kv/KeyValueDB, os/bluestore: Made fsck faster by not caching results

Added ability to pass extra options when creating iterators to KV.
Used that ability to perform fsck operations without caching results.

Signed-off-by: Adam Kupczyk <akupczyk@redhat.com>

diff --git a/src/kv/KeyValueDB.h b/src/kv/KeyValueDB.h
index ad93631121accc64212dd54c2cf1e3c39f1bd413..6e4616e3a6c46eac6b1b97613610441433940d4c 100644 (file)
--- a/src/kv/KeyValueDB.h
+++ b/src/kv/KeyValueDB.h
@@ -312,12 +312,13 @@ private:
     }
   };
 public:
-
-  virtual WholeSpaceIterator get_wholespace_iterator() = 0;
-  virtual Iterator get_iterator(const std::string &prefix) {
+  typedef uint32_t IteratorOpts;
+  static const uint32_t ITERATOR_NOCACHE = 1;
+  virtual WholeSpaceIterator get_wholespace_iterator(IteratorOpts opts = 0) = 0;
+  virtual Iterator get_iterator(const std::string &prefix, IteratorOpts opts = 0) {
     return std::make_shared<PrefixIteratorImpl>(
       prefix,
-      get_wholespace_iterator());
+      get_wholespace_iterator(opts));
   }
 
   virtual uint64_t get_estimated_size(std::map<std::string,uint64_t> &extra) = 0;

diff --git a/src/kv/LevelDBStore.h b/src/kv/LevelDBStore.h
index 7876904e82db62cf0b3d0f0ef612053628274750..085193ee03321a5859928cb33371857be272690b 100644 (file)
--- a/src/kv/LevelDBStore.h
+++ b/src/kv/LevelDBStore.h
@@ -402,7 +402,7 @@ err:
   }
 
 
-  WholeSpaceIterator get_wholespace_iterator() override {
+  WholeSpaceIterator get_wholespace_iterator(IteratorOpts opts = 0) override {
     return std::make_shared<LevelDBWholeSpaceIteratorImpl>(
        db->NewIterator(leveldb::ReadOptions()));
   }

diff --git a/src/kv/MemDB.h b/src/kv/MemDB.h
index 71558ad0c8b4ff3d40e68ebbfa06d59cf69eb476..32d81db225ed22b887377f84aadc699b7bcd6405 100644 (file)
--- a/src/kv/MemDB.h
+++ b/src/kv/MemDB.h
@@ -212,7 +212,7 @@ public:
     return 0;
   }
 
-  WholeSpaceIterator get_wholespace_iterator() override {
+  WholeSpaceIterator get_wholespace_iterator(IteratorOpts opts = 0) override {
     return std::shared_ptr<KeyValueDB::WholeSpaceIteratorImpl>(
       new MDBWholeSpaceIteratorImpl(&m_map, &m_lock, &iterator_seq_no, m_using_btree));
   }

diff --git a/src/kv/RocksDBStore.cc b/src/kv/RocksDBStore.cc
index 07d7a7f71184fedd00f05a86ef6183c3d7034e0d..6360536c2c4a74d76c557f877ee55211d42300d8 100644 (file)
--- a/src/kv/RocksDBStore.cc
+++ b/src/kv/RocksDBStore.cc
@@ -2625,7 +2625,7 @@ public:
   }
 };
 
-KeyValueDB::Iterator RocksDBStore::get_iterator(const std::string& prefix)
+KeyValueDB::Iterator RocksDBStore::get_iterator(const std::string& prefix, IteratorOpts opts)
 {
   auto cf_it = cf_handles.find(prefix);
   if (cf_it != cf_handles.end()) {
@@ -2640,7 +2640,7 @@ KeyValueDB::Iterator RocksDBStore::get_iterator(const std::string& prefix)
         cf_it->second.handles);
     }
   } else {
-    return KeyValueDB::get_iterator(prefix);
+    return KeyValueDB::get_iterator(prefix, opts);
   }
 }
 
@@ -2649,11 +2649,14 @@ rocksdb::Iterator* RocksDBStore::new_shard_iterator(rocksdb::ColumnFamilyHandle*
   return db->NewIterator(rocksdb::ReadOptions(), cf);
 }
 
-RocksDBStore::WholeSpaceIterator RocksDBStore::get_wholespace_iterator()
+RocksDBStore::WholeSpaceIterator RocksDBStore::get_wholespace_iterator(IteratorOpts opts)
 {
   if (cf_handles.size() == 0) {
+    rocksdb::ReadOptions opt = rocksdb::ReadOptions();
+    if (opts & ITERATOR_NOCACHE)
+      opt.fill_cache=false;
     return std::make_shared<RocksDBWholeSpaceIteratorImpl>(
-      db->NewIterator(rocksdb::ReadOptions(), default_cf));
+      db->NewIterator(opt, default_cf));
   } else {
     return std::make_shared<WholeMergeIteratorImpl>(this);
   }

diff --git a/src/kv/RocksDBStore.h b/src/kv/RocksDBStore.h
index fe749145b32df039112ac0d8dd404904bdeebf86..43474acf88ab098596fd04303203c9359d537b9e 100644 (file)
--- a/src/kv/RocksDBStore.h
+++ b/src/kv/RocksDBStore.h
@@ -450,7 +450,7 @@ public:
     size_t value_size() override;
   };
 
-  Iterator get_iterator(const std::string& prefix) override;
+  Iterator get_iterator(const std::string& prefix, IteratorOpts opts = 0) override;
 private:
   /// this iterator spans single cf
   rocksdb::Iterator* new_shard_iterator(rocksdb::ColumnFamilyHandle* cf);
@@ -565,7 +565,7 @@ err:
         bbt_opts.block_cache);
   }
 
-  WholeSpaceIterator get_wholespace_iterator() override;
+  WholeSpaceIterator get_wholespace_iterator(IteratorOpts opts = 0) override;
 private:
   WholeSpaceIterator get_default_cf_iterator();
 };

diff --git a/src/os/bluestore/BlueStore.cc b/src/os/bluestore/BlueStore.cc
index fe617344530aafb635b9c302561965294f34295e..2ca87eda32743bf21e2536520341f5daa0639e10 100644 (file)
--- a/src/os/bluestore/BlueStore.cc
+++ b/src/os/bluestore/BlueStore.cc
@@ -6138,7 +6138,7 @@ void BlueStore::_fsck_collections(int64_t* errors)
 {
   if (collections_had_errors) {
     dout(10) << __func__ << dendl;
-    KeyValueDB::Iterator it = db->get_iterator(PREFIX_COLL);
+    KeyValueDB::Iterator it = db->get_iterator(PREFIX_COLL, KeyValueDB::ITERATOR_NOCACHE);
     for (it->upper_bound(string());
       it->valid();
       it->next()) {
@@ -6187,7 +6187,7 @@ void BlueStore::_open_statfs()
   } else {
     per_pool_stat_collection = true;
     dout(10) << __func__ << " per-pool statfs is enabled" << dendl;
-    KeyValueDB::Iterator it = db->get_iterator(PREFIX_STAT);
+    KeyValueDB::Iterator it = db->get_iterator(PREFIX_STAT, KeyValueDB::ITERATOR_NOCACHE);
     for (it->upper_bound(string());
         it->valid();
         it->next()) {
@@ -7236,7 +7236,7 @@ void BlueStore::_fsck_check_pool_statfs(
   int64_t& warnings,
   BlueStoreRepairer* repairer)
 {
-  auto it = db->get_iterator(PREFIX_STAT);
+  auto it = db->get_iterator(PREFIX_STAT, KeyValueDB::ITERATOR_NOCACHE);
   if (it) {
     for (it->lower_bound(string()); it->valid(); it->next()) {
       string key = it->key();
@@ -7864,7 +7864,7 @@ void BlueStore::_fsck_check_objects(FSCKDepth depth,
 
   size_t processed_myself = 0;
 
-  auto it = db->get_iterator(PREFIX_OBJ);
+  auto it = db->get_iterator(PREFIX_OBJ, KeyValueDB::ITERATOR_NOCACHE);
   mempool::bluestore_fsck::list<string> expecting_shards;
   if (it) {
     const size_t thread_count = cct->_conf->bluestore_fsck_quick_fix_threads;
@@ -8369,7 +8369,7 @@ int BlueStore::_fsck_on_open(BlueStore::FSCKDepth depth, bool repair)
   }
 
   dout(1) << __func__ << " checking shared_blobs" << dendl;
-  it = db->get_iterator(PREFIX_SHARED_BLOB);
+  it = db->get_iterator(PREFIX_SHARED_BLOB, KeyValueDB::ITERATOR_NOCACHE);
   if (it) {
     // FIXME minor: perhaps simplify for shallow mode?
     // fill global if not overriden below
@@ -8453,7 +8453,7 @@ int BlueStore::_fsck_on_open(BlueStore::FSCKDepth depth, bool repair)
     auto& space_tracker = repairer.get_space_usage_tracker();
     auto& misref_extents = repairer.get_misreferences();
     interval_set<uint64_t> to_release;
-    it = db->get_iterator(PREFIX_OBJ);
+    it = db->get_iterator(PREFIX_OBJ, KeyValueDB::ITERATOR_NOCACHE);
     if (it) {
       // fill global if not overriden below
       auto expected_statfs = &expected_store_statfs;
@@ -8690,7 +8690,7 @@ int BlueStore::_fsck_on_open(BlueStore::FSCKDepth depth, bool repair)
 
   if (depth != FSCK_SHALLOW) {
     dout(1) << __func__ << " checking for stray omap data " << dendl;
-    it = db->get_iterator(PREFIX_OMAP);
+    it = db->get_iterator(PREFIX_OMAP, KeyValueDB::ITERATOR_NOCACHE);
     if (it) {
       uint64_t last_omap_head = 0;
       for (it->lower_bound(string()); it->valid(); it->next()) {
@@ -8706,7 +8706,7 @@ int BlueStore::_fsck_on_open(BlueStore::FSCKDepth depth, bool repair)
         }
       }
     }
-    it = db->get_iterator(PREFIX_PGMETA_OMAP);
+    it = db->get_iterator(PREFIX_PGMETA_OMAP, KeyValueDB::ITERATOR_NOCACHE);
     if (it) {
       uint64_t last_omap_head = 0;
       for (it->lower_bound(string()); it->valid(); it->next()) {
@@ -8722,7 +8722,7 @@ int BlueStore::_fsck_on_open(BlueStore::FSCKDepth depth, bool repair)
         }
       }
     }
-    it = db->get_iterator(PREFIX_PERPOOL_OMAP);
+    it = db->get_iterator(PREFIX_PERPOOL_OMAP, KeyValueDB::ITERATOR_NOCACHE);
     if (it) {
       uint64_t last_omap_head = 0;
       for (it->lower_bound(string()); it->valid(); it->next()) {
@@ -8743,7 +8743,7 @@ int BlueStore::_fsck_on_open(BlueStore::FSCKDepth depth, bool repair)
       }
     }
     dout(1) << __func__ << " checking deferred events" << dendl;
-    it = db->get_iterator(PREFIX_DEFERRED);
+    it = db->get_iterator(PREFIX_DEFERRED, KeyValueDB::ITERATOR_NOCACHE);
     if (it) {
       for (it->lower_bound(string()); it->valid(); it->next()) {
         bufferlist bl = it->value();

diff --git a/src/test/ObjectMap/KeyValueDBMemory.cc b/src/test/ObjectMap/KeyValueDBMemory.cc
index a4b2a4e1ca3581ed4958b544f57b96073ed616b9..234e963397e31ea92c0748685d45f0995b00fecb 100644 (file)
--- a/src/test/ObjectMap/KeyValueDBMemory.cc
+++ b/src/test/ObjectMap/KeyValueDBMemory.cc
@@ -234,7 +234,7 @@ int KeyValueDBMemory::rm_range_keys(const string &prefix, const string &start, c
   return 0;
 }
 
-KeyValueDB::WholeSpaceIterator KeyValueDBMemory::get_wholespace_iterator() {
+KeyValueDB::WholeSpaceIterator KeyValueDBMemory::get_wholespace_iterator(IteratorOpts opts) {
   return std::shared_ptr<KeyValueDB::WholeSpaceIteratorImpl>(
     new WholeSpaceMemIterator(this)
   );

diff --git a/src/test/ObjectMap/KeyValueDBMemory.h b/src/test/ObjectMap/KeyValueDBMemory.h
index 75e5d64232675b3d8da5377d6ee76934b32b5cfe..5e7d3f0e1675f33f7c171ba1ce8d5df5672e3577 100644 (file)
--- a/src/test/ObjectMap/KeyValueDBMemory.h
+++ b/src/test/ObjectMap/KeyValueDBMemory.h
@@ -186,5 +186,5 @@ private:
   friend class WholeSpaceMemIterator;
 
 public:
-  WholeSpaceIterator get_wholespace_iterator() override;
+  WholeSpaceIterator get_wholespace_iterator(IteratorOpts opts = 0) override;
 };