commit for note

benchmarks/db_bench.cc

@@ -50,6 +50,8 @@ static const char* FLAGS_benchmarks =
     "fillsync,"
     "fillrandom,"
     "overwrite,"
+    "deleteseq,"
+    "deleterandom,"
     "readrandom,"
     "readrandom,"  // Extra run to allow previous compactions to quiesce
     "readseq,"
@@ -59,6 +61,8 @@ static const char* FLAGS_benchmarks =
     "findkeysbyfield,"
     "readseq,"
     "readreverse,"
+    "seekrandom,"
+    "seekordered,"
     "fill100K,";
 // "crc32c,"
 // "snappycomp,"

db/db_impl.cc

@@ -954,8 +954,8 @@ void DBImpl::MaybeScheduleGarbageCollection() {
     // 如果后台线程出错，也不调度。
   } else {
     //设置调度变量,通过detach线程调度;detach线程即使主线程退出,依然可以正常执行完成
-    background_GarbageCollection_scheduled_ = true;
-    env_->ScheduleForGarbageCollection(&DBImpl::GarbageCollectionBGWork, this);
+//    background_GarbageCollection_scheduled_ = true;
+//    env_->ScheduleForGarbageCollection(&DBImpl::GarbageCollectionBGWork, this);
   }
 }
 // 后台gc线程中执行的任务
@@ -1441,6 +1441,7 @@ int64_t DBImpl::TEST_MaxNextLevelOverlappingBytes() {
   return versions_->MaxNextLevelOverlappingBytes();
 }
 
+/* 从 VlogReader 读取的 VLog kv 数据对中解析出 value */
 bool DBImpl::ParseVlogValue(Slice key_value, Slice key,
                             std::string& value, uint64_t val_size) {
   Slice k_v = key_value;
@@ -1564,9 +1565,12 @@ Status DBImpl::Get(const ReadOptions& options, const Slice& key,
     uint64_t val_size;
     size_t key_size = key.size();
 
+    /* 已知该 value 保存在 VLog，解码出 vlog_ptr（fid, offset, val_size）*/
     GetVarint64(&vlog_ptr, &file_no);
     GetVarint64(&vlog_ptr, &offset);
     GetVarint64(&vlog_ptr, &val_size);
+
+    /* VLog 内部 kv 对的编码长度，1B 为 type */
     uint64_t encoded_len = 1 + VarintLength(key_size) + key.size() + VarintLength(val_size) + val_size;
 
     std::string fname = LogFileName(dbname_, file_no);
@@ -2022,7 +2026,7 @@ Status DB::Open(const Options& options, const std::string& dbname, DB** dbptr) {
     impl->RemoveObsoleteFiles();
     impl->MaybeScheduleCompaction();
   }
-  // TODO beigin 开始全盘的回收。
+  // TODO begin 开始全盘的回收。
 
   if( s.ok() && impl->options_.start_garbage_collection ){
       if( s.ok() ){

db/log_format.h

@@ -29,11 +29,6 @@ static const int kBlockSize = 32768;
 // Header is checksum (4 bytes), length (2 bytes), type (1 byte).
 static const int kHeaderSize = 4 + 2 + 1;
 
-
-/* 需要再研究下 */
-//4M 每次读取 vlog 的大小，GC 和 恢复 会用到
-//static const int vBlockSize = 4*1024*1024;
-
 // VlogHeader 只包含 length (4 bytes).
 static const int vHeaderSize = 4;
 // write_batch Header

db/vlog_reader.cc

@@ -30,6 +30,7 @@ bool VlogReader::ReadValue(uint64_t offset, size_t length, Slice *key_value, cha
   if (file_random_ == nullptr) {
     return false;
   }
+  /* 随机读取一个 RandomAccessFile，使用对应的读接口 Read 函数 */
   Status status = file_random_->Read(offset, length, key_value, scratch);
   if (!status.ok()) {
     return false;
@@ -60,6 +61,7 @@ bool VlogReader::ReadPhysicalRecord(std::string *result) {
   buffer_.clear();
 
   char* tmp_head = new char[vHeaderSize];
+  /* 顺序读取一个 SequentialFile，使用对应的读接口 Read 函数 */
   Status status = file_->Read(vHeaderSize, &buffer_, tmp_head);
   if (!status.ok()) {
     buffer_.clear();

db/vlog_reader.h

@@ -22,6 +22,7 @@ class VlogReader {
                 virtual void Corruption(size_t bytes, const Status& status) = 0;
         };
 
+        // 支持随机读与顺序读，提升顺序读取的效率
         explicit VlogReader(SequentialFile* file, Reporter* reporter);
         explicit VlogReader(RandomAccessFile* file, Reporter* reporter);
 

include/leveldb/options.h

@@ -177,7 +177,7 @@ struct LEVELDB_EXPORT ReadOptions {
 
 // Options that control write operations
 struct LEVELDB_EXPORT WriteOptions {
-  explicit WriteOptions(size_t separateThreshold = 16)
+  explicit WriteOptions(size_t separateThreshold = 32)
       : separate_threshold(separateThreshold) {}
 //  WriteOptions() = default;
 

test/bench_test.cc

@@ -153,11 +153,11 @@ void RunBenchmark(const char* name, Func func, bool setup_data = true, bool setu
   delete db;
 }
 
-// TEST(BenchTest, PutLatency) { RunBenchmark("Put", InsertData, false, false); }
-// TEST(BenchTest, PutFieldsLatency) { RunBenchmark("PutFields", InsertFields, false, false); }
+ TEST(BenchTest, PutLatency) { RunBenchmark("Put", InsertData, false, false); }
+ TEST(BenchTest, PutFieldsLatency) { RunBenchmark("PutFields", InsertFields, false, false); }
 
-// TEST(BenchTest, GetLatency) { RunBenchmark("Get", GetData, true, false); }
-// TEST(BenchTest, IteratorLatency) { RunBenchmark("Iterator", ReadOrdered, true, false); }
+ TEST(BenchTest, GetLatency) { RunBenchmark("Get", GetData, true, false); }
+ TEST(BenchTest, IteratorLatency) { RunBenchmark("Iterator", ReadOrdered, true, false); }
 
 TEST(BenchTest, FindKeysByFieldLatency) {
   RunBenchmark("FindKeysByFields", FindKeys, false, true);