return ret;
+}
 
 bool OSDMonitor::preprocess_command(MMonCommand *m)
 {
diff --git a/src/mon/OSDMonitor.h b/src/mon/OSDMonitor.h
index 9be1ab30750c..193c209e9dc8 100644
--- a/src/mon/OSDMonitor.h
+++ b/src/mon/OSDMonitor.h
@@ -158,6 +158,7 @@ private:
 
   void tick();  // check state, take actions
 
+  enum health_status_t get_health(std::ostream &ss);
   bool preprocess_command(MMonCommand *m);
   bool prepare_command(MMonCommand *m);