Merge remote-tracking branch 'gh/next'

3 files changed, 21 insertions, 3 deletions

diff --git a/src/common/config_opts.h b/src/common/config_opts.h
index d3e21105e79..b021651bd4d 100644
--- a/src/common/config_opts.h
+++ b/src/common/config_opts.h
@@ -491,6 +491,9 @@ OPTION(osd_leveldb_compression, OPT_BOOL, true) // OSD's leveldb uses compressio
 OPTION(osd_leveldb_paranoid, OPT_BOOL, false) // OSD's leveldb paranoid flag
 OPTION(osd_leveldb_log, OPT_STR, "")  // enable OSD leveldb log file
 
+// determines whether PGLog::check() compares written out log to stored log
+OPTION(osd_debug_pg_log_writeout, OPT_BOOL, false)
+
 /**
  * osd_client_op_priority and osd_recovery_op_priority adjust the relative
  * priority of client io vs recovery io.
diff --git a/src/osd/PG.cc b/src/osd/PG.cc
index 63be6a34b03..49ea61a603a 100644
--- a/src/osd/PG.cc
+++ b/src/osd/PG.cc
@@ -158,7 +158,7 @@ PG::PG(OSDService *o, OSDMapRef curmap,
   deleting(false), dirty_info(false), dirty_big_info(false),
   info(p),
   info_struct_v(0),
-  coll(p), log_oid(loid), biginfo_oid(ioid),
+  coll(p), pg_log(g_ceph_context), log_oid(loid), biginfo_oid(ioid),
   recovery_item(this), scrub_item(this), scrub_finalize_item(this), snap_trim_item(this), stat_queue_item(this),
   recovery_ops_active(0),
   waiting_on_backfill(0),
@@ -2438,7 +2438,9 @@ void PG::read_state(ObjectStore *store, bufferlist &bl)
     /* We don't want to leave the old format around in case the next log
      * write happens to be an append_log()
      */
+    pg_log.mark_log_for_rewrite();
     ObjectStore::Transaction t;
+    t.remove(coll_t(), log_oid); // remove old version
     pg_log.write_log(t, log_oid);
     int r = osd->store->apply_transaction(t);
     assert(!r);
diff --git a/src/osd/PGLog.h b/src/osd/PGLog.h
index 68317750332..552f9b0cee9 100644
--- a/src/osd/PGLog.h
+++ b/src/osd/PGLog.h
@@ -21,6 +21,7 @@
 #include "include/assert.h" 
 #include "osd_types.h"
 #include "os/ObjectStore.h"
+#include "common/ceph_context.h"
 #include <list>
 using namespace std;
 
@@ -159,6 +160,7 @@ protected:
   eversion_t dirty_to;
   eversion_t dirty_from;
   bool dirty_divergent_priors;
+  CephContext *cct;
 
   bool is_dirty() const {
     return !touched_log ||
@@ -178,6 +180,13 @@ protected:
     divergent_priors.insert(make_pair(version, obj));
     dirty_divergent_priors = true;
   }
+public:
+  void mark_log_for_rewrite() {
+    mark_dirty_to(eversion_t::max());
+    mark_dirty_from(eversion_t());
+    touched_log = false;
+  }
+protected:
 
   /// DEBUG
   set<string> log_keys_debug;
@@ -197,6 +206,10 @@ protected:
   }
   void check() {
     assert(log.log.size() == log_keys_debug.size());
+    if (cct &&
+        !(cct->_conf->osd_debug_pg_log_writeout)) {
+      return;
+    }
     for (list<pg_log_entry_t>::iterator i = log.log.begin();
 	 i != log.log.end();
 	 ++i) {
@@ -212,9 +225,9 @@ protected:
     check();
   }
 public:
-  PGLog() :
+  PGLog(CephContext *cct = 0) :
     touched_log(false), dirty_from(eversion_t::max()),
-    dirty_divergent_priors(false) {}
+    dirty_divergent_priors(false), cct(cct) {}
 
 
   void reset_backfill();