From 3b782a488e6f9b5018488615eca64a54d9e126c5 Mon Sep 17 00:00:00 2001
From: Sage Weil <sage@newdream.net>
Date: Tue, 2 Dec 2008 14:25:54 -0800
Subject: [PATCH] osd: log scrub errors to system log

---
 src/osd/OSD.cc          | 54 ++++++++++++++++++++++++++++++++++++++---
 src/osd/OSD.h           | 10 ++++++++
 src/osd/ReplicatedPG.cc | 25 ++++++++++++++++++-
 3 files changed, 84 insertions(+), 5 deletions(-)

diff --git a/src/osd/OSD.cc b/src/osd/OSD.cc
index 636799e5533f3..af1c0b50f0a25 100644
--- a/src/osd/OSD.cc
+++ b/src/osd/OSD.cc
@@ -36,6 +36,8 @@
 #include "msg/Messenger.h"
 #include "msg/Message.h"
 
+#include "messages/MLog.h"
+
 #include "messages/MGenericMessage.h"
 #include "messages/MPing.h"
 #include "messages/MOSDPing.h"
@@ -271,8 +273,12 @@ OSD::OSD(int id, Messenger *m, Messenger *hbm, MonMap *mm, const char *dev) :
   osdmap(NULL),
   map_lock("OSD::map_lock"),
   map_cache_lock("OSD::map_cache_lock"),
+  log_lock("OSD::log_lock"), last_log(0),
+  up_thru_wanted(0), up_thru_pending(0),
   pg_stat_queue_lock("OSD::pg_stat_queue_lock"),
+  last_tid(0),
   tid_lock("OSD::tid_lock"),
+  num_pulling(0),
   recovery_ops_active(0),
   recovery_wq(this),
   remove_list_lock("OSD::remove_list_lock"),
@@ -281,12 +287,8 @@ OSD::OSD(int id, Messenger *m, Messenger *hbm, MonMap *mm, const char *dev) :
 {
   osdmap = 0;
 
-  last_tid = 0;
-  num_pulling = 0;
-
   memset(&my_stat, 0, sizeof(my_stat));
 
-  up_thru_wanted = up_thru_pending = 0;
   osd_stat_updated = osd_stat_pending = false;
 
   stat_ops = 0;
@@ -1213,6 +1215,7 @@ void OSD::do_mon_report()
   }
 
   // do any pending reports
+  send_log();
   if (is_booting())
     send_boot();
   send_alive();
@@ -1220,6 +1223,45 @@ void OSD::do_mon_report()
   send_pg_stats();
 }
 
+
+void OSD::log(__u8 level, string s)
+{
+  Mutex::Locker l(log_lock);
+  dout(10) << "log " << (int)level << " : " << s << dendl;
+  LogEntry e;
+  e.who = messenger->get_myinst();
+  e.stamp = g_clock.now();
+  e.seq = ++last_log;
+  e.level = level;
+  e.msg = s;
+  log_queue.push_back(e);
+}
+
+void OSD::send_log()
+{
+  Mutex::Locker l(log_lock);
+  if (log_queue.empty())
+    return;
+  MLog *log = new MLog(osdmap->get_fsid());
+  log->entries = log_queue;
+  int mon = monmap->pick_mon();
+  dout(10) << "send_log to mon" << mon << dendl;
+  messenger->send_message(log, monmap->get_inst(mon));
+}
+
+void OSD::handle_log(MLog *m)
+{
+  Mutex::Locker l(log_lock);
+  dout(10) << "handle_log " << *m << dendl;
+
+  version_t last = m->entries.rbegin()->seq;
+  while (log_queue.size() && log_queue.begin()->seq <= last) {
+    dout(10) << " logged " << log_queue.front() << dendl;
+    log_queue.pop_front();
+  }
+  delete m;
+}
+
 void OSD::send_boot()
 {
   int mon = monmap->pick_mon(true);
@@ -1453,6 +1495,10 @@ void OSD::dispatch(Message *m)
     delete m;
     break;
 
+  case MSG_LOG:
+    handle_log((MLog*)m);
+    break;
+
     // -- don't need OSDMap --
 
     // map and replication
diff --git a/src/osd/OSD.h b/src/osd/OSD.h
index 80c98fc895ed2..8bd47429c47ca 100644
--- a/src/osd/OSD.h
+++ b/src/osd/OSD.h
@@ -30,6 +30,7 @@
 
 #include "common/DecayCounter.h"
 
+#include "include/LogEntry.h"
 
 #include <map>
 using namespace std;
@@ -44,6 +45,7 @@ class Message;
 class Logger;
 class ObjectStore;
 class OSDMap;
+class MLog;
 
 class OSD : public Dispatcher {
 
@@ -366,6 +368,14 @@ private:
 
   void do_mon_report();
 
+  // -- log --
+  Mutex log_lock;
+  deque<LogEntry> log_queue;
+  version_t last_log;
+
+  void log(__u8 level, string s);
+  void send_log();
+  void handle_log(MLog *m);
 
   // -- boot --
   void send_boot();
diff --git a/src/osd/ReplicatedPG.cc b/src/osd/ReplicatedPG.cc
index cb5783465897f..50aaba1304f27 100644
--- a/src/osd/ReplicatedPG.cc
+++ b/src/osd/ReplicatedPG.cc
@@ -39,9 +39,9 @@ static ostream& _prefix(PG *pg, int whoami, OSDMap *osdmap) {
 }
 
 
+#include <sstream>
 
 #include <errno.h>
-#include <sys/stat.h>
 
 static const int LOAD_LATENCY    = 1;
 static const int LOAD_QUEUE_SIZE = 2;
@@ -2942,6 +2942,29 @@ void ReplicatedPG::scrub()
 	   << stat.num_kb << "/" << info.stats.num_kb << " kb."
 	   << dendl;
 
+  if (stat.num_objects != info.stats.num_objects ||
+      stat.num_object_clones != info.stats.num_object_clones ||
+      stat.num_bytes != info.stats.num_bytes ||
+      stat.num_kb != info.stats.num_kb) {
+    stringstream ss;
+    ss << info.pgid << " scrub got "
+       << stat.num_objects << "/" << info.stats.num_objects << " objects, "
+       << stat.num_object_clones << "/" << info.stats.num_object_clones << " clones, "
+       << stat.num_bytes << "/" << info.stats.num_bytes << " bytes, "
+       << stat.num_kb << "/" << info.stats.num_kb << " kb.";
+    string s;
+    getline(ss, s);
+    osd->log(10, s);
+    /*
+  } else {
+    stringstream ss;
+    ss << info.pgid << " scrub ok";
+    string s;
+    getline(ss, s);
+    osd->log(0, s);
+    */
+  }
+
   dout(10) << "scrub finish" << dendl;
   unlock();
 }
-- 
2.39.5