OSD.cc: track osdmap refs using an LRU

Signed-off-by: Samuel Just <samuel.just@dreamhost.com>

diff --git a/src/common/config_opts.h b/src/common/config_opts.h
index f39a5f3307fb2c8bbb287f01cd5c8aeed7261649..ae177a1e1a19c9a51513bd6e732fe9e33fc7c691 100644 (file)
--- a/src/common/config_opts.h
+++ b/src/common/config_opts.h
@@ -269,7 +269,7 @@ OPTION(osd_pool_default_size, OPT_INT, 2)
 OPTION(osd_pool_default_pg_num, OPT_INT, 8)
 OPTION(osd_pool_default_pgp_num, OPT_INT, 8)
 OPTION(osd_map_dedup, OPT_BOOL, true)
-OPTION(osd_map_cache_max, OPT_INT, 250)
+OPTION(osd_map_cache_size, OPT_INT, 500)
 OPTION(osd_map_message_max, OPT_INT, 100)  // max maps per MOSDMap message
 OPTION(osd_op_threads, OPT_INT, 2)    // 0 == no threading
 OPTION(osd_disk_threads, OPT_INT, 1)

diff --git a/src/osd/OSD.cc b/src/osd/OSD.cc
index 81143d580cd4ebea88221760f2cd1c2d292eb9af..51ee8806818d9f9d4a4a44821ebc73a6d36c312a 100644 (file)
--- a/src/osd/OSD.cc
+++ b/src/osd/OSD.cc
@@ -556,6 +556,9 @@ OSD::OSD(int id, Messenger *internal_messenger, Messenger *external_messenger,
   map_lock("OSD::map_lock"),
   peer_map_epoch_lock("OSD::peer_map_epoch_lock"),
   map_cache_lock("OSD::map_cache_lock"),
+  map_cache(g_conf->osd_map_cache_size),
+  map_bl_cache(g_conf->osd_map_cache_size),
+  map_bl_inc_cache(g_conf->osd_map_cache_size),
   outstanding_pg_stats(false),
   up_thru_wanted(0), up_thru_pending(0),
   pg_stat_queue_lock("OSD::pg_stat_queue_lock"),
@@ -956,8 +959,6 @@ int OSD::shutdown()
 
   delete watch;
 
-  clear_map_cache();
-
   return r;
 }
 
@@ -3046,6 +3047,7 @@ void OSD::note_up_osd(int peer)
 void OSD::handle_osd_map(MOSDMap *m)
 {
   assert(osd_lock.is_locked());
+  list<OSDMapRef> pinned_maps;
   if (m->fsid != monc->get_fsid()) {
     dout(0) << "handle_osd_map fsid " << m->fsid << " != " << monc->get_fsid() << dendl;
     m->put();
@@ -3159,7 +3161,7 @@ void OSD::handle_osd_map(MOSDMap *m)
       bufferlist& bl = p->second;
       
       o->decode(bl);
-      add_map(o);
+      pinned_maps.push_back(add_map(o));
 
       hobject_t fulloid = get_osdmap_pobject_name(e);
       t.write(coll_t::META_COLL, fulloid, 0, bl.length(), bl);
@@ -3191,7 +3193,7 @@ void OSD::handle_osd_map(MOSDMap *m)
        assert(0 == "bad fsid");
       }
 
-      add_map(o);
+      pinned_maps.push_back(add_map(o));
 
       bufferlist fbl;
       o->encode(fbl);
@@ -3387,11 +3389,6 @@ void OSD::handle_osd_map(MOSDMap *m)
   ulock.Unlock();
   osd_lock.Lock();
 
-  // everything through current epoch now on disk; keep anything after
-  // that in cache
-  trim_map_bl_cache(osdmap->get_epoch()+1);
-  trim_map_cache(0);
-
   op_tp.unpause();
   recovery_tp.unpause();
   disk_tp.unpause();
@@ -3621,7 +3618,6 @@ void OSD::activate_map(ObjectStore::Transaction& t, list<Context*>& tfin)
   logger->set(l_osd_pg_stray, num_pg_stray);
 
   wake_all_pg_waiters();   // the pg mapping may have shifted
-  trim_map_cache(oldest_last_clean);
   maybe_update_heartbeat_peers();
 
   send_pg_temp();
@@ -3694,135 +3690,79 @@ void OSD::send_incremental_map(epoch_t since, const entity_inst_t& inst, bool la
   }
 }
 
-bool OSD::get_map_bl(epoch_t e, bufferlist& bl)
+bool OSD::_get_map_bl(epoch_t e, bufferlist& bl)
 {
-  {
-    Mutex::Locker l(map_cache_lock);
-    map<epoch_t,bufferlist>::iterator p = map_bl.find(e);
-    if (p != map_bl.end()) {
-      bl = p->second;
-      return true;
-    }
-  }
-  return store->read(coll_t::META_COLL, get_osdmap_pobject_name(e), 0, 0, bl) >= 0;
+  bool found = map_bl_cache.lookup(e, &bl);
+  if (found)
+    return true;
+  found = store->read(
+    coll_t::META_COLL, get_osdmap_pobject_name(e), 0, 0, bl) >= 0;
+  if (found)
+    _add_map_bl(e, bl);
+  return found;
 }
 
 bool OSD::get_inc_map_bl(epoch_t e, bufferlist& bl)
-{
-  {
-    Mutex::Locker l(map_cache_lock);
-    map<epoch_t,bufferlist>::iterator p = map_inc_bl.find(e);
-    if (p != map_inc_bl.end()) {
-      bl = p->second;
-      return true;
-    }
-  }
-  return store->read(coll_t::META_COLL, get_inc_osdmap_pobject_name(e), 0, 0, bl) >= 0;
-}
-
-OSDMapRef OSD::add_map(OSDMap *o)
 {
   Mutex::Locker l(map_cache_lock);
-  epoch_t e = o->get_epoch();
-  if (map_cache.count(e) == 0) {
-    dout(10) << "add_map " << e << " " << o << dendl;
-
-    if (g_conf->osd_map_dedup) {
-      // dedup against an existing map at nearby epoch
-      map<epoch_t,OSDMapRef>::iterator p = map_cache.lower_bound(e);
-      if (p == map_cache.end() && !map_cache.empty())
-       p--;
-      if (p != map_cache.end())
-       OSDMap::dedup(p->second.get(), o);
-    }
-
-    map_cache.insert(make_pair(e, OSDMapRef(o)));
-  } else {
-    dout(10) << "add_map " << e << " already have it" << dendl;
-  }
-  return map_cache[e];
+  bool found = map_bl_inc_cache.lookup(e, &bl);
+  if (found)
+    return true;
+  found = store->read(
+    coll_t::META_COLL, get_inc_osdmap_pobject_name(e), 0, 0, bl) >= 0;
+  if (found)
+    _add_map_inc_bl(e, bl);
+  return found;
 }
 
-void OSD::add_map_bl(epoch_t e, bufferlist& bl)
+void OSD::_add_map_bl(epoch_t e, bufferlist& bl)
 {
-  Mutex::Locker l(map_cache_lock);
   dout(10) << "add_map_bl " << e << " " << bl.length() << " bytes" << dendl;
-  map_bl[e] = bl;
+  map_bl_cache.add(e, bl);
 }
 
-void OSD::add_map_inc_bl(epoch_t e, bufferlist& bl)
+void OSD::_add_map_inc_bl(epoch_t e, bufferlist& bl)
 {
-  Mutex::Locker l(map_cache_lock);
   dout(10) << "add_map_inc_bl " << e << " " << bl.length() << " bytes" << dendl;
-  map_inc_bl[e] = bl;
+  map_bl_inc_cache.add(e, bl);
 }
 
-OSDMapRef OSD::get_map(epoch_t epoch)
+OSDMapRef OSD::_add_map(OSDMap *o)
 {
-  {
-    Mutex::Locker l(map_cache_lock);
-    map<epoch_t,OSDMapRef>::iterator p = map_cache.find(epoch);
-    if (p != map_cache.end()) {
-      dout(30) << "get_map " << epoch << " - cached " << p->second << dendl;
-      return p->second;
+  epoch_t e = o->get_epoch();
+
+  if (g_conf->osd_map_dedup) {
+    // Dedup against an existing map at a nearby epoch
+    OSDMapRef for_dedup = map_cache.lower_bound(e);
+    if (for_dedup) {
+      OSDMap::dedup(for_dedup.get(), o);
     }
   }
+  OSDMapRef l = map_cache.add(e, o);
+  return l;
+}
+
+OSDMapRef OSD::get_map(epoch_t epoch)
+{
+  Mutex::Locker l(map_cache_lock);
+  OSDMapRef retval = map_cache.lookup(epoch);
+  if (retval) {
+    dout(30) << "get_map " << epoch << " -cached" << dendl;
+    return retval;
+  }
 
   OSDMap *map = new OSDMap;
   if (epoch > 0) {
     dout(20) << "get_map " << epoch << " - loading and decoding " << map << dendl;
     bufferlist bl;
-    get_map_bl(epoch, bl);
+    assert(_get_map_bl(epoch, bl));
     map->decode(bl);
   } else {
     dout(20) << "get_map " << epoch << " - return initial " << map << dendl;
   }
-  return add_map(map);
-}
-
-void OSD::trim_map_bl_cache(epoch_t oldest)
-{
-  Mutex::Locker l(map_cache_lock);
-  dout(10) << "trim_map_bl_cache up to " << oldest << dendl;
-  while (!map_inc_bl.empty() && map_inc_bl.begin()->first < oldest)
-    map_inc_bl.erase(map_inc_bl.begin());
-  while (!map_bl.empty() && map_bl.begin()->first < oldest)
-    map_bl.erase(map_bl.begin());
-}
-
-void OSD::trim_map_cache(epoch_t oldest)
-{
-  Mutex::Locker l(map_cache_lock);
-  dout(10) << "trim_map_cache prior to " << oldest << dendl;
-  while (!map_cache.empty() &&
-        (map_cache.begin()->first < oldest ||
-         (int)map_cache.size() > g_conf->osd_map_cache_max)) {
-    epoch_t e = map_cache.begin()->first;
-    OSDMapRef o = map_cache.begin()->second;
-    dout(10) << "trim_map_cache " << e << " " << o << dendl;
-    map_cache.erase(map_cache.begin());
-  }
+  return _add_map(map);
 }
 
-void OSD::clear_map_cache()
-{
-  while (!map_cache.empty()) {
-    map_cache.erase(map_cache.begin());
-  }
-}
-
-bool OSD::get_inc_map(epoch_t e, OSDMap::Incremental &inc)
-{
-  bufferlist bl;
-  if (!get_inc_map_bl(e, bl)) 
-    return false;
-  bufferlist::iterator p = bl.begin();
-  inc.decode(p);
-  return true;
-}
-
-
-
 bool OSD::require_mon_peer(Message *m)
 {
   if (!m->get_connection()->peer_is_mon()) {

diff --git a/src/osd/OSD.h b/src/osd/OSD.h
index 562aad5f77e4cb913c80ded77ae8c064646227fe..155f4048a250eba331d15f10d367199a6b411265 100644 (file)
--- a/src/osd/OSD.h
+++ b/src/osd/OSD.h
@@ -47,6 +47,8 @@ using namespace std;
 using namespace __gnu_cxx;
 
 #include "OpRequest.h"
+#include "common/shared_cache.hpp"
+#include "common/simple_cache.hpp"
 
 #define CEPH_OSD_PROTOCOL    10 /* cluster internal */
 
@@ -409,22 +411,36 @@ private:
   void activate_map(ObjectStore::Transaction& t, list<Context*>& tfin);
 
   // osd map cache (past osd maps)
-  map<epoch_t,OSDMapRef > map_cache;
-  map<epoch_t,bufferlist> map_inc_bl;
-  map<epoch_t,bufferlist> map_bl;
   Mutex map_cache_lock;
+  SharedLRU<epoch_t, OSDMap> map_cache;
+  SimpleLRU<epoch_t, bufferlist> map_bl_cache;
+  SimpleLRU<epoch_t, bufferlist> map_bl_inc_cache;
+
 
   OSDMapRef get_map(epoch_t e);
-  OSDMapRef add_map(OSDMap *o);
-  void add_map_bl(epoch_t e, bufferlist& bl);
-  void add_map_inc_bl(epoch_t e, bufferlist& bl);
-  void trim_map_cache(epoch_t oldest);
-  void trim_map_bl_cache(epoch_t oldest);
-  void clear_map_cache();
-
-  bool get_map_bl(epoch_t e, bufferlist& bl);
+  OSDMapRef add_map(OSDMap *o) {
+    Mutex::Locker l(map_cache_lock);
+    return _add_map(o);
+  }
+  OSDMapRef _add_map(OSDMap *o);
+
+  void add_map_bl(epoch_t e, bufferlist& bl) {
+    Mutex::Locker l(map_cache_lock);
+    return _add_map_bl(e, bl);
+  }
+  void _add_map_bl(epoch_t e, bufferlist& bl);
+  bool get_map_bl(epoch_t e, bufferlist& bl) {
+    Mutex::Locker l(map_cache_lock);
+    return _get_map_bl(e, bl);
+  }
+  bool _get_map_bl(epoch_t e, bufferlist& bl);
+
+  void add_map_inc_bl(epoch_t e, bufferlist& bl) {
+    Mutex::Locker l(map_cache_lock);
+    return _add_map_inc_bl(e, bl);
+  }
+  void _add_map_inc_bl(epoch_t e, bufferlist& bl);
   bool get_inc_map_bl(epoch_t e, bufferlist& bl);
-  bool get_inc_map(epoch_t e, OSDMap::Incremental &inc);
   
   MOSDMap *build_incremental_map_msg(epoch_t from, epoch_t to);
   void send_incremental_map(epoch_t since, const entity_inst_t& inst, bool lazy=false);