leveldb-ruby 0.1

data/leveldb/db/builder.h

@@ -0,0 +1,36 @@
+// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file. See the AUTHORS file for names of contributors.
+
+#ifndef STORAGE_LEVELDB_DB_BUILDER_H_
+#define STORAGE_LEVELDB_DB_BUILDER_H_
+
+#include "leveldb/status.h"
+
+namespace leveldb {
+
+struct Options;
+struct FileMetaData;
+
+class Env;
+class Iterator;
+class TableCache;
+class VersionEdit;
+
+// Build a Table file from the contents of *iter.  The generated file
+// will be named according to meta->number.  On success, the rest of
+// *meta will be filled with metadata about the generated table, and
+// the file information will be added to *edit.  If no data is present
+// in *iter, meta->file_size will be set to zero, and no Table file
+// will be produced.
+extern Status BuildTable(const std::string& dbname,
+                         Env* env,
+                         const Options& options,
+                         TableCache* table_cache,
+                         Iterator* iter,
+                         FileMetaData* meta,
+                         VersionEdit* edit);
+
+}
+
+#endif  // STORAGE_LEVELDB_DB_BUILDER_H_

data/leveldb/db/corruption_test.cc

@@ -0,0 +1,354 @@
+// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file. See the AUTHORS file for names of contributors.
+
+#include "leveldb/db.h"
+
+#include <errno.h>
+#include <fcntl.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include "leveldb/cache.h"
+#include "leveldb/env.h"
+#include "leveldb/table.h"
+#include "leveldb/write_batch.h"
+#include "db/db_impl.h"
+#include "db/filename.h"
+#include "db/log_format.h"
+#include "db/version_set.h"
+#include "util/logging.h"
+#include "util/testharness.h"
+#include "util/testutil.h"
+
+namespace leveldb {
+
+static const int kValueSize = 1000;
+
+class CorruptionTest {
+ public:
+  test::ErrorEnv env_;
+  Random rnd_;
+  std::string dbname_;
+  Cache* tiny_cache_;
+  Options options_;
+  DB* db_;
+
+  CorruptionTest() : rnd_(test::RandomSeed()) {
+    tiny_cache_ = NewLRUCache(100);
+    options_.env = &env_;
+    dbname_ = test::TmpDir() + "/db_test";
+    DestroyDB(dbname_, options_);
+
+    db_ = NULL;
+    options_.create_if_missing = true;
+    Reopen();
+    options_.create_if_missing = false;
+  }
+
+  ~CorruptionTest() {
+     delete db_;
+     DestroyDB(dbname_, Options());
+     delete tiny_cache_;
+  }
+
+  Status TryReopen(Options* options = NULL) {
+    delete db_;
+    db_ = NULL;
+    Options opt = (options ? *options : options_);
+    opt.env = &env_;
+    opt.block_cache = tiny_cache_;
+    return DB::Open(opt, dbname_, &db_);
+  }
+
+  void Reopen(Options* options = NULL) {
+    ASSERT_OK(TryReopen(options));
+  }
+
+  void RepairDB() {
+    delete db_;
+    db_ = NULL;
+    ASSERT_OK(::leveldb::RepairDB(dbname_, options_));
+  }
+
+  void Build(int n) {
+    std::string key_space, value_space;
+    WriteBatch batch;
+    for (int i = 0; i < n; i++) {
+      //if ((i % 100) == 0) fprintf(stderr, "@ %d of %d\n", i, n);
+      Slice key = Key(i, &key_space);
+      batch.Clear();
+      batch.Put(key, Value(i, &value_space));
+      ASSERT_OK(db_->Write(WriteOptions(), &batch));
+    }
+  }
+
+  void Check(int min_expected, int max_expected) {
+    int next_expected = 0;
+    int missed = 0;
+    int bad_keys = 0;
+    int bad_values = 0;
+    int correct = 0;
+    std::string value_space;
+    Iterator* iter = db_->NewIterator(ReadOptions());
+    for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
+      uint64_t key;
+      Slice in(iter->key());
+      if (!ConsumeDecimalNumber(&in, &key) ||
+          !in.empty() ||
+          key < next_expected) {
+        bad_keys++;
+        continue;
+      }
+      missed += (key - next_expected);
+      next_expected = key + 1;
+      if (iter->value() != Value(key, &value_space)) {
+        bad_values++;
+      } else {
+        correct++;
+      }
+    }
+    delete iter;
+
+    fprintf(stderr,
+            "expected=%d..%d; got=%d; bad_keys=%d; bad_values=%d; missed=%d\n",
+            min_expected, max_expected, correct, bad_keys, bad_values, missed);
+    ASSERT_LE(min_expected, correct);
+    ASSERT_GE(max_expected, correct);
+  }
+
+  void Corrupt(FileType filetype, int offset, int bytes_to_corrupt) {
+    // Pick file to corrupt
+    std::vector<std::string> filenames;
+    ASSERT_OK(env_.GetChildren(dbname_, &filenames));
+    uint64_t number;
+    FileType type;
+    std::vector<std::string> candidates;
+    for (int i = 0; i < filenames.size(); i++) {
+      if (ParseFileName(filenames[i], &number, &type) &&
+          type == filetype) {
+        candidates.push_back(dbname_ + "/" + filenames[i]);
+      }
+    }
+    ASSERT_TRUE(!candidates.empty()) << filetype;
+    std::string fname = candidates[rnd_.Uniform(candidates.size())];
+
+    struct stat sbuf;
+    if (stat(fname.c_str(), &sbuf) != 0) {
+      const char* msg = strerror(errno);
+      ASSERT_TRUE(false) << fname << ": " << msg;
+    }
+
+    if (offset < 0) {
+      // Relative to end of file; make it absolute
+      if (-offset > sbuf.st_size) {
+        offset = 0;
+      } else {
+        offset = sbuf.st_size + offset;
+      }
+    }
+    if (offset > sbuf.st_size) {
+      offset = sbuf.st_size;
+    }
+    if (offset + bytes_to_corrupt > sbuf.st_size) {
+      bytes_to_corrupt = sbuf.st_size - offset;
+    }
+
+    // Do it
+    std::string contents;
+    Status s = ReadFileToString(Env::Default(), fname, &contents);
+    ASSERT_TRUE(s.ok()) << s.ToString();
+    for (int i = 0; i < bytes_to_corrupt; i++) {
+      contents[i + offset] ^= 0x80;
+    }
+    s = WriteStringToFile(Env::Default(), contents, fname);
+    ASSERT_TRUE(s.ok()) << s.ToString();
+  }
+
+  int Property(const std::string& name) {
+    std::string property;
+    int result;
+    if (db_->GetProperty(name, &property) &&
+        sscanf(property.c_str(), "%d", &result) == 1) {
+      return result;
+    } else {
+      return -1;
+    }
+  }
+
+  // Return the ith key
+  Slice Key(int i, std::string* storage) {
+    char buf[100];
+    snprintf(buf, sizeof(buf), "%016d", i);
+    storage->assign(buf, strlen(buf));
+    return Slice(*storage);
+  }
+
+  // Return the value to associate with the specified key
+  Slice Value(int k, std::string* storage) {
+    Random r(k);
+    return test::RandomString(&r, kValueSize, storage);
+  }
+};
+
+TEST(CorruptionTest, Recovery) {
+  Build(100);
+  Check(100, 100);
+  Corrupt(kLogFile, 19, 1);      // WriteBatch tag for first record
+  Corrupt(kLogFile, log::kBlockSize + 1000, 1);  // Somewhere in second block
+  Reopen();
+
+  // The 64 records in the first two log blocks are completely lost.
+  Check(36, 36);
+}
+
+TEST(CorruptionTest, RecoverWriteError) {
+  env_.writable_file_error_ = true;
+  Status s = TryReopen();
+  ASSERT_TRUE(!s.ok());
+}
+
+TEST(CorruptionTest, NewFileErrorDuringWrite) {
+  // Do enough writing to force minor compaction
+  env_.writable_file_error_ = true;
+  const int num = 3 + (Options().write_buffer_size / kValueSize);
+  std::string value_storage;
+  Status s;
+  for (int i = 0; s.ok() && i < num; i++) {
+    WriteBatch batch;
+    batch.Put("a", Value(100, &value_storage));
+    s = db_->Write(WriteOptions(), &batch);
+  }
+  ASSERT_TRUE(!s.ok());
+  ASSERT_GE(env_.num_writable_file_errors_, 1);
+  env_.writable_file_error_ = false;
+  Reopen();
+}
+
+TEST(CorruptionTest, TableFile) {
+  Build(100);
+  DBImpl* dbi = reinterpret_cast<DBImpl*>(db_);
+  dbi->TEST_CompactMemTable();
+  dbi->TEST_CompactRange(0, "", "~");
+  dbi->TEST_CompactRange(1, "", "~");
+
+  Corrupt(kTableFile, 100, 1);
+  Check(99, 99);
+}
+
+TEST(CorruptionTest, TableFileIndexData) {
+  Build(10000);  // Enough to build multiple Tables
+  DBImpl* dbi = reinterpret_cast<DBImpl*>(db_);
+  dbi->TEST_CompactMemTable();
+  dbi->TEST_CompactRange(0, "", "~");
+  dbi->TEST_CompactRange(1, "", "~");
+
+  Corrupt(kTableFile, -2000, 500);
+  Reopen();
+  Check(5000, 9999);
+}
+
+TEST(CorruptionTest, MissingDescriptor) {
+  Build(1000);
+  RepairDB();
+  Reopen();
+  Check(1000, 1000);
+}
+
+TEST(CorruptionTest, SequenceNumberRecovery) {
+  ASSERT_OK(db_->Put(WriteOptions(), "foo", "v1"));
+  ASSERT_OK(db_->Put(WriteOptions(), "foo", "v2"));
+  ASSERT_OK(db_->Put(WriteOptions(), "foo", "v3"));
+  ASSERT_OK(db_->Put(WriteOptions(), "foo", "v4"));
+  ASSERT_OK(db_->Put(WriteOptions(), "foo", "v5"));
+  RepairDB();
+  Reopen();
+  std::string v;
+  ASSERT_OK(db_->Get(ReadOptions(), "foo", &v));
+  ASSERT_EQ("v5", v);
+  // Write something.  If sequence number was not recovered properly,
+  // it will be hidden by an earlier write.
+  ASSERT_OK(db_->Put(WriteOptions(), "foo", "v6"));
+  ASSERT_OK(db_->Get(ReadOptions(), "foo", &v));
+  ASSERT_EQ("v6", v);
+  Reopen();
+  ASSERT_OK(db_->Get(ReadOptions(), "foo", &v));
+  ASSERT_EQ("v6", v);
+}
+
+TEST(CorruptionTest, CorruptedDescriptor) {
+  ASSERT_OK(db_->Put(WriteOptions(), "foo", "hello"));
+  DBImpl* dbi = reinterpret_cast<DBImpl*>(db_);
+  dbi->TEST_CompactMemTable();
+  dbi->TEST_CompactRange(0, "", "~");
+
+  Corrupt(kDescriptorFile, 0, 1000);
+  Status s = TryReopen();
+  ASSERT_TRUE(!s.ok());
+
+  RepairDB();
+  Reopen();
+  std::string v;
+  ASSERT_OK(db_->Get(ReadOptions(), "foo", &v));
+  ASSERT_EQ("hello", v);
+}
+
+TEST(CorruptionTest, CompactionInputError) {
+  Build(10);
+  DBImpl* dbi = reinterpret_cast<DBImpl*>(db_);
+  dbi->TEST_CompactMemTable();
+  ASSERT_EQ(1, Property("leveldb.num-files-at-level0"));
+
+  Corrupt(kTableFile, 100, 1);
+  Check(9, 9);
+
+  // Force compactions by writing lots of values
+  Build(10000);
+  Check(10000, 10000);
+  dbi->TEST_CompactRange(0, "", "~");
+  ASSERT_EQ(0, Property("leveldb.num-files-at-level0"));
+}
+
+TEST(CorruptionTest, CompactionInputErrorParanoid) {
+  Options options;
+  options.paranoid_checks = true;
+  options.write_buffer_size = 1048576;
+  Reopen(&options);
+
+  Build(10);
+  DBImpl* dbi = reinterpret_cast<DBImpl*>(db_);
+  dbi->TEST_CompactMemTable();
+  ASSERT_EQ(1, Property("leveldb.num-files-at-level0"));
+
+  Corrupt(kTableFile, 100, 1);
+  Check(9, 9);
+
+  // Write must eventually fail because of corrupted table
+  Status s;
+  std::string tmp1, tmp2;
+  for (int i = 0; i < 10000 && s.ok(); i++) {
+    s = db_->Put(WriteOptions(), Key(i, &tmp1), Value(i, &tmp2));
+  }
+  ASSERT_TRUE(!s.ok()) << "write did not fail in corrupted paranoid db";
+}
+
+TEST(CorruptionTest, UnrelatedKeys) {
+  Build(10);
+  DBImpl* dbi = reinterpret_cast<DBImpl*>(db_);
+  dbi->TEST_CompactMemTable();
+  Corrupt(kTableFile, 100, 1);
+
+  std::string tmp1, tmp2;
+  ASSERT_OK(db_->Put(WriteOptions(), Key(1000, &tmp1), Value(1000, &tmp2)));
+  std::string v;
+  ASSERT_OK(db_->Get(ReadOptions(), Key(1000, &tmp1), &v));
+  ASSERT_EQ(Value(1000, &tmp2).ToString(), v);
+  dbi->TEST_CompactMemTable();
+  ASSERT_OK(db_->Get(ReadOptions(), Key(1000, &tmp1), &v));
+  ASSERT_EQ(Value(1000, &tmp2).ToString(), v);
+}
+
+}
+
+int main(int argc, char** argv) {
+  return leveldb::test::RunAllTests();
+}

data/leveldb/db/db_bench.cc

@@ -0,0 +1,677 @@
+// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file. See the AUTHORS file for names of contributors.
+
+#include <sys/types.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include "db/db_impl.h"
+#include "db/version_set.h"
+#include "leveldb/cache.h"
+#include "leveldb/db.h"
+#include "leveldb/env.h"
+#include "leveldb/write_batch.h"
+#include "port/port.h"
+#include "util/crc32c.h"
+#include "util/histogram.h"
+#include "util/random.h"
+#include "util/testutil.h"
+
+// Comma-separated list of operations to run in the specified order
+//   Actual benchmarks:
+//      fillseq       -- write N values in sequential key order in async mode
+//      fillrandom    -- write N values in random key order in async mode
+//      overwrite     -- overwrite N values in random key order in async mode
+//      fillsync      -- write N/100 values in random key order in sync mode
+//      fill100K      -- write N/1000 100K values in random order in async mode
+//      readseq       -- read N times sequentially
+//      readreverse   -- read N times in reverse order
+//      readrandom    -- read N times in random order
+//      readhot       -- read N times in random order from 1% section of DB
+//      crc32c        -- repeated crc32c of 4K of data
+//      acquireload   -- load N*1000 times
+//   Meta operations:
+//      compact     -- Compact the entire DB
+//      stats       -- Print DB stats
+//      heapprofile -- Dump a heap profile (if supported by this port)
+static const char* FLAGS_benchmarks =
+    "fillseq,"
+    "fillsync,"
+    "fillrandom,"
+    "overwrite,"
+    "readrandom,"
+    "readrandom,"  // Extra run to allow previous compactions to quiesce
+    "readseq,"
+    "readreverse,"
+    "compact,"
+    "readrandom,"
+    "readseq,"
+    "readreverse,"
+    "fill100K,"
+    "crc32c,"
+    "snappycomp,"
+    "snappyuncomp,"
+    "acquireload,"
+    ;
+
+// Number of key/values to place in database
+static int FLAGS_num = 1000000;
+
+// Number of read operations to do.  If negative, do FLAGS_num reads.
+static int FLAGS_reads = -1;
+
+// Size of each value
+static int FLAGS_value_size = 100;
+
+// Arrange to generate values that shrink to this fraction of
+// their original size after compression
+static double FLAGS_compression_ratio = 0.5;
+
+// Print histogram of operation timings
+static bool FLAGS_histogram = false;
+
+// Number of bytes to buffer in memtable before compacting
+// (initialized to default value by "main")
+static int FLAGS_write_buffer_size = 0;
+
+// Number of bytes to use as a cache of uncompressed data.
+// Negative means use default settings.
+static int FLAGS_cache_size = -1;
+
+// Maximum number of files to keep open at the same time (use default if == 0)
+static int FLAGS_open_files = 0;
+
+// If true, do not destroy the existing database.  If you set this
+// flag and also specify a benchmark that wants a fresh database, that
+// benchmark will fail.
+static bool FLAGS_use_existing_db = false;
+
+namespace leveldb {
+
+// Helper for quickly generating random data.
+namespace {
+class RandomGenerator {
+ private:
+  std::string data_;
+  int pos_;
+
+ public:
+  RandomGenerator() {
+    // We use a limited amount of data over and over again and ensure
+    // that it is larger than the compression window (32KB), and also
+    // large enough to serve all typical value sizes we want to write.
+    Random rnd(301);
+    std::string piece;
+    while (data_.size() < 1048576) {
+      // Add a short fragment that is as compressible as specified
+      // by FLAGS_compression_ratio.
+      test::CompressibleString(&rnd, FLAGS_compression_ratio, 100, &piece);
+      data_.append(piece);
+    }
+    pos_ = 0;
+  }
+
+  Slice Generate(int len) {
+    if (pos_ + len > data_.size()) {
+      pos_ = 0;
+      assert(len < data_.size());
+    }
+    pos_ += len;
+    return Slice(data_.data() + pos_ - len, len);
+  }
+};
+
+static Slice TrimSpace(Slice s) {
+  int start = 0;
+  while (start < s.size() && isspace(s[start])) {
+    start++;
+  }
+  int limit = s.size();
+  while (limit > start && isspace(s[limit-1])) {
+    limit--;
+  }
+  return Slice(s.data() + start, limit - start);
+}
+
+}
+
+class Benchmark {
+ private:
+  Cache* cache_;
+  DB* db_;
+  int num_;
+  int reads_;
+  int heap_counter_;
+  double start_;
+  double last_op_finish_;
+  int64_t bytes_;
+  std::string message_;
+  std::string post_message_;
+  Histogram hist_;
+  RandomGenerator gen_;
+  Random rand_;
+
+  // State kept for progress messages
+  int done_;
+  int next_report_;     // When to report next
+
+  void PrintHeader() {
+    const int kKeySize = 16;
+    PrintEnvironment();
+    fprintf(stdout, "Keys:       %d bytes each\n", kKeySize);
+    fprintf(stdout, "Values:     %d bytes each (%d bytes after compression)\n",
+            FLAGS_value_size,
+            static_cast<int>(FLAGS_value_size * FLAGS_compression_ratio + 0.5));
+    fprintf(stdout, "Entries:    %d\n", num_);
+    fprintf(stdout, "RawSize:    %.1f MB (estimated)\n",
+            ((static_cast<int64_t>(kKeySize + FLAGS_value_size) * num_)
+             / 1048576.0));
+    fprintf(stdout, "FileSize:   %.1f MB (estimated)\n",
+            (((kKeySize + FLAGS_value_size * FLAGS_compression_ratio) * num_)
+             / 1048576.0));
+    PrintWarnings();
+    fprintf(stdout, "------------------------------------------------\n");
+  }
+
+  void PrintWarnings() {
+#if defined(__GNUC__) && !defined(__OPTIMIZE__)
+    fprintf(stdout,
+            "WARNING: Optimization is disabled: benchmarks unnecessarily slow\n"
+            );
+#endif
+#ifndef NDEBUG
+    fprintf(stdout,
+            "WARNING: Assertions are enabled; benchmarks unnecessarily slow\n");
+#endif
+
+    // See if snappy is working by attempting to compress a compressible string
+    const char text[] = "yyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyy";
+    std::string compressed;
+    if (!port::Snappy_Compress(text, sizeof(text), &compressed)) {
+      fprintf(stdout, "WARNING: Snappy compression is not enabled\n");
+    } else if (compressed.size() >= sizeof(text)) {
+      fprintf(stdout, "WARNING: Snappy compression is not effective\n");
+    }
+  }
+
+  void PrintEnvironment() {
+    fprintf(stderr, "LevelDB:    version %d.%d\n",
+            kMajorVersion, kMinorVersion);
+
+#if defined(__linux)
+    time_t now = time(NULL);
+    fprintf(stderr, "Date:       %s", ctime(&now));  // ctime() adds newline
+
+    FILE* cpuinfo = fopen("/proc/cpuinfo", "r");
+    if (cpuinfo != NULL) {
+      char line[1000];
+      int num_cpus = 0;
+      std::string cpu_type;
+      std::string cache_size;
+      while (fgets(line, sizeof(line), cpuinfo) != NULL) {
+        const char* sep = strchr(line, ':');
+        if (sep == NULL) {
+          continue;
+        }
+        Slice key = TrimSpace(Slice(line, sep - 1 - line));
+        Slice val = TrimSpace(Slice(sep + 1));
+        if (key == "model name") {
+          ++num_cpus;
+          cpu_type = val.ToString();
+        } else if (key == "cache size") {
+          cache_size = val.ToString();
+        }
+      }
+      fclose(cpuinfo);
+      fprintf(stderr, "CPU:        %d * %s\n", num_cpus, cpu_type.c_str());
+      fprintf(stderr, "CPUCache:   %s\n", cache_size.c_str());
+    }
+#endif
+  }
+
+  void Start() {
+    start_ = Env::Default()->NowMicros() * 1e-6;
+    bytes_ = 0;
+    message_.clear();
+    last_op_finish_ = start_;
+    hist_.Clear();
+    done_ = 0;
+    next_report_ = 100;
+  }
+
+  void FinishedSingleOp() {
+    if (FLAGS_histogram) {
+      double now = Env::Default()->NowMicros() * 1e-6;
+      double micros = (now - last_op_finish_) * 1e6;
+      hist_.Add(micros);
+      if (micros > 20000) {
+        fprintf(stderr, "long op: %.1f micros%30s\r", micros, "");
+        fflush(stderr);
+      }
+      last_op_finish_ = now;
+    }
+
+    done_++;
+    if (done_ >= next_report_) {
+      if      (next_report_ < 1000)   next_report_ += 100;
+      else if (next_report_ < 5000)   next_report_ += 500;
+      else if (next_report_ < 10000)  next_report_ += 1000;
+      else if (next_report_ < 50000)  next_report_ += 5000;
+      else if (next_report_ < 100000) next_report_ += 10000;
+      else if (next_report_ < 500000) next_report_ += 50000;
+      else                            next_report_ += 100000;
+      fprintf(stderr, "... finished %d ops%30s\r", done_, "");
+      fflush(stderr);
+    }
+  }
+
+  void Stop(const Slice& name) {
+    double finish = Env::Default()->NowMicros() * 1e-6;
+
+    // Pretend at least one op was done in case we are running a benchmark
+    // that does nto call FinishedSingleOp().
+    if (done_ < 1) done_ = 1;
+
+    if (bytes_ > 0) {
+      char rate[100];
+      snprintf(rate, sizeof(rate), "%6.1f MB/s",
+               (bytes_ / 1048576.0) / (finish - start_));
+      if (!message_.empty()) {
+        message_  = std::string(rate) + " " + message_;
+      } else {
+        message_ = rate;
+      }
+    }
+
+    fprintf(stdout, "%-12s : %11.3f micros/op;%s%s\n",
+            name.ToString().c_str(),
+            (finish - start_) * 1e6 / done_,
+            (message_.empty() ? "" : " "),
+            message_.c_str());
+    if (FLAGS_histogram) {
+      fprintf(stdout, "Microseconds per op:\n%s\n", hist_.ToString().c_str());
+    }
+    fflush(stdout);
+
+    if (!post_message_.empty()) {
+      fprintf(stdout, "\n%s\n", post_message_.c_str());
+      post_message_.clear();
+    }
+  }
+
+ public:
+  enum Order {
+    SEQUENTIAL,
+    RANDOM
+  };
+  enum DBState {
+    FRESH,
+    EXISTING
+  };
+
+  Benchmark()
+  : cache_(FLAGS_cache_size >= 0 ? NewLRUCache(FLAGS_cache_size) : NULL),
+    db_(NULL),
+    num_(FLAGS_num),
+    reads_(FLAGS_reads < 0 ? FLAGS_num : FLAGS_reads),
+    heap_counter_(0),
+    bytes_(0),
+    rand_(301) {
+    std::vector<std::string> files;
+    Env::Default()->GetChildren("/tmp/dbbench", &files);
+    for (int i = 0; i < files.size(); i++) {
+      if (Slice(files[i]).starts_with("heap-")) {
+        Env::Default()->DeleteFile("/tmp/dbbench/" + files[i]);
+      }
+    }
+    if (!FLAGS_use_existing_db) {
+      DestroyDB("/tmp/dbbench", Options());
+    }
+  }
+
+  ~Benchmark() {
+    delete db_;
+    delete cache_;
+  }
+
+  void Run() {
+    PrintHeader();
+    Open();
+
+    const char* benchmarks = FLAGS_benchmarks;
+    while (benchmarks != NULL) {
+      const char* sep = strchr(benchmarks, ',');
+      Slice name;
+      if (sep == NULL) {
+        name = benchmarks;
+        benchmarks = NULL;
+      } else {
+        name = Slice(benchmarks, sep - benchmarks);
+        benchmarks = sep + 1;
+      }
+
+      Start();
+
+      WriteOptions write_options;
+      bool known = true;
+      if (name == Slice("fillseq")) {
+        Write(write_options, SEQUENTIAL, FRESH, num_, FLAGS_value_size, 1);
+      } else if (name == Slice("fillbatch")) {
+        Write(write_options, SEQUENTIAL, FRESH, num_, FLAGS_value_size, 1000);
+      } else if (name == Slice("fillrandom")) {
+        Write(write_options, RANDOM, FRESH, num_, FLAGS_value_size, 1);
+      } else if (name == Slice("overwrite")) {
+        Write(write_options, RANDOM, EXISTING, num_, FLAGS_value_size, 1);
+      } else if (name == Slice("fillsync")) {
+        write_options.sync = true;
+        Write(write_options, RANDOM, FRESH, num_ / 100, FLAGS_value_size, 1);
+      } else if (name == Slice("fill100K")) {
+        Write(write_options, RANDOM, FRESH, num_ / 1000, 100 * 1000, 1);
+      } else if (name == Slice("readseq")) {
+        ReadSequential();
+      } else if (name == Slice("readreverse")) {
+        ReadReverse();
+      } else if (name == Slice("readrandom")) {
+        ReadRandom();
+      } else if (name == Slice("readhot")) {
+        ReadHot();
+      } else if (name == Slice("readrandomsmall")) {
+        int n = reads_;
+        reads_ /= 1000;
+        ReadRandom();
+        reads_ = n;
+      } else if (name == Slice("compact")) {
+        Compact();
+      } else if (name == Slice("crc32c")) {
+        Crc32c(4096, "(4K per op)");
+      } else if (name == Slice("acquireload")) {
+        AcquireLoad();
+      } else if (name == Slice("snappycomp")) {
+        SnappyCompress();
+      } else if (name == Slice("snappyuncomp")) {
+        SnappyUncompress();
+      } else if (name == Slice("heapprofile")) {
+        HeapProfile();
+      } else if (name == Slice("stats")) {
+        PrintStats();
+      } else {
+        known = false;
+        if (name != Slice()) {  // No error message for empty name
+          fprintf(stderr, "unknown benchmark '%s'\n", name.ToString().c_str());
+        }
+      }
+      if (known) {
+        Stop(name);
+      }
+    }
+  }
+
+ private:
+  void Crc32c(int size, const char* label) {
+    // Checksum about 500MB of data total
+    std::string data(size, 'x');
+    int64_t bytes = 0;
+    uint32_t crc = 0;
+    while (bytes < 500 * 1048576) {
+      crc = crc32c::Value(data.data(), size);
+      FinishedSingleOp();
+      bytes += size;
+    }
+    // Print so result is not dead
+    fprintf(stderr, "... crc=0x%x\r", static_cast<unsigned int>(crc));
+
+    bytes_ = bytes;
+    message_ = label;
+  }
+
+  void AcquireLoad() {
+    int dummy;
+    port::AtomicPointer ap(&dummy);
+    int count = 0;
+    void *ptr = NULL;
+    message_ = "(each op is 1000 loads)";
+    while (count < 100000) {
+      for (int i = 0; i < 1000; i++) {
+        ptr = ap.Acquire_Load();
+      }
+      count++;
+      FinishedSingleOp();
+    }
+    if (ptr == NULL) exit(1); // Disable unused variable warning.
+  }
+
+  void SnappyCompress() {
+    Slice input = gen_.Generate(Options().block_size);
+    int64_t bytes = 0;
+    int64_t produced = 0;
+    bool ok = true;
+    std::string compressed;
+    while (ok && bytes < 1024 * 1048576) {  // Compress 1G
+      ok = port::Snappy_Compress(input.data(), input.size(), &compressed);
+      produced += compressed.size();
+      bytes += input.size();
+      FinishedSingleOp();
+    }
+
+    if (!ok) {
+      message_ = "(snappy failure)";
+    } else {
+      char buf[100];
+      snprintf(buf, sizeof(buf), "(output: %.1f%%)",
+               (produced * 100.0) / bytes);
+      message_ = buf;
+      bytes_ = bytes;
+    }
+  }
+
+  void SnappyUncompress() {
+    Slice input = gen_.Generate(Options().block_size);
+    std::string compressed;
+    bool ok = port::Snappy_Compress(input.data(), input.size(), &compressed);
+    int64_t bytes = 0;
+    std::string uncompressed;
+    while (ok && bytes < 1024 * 1048576) {  // Compress 1G
+      ok =  port::Snappy_Uncompress(compressed.data(), compressed.size(),
+                                    &uncompressed);
+      bytes += uncompressed.size();
+      FinishedSingleOp();
+    }
+
+    if (!ok) {
+      message_ = "(snappy failure)";
+    } else {
+      bytes_ = bytes;
+    }
+  }
+
+  void Open() {
+    assert(db_ == NULL);
+    Options options;
+    options.create_if_missing = !FLAGS_use_existing_db;
+    options.block_cache = cache_;
+    options.write_buffer_size = FLAGS_write_buffer_size;
+    Status s = DB::Open(options, "/tmp/dbbench", &db_);
+    if (!s.ok()) {
+      fprintf(stderr, "open error: %s\n", s.ToString().c_str());
+      exit(1);
+    }
+  }
+
+  void Write(const WriteOptions& options, Order order, DBState state,
+             int num_entries, int value_size, int entries_per_batch) {
+    if (state == FRESH) {
+      if (FLAGS_use_existing_db) {
+        message_ = "skipping (--use_existing_db is true)";
+        return;
+      }
+      delete db_;
+      db_ = NULL;
+      DestroyDB("/tmp/dbbench", Options());
+      Open();
+      Start();  // Do not count time taken to destroy/open
+    }
+
+    if (num_entries != num_) {
+      char msg[100];
+      snprintf(msg, sizeof(msg), "(%d ops)", num_entries);
+      message_ = msg;
+    }
+
+    WriteBatch batch;
+    Status s;
+    std::string val;
+    for (int i = 0; i < num_entries; i += entries_per_batch) {
+      batch.Clear();
+      for (int j = 0; j < entries_per_batch; j++) {
+        const int k = (order == SEQUENTIAL) ? i+j : (rand_.Next() % FLAGS_num);
+        char key[100];
+        snprintf(key, sizeof(key), "%016d", k);
+        batch.Put(key, gen_.Generate(value_size));
+        bytes_ += value_size + strlen(key);
+        FinishedSingleOp();
+      }
+      s = db_->Write(options, &batch);
+      if (!s.ok()) {
+        fprintf(stderr, "put error: %s\n", s.ToString().c_str());
+        exit(1);
+      }
+    }
+  }
+
+  void ReadSequential() {
+    Iterator* iter = db_->NewIterator(ReadOptions());
+    int i = 0;
+    for (iter->SeekToFirst(); i < reads_ && iter->Valid(); iter->Next()) {
+      bytes_ += iter->key().size() + iter->value().size();
+      FinishedSingleOp();
+      ++i;
+    }
+    delete iter;
+  }
+
+  void ReadReverse() {
+    Iterator* iter = db_->NewIterator(ReadOptions());
+    int i = 0;
+    for (iter->SeekToLast(); i < reads_ && iter->Valid(); iter->Prev()) {
+      bytes_ += iter->key().size() + iter->value().size();
+      FinishedSingleOp();
+      ++i;
+    }
+    delete iter;
+  }
+
+  void ReadRandom() {
+    ReadOptions options;
+    std::string value;
+    for (int i = 0; i < reads_; i++) {
+      char key[100];
+      const int k = rand_.Next() % FLAGS_num;
+      snprintf(key, sizeof(key), "%016d", k);
+      db_->Get(options, key, &value);
+      FinishedSingleOp();
+    }
+  }
+
+  void ReadHot() {
+    ReadOptions options;
+    std::string value;
+    const int range = (FLAGS_num + 99) / 100;
+    for (int i = 0; i < reads_; i++) {
+      char key[100];
+      const int k = rand_.Next() % range;
+      snprintf(key, sizeof(key), "%016d", k);
+      db_->Get(options, key, &value);
+      FinishedSingleOp();
+    }
+  }
+
+  void Compact() {
+    DBImpl* dbi = reinterpret_cast<DBImpl*>(db_);
+    dbi->TEST_CompactMemTable();
+    int max_level_with_files = 1;
+    for (int level = 1; level < config::kNumLevels; level++) {
+      std::string property;
+      char name[100];
+      snprintf(name, sizeof(name), "leveldb.num-files-at-level%d", level);
+      if (db_->GetProperty(name, &property) && atoi(property.c_str()) > 0) {
+        max_level_with_files = level;
+      }
+    }
+    for (int level = 0; level < max_level_with_files; level++) {
+      dbi->TEST_CompactRange(level, "", "~");
+    }
+  }
+
+  void PrintStats() {
+    std::string stats;
+    if (!db_->GetProperty("leveldb.stats", &stats)) {
+      message_ = "(failed)";
+    } else {
+      post_message_ = stats;
+    }
+  }
+
+  static void WriteToFile(void* arg, const char* buf, int n) {
+    reinterpret_cast<WritableFile*>(arg)->Append(Slice(buf, n));
+  }
+
+  void HeapProfile() {
+    char fname[100];
+    snprintf(fname, sizeof(fname), "/tmp/dbbench/heap-%04d", ++heap_counter_);
+    WritableFile* file;
+    Status s = Env::Default()->NewWritableFile(fname, &file);
+    if (!s.ok()) {
+      message_ = s.ToString();
+      return;
+    }
+    bool ok = port::GetHeapProfile(WriteToFile, file);
+    delete file;
+    if (!ok) {
+      message_ = "not supported";
+      Env::Default()->DeleteFile(fname);
+    }
+  }
+};
+
+}
+
+int main(int argc, char** argv) {
+  FLAGS_write_buffer_size = leveldb::Options().write_buffer_size;
+  FLAGS_open_files = leveldb::Options().max_open_files;
+
+  for (int i = 1; i < argc; i++) {
+    double d;
+    int n;
+    char junk;
+    if (leveldb::Slice(argv[i]).starts_with("--benchmarks=")) {
+      FLAGS_benchmarks = argv[i] + strlen("--benchmarks=");
+    } else if (sscanf(argv[i], "--compression_ratio=%lf%c", &d, &junk) == 1) {
+      FLAGS_compression_ratio = d;
+    } else if (sscanf(argv[i], "--histogram=%d%c", &n, &junk) == 1 &&
+               (n == 0 || n == 1)) {
+      FLAGS_histogram = n;
+    } else if (sscanf(argv[i], "--use_existing_db=%d%c", &n, &junk) == 1 &&
+               (n == 0 || n == 1)) {
+      FLAGS_use_existing_db = n;
+    } else if (sscanf(argv[i], "--num=%d%c", &n, &junk) == 1) {
+      FLAGS_num = n;
+    } else if (sscanf(argv[i], "--reads=%d%c", &n, &junk) == 1) {
+      FLAGS_reads = n;
+    } else if (sscanf(argv[i], "--value_size=%d%c", &n, &junk) == 1) {
+      FLAGS_value_size = n;
+    } else if (sscanf(argv[i], "--write_buffer_size=%d%c", &n, &junk) == 1) {
+      FLAGS_write_buffer_size = n;
+    } else if (sscanf(argv[i], "--cache_size=%d%c", &n, &junk) == 1) {
+      FLAGS_cache_size = n;
+    } else if (sscanf(argv[i], "--open_files=%d%c", &n, &junk) == 1) {
+      FLAGS_open_files = n;
+    } else {
+      fprintf(stderr, "Invalid flag '%s'\n", argv[i]);
+      exit(1);
+    }
+  }
+
+  leveldb::Benchmark benchmark;
+  benchmark.Run();
+  return 0;
+}