Enable prefix bloom filter by default (vesoft-inc#2860)

* enable prefix bloom filter by default * GetProperty * make rate limiter dynamic configurable Co-authored-by: Yee <[email protected]>
CPWstatic · Sep 25, 2021 · 5fe655a · 5fe655a
1 parent 21c1b5c
commit 5fe655a
Show file tree

Hide file tree

Showing 23 changed files with 923 additions and 229 deletions.
diff --git a/conf/nebula-storaged.conf.default b/conf/nebula-storaged.conf.default
@@ -90,8 +90,10 @@
 #   * kAll, Collect all stats
 --rocksdb_stats_level=kExceptHistogramOrTimers
 
-# Whether or not to enable rocksdb's prefix bloom filter, disabled by default.
---enable_rocksdb_prefix_filtering=false
+# Whether or not to enable rocksdb's prefix bloom filter, enabled by default.
+--enable_rocksdb_prefix_filtering=true
+# Whether or not to enable rocksdb's whole key bloom filter, disabled by default.
+--enable_rocksdb_whole_key_filtering=false
 
 ############## rocksdb Options ##############
 # rocksdb DBOptions in json, each name and value of option is a string, given as "option_name":"option_value" separated by comma

diff --git a/conf/nebula-storaged.conf.production b/conf/nebula-storaged.conf.production
@@ -96,8 +96,10 @@
 #   * kAll, Collect all stats
 --rocksdb_stats_level=kExceptHistogramOrTimers
 
-# Whether or not to enable rocksdb's prefix bloom filter, disabled by default.
---enable_rocksdb_prefix_filtering=false
+# Whether or not to enable rocksdb's prefix bloom filter, enabled by default.
+--enable_rocksdb_prefix_filtering=true
+# Whether or not to enable rocksdb's whole key bloom filter, disabled by default.
+--enable_rocksdb_whole_key_filtering=false
 
 ############### misc ####################
 --snapshot_part_rate_limit=8388608

diff --git a/src/kvstore/KVEngine.h b/src/kvstore/KVEngine.h
@@ -111,6 +111,10 @@ class KVEngine {
   virtual nebula::cpp2::ErrorCode setDBOption(const std::string& configKey,
                                               const std::string& configValue) = 0;
 
+  // Get DB Property
+  virtual ErrorOr<nebula::cpp2::ErrorCode, std::string> getProperty(
+      const std::string& property) = 0;
+
   virtual nebula::cpp2::ErrorCode compact() = 0;
 
   virtual nebula::cpp2::ErrorCode flush() = 0;

diff --git a/src/kvstore/KVStore.h b/src/kvstore/KVStore.h
@@ -216,6 +216,9 @@ class KVStore {
 
   virtual std::vector<std::string> getDataRoot() const = 0;
 
+  virtual ErrorOr<nebula::cpp2::ErrorCode, std::string> getProperty(
+      GraphSpaceID spaceId, const std::string& property) = 0;
+
  protected:
   KVStore() = default;
 };

diff --git a/src/kvstore/NebulaSnapshotManager.cpp b/src/kvstore/NebulaSnapshotManager.cpp
@@ -11,7 +11,7 @@
 #include "kvstore/RateLimiter.h"
 
 DEFINE_uint32(snapshot_part_rate_limit,
-              1024 * 1024 * 2,
+              1024 * 1024 * 8,
               "max bytes of pulling snapshot for each partition in one second");
 DEFINE_uint32(snapshot_batch_size, 1024 * 512, "batch size for snapshot, in bytes");
 
@@ -22,16 +22,15 @@ const int32_t kReserveNum = 1024 * 4;
 
 NebulaSnapshotManager::NebulaSnapshotManager(NebulaStore* kv) : store_(kv) {
   // Snapshot rate is limited to FLAGS_snapshot_worker_threads * FLAGS_snapshot_part_rate_limit.
-  // So by default, the total send rate is limited to 4 * 2Mb = 8Mb.
+  // So by default, the total send rate is limited to 4 * 8Mb = 32Mb.
   LOG(INFO) << "Send snapshot is rate limited to " << FLAGS_snapshot_part_rate_limit
-            << " for each part";
+            << " for each part by default";
 }
 
 void NebulaSnapshotManager::accessAllRowsInSnapshot(GraphSpaceID spaceId,
                                                     PartitionID partId,
                                                     raftex::SnapshotCallback cb) {
-  auto rateLimiter = std::make_unique<kvstore::RateLimiter>(FLAGS_snapshot_part_rate_limit,
-                                                            FLAGS_snapshot_part_rate_limit);
+  auto rateLimiter = std::make_unique<kvstore::RateLimiter>();
   CHECK_NOTNULL(store_);
   auto tables = NebulaKeyUtils::snapshotPrefix(partId);
   std::vector<std::string> data;
@@ -74,7 +73,9 @@ bool NebulaSnapshotManager::accessTable(GraphSpaceID spaceId,
   size_t batchSize = 0;
   while (iter && iter->valid()) {
     if (batchSize >= FLAGS_snapshot_batch_size) {
-      rateLimiter->consume(batchSize);
+      rateLimiter->consume(static_cast<double>(batchSize),                        // toConsume
+                           static_cast<double>(FLAGS_snapshot_part_rate_limit),   // rate
+                           static_cast<double>(FLAGS_snapshot_part_rate_limit));  // burstSize
       if (cb(data, totalCount, totalSize, raftex::SnapshotStatus::IN_PROGRESS)) {
         data.clear();
         batchSize = 0;

diff --git a/src/kvstore/NebulaStore.cpp b/src/kvstore/NebulaStore.cpp
@@ -1168,5 +1168,26 @@ nebula::cpp2::ErrorCode NebulaStore::multiPutWithoutReplicator(GraphSpaceID spac
   return nebula::cpp2::ErrorCode::SUCCEEDED;
 }
 
+ErrorOr<nebula::cpp2::ErrorCode, std::string> NebulaStore::getProperty(
+    GraphSpaceID spaceId, const std::string& property) {
+  auto spaceRet = space(spaceId);
+  if (!ok(spaceRet)) {
+    LOG(ERROR) << "Get Space " << spaceId << " Failed";
+    return error(spaceRet);
+  }
+  auto space = nebula::value(spaceRet);
+
+  folly::dynamic obj = folly::dynamic::object;
+  for (size_t i = 0; i < space->engines_.size(); i++) {
+    auto val = space->engines_[i]->getProperty(property);
+    if (!ok(val)) {
+      return error(val);
+    }
+    auto eng = folly::stringPrintf("Engine %zu", i);
+    obj[eng] = std::move(value(val));
+  }
+  return folly::toJson(obj);
+}
+
 }  // namespace kvstore
 }  // namespace nebula
diff --git a/src/kvstore/NebulaStore.h b/src/kvstore/NebulaStore.h
@@ -275,6 +275,9 @@ class NebulaStore : public KVStore, public Handler {
   nebula::cpp2::ErrorCode multiPutWithoutReplicator(GraphSpaceID spaceId,
                                                     std::vector<KV> keyValues) override;
 
+  ErrorOr<nebula::cpp2::ErrorCode, std::string> getProperty(GraphSpaceID spaceId,
+                                                            const std::string& property) override;
+
  private:
   void loadPartFromDataPath();
 

diff --git a/src/kvstore/RateLimiter.h b/src/kvstore/RateLimiter.h
@@ -21,32 +21,29 @@ namespace kvstore {
 // For now, there are two major cases: snapshot (both for balance or catch up) and rebuild index.
 class RateLimiter {
  public:
-  RateLimiter(int32_t rate, int32_t burstSize)
-      : rate_(static_cast<double>(rate)), burstSize_(static_cast<double>(burstSize)) {
+  RateLimiter() {
     // token will be available after 1 second, to prevent speed spike at the beginning
     auto now = time::WallClock::fastNowInSec();
     int64_t waitInSec = FLAGS_skip_wait_in_rate_limiter ? 0 : 1;
-    bucket_.reset(new folly::TokenBucket(rate_, burstSize_, static_cast<double>(now + waitInSec)));
+    bucket_.reset(new folly::DynamicTokenBucket(static_cast<double>(now + waitInSec)));
   }
 
   // Caller must make sure the **the parition has been add, and won't be removed during consume.**
   // Snaphot and rebuild index follow this principle by design.
-  void consume(size_t toConsume) {
-    if (toConsume > burstSize_) {
+  void consume(double toConsume, double rate, double burstSize) {
+    if (toConsume > burstSize) {
       // consumeWithBorrowAndWait do nothing when toConsume > burstSize_, we sleep 1s instead
       std::this_thread::sleep_for(std::chrono::seconds(1));
     } else {
       // If there are enouth tokens, consume and return immediately.
       // If not, cosume anyway, but sleep enough time before return.
       auto now = time::WallClock::fastNowInSec();
-      bucket_->consumeWithBorrowAndWait(static_cast<double>(toConsume), static_cast<double>(now));
+      bucket_->consumeWithBorrowAndWait(toConsume, rate, burstSize, static_cast<double>(now));
     }
   }
 
  private:
-  std::unique_ptr<folly::TokenBucket> bucket_;
-  double rate_{1 << 20};
-  double burstSize_{1 << 20};
+  std::unique_ptr<folly::DynamicTokenBucket> bucket_;
 };
 
 }  // namespace kvstore

diff --git a/src/kvstore/RocksEngine.cpp b/src/kvstore/RocksEngine.cpp
@@ -13,7 +13,6 @@
 #include "common/fs/FileUtils.h"
 #include "common/utils/NebulaKeyUtils.h"
 #include "kvstore/KVStore.h"
-#include "kvstore/RocksEngineConfig.h"
 
 DEFINE_bool(move_files, false, "Move the SST files instead of copy when ingest into dataset");
 
@@ -126,6 +125,7 @@ RocksEngine::RocksEngine(GraphSpaceID spaceId,
   CHECK(status.ok()) << status.ToString();
   db_.reset(db);
   partsNum_ = allParts().size();
+  extractorLen_ = sizeof(PartitionID) + vIdLen;
   LOG(INFO) << "open rocksdb on " << path;
 
   backup();
@@ -202,7 +202,7 @@ nebula::cpp2::ErrorCode RocksEngine::range(const std::string& start,
                                            const std::string& end,
                                            std::unique_ptr<KVIterator>* storageIter) {
   rocksdb::ReadOptions options;
-  options.total_order_seek = true;
+  options.total_order_seek = FLAGS_enable_rocksdb_prefix_filtering;
   rocksdb::Iterator* iter = db_->NewIterator(options);
   if (iter) {
     iter->Seek(rocksdb::Slice(start));
@@ -213,7 +213,32 @@ nebula::cpp2::ErrorCode RocksEngine::range(const std::string& start,
 
 nebula::cpp2::ErrorCode RocksEngine::prefix(const std::string& prefix,
                                             std::unique_ptr<KVIterator>* storageIter) {
+  // In fact, we don't need to check prefix.size() >= extractorLen_, which is caller's duty to make
+  // sure the prefix bloom filter exists. But this is quite error-proning, so we do a check here.
+  if (FLAGS_enable_rocksdb_prefix_filtering && prefix.size() >= extractorLen_) {
+    return prefixWithExtractor(prefix, storageIter);
+  } else {
+    return prefixWithoutExtractor(prefix, storageIter);
+  }
+}
+
+nebula::cpp2::ErrorCode RocksEngine::prefixWithExtractor(const std::string& prefix,
+                                                         std::unique_ptr<KVIterator>* storageIter) {
+  rocksdb::ReadOptions options;
+  options.prefix_same_as_start = true;
+  rocksdb::Iterator* iter = db_->NewIterator(options);
+  if (iter) {
+    iter->Seek(rocksdb::Slice(prefix));
+  }
+  storageIter->reset(new RocksPrefixIter(iter, prefix));
+  return nebula::cpp2::ErrorCode::SUCCEEDED;
+}
+
+nebula::cpp2::ErrorCode RocksEngine::prefixWithoutExtractor(
+    const std::string& prefix, std::unique_ptr<KVIterator>* storageIter) {
   rocksdb::ReadOptions options;
+  // prefix_same_as_start is false by default
+  options.total_order_seek = FLAGS_enable_rocksdb_prefix_filtering;
   rocksdb::Iterator* iter = db_->NewIterator(options);
   if (iter) {
     iter->Seek(rocksdb::Slice(prefix));
@@ -226,6 +251,8 @@ nebula::cpp2::ErrorCode RocksEngine::rangeWithPrefix(const std::string& start,
                                                      const std::string& prefix,
                                                      std::unique_ptr<KVIterator>* storageIter) {
   rocksdb::ReadOptions options;
+  // prefix_same_as_start is false by default
+  options.total_order_seek = FLAGS_enable_rocksdb_prefix_filtering;
   rocksdb::Iterator* iter = db_->NewIterator(options);
   if (iter) {
     iter->Seek(rocksdb::Slice(start));
@@ -397,6 +424,16 @@ nebula::cpp2::ErrorCode RocksEngine::setDBOption(const std::string& configKey,
   }
 }
 
+ErrorOr<nebula::cpp2::ErrorCode, std::string> RocksEngine::getProperty(
+    const std::string& property) {
+  std::string value;
+  if (!db_->GetProperty(property, &value)) {
+    return nebula::cpp2::ErrorCode::E_INVALID_PARM;
+  } else {
+    return value;
+  }
+}
+
 nebula::cpp2::ErrorCode RocksEngine::compact() {
   rocksdb::CompactRangeOptions options;
   options.change_level = FLAGS_rocksdb_compact_change_level;

diff --git a/src/kvstore/RocksEngine.h b/src/kvstore/RocksEngine.h
@@ -15,6 +15,7 @@
 #include "common/base/Base.h"
 #include "kvstore/KVEngine.h"
 #include "kvstore/KVIterator.h"
+#include "kvstore/RocksEngineConfig.h"
 
 namespace nebula {
 namespace kvstore {
@@ -128,6 +129,12 @@ class RocksEngine : public KVEngine {
                                           const std::string& prefix,
                                           std::unique_ptr<KVIterator>* iter) override;
 
+  nebula::cpp2::ErrorCode prefixWithExtractor(const std::string& prefix,
+                                              std::unique_ptr<KVIterator>* storageIter);
+
+  nebula::cpp2::ErrorCode prefixWithoutExtractor(const std::string& prefix,
+                                                 std::unique_ptr<KVIterator>* storageIter);
+
   /*********************
    * Data modification
    ********************/
@@ -161,6 +168,8 @@ class RocksEngine : public KVEngine {
   nebula::cpp2::ErrorCode setDBOption(const std::string& configKey,
                                       const std::string& configValue) override;
 
+  ErrorOr<nebula::cpp2::ErrorCode, std::string> getProperty(const std::string& property) override;
+
   nebula::cpp2::ErrorCode compact() override;
 
   nebula::cpp2::ErrorCode flush() override;
@@ -190,6 +199,7 @@ class RocksEngine : public KVEngine {
   std::string backupPath_;
   std::unique_ptr<rocksdb::BackupEngine> backupDb_{nullptr};
   int32_t partsNum_ = -1;
+  size_t extractorLen_;
 };
 
 }  // namespace kvstore

diff --git a/src/kvstore/RocksEngineConfig.cpp b/src/kvstore/RocksEngineConfig.cpp
@@ -51,7 +51,7 @@ DEFINE_int64(rocksdb_block_cache,
              1024,
              "The default block cache size used in BlockBasedTable. The unit is MB");
 
-DEFINE_int32(row_cache_num, 16 * 1000 * 1000, "Total keys inside the cache");
+DEFINE_int32(rocksdb_row_cache_num, 16 * 1000 * 1000, "Total keys inside the cache");
 
 DEFINE_int32(cache_bucket_exp, 8, "Total buckets number is 1 << cache_bucket_exp");
 
@@ -78,15 +78,13 @@ DEFINE_int32(rocksdb_rate_limit,
              0,
              "write limit in bytes per sec. The unit is MB. 0 means unlimited.");
 
-DEFINE_bool(enable_rocksdb_prefix_filtering,
+DEFINE_bool(enable_rocksdb_whole_key_filtering,
             false,
+            "Whether or not to enable rocksdb's whole key bloom filter");
+
+DEFINE_bool(enable_rocksdb_prefix_filtering,
+            true,
             "Whether or not to enable rocksdb's prefix bloom filter.");
-DEFINE_bool(rocksdb_prefix_bloom_filter_length_flag,
-            false,
-            "If true, prefix bloom filter will be sizeof(PartitionID) + vidLen + "
-            "sizeof(EdgeType). "
-            "If false, prefix bloom filter will be sizeof(PartitionID) + vidLen. ");
-DEFINE_int32(rocksdb_plain_table_prefix_length, 4, "PlainTable prefix size");
 
 DEFINE_bool(rocksdb_compact_change_level,
             true,
@@ -118,34 +116,6 @@ DEFINE_int32(rocksdb_backup_interval_secs,
 namespace nebula {
 namespace kvstore {
 
-class GraphPrefixTransform : public rocksdb::SliceTransform {
- private:
-  size_t prefixLen_;
-  std::string name_;
-
- public:
-  explicit GraphPrefixTransform(size_t prefixLen)
-      : prefixLen_(prefixLen), name_("nebula.GraphPrefix." + std::to_string(prefixLen_)) {}
-
-  const char* Name() const override { return name_.c_str(); }
-
-  rocksdb::Slice Transform(const rocksdb::Slice& src) const override {
-    return rocksdb::Slice(src.data(), prefixLen_);
-  }
-
-  bool InDomain(const rocksdb::Slice& key) const override {
-    if (key.size() < prefixLen_) {
-      return false;
-    }
-    // And we should not use NebulaKeyUtils::isVertex or isEdge here, because it
-    // will regard the prefix itself not in domain since its length does not
-    // satisfy
-    constexpr int32_t len = static_cast<int32_t>(sizeof(NebulaKeyType));
-    auto type = static_cast<NebulaKeyType>(readInt<uint32_t>(key.data(), len) & kTypeMask);
-    return type == NebulaKeyType::kEdge || type == NebulaKeyType::kVertex;
-  }
-};
-
 static rocksdb::Status initRocksdbCompression(rocksdb::Options& baseOpts) {
   static std::unordered_map<std::string, rocksdb::CompressionType> m = {
       {"no", rocksdb::kNoCompression},
@@ -256,10 +226,8 @@ rocksdb::Status initRocksdbOptions(rocksdb::Options& baseOpts,
     baseOpts.rate_limiter = rate_limiter;
   }
 
+  size_t prefixLength = sizeof(PartitionID) + vidLen;
   if (FLAGS_rocksdb_table_format == "BlockBasedTable") {
-    size_t prefixLength = FLAGS_rocksdb_prefix_bloom_filter_length_flag
-                              ? sizeof(PartitionID) + vidLen + sizeof(EdgeType)
-                              : sizeof(PartitionID) + vidLen;
     // BlockBasedTableOptions
     std::unordered_map<std::string, std::string> bbtOptsMap;
     if (!loadOptionsMap(bbtOptsMap, FLAGS_rocksdb_block_based_table_options)) {
@@ -279,9 +247,9 @@ rocksdb::Status initRocksdbOptions(rocksdb::Options& baseOpts,
       bbtOpts.block_cache = blockCache;
     }
 
-    if (FLAGS_row_cache_num) {
+    if (FLAGS_rocksdb_row_cache_num) {
       static std::shared_ptr<rocksdb::Cache> rowCache =
-          rocksdb::NewLRUCache(FLAGS_row_cache_num, FLAGS_cache_bucket_exp);
+          rocksdb::NewLRUCache(FLAGS_rocksdb_row_cache_num, FLAGS_cache_bucket_exp);
       baseOpts.row_cache = rowCache;
     }
 
@@ -296,8 +264,9 @@ rocksdb::Status initRocksdbOptions(rocksdb::Options& baseOpts,
           baseOpts.compaction_style == rocksdb::CompactionStyle::kCompactionStyleLevel;
     }
     if (FLAGS_enable_rocksdb_prefix_filtering) {
-      baseOpts.prefix_extractor.reset(new GraphPrefixTransform(prefixLength));
+      baseOpts.prefix_extractor.reset(rocksdb::NewCappedPrefixTransform(prefixLength));
     }
+    bbtOpts.whole_key_filtering = FLAGS_enable_rocksdb_whole_key_filtering;
     baseOpts.table_factory.reset(NewBlockBasedTableFactory(bbtOpts));
     baseOpts.create_if_missing = true;
   } else if (FLAGS_rocksdb_table_format == "PlainTable") {
@@ -308,8 +277,10 @@ rocksdb::Status initRocksdbOptions(rocksdb::Options& baseOpts,
     // by default. WAL_ttl_seconds and rocksdb_backup_interval_secs need to be
     // modify together if necessary
     FLAGS_rocksdb_disable_wal = false;
-    baseOpts.prefix_extractor.reset(
-        rocksdb::NewCappedPrefixTransform(FLAGS_rocksdb_plain_table_prefix_length));
+    if (!FLAGS_enable_rocksdb_prefix_filtering) {
+      return rocksdb::Status::InvalidArgument("PlainTable should use prefix bloom filter");
+    }
+    baseOpts.prefix_extractor.reset(rocksdb::NewCappedPrefixTransform(prefixLength));
     baseOpts.table_factory.reset(rocksdb::NewPlainTableFactory());
     baseOpts.create_if_missing = true;
   } else {