os/bluestore: Keep separate onode cache pinned list.

Signed-off-by: Mark Nelson <mnelson@redhat.com>

diff --git a/src/os/bluestore/BlueStore.cc b/src/os/bluestore/BlueStore.cc
index ee9556a483746364d209a30be7d07fd2251b34c1..ccbfa25d2b360cee9020acfaba9de3a714b3f0a2 100644 (file)
--- a/src/os/bluestore/BlueStore.cc
+++ b/src/os/bluestore/BlueStore.cc
@@ -831,58 +831,81 @@ struct LruOnodeCacheShard : public BlueStore::OnodeCacheShard {
       BlueStore::Onode,
       boost::intrusive::list_member_hook<>,
       &BlueStore::Onode::lru_item> > list_t;
+  typedef boost::intrusive::list<
+    BlueStore::Onode,
+    boost::intrusive::member_hook<
+      BlueStore::Onode,
+      boost::intrusive::list_member_hook<>,
+      &BlueStore::Onode::pin_item> > pin_list_t;
+
   list_t lru;
+  pin_list_t pin_list;
 
   explicit LruOnodeCacheShard(CephContext *cct) : BlueStore::OnodeCacheShard(cct) {}
 
   void _add(BlueStore::OnodeRef& o, int level) override
   {
     (level > 0) ? lru.push_front(*o) : lru.push_back(*o);
+    o->s = this;
     num = lru.size();
   }
   void _rm(BlueStore::OnodeRef& o) override
   {
-    lru.erase(lru.iterator_to(*o));
+    o->s = nullptr;
+    if (o->pinned) {
+      o->pinned = false;
+      pin_list.erase(pin_list.iterator_to(*o));
+    } else {
+      lru.erase(lru.iterator_to(*o));
+    }
     num = lru.size();
+    num_pinned = pin_list.size();
   }
   void _touch(BlueStore::OnodeRef& o) override
   {
+    if (o->pinned) {
+      return;
+    }
     lru.erase(lru.iterator_to(*o));
     lru.push_front(*o);
     num = lru.size();
   }
-  void _trim_to(uint64_t max) override
+  void _pin(BlueStore::Onode& o) override
+  {
+    if (o.pinned == true) {
+      return;
+    }
+    lru.erase(lru.iterator_to(o));
+    pin_list.push_front(o);
+    o.pinned = true;
+    num = lru.size();
+    num_pinned = pin_list.size();
+    dout(30) << __func__ << " " << o.oid << " pinned" << dendl;
+
+  } 
+  void _unpin(BlueStore::Onode& o) override
+  {
+    if (o.pinned == false) {
+      return;
+    }
+    pin_list.erase(pin_list.iterator_to(o));
+    lru.push_front(o);
+    o.pinned = false;
+    num = lru.size();
+    num_pinned = pin_list.size();
+    dout(30) << __func__ << " " << o.oid << " unpinned" << dendl;
+  }
+  void _trim_to(uint64_t new_size) override
   {
-    if (max >= lru.size()) {
+    if (new_size >= lru.size()) {
       return; // don't even try
     } 
-    uint64_t n = lru.size() - max;
-
+    uint64_t n = lru.size() - new_size;
     auto p = lru.end();
     ceph_assert(p != lru.begin());
     --p;
-    int skipped = 0;
-    int max_skipped = g_conf()->bluestore_cache_trim_max_skip_pinned;
     while (n > 0) {
       BlueStore::Onode *o = &*p;
-      int refs = o->nref.load();
-      if (refs > 1) {
-        dout(20) << __func__ << "  " << o->oid << " has " << refs
-                 << " refs, skipping" << dendl;
-        if (++skipped >= max_skipped) {
-          dout(20) << __func__ << " maximum skip pinned reached; stopping with "
-                   << n << " left to trim" << dendl;
-          break;
-        }
-
-        if (p == lru.begin()) {
-          break;
-        } else {
-          p--;
-          n--;
-          continue;
-        }
-      }
       dout(30) << __func__ << "  rm " << o->oid << dendl;
       if (p != lru.begin()) {
         lru.erase(p--);
@@ -890,6 +913,7 @@ struct LruOnodeCacheShard : public BlueStore::OnodeCacheShard {
         lru.erase(p);
         ceph_assert(n == 1);
       }
+      o->s = nullptr;
       o->get();  // paranoia
       o->c->onode_map.remove(o->oid);
       o->put();
@@ -897,9 +921,10 @@ struct LruOnodeCacheShard : public BlueStore::OnodeCacheShard {
     }
     num = lru.size();
   }
-  void add_stats(uint64_t *onodes) override
+  void add_stats(uint64_t *onodes, uint64_t *pinned_onodes) override
   {
-    *onodes += num;
+    *onodes += num + num_pinned;
+    *pinned_onodes += num_pinned;
   }
 };
 
@@ -4565,6 +4590,8 @@ void BlueStore::_init_logger()
 
   b.add_u64(l_bluestore_onodes, "bluestore_onodes",
            "Number of onodes in cache");
+  b.add_u64(l_bluestore_pinned_onodes, "bluestore_pinned_onodes",
+            "Number of pinned onodes in cache");
   b.add_u64_counter(l_bluestore_onode_hits, "bluestore_onode_hits",
                    "Sum for onode-lookups hit in the cache");
   b.add_u64_counter(l_bluestore_onode_misses, "bluestore_onode_misses",
@@ -9166,18 +9193,20 @@ void BlueStore::_reap_collections()
 void BlueStore::_update_cache_logger()
 {
   uint64_t num_onodes = 0;
+  uint64_t num_pinned_onodes = 0;
   uint64_t num_extents = 0;
   uint64_t num_blobs = 0;
   uint64_t num_buffers = 0;
   uint64_t num_buffer_bytes = 0;
   for (auto c : onode_cache_shards) {
-    c->add_stats(&num_onodes);
+    c->add_stats(&num_onodes, &num_pinned_onodes);
   }
   for (auto c : buffer_cache_shards) {
     c->add_stats(&num_extents, &num_blobs,
                  &num_buffers, &num_buffer_bytes);
   }
   logger->set(l_bluestore_onodes, num_onodes);
+  logger->set(l_bluestore_pinned_onodes, num_pinned_onodes);
   logger->set(l_bluestore_extents, num_extents);
   logger->set(l_bluestore_blobs, num_blobs);
   logger->set(l_bluestore_buffers, num_buffers);

diff --git a/src/os/bluestore/BlueStore.h b/src/os/bluestore/BlueStore.h
index 8976c03b0278b627fd83a5dbecdce657346d6781..61c86eb4e7fcca4e8e7125e10a86893f0cf210b1 100644 (file)
--- a/src/os/bluestore/BlueStore.h
+++ b/src/os/bluestore/BlueStore.h
@@ -99,6 +99,7 @@ enum {
   l_bluestore_compressed_allocated,
   l_bluestore_compressed_original,
   l_bluestore_onodes,
+  l_bluestore_pinned_onodes,
   l_bluestore_onode_hits,
   l_bluestore_onode_misses,
   l_bluestore_onode_shard_hits,
@@ -1045,20 +1046,22 @@ public:
   };
 
   struct OnodeSpace;
-
+  struct OnodeCacheShard;
   /// an in-memory object
   struct Onode {
     MEMPOOL_CLASS_HELPERS();
+    // Not persisted and updated on cache insertion/removal
+    OnodeCacheShard *s;
+    bool pinned = false; // Only to be used by the onode cache shard
 
     std::atomic_int nref;  ///< reference count
     Collection *c;
-
     ghobject_t oid;
 
     /// key under PREFIX_OBJ where we are stored
     mempool::bluestore_cache_other::string key;
 
-    boost::intrusive::list_member_hook<> lru_item;
+    boost::intrusive::list_member_hook<> lru_item, pin_item;
 
     bluestore_onode_t onode;  ///< metadata stored as value in kv store
     bool exists;              ///< true if object logically exists
@@ -1075,7 +1078,8 @@ public:
 
     Onode(Collection *c, const ghobject_t& o,
          const mempool::bluestore_cache_other::string& k)
-      : nref(0),
+      : s(nullptr),
+        nref(0),
        c(c),
        oid(o),
        key(k),
@@ -1084,7 +1088,8 @@ public:
     }
     Onode(Collection* c, const ghobject_t& o,
       const string& k)
-      : nref(0),
+      : s(nullptr),
+      nref(0),
       c(c),
       oid(o),
       key(k),
@@ -1093,7 +1098,8 @@ public:
     }
     Onode(Collection* c, const ghobject_t& o,
       const char* k)
-      : nref(0),
+      : s(nullptr),
+      nref(0),
       c(c),
       oid(o),
       key(k),
@@ -1111,11 +1117,18 @@ public:
 
     void flush();
     void get() {
-      ++nref;
+      if (++nref == 2 && s != nullptr) {
+        s->pin(*this);
+      }
     }
     void put() {
-      if (--nref == 0)
+      int n = --nref;
+      if (n == 1 && s != nullptr) {
+        s->unpin(*this);
+      }
+      if (n == 0) {
        delete this;
+      }
     }
 
     const string& get_omap_prefix();
@@ -1150,7 +1163,7 @@ public:
       return num;
     }
 
-    virtual void _trim_to(uint64_t max) = 0;
+    virtual void _trim_to(uint64_t new_size) = 0;
     void _trim() {
       if (cct->_conf->objectstore_blackhole) {
        // do not trim if we are throwing away IOs a layer down
@@ -1158,6 +1171,7 @@ public:
       }
       _trim_to(max);
     }
+
     void trim() {
       std::lock_guard l(lock);
       _trim();    
@@ -1178,6 +1192,8 @@ public:
 
   /// A Generic onode Cache Shard
   struct OnodeCacheShard : public CacheShard {
+    std::atomic<uint64_t> num_pinned = {0};
+
     std::array<std::pair<ghobject_t, mono_clock::time_point>, 64> dumped_onodes;
   public:
     OnodeCacheShard(CephContext* cct) : CacheShard(cct) {}
@@ -1186,8 +1202,20 @@ public:
     virtual void _add(OnodeRef& o, int level) = 0;
     virtual void _rm(OnodeRef& o) = 0;
     virtual void _touch(OnodeRef& o) = 0;
-    virtual void add_stats(uint64_t *onodes) = 0;
+    virtual void _pin(Onode& o) = 0;
+    virtual void _unpin(Onode& o) = 0;
+
+    void pin(Onode& o) {
+      std::lock_guard l(lock);
+      _pin(o);
+    }
+
+    void unpin(Onode& o) {
+      std::lock_guard l(lock);
+      _unpin(o);
+    }
 
+    virtual void add_stats(uint64_t *onodes, uint64_t *pinned_onodes) = 0;
     bool empty() {
       return _get_num() == 0;
     }