Merge pull request #550 from ceph/wip-6040

Wip 6040

Reviewed-by: Sage Weil <sage@inktank.com>
Reviewed-by: Loic Dachary <loic@dachary.com>

2 files changed, 61 insertions, 28 deletions

diff --git a/src/osd/PGLog.cc b/src/osd/PGLog.cc
index 6fcca1b41bd..486d64302b9 100644
--- a/src/osd/PGLog.cc
+++ b/src/osd/PGLog.cc
@@ -64,7 +64,7 @@ void PGLog::IndexedLog::split_into(
   index();
 }
 
-void PGLog::IndexedLog::trim(eversion_t s)
+void PGLog::IndexedLog::trim(eversion_t s, set<eversion_t> *trimmed)
 {
   if (complete_to != log.end() &&
       complete_to->version <= s) {
@@ -77,6 +77,8 @@ void PGLog::IndexedLog::trim(eversion_t s)
     if (e.version > s)
       break;
     generic_dout(20) << "trim " << e << dendl;
+    if (trimmed)
+      trimmed->insert(e.version);
     unindex(e);         // remove from index,
     log.pop_front();    // from log
   }
@@ -142,14 +144,8 @@ void PGLog::trim(eversion_t trim_to, pg_info_t &info)
     assert(trim_to <= info.last_complete);
 
     dout(10) << "trim " << log << " to " << trim_to << dendl;
-    log.trim(trim_to);
+    log.trim(trim_to, &trimmed);
     info.log_tail = log.tail;
-
-    if (log.log.empty()) {
-      mark_dirty_to(eversion_t::max());
-    } else {
-      mark_dirty_to(log.log.front().version);
-    }
   }
 }
 
@@ -542,13 +538,18 @@ void PGLog::write_log(
 	     << "dirty_to: " << dirty_to
 	     << ", dirty_from: " << dirty_from
 	     << ", dirty_divergent_priors: " << dirty_divergent_priors
+	     << ", writeout_from: " << writeout_from
+	     << ", trimmed: " << trimmed
 	     << dendl;
-    _write_log(t, log, log_oid, divergent_priors,
-	       dirty_to,
-	       dirty_from,
-	       dirty_divergent_priors,
-	       !touched_log,
-               &log_keys_debug);
+    _write_log(
+      t, log, log_oid, divergent_priors,
+      dirty_to,
+      dirty_from,
+      writeout_from,
+      trimmed,
+      dirty_divergent_priors,
+      !touched_log,
+      (pg_log_debug ? &log_keys_debug : 0));
     undirty();
   } else {
     dout(10) << "log is not dirty" << dendl;
@@ -558,8 +559,11 @@ void PGLog::write_log(
 void PGLog::write_log(ObjectStore::Transaction& t, pg_log_t &log,
     const hobject_t &log_oid, map<eversion_t, hobject_t> &divergent_priors)
 {
-  _write_log(t, log, log_oid, divergent_priors, eversion_t::max(), eversion_t(),
-	     true, true, 0);
+  _write_log(
+    t, log, log_oid,
+    divergent_priors, eversion_t::max(), eversion_t(), eversion_t(),
+    set<eversion_t>(),
+    true, true, 0);
 }
 
 void PGLog::_write_log(
@@ -567,11 +571,24 @@ void PGLog::_write_log(
   const hobject_t &log_oid, map<eversion_t, hobject_t> &divergent_priors,
   eversion_t dirty_to,
   eversion_t dirty_from,
+  eversion_t writeout_from,
+  const set<eversion_t> &trimmed,
   bool dirty_divergent_priors,
   bool touch_log,
   set<string> *log_keys_debug
   )
 {
+  set<string> to_remove;
+  for (set<eversion_t>::const_iterator i = trimmed.begin();
+       i != trimmed.end();
+       ++i) {
+    to_remove.insert(i->get_key_name());
+    if (log_keys_debug) {
+      assert(log_keys_debug->count(i->get_key_name()));
+      log_keys_debug->erase(i->get_key_name());
+    }
+  }
+
 //dout(10) << "write_log, clearing up to " << dirty_to << dendl;
   if (touch_log)
     t.touch(coll_t(), log_oid);
@@ -599,7 +616,8 @@ void PGLog::_write_log(
   }
 
   for (list<pg_log_entry_t>::reverse_iterator p = log.log.rbegin();
-       p != log.log.rend() && p->version >= dirty_from &&
+       p != log.log.rend() &&
+	 (p->version >= dirty_from || p->version >= writeout_from) &&
 	 p->version >= dirty_to;
        ++p) {
     bufferlist bl(sizeof(*p) * 2);
@@ -621,6 +639,7 @@ void PGLog::_write_log(
     ::encode(divergent_priors, keys["divergent_priors"]);
   }
 
+  t.omap_rmkeys(coll_t::META_COLL, log_oid, to_remove);
   t.omap_setkeys(coll_t::META_COLL, log_oid, keys);
 }
 
diff --git a/src/osd/PGLog.h b/src/osd/PGLog.h
index 8f192a8eac0..441b56706c4 100644
--- a/src/osd/PGLog.h
+++ b/src/osd/PGLog.h
@@ -142,7 +142,7 @@ struct PGLog {
 	caller_ops[e.reqid] = &(log.back());
     }
 
-    void trim(eversion_t s);
+    void trim(eversion_t s, set<eversion_t> *trimmed);
 
     ostream& print(ostream& out) const;
   };
@@ -150,6 +150,7 @@ struct PGLog {
 
 protected:
   //////////////////// data members ////////////////////
+  bool pg_log_debug;
 
   map<eversion_t, hobject_t> divergent_priors;
   pg_missing_t     missing;
@@ -157,8 +158,10 @@ protected:
 
   /// Log is clean on [dirty_to, dirty_from)
   bool touched_log;
-  eversion_t dirty_to;
-  eversion_t dirty_from;
+  eversion_t dirty_to;         ///< must clear/writeout all keys up to dirty_to
+  eversion_t dirty_from;       ///< must clear/writeout all keys past dirty_from
+  eversion_t writeout_from;    ///< must writout keys past writeout_from
+  set<eversion_t> trimmed;     ///< must clear keys in trimmed
   bool dirty_divergent_priors;
   CephContext *cct;
 
@@ -166,7 +169,9 @@ protected:
     return !touched_log ||
       (dirty_to != eversion_t()) ||
       (dirty_from != eversion_t::max()) ||
-      dirty_divergent_priors;
+      dirty_divergent_priors ||
+      (writeout_from != eversion_t::max()) ||
+      !(trimmed.empty());
   }
   void mark_dirty_to(eversion_t to) {
     if (to > dirty_to)
@@ -176,6 +181,10 @@ protected:
     if (from < dirty_from)
       dirty_from = from;
   }
+  void mark_writeout_from(eversion_t from) {
+    if (from < writeout_from)
+      writeout_from = from;
+  }
   void add_divergent_prior(eversion_t version, hobject_t obj) {
     divergent_priors.insert(make_pair(version, obj));
     dirty_divergent_priors = true;
@@ -205,11 +214,9 @@ protected:
 	 log_keys_debug->erase(i++));
   }
   void check() {
-    assert(log.log.size() == log_keys_debug.size());
-    if (cct &&
-        !(cct->_conf->osd_debug_pg_log_writeout)) {
+    if (!pg_log_debug)
       return;
-    }
+    assert(log.log.size() == log_keys_debug.size());
     for (list<pg_log_entry_t>::iterator i = log.log.begin();
 	 i != log.log.end();
 	 ++i) {
@@ -222,10 +229,13 @@ protected:
     dirty_from = eversion_t::max();
     dirty_divergent_priors = false;
     touched_log = true;
+    trimmed.clear();
+    writeout_from = eversion_t::max();
     check();
   }
 public:
   PGLog(CephContext *cct = 0) :
+    pg_log_debug(!(cct && !(cct->_conf->osd_debug_pg_log_writeout))),
     touched_log(false), dirty_from(eversion_t::max()),
     dirty_divergent_priors(false), cct(cct) {}
 
@@ -281,7 +291,7 @@ public:
   void unindex() { log.unindex(); }
 
   void add(pg_log_entry_t& e) {
-    mark_dirty_from(e.version);
+    mark_writeout_from(e.version);
     log.add(e);
   }
 
@@ -374,6 +384,8 @@ public:
     const hobject_t &log_oid, map<eversion_t, hobject_t> &divergent_priors,
     eversion_t dirty_to,
     eversion_t dirty_from,
+    eversion_t writeout_from,
+    const set<eversion_t> &trimmed,
     bool dirty_divergent_priors,
     bool touch_log,
     set<string> *log_keys_debug
@@ -381,8 +393,10 @@ public:
 
   bool read_log(ObjectStore *store, coll_t coll, hobject_t log_oid,
 		const pg_info_t &info, ostringstream &oss) {
-    return read_log(store, coll, log_oid, info, divergent_priors,
-		    log, missing, oss, &log_keys_debug);
+    return read_log(
+      store, coll, log_oid, info, divergent_priors,
+      log, missing, oss,
+      (pg_log_debug ? &log_keys_debug : 0));
   }
 
   /// return true if the log should be rewritten