Add a tool to change number of levels

Summary: as subject.

Test Plan: manually test it, will add a testcase

Reviewers: dhruba, MarkCallaghan

Differential Revision: https://reviews.facebook.net/D6345

Makefile

@@ -51,6 +51,7 @@ TESTS = \
 	table_test \
 	version_edit_test \
 	version_set_test \
+	reduce_levels_test \
 	write_batch_test \
 	filelock_test
 
@@ -173,6 +174,9 @@ version_edit_test: db/version_edit_test.o $(LIBOBJECTS) $(TESTHARNESS)
 version_set_test: db/version_set_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) db/version_set_test.o $(LIBOBJECTS) $(TESTHARNESS) -o $@ $(LDFLAGS)
 
+reduce_levels_test: tools/reduce_levels_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(CXX) tools/reduce_levels_test.o $(LIBOBJECTS) $(TESTHARNESS) -o $@ $(LDFLAGS)
+
 write_batch_test: db/write_batch_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) db/write_batch_test.o $(LIBOBJECTS) $(TESTHARNESS) -o $@ $(LDFLAGS)
 

db/version_set.cc

@@ -437,8 +437,8 @@ int Version::PickLevelForMemTableOutput(
         break;
       }
       if (level + 2 >= vset_->NumberLevels()) {
-	level++;
+        level++;
-	break;
+        break;
       }
       GetOverlappingInputs(level + 2, &start, &limit, &overlaps);
       const int64_t sum = TotalFileSize(overlaps);
@@ -714,21 +714,10 @@ VersionSet::VersionSet(const std::string& dbname,
       descriptor_file_(NULL),
       descriptor_log_(NULL),
       dummy_versions_(this),
-      current_(NULL) {
+      current_(NULL),
+      num_levels_(options_->num_levels) {
   compact_pointer_ = new std::string[options_->num_levels];
-  max_file_size_ = new uint64_t[options_->num_levels];
+  Init(options_->num_levels);
-  level_max_bytes_ = new uint64_t[options->num_levels];
-  int target_file_size_multiplier = options_->target_file_size_multiplier;
-  int max_bytes_multiplier = options_->max_bytes_for_level_multiplier;
-  for (int i = 0; i < options_->num_levels; i++) {
-    if (i > 1) {
-      max_file_size_[i] = max_file_size_[i-1] * target_file_size_multiplier;
-      level_max_bytes_[i] = level_max_bytes_[i-1] * max_bytes_multiplier;
-    } else {
-      max_file_size_[i] = options_->target_file_size_base;
-      level_max_bytes_[i] = options_->max_bytes_for_level_base;
-    }
-  }
   AppendVersion(new Version(this));
 }
 
@@ -742,6 +731,22 @@ VersionSet::~VersionSet() {
   delete descriptor_file_;
 }
 
+void VersionSet::Init(int num_levels) {
+  max_file_size_ = new uint64_t[num_levels];
+  level_max_bytes_ = new uint64_t[num_levels];
+  int target_file_size_multiplier = options_->target_file_size_multiplier;
+  int max_bytes_multiplier = options_->max_bytes_for_level_multiplier;
+  for (int i = 0; i < num_levels; i++) {
+    if (i > 1) {
+      max_file_size_[i] = max_file_size_[i-1] * target_file_size_multiplier;
+      level_max_bytes_[i] = level_max_bytes_[i-1] * max_bytes_multiplier;
+    } else {
+      max_file_size_[i] = options_->target_file_size_base;
+      level_max_bytes_[i] = options_->max_bytes_for_level_base;
+    }
+  }
+}
+
 void VersionSet::AppendVersion(Version* v) {
   // Make "v" current
   assert(v->refs_ == 0);
@@ -759,7 +764,8 @@ void VersionSet::AppendVersion(Version* v) {
   v->next_->prev_ = v;
 }
 
-Status VersionSet::LogAndApply(VersionEdit* edit, port::Mutex* mu) {
+Status VersionSet::LogAndApply(VersionEdit* edit, port::Mutex* mu,
+    bool new_descriptor_log) {
   if (edit->has_log_number_) {
     assert(edit->log_number_ >= log_number_);
     assert(edit->log_number_ < next_file_number_);
@@ -787,10 +793,10 @@ Status VersionSet::LogAndApply(VersionEdit* edit, port::Mutex* mu) {
   std::string new_manifest_file;
   uint64_t new_manifest_file_size = 0;
   Status s;
-  if (descriptor_log_ == NULL) {
+  if (descriptor_log_ == NULL || new_descriptor_log) {
     // No reason to unlock *mu here since we only hit this path in the
     // first call to LogAndApply (when opening the database).
-    assert(descriptor_file_ == NULL);
+    assert(descriptor_file_ == NULL || new_descriptor_log)
     new_manifest_file = DescriptorFileName(dbname_, manifest_file_number_);
     edit->SetNextFile(next_file_number_);
     s = env_->NewWritableFile(new_manifest_file, &descriptor_file_);
@@ -1090,7 +1096,6 @@ Status VersionSet::DumpManifest(Options& options, std::string& dscname) {
     printf("%s \n", v->DebugString().c_str());
   }
 
-
   return s;
 }
 

db/version_set.h

@@ -21,6 +21,7 @@
 #include "db/dbformat.h"
 #include "db/version_edit.h"
 #include "port/port.h"
+#include "db/table_cache.h"
 
 namespace leveldb {
 
@@ -156,11 +157,20 @@ class VersionSet {
   // current version.  Will release *mu while actually writing to the file.
   // REQUIRES: *mu is held on entry.
   // REQUIRES: no other thread concurrently calls LogAndApply()
-  Status LogAndApply(VersionEdit* edit, port::Mutex* mu);
+  Status LogAndApply(VersionEdit* edit, port::Mutex* mu,
+      bool new_descriptor_log = false);
 
   // Recover the last saved descriptor from persistent storage.
   Status Recover();
 
+  // Try to reduce the number of levels. This call is valid when
+  // only one level from the new max level to the old
+  // max level containing files.
+  // For example, a db currently has 7 levels [0-6], and a call to
+  // to reduce to 5 [0-4] can only be executed when only one level
+  // among [4-6] contains files.
+  Status ReduceNumberOfLevels(int new_levels, port::Mutex* mu);
+
   // Return the current version.
   Version* current() const { return current_; }
 
@@ -204,7 +214,7 @@ class VersionSet {
   // being compacted, or zero if there is no such log file.
   uint64_t PrevLogNumber() const { return prev_log_number_; }
 
-  int NumberLevels() const { return options_->num_levels; }
+  int NumberLevels() const { return num_levels_; }
 
   // Pick level and inputs for a new compaction.
   // Returns NULL if there is no compaction to be done.
@@ -274,6 +284,8 @@ class VersionSet {
   friend class Compaction;
   friend class Version;
 
+  void Init(int num_levels);
+
   void Finalize(Version* v);
 
   void GetRange(const std::vector<FileMetaData*>& inputs,
@@ -311,6 +323,8 @@ class VersionSet {
   uint64_t log_number_;
   uint64_t prev_log_number_;  // 0 or backing store for memtable being compacted
 
+  int num_levels_;
+
   // Opened lazily
   WritableFile* descriptor_file_;
   log::Writer* descriptor_log_;

db/version_set_reduce_num_levels.cc

@@ -0,0 +1,70 @@
+// Copyright (c) 2012 Facebook. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "db/version_set.h"
+
+#include <algorithm>
+#include <stdio.h>
+#include "db/log_reader.h"
+#include "db/log_writer.h"
+#include "util/logging.h"
+
+namespace leveldb {
+
+Status VersionSet::ReduceNumberOfLevels(int new_levels, port::Mutex* mu) {
+
+  if(new_levels <= 1) {
+    return Status::InvalidArgument(
+        "Number of levels needs to be bigger than 1");
+  }
+
+  Version* current_version = current_;
+  int current_levels = NumberLevels();
+
+  // Make sure there are file only on one level from
+  // (new_levels-1) to (current_levels-1)
+  int first_nonempty_level = -1;
+  int first_nonempty_level_filenum = 0;
+  for (int i = new_levels - 1; i < current_levels; i++) {
+    int file_num = NumLevelFiles(i);
+    if (file_num != 0) {
+      if (first_nonempty_level < 0) {
+        first_nonempty_level = i;
+        first_nonempty_level_filenum = file_num;
+      } else {
+        char msg[255];
+        sprintf(msg, "Found at least two levels containing files: "
+            "[%d:%d],[%d:%d].\n",
+            first_nonempty_level, first_nonempty_level_filenum, i, file_num);
+        return Status::InvalidArgument(msg);
+      }
+    }
+  }
+
+  Status st;
+  std::vector<FileMetaData*>*  old_files_list = current_version->files_;
+  std::vector<FileMetaData*>* new_files_list =
+      new std::vector<FileMetaData*>[new_levels];
+  for (int i = 0; i < new_levels - 1; i++) {
+    new_files_list[i] = old_files_list[i];
+  }
+
+  if (first_nonempty_level > 0) {
+    new_files_list[new_levels - 1] = old_files_list[first_nonempty_level];
+  }
+
+  delete[] current_version->files_;
+  current_version->files_ = new_files_list;
+
+  delete[] compact_pointer_;
+  delete[] max_file_size_;
+  delete[] level_max_bytes_;
+  num_levels_ = new_levels;
+  compact_pointer_ = new std::string[new_levels];
+  Init(new_levels);
+  st = LogAndApply(new VersionEdit(new_levels), mu, true);
+  return st;
+}
+
+}

tools/ldb.cc

@@ -2,234 +2,74 @@
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
-#include <string>
+#include "util/ldb_cmd.h"
-#include <iostream>
-#include <sstream>
-#include <stdlib.h>
 
-#include "leveldb/db.h"
+namespace leveldb {
-#include "leveldb/options.h"
-#include "leveldb/iterator.h"
-#include "leveldb/slice.h"
 
-std::string HexToString(const std::string& str) {
+class LDBCommandRunner {
-  std::string parsed;
+public:
-  for (int i = 0; i < str.length(); ) {
-    int c;
-    sscanf(str.c_str() + i, "%2X", &c);
-    parsed.push_back(c);
-    i += 2;
-  }
-  return parsed;
-}
 
+  static void PrintHelp(const char* exec_name) {
+    std::string ret;
+    ret.append("--- compact ----:\n");
+    ret.append(exec_name);
+    ret.append(" compact ");
+    Compactor::Help(ret);
 
-static void print_usage() {
+    ret.append("\n--- dump ----:\n");
-  fprintf(stderr,
+    ret.append(exec_name);
-    "ldb [compact|dump] "
+    ret.append(" dump ");
-    "--db-path=database_path "
+    DBDumper::Help(ret);
-    "[--from=START KEY] "
-    "[--to=END KEY ] "
-    "[--max_keys=[NUM] (only for dump)] "
-    "[--hex ] "
-    "[--count_only (only for dump)] "
-    "[--stats (only for dump) ] \n");
-}
 
+    ret.append("\n---reduce_levels ----:\n");
+    ret.append(exec_name);
+    ret.append(" reduce_levels ");
+    ReduceDBLevels::Help(ret);
 
-static void safe_open_db(const std::string& dbname, leveldb::DB** db) {
+    fprintf(stderr, "%s\n", ret.c_str());
-  leveldb::Options options;
-  options.create_if_missing = false;
-  leveldb::Status status = leveldb::DB::Open(options, dbname, db);
-
-  if(!status.ok()) {
-    fprintf(
-      stderr,
-      "Could not open db at %s\nERROR: %s",
-      dbname.data(),
-      status.ToString().data()
-    );
-    exit(1);
-  }
-}
-
-
-static void dump_db(
-  const std::string& db_path,
-  std::string& start,
-  std::string& end,
-  int64_t max_keys,
-  const bool hex,
-  const bool print_stats,
-  const bool count_only
-) {
-    // Parse command line args
-  uint64_t count = 0;
-
-
-  if (hex) {
-    start = HexToString(start);
-    end = HexToString(end);
   }
 
-  leveldb::DB *db;
+  static void RunCommand(int argc, char** argv) {
-  safe_open_db(db_path, &db);
+    if (argc <= 2) {
-
+      PrintHelp(argv[0]);
-  if (print_stats) {
+      exit(1);
-    std::string stats;
-    if (db->GetProperty("leveldb.stats", &stats)) {
-      fprintf(stdout, "%s\n", stats.c_str());
     }
-  }
+    const char* cmd = argv[1];
-
+    std::string db_name;
-  // Setup key iterator
+    std::vector<std::string> args;
-  leveldb::Iterator* iter = db->NewIterator(leveldb::ReadOptions());
+    for (int i = 2; i < argc; i++) {
-  leveldb::Status status = iter->status();
+      if (strncmp(argv[i], "--db=", strlen("--db=")) == 0) {
-  if (!status.ok()) {
+        db_name = argv[i] + strlen("--db=");
-    fprintf(stderr, "%s\n", status.ToString().c_str());
-    delete db;
-    exit(1);
-  }
-
-  for (iter->Seek(start); iter->Valid(); iter->Next()) {
-    // If end marker was specified, we stop before it
-    if (!end.empty() && (iter->key().ToString() >= end))
-      break;
-
-    // Terminate if maximum number of keys have been dumped
-    if (max_keys == 0)
-      break;
-
-    --max_keys;
-    ++count;
-
-    if (!count_only) {
-      if (hex) {
-        std::string str = iter->key().ToString();
-        for (int i = 0; i < str.length(); ++i) {
-          fprintf(stdout, "%X", str[i]);
-        }
-        fprintf(stdout, " ==> ");
-        str = iter->value().ToString();
-        for (int i = 0; i < str.length(); ++i) {
-          fprintf(stdout, "%X", str[i]);
-        }
-        fprintf(stdout, "\n");
       } else {
-        fprintf(stdout, "%s ==> %s\n",
+        args.push_back(argv[i]);
-            iter->key().ToString().c_str(),
-            iter->value().ToString().c_str());
       }
     }
+
+    LDBCommand* cmdObj = NULL;
+    if (strncmp(cmd, "compact", strlen("compact")) == 0) {
+      // run compactor
+      cmdObj = new Compactor(db_name, args);
+    } else if (strncmp(cmd, "dump", strlen("dump")) == 0) {
+      // run dump
+      cmdObj = new DBDumper(db_name, args);
+    } else if (strncmp(cmd, "reduce_levels", strlen("reduce_levels")) == 0) {
+      // reduce db levels
+      cmdObj = new ReduceDBLevels(db_name, args);
+    } else {
+      fprintf(stderr, "Unknown command: %s\n", cmd);
+      PrintHelp(argv[0]);
+      exit(1);
+    }
+
+    cmdObj->Run();
+    LDBCommandExecuteResult ret = cmdObj->GetExecuteState();
+    fprintf(stderr, "%s\n", ret.ToString().c_str());
+    delete cmdObj;
   }
+};
 
-  fprintf(stdout, "Keys in range: %lld\n", (long long) count);
-
-  // Clean up
-  delete iter;
-  delete db;
-}
-
-static void compact(
-  const std::string dbname,
-  std::string from,
-  std::string to,
-  const bool hex
-) {
-
-  leveldb::DB* db;
-  safe_open_db(dbname, &db);
-
-  if(hex) {
-    from = HexToString(from);
-    to = HexToString(to);
-  }
-
-  leveldb::Slice* begin = from.empty() ? NULL : new leveldb::Slice(from);
-  leveldb::Slice* end = to.empty() ? NULL : new leveldb::Slice(to);
-  db->CompactRange(begin, end);
-  delete db;
 }
 
 int main(int argc, char** argv) {
-
+  leveldb::LDBCommandRunner::RunCommand(argc, argv);
-  enum {
-    DUMP, COMPACT
-  } command;
-
-  if (argc < 2) {
-    print_usage();
-    exit(1);
-  }
-
-  size_t n;
-
-  const std::string dbnameKey = "--db-path=";
-  const std::string toKey = "--to=";
-  const std::string fromKey = "--from=";
-  std::string dbname;
-  bool dbnameFound = false;
-  std::string from;
-  std::string to;
-  int64_t temp;
-  int64_t max_keys = -1;
-
-  bool print_stats = false;
-  bool count_only = false;
-  bool hex = false;
-  char junk;
-
-
-  std::string commandString = argv[1];
-  if (commandString == "dump") {
-    command = DUMP;
-  } else if (commandString == "compact") {
-    command = COMPACT;
-  } else {
-    print_usage();
-    exit(1);
-  }
-
-  for (int i = 2; i < argc; i++) {
-    std::string param(argv[i]);
-    if ((n = param.find(dbnameKey)) != std::string::npos) {
-      dbname = param.substr(dbnameKey.size());
-      dbnameFound = true;
-    } else if ((n = param.find(fromKey)) != std::string::npos) {
-      from = param.substr(fromKey.size());
-    } else if ((n = param.find(toKey)) != std::string::npos) {
-      to = param.substr(toKey.size());
-    } else if (sscanf(argv[i], "--max_keys=%ld%c", &temp, &junk) == 1) {
-      max_keys = temp;
-    } else if (strncmp(argv[i], "--stats", 7) == 0) {
-      print_stats = true;
-    } else if (strncmp(argv[i], "--count_only", 12) == 0) {
-      count_only = true;
-    } else if (strncmp(argv[i], "--hex", 5) == 0) {
-      hex = true;
-    } else {
-      print_usage();
-      exit(1);
-    }
-  }
-
-  if (!dbnameFound || dbname.empty()) {
-    fprintf(stderr, "DB path required. See help\n");
-    print_usage();
-    exit(1);
-  }
-
-  switch(command) {
-    case DUMP:
-      dump_db(dbname, from, to, max_keys, hex, print_stats, count_only);
-      break;
-    case COMPACT:
-      compact(dbname, from, to, hex);
-      break;
-    default:
-      print_usage();
-      exit(1);
-  }
-
-  return 0;
 }

tools/reduce_levels_test.cc

@@ -0,0 +1,200 @@
+// Copyright (c) 2012 Facebook. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+
+#include "leveldb/db.h"
+#include "db/db_impl.h"
+#include "db/version_set.h"
+#include "util/logging.h"
+#include "util/testutil.h"
+#include "util/testharness.h"
+#include "util/ldb_cmd.h"
+
+namespace leveldb {
+
+class ReduceLevelTest {
+public:
+  ReduceLevelTest() {
+    dbname_ = test::TmpDir() + "/db_reduce_levels_test";
+    DestroyDB(dbname_, Options());
+    db_ = NULL;
+  }
+
+  Status OpenDB(bool create_if_missing, int levels,
+      int mem_table_compact_level);
+
+  Status Put(const std::string& k, const std::string& v) {
+    return db_->Put(WriteOptions(), k, v);
+  }
+
+  std::string Get(const std::string& k) {
+    ReadOptions options;
+    std::string result;
+    Status s = db_->Get(options, k, &result);
+    if (s.IsNotFound()) {
+      result = "NOT_FOUND";
+    } else if (!s.ok()) {
+      result = s.ToString();
+    }
+    return result;
+  }
+
+  Status CompactMemTable() {
+    if (db_ == NULL) {
+      return Status::InvalidArgument("DB not opened.");
+    }
+    DBImpl* db_impl = reinterpret_cast<DBImpl*>(db_);
+    return db_impl->TEST_CompactMemTable();
+  }
+
+  void CloseDB() {
+    if (db_ != NULL) {
+      delete db_;
+      db_ = NULL;
+    }
+  }
+
+  bool ReduceLevels(int target_level);
+
+  int FilesOnLevel(int level) {
+    std::string property;
+    ASSERT_TRUE(
+        db_->GetProperty("leveldb.num-files-at-level" + NumberToString(level),
+                         &property));
+    return atoi(property.c_str());
+  }
+
+private:
+  std::string dbname_;
+  DB* db_;
+};
+
+Status ReduceLevelTest::OpenDB(bool create_if_missing, int num_levels,
+    int mem_table_compact_level) {
+  leveldb::Options opt;
+  opt.num_levels = num_levels;
+  opt.create_if_missing = create_if_missing;
+  opt.max_mem_compaction_level = mem_table_compact_level;
+  leveldb::Status st = leveldb::DB::Open(opt, dbname_, &db_);
+  if (!st.ok()) {
+    fprintf(stderr, "Can't open the db:%s\n", st.ToString().c_str());
+  }
+  return st;
+}
+
+bool ReduceLevelTest::ReduceLevels(int target_level) {
+  std::vector<std::string> args = leveldb::ReduceDBLevels::PrepareArgs(
+      target_level, false);
+  ReduceDBLevels level_reducer(dbname_, args);
+  level_reducer.Run();
+  return level_reducer.GetExecuteState().IsSucceed();
+}
+
+TEST(ReduceLevelTest, Last_Level) {
+  // create files on all levels;
+  ASSERT_OK(OpenDB(true, 4, 3));
+  ASSERT_OK(Put("aaaa", "11111"));
+  ASSERT_OK(CompactMemTable());
+  ASSERT_EQ(FilesOnLevel(3), 1);
+  CloseDB();
+
+  ASSERT_TRUE(ReduceLevels(3));
+  ASSERT_OK(OpenDB(true, 3, 1));
+  ASSERT_EQ(FilesOnLevel(2), 1);
+  CloseDB();
+
+  ASSERT_TRUE(ReduceLevels(2));
+  ASSERT_OK(OpenDB(true, 2, 1));
+  ASSERT_EQ(FilesOnLevel(1), 1);
+  CloseDB();
+}
+
+TEST(ReduceLevelTest, Top_Level) {
+  // create files on all levels;
+  ASSERT_OK(OpenDB(true, 5, 0));
+  ASSERT_OK(Put("aaaa", "11111"));
+  ASSERT_OK(CompactMemTable());
+  ASSERT_EQ(FilesOnLevel(0), 1);
+  CloseDB();
+
+  // The CompactRange(NULL, NULL) call in ReduceLevels
+  // will push this file to level-1
+  ASSERT_TRUE(ReduceLevels(4));
+  ASSERT_OK(OpenDB(true, 4, 0));
+  ASSERT_EQ(FilesOnLevel(1), 1);
+  CloseDB();
+
+  ASSERT_TRUE(ReduceLevels(3));
+  ASSERT_OK(OpenDB(true, 3, 0));
+  ASSERT_EQ(FilesOnLevel(1), 1);
+  CloseDB();
+
+  ASSERT_TRUE(ReduceLevels(2));
+  ASSERT_OK(OpenDB(true, 2, 0));
+  ASSERT_EQ(FilesOnLevel(1), 1);
+  CloseDB();
+}
+
+TEST(ReduceLevelTest, All_Levels) {
+  // create files on all levels;
+  ASSERT_OK(OpenDB(true, 5, 1));
+  ASSERT_OK(Put("a", "a11111"));
+  ASSERT_OK(CompactMemTable());
+  ASSERT_EQ(FilesOnLevel(1), 1);
+  CloseDB();
+
+  ASSERT_OK(OpenDB(true, 5, 2));
+  ASSERT_OK(Put("b", "b11111"));
+  ASSERT_OK(CompactMemTable());
+  ASSERT_EQ(FilesOnLevel(1), 1);
+  ASSERT_EQ(FilesOnLevel(2), 1);
+  CloseDB();
+
+  ASSERT_OK(OpenDB(true, 5, 3));
+  ASSERT_OK(Put("c", "c11111"));
+  ASSERT_OK(CompactMemTable());
+  ASSERT_EQ(FilesOnLevel(1), 1);
+  ASSERT_EQ(FilesOnLevel(2), 1);
+  ASSERT_EQ(FilesOnLevel(3), 1);
+  CloseDB();
+
+  ASSERT_OK(OpenDB(true, 5, 4));
+  ASSERT_OK(Put("d", "d11111"));
+  ASSERT_OK(CompactMemTable());
+  ASSERT_EQ(FilesOnLevel(1), 1);
+  ASSERT_EQ(FilesOnLevel(2), 1);
+  ASSERT_EQ(FilesOnLevel(3), 1);
+  ASSERT_EQ(FilesOnLevel(4), 1);
+  CloseDB();
+
+  ASSERT_TRUE(ReduceLevels(4));
+  ASSERT_OK(OpenDB(true, 4, 0));
+  ASSERT_EQ("a11111", Get("a"));
+  ASSERT_EQ("b11111", Get("b"));
+  ASSERT_EQ("c11111", Get("c"));
+  ASSERT_EQ("d11111", Get("d"));
+  CloseDB();
+
+  ASSERT_TRUE(ReduceLevels(3));
+  ASSERT_OK(OpenDB(true, 3, 0));
+  ASSERT_EQ("a11111", Get("a"));
+  ASSERT_EQ("b11111", Get("b"));
+  ASSERT_EQ("c11111", Get("c"));
+  ASSERT_EQ("d11111", Get("d"));
+  CloseDB();
+
+  ASSERT_TRUE(ReduceLevels(2));
+  ASSERT_OK(OpenDB(true, 2, 0));
+  ASSERT_EQ("a11111", Get("a"));
+  ASSERT_EQ("b11111", Get("b"));
+  ASSERT_EQ("c11111", Get("c"));
+  ASSERT_EQ("d11111", Get("d"));
+  CloseDB();
+}
+
+}
+
+int main(int argc, char** argv) {
+  return leveldb::test::RunAllTests();
+}

util/ldb_cmd.cc

@@ -0,0 +1,294 @@
+// Copyright (c) 2012 Facebook. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "util/ldb_cmd.h"
+
+namespace leveldb {
+
+const char* LDBCommand::FROM_ARG = "--from=";
+const char* LDBCommand::END_ARG = "--to=";
+const char* LDBCommand::HEX_ARG = "--hex";
+
+Compactor::Compactor(std::string& db_name, std::vector<std::string>& args) :
+  LDBCommand(db_name, args), null_from_(true), null_to_(true), hex_(false) {
+  for (int i = 0; i < args.size(); i++) {
+    std::string& arg = args.at(i);
+    if (arg.find(FROM_ARG) == 0) {
+      null_from_ = false;
+      from_ = arg.substr(strlen(FROM_ARG));
+    } else if (arg.find(END_ARG) == 0) {
+      null_to_ = false;
+      to_ = arg.substr(strlen(END_ARG));
+    } else if (arg.find(HEX_ARG) == 0) {
+      hex_ = true;
+    } else {
+      exec_state_ = LDBCommandExecuteResult::FAILED("Unknown argument." + arg);
+    }
+  }
+
+  if (hex_) {
+    if (!null_from_) {
+      from_ = HexToString(from_);
+    }
+    if (!null_to_) {
+      to_ = HexToString(to_);
+    }
+  }
+}
+
+void Compactor::Help(std::string& ret) {
+  LDBCommand::Help(ret);
+  ret.append("[--from=START KEY] ");
+  ret.append("[--to=START KEY] ");
+  ret.append("[--hex] ");
+}
+
+void Compactor::DoCommand() {
+
+  leveldb::Slice* begin = NULL;
+  leveldb::Slice* end = NULL;
+  if (!null_from_) {
+    begin = new leveldb::Slice(from_);
+  }
+  if (!null_to_) {
+    end = new leveldb::Slice(to_);
+  }
+
+  db_->CompactRange(begin, end);
+  exec_state_ = LDBCommandExecuteResult::SUCCEED("");
+
+  delete begin;
+  delete end;
+}
+
+const char* DBDumper::MAX_KEYS_ARG = "--max_keys=";
+const char* DBDumper::COUNT_ONLY_ARG = "--count_only";
+const char* DBDumper::STATS_ARG = "--stats";
+const char* DBDumper::HEX_OUTPUT_ARG = "--output_hex";
+
+DBDumper::DBDumper(std::string& db_name, std::vector<std::string>& args) :
+  LDBCommand(db_name, args), null_from_(true), null_to_(true), hex_(false),
+      count_only_(false), print_stats_(false), max_keys_(-1),
+      hex_output_(false) {
+  for (int i = 0; i < args.size(); i++) {
+    std::string& arg = args.at(i);
+    if (arg.find(FROM_ARG) == 0) {
+      null_from_ = false;
+      from_ = arg.substr(strlen(FROM_ARG));
+    } else if (arg.find(END_ARG) == 0) {
+      null_to_ = false;
+      to_ = arg.substr(strlen(END_ARG));
+    } else if (arg.find(HEX_ARG) == 0) {
+      hex_ = true;
+    } else if (arg.find(MAX_KEYS_ARG) == 0) {
+      max_keys_ = atoi(arg.substr(strlen(MAX_KEYS_ARG)).c_str());
+    } else if (arg.find(STATS_ARG) == 0) {
+      print_stats_ = true;
+    } else if (arg.find(COUNT_ONLY_ARG) == 0) {
+      count_only_ = true;
+    } else if (arg.find(HEX_OUTPUT_ARG) == 0) {
+      hex_output_ = true;
+    } else {
+      exec_state_ = LDBCommandExecuteResult::FAILED("Unknown argument:" + arg);
+    }
+  }
+
+  if (hex_) {
+    if (!null_from_) {
+      from_ = HexToString(from_);
+    }
+    if (!null_to_) {
+      to_ = HexToString(to_);
+    }
+  }
+}
+
+void DBDumper::Help(std::string& ret) {
+  LDBCommand::Help(ret);
+  ret.append("[--from=START KEY] ");
+  ret.append("[--to=END Key] ");
+  ret.append("[--hex] ");
+  ret.append("[--output_hex] ");
+  ret.append("[--max_keys=NUM] ");
+  ret.append("[--count_only] ");
+  ret.append("[--stats] ");
+}
+
+void DBDumper::DoCommand() {
+  // Parse command line args
+  uint64_t count = 0;
+  if (print_stats_) {
+    std::string stats;
+    if (db_->GetProperty("leveldb.stats", &stats)) {
+      fprintf(stdout, "%s\n", stats.c_str());
+    }
+  }
+
+  // Setup key iterator
+  leveldb::Iterator* iter = db_->NewIterator(leveldb::ReadOptions());
+  leveldb::Status st = iter->status();
+  if (!st.ok()) {
+    exec_state_ = LDBCommandExecuteResult::FAILED("Iterator error."
+        + st.ToString());
+  }
+
+  if (!null_from_) {
+    iter->Seek(from_);
+  } else {
+    iter->SeekToFirst();
+  }
+
+  int max_keys = max_keys_;
+  for (; iter->Valid(); iter->Next()) {
+    // If end marker was specified, we stop before it
+    if (!null_to_ && (iter->key().ToString() >= to_))
+      break;
+    // Terminate if maximum number of keys have been dumped
+    if (max_keys == 0)
+      break;
+    if (max_keys > 0) {
+      --max_keys;
+    }
+    ++count;
+    if (!count_only_) {
+      if (hex_output_) {
+        std::string str = iter->key().ToString();
+        for (int i = 0; i < str.length(); ++i) {
+          fprintf(stdout, "%X", str[i]);
+        }
+        fprintf(stdout, " ==> ");
+        str = iter->value().ToString();
+        for (int i = 0; i < str.length(); ++i) {
+          fprintf(stdout, "%X", str[i]);
+        }
+        fprintf(stdout, "\n");
+      } else {
+        fprintf(stdout, "%s ==> %s\n", iter->key().ToString().c_str(),
+            iter->value().ToString().c_str());
+      }
+    }
+  }
+  fprintf(stdout, "Keys in range: %lld\n", (long long) count);
+  // Clean up
+  delete iter;
+}
+
+
+const char* ReduceDBLevels::NEW_LEVLES_ARG = "--new_levels=";
+const char* ReduceDBLevels::PRINT_OLD_LEVELS_ARG = "--print_old_levels";
+
+ReduceDBLevels::ReduceDBLevels(std::string& db_name,
+    std::vector<std::string>& args)
+: LDBCommand(db_name, args),
+  new_levels_(-1),
+  print_old_levels_(false) {
+  for (int i = 0; i < args.size(); i++) {
+    std::string& arg = args.at(i);
+    if (arg.find(NEW_LEVLES_ARG) == 0) {
+      new_levels_ = atoi(arg.substr(strlen(NEW_LEVLES_ARG)).c_str());
+    } else if (arg.find(PRINT_OLD_LEVELS_ARG) == 0) {
+      print_old_levels_ = true;
+    } else {
+      exec_state_ = LDBCommandExecuteResult::FAILED(
+          "Unknown argument." + arg);
+    }
+  }
+
+  if(new_levels_ <= 0) {
+    exec_state_ = LDBCommandExecuteResult::FAILED(
+           " Use --new_levels to specify a new level number\n");
+  }
+}
+
+std::vector<std::string> ReduceDBLevels::PrepareArgs(int new_levels,
+    bool print_old_level) {
+  std::vector<std::string> ret;
+  char arg[100];
+  sprintf(arg, "%s%d", NEW_LEVLES_ARG, new_levels);
+  ret.push_back(arg);
+  if(print_old_level) {
+    sprintf(arg, "%s", PRINT_OLD_LEVELS_ARG);
+    ret.push_back(arg);
+  }
+  return ret;
+}
+
+void ReduceDBLevels::Help(std::string& msg) {
+    LDBCommand::Help(msg);
+    msg.append("[--new_levels=New number of levels] ");
+    msg.append("[--print_old_levels] ");
+}
+
+leveldb::Options ReduceDBLevels::PrepareOptionsForOpenDB() {
+  leveldb::Options opt = LDBCommand::PrepareOptionsForOpenDB();
+  // Set to a big value to make sure we can open the db
+  opt.num_levels = 1 << 16;
+  return opt;
+}
+
+void ReduceDBLevels::DoCommand() {
+  if (new_levels_ <= 1) {
+    exec_state_ = LDBCommandExecuteResult::FAILED(
+        "Invalid number of levels.\n");
+    return;
+  }
+
+  leveldb::Status st;
+
+  leveldb::Options opt = PrepareOptionsForOpenDB();
+  if (print_old_levels_) {
+    TableCache* tc = new TableCache(db_path_, &opt, 10);
+    const InternalKeyComparator* cmp = new InternalKeyComparator(
+        opt.comparator);
+    VersionSet* versions = new VersionSet(db_path_, &opt,
+                                     tc, cmp);
+    // We rely the VersionSet::Recover to tell us the internal data structures
+    // in the db. And the Recover() should never do any change
+    // (like LogAndApply) to the manifest file.
+    st = versions->Recover();
+    int max = -1;
+    for(int i = 0; i<versions->NumberLevels(); i++) {
+      if (versions->NumLevelFiles(i)) {
+        max = i;
+      }
+    }
+    fprintf(stdout, "The old number of levels in use is %d\n", max + 1);
+    delete versions;
+
+    if (!st.ok()) {
+      exec_state_ = LDBCommandExecuteResult::FAILED(st.ToString());
+      return;
+    }
+  }
+
+  // Compact the whole DB to put all files to the highest level.
+  db_->CompactRange(NULL, NULL);
+  CloseDB();
+
+  TableCache* tc = new TableCache(db_path_, &opt, 10);
+  const InternalKeyComparator* cmp = new InternalKeyComparator(
+      opt.comparator);
+  VersionSet* versions = new VersionSet(db_path_, &opt,
+                                   tc, cmp);
+  // We rely the VersionSet::Recover to tell us the internal data structures
+  // in the db. And the Recover() should never do any change (like LogAndApply)
+  // to the manifest file.
+  st = versions->Recover();
+  if (!st.ok()) {
+    exec_state_ = LDBCommandExecuteResult::FAILED(st.ToString());
+    return;
+  }
+
+  port::Mutex mu;
+  mu.Lock();
+  st = versions->ReduceNumberOfLevels(new_levels_, &mu);
+  mu.Unlock();
+
+  if (!st.ok()) {
+    exec_state_ = LDBCommandExecuteResult::FAILED(st.ToString());
+    return;
+  }
+}
+
+}

util/ldb_cmd.h

@@ -0,0 +1,248 @@
+// Copyright (c) 2012 Facebook. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef LEVELDB_UTIL_LDB_H_
+#define LEVELDB_UTIL_LDB_H_
+
+#include <string>
+#include <iostream>
+#include <sstream>
+#include <stdlib.h>
+#include <algorithm>
+#include <stdio.h>
+
+#include "leveldb/db.h"
+#include "leveldb/options.h"
+#include "leveldb/iterator.h"
+#include "leveldb/slice.h"
+#include "db/version_set.h"
+#include "util/logging.h"
+
+namespace leveldb {
+
+class LDBCommandExecuteResult {
+public:
+  enum State {
+    EXEC_NOT_STARTED = 0, EXEC_SUCCEED = 1, EXEC_FAILED = 2,
+  };
+
+  LDBCommandExecuteResult() {
+    state_ = EXEC_NOT_STARTED;
+    message_ = "";
+  }
+
+  LDBCommandExecuteResult(State state, std::string& msg) {
+    state_ = state;
+    message_ = msg;
+  }
+
+  std::string ToString() {
+    std::string ret;
+    switch (state_) {
+    case EXEC_SUCCEED:
+      ret.append("Succeeded.");
+      break;
+    case EXEC_FAILED:
+      ret.append("Failed.");
+      break;
+    case EXEC_NOT_STARTED:
+      ret.append("Not started.");
+    }
+    if (!message_.empty()) {
+      ret.append(message_);
+    }
+    return ret;
+  }
+
+  void Reset() {
+    state_ = EXEC_NOT_STARTED;
+    message_ = "";
+  }
+
+  bool IsSucceed() {
+    return state_ == EXEC_SUCCEED;
+  }
+
+  bool IsNotStarted() {
+    return state_ == EXEC_NOT_STARTED;
+  }
+
+  bool IsFailed() {
+    return state_ == EXEC_FAILED;
+  }
+
+  static LDBCommandExecuteResult SUCCEED(std::string msg) {
+    return LDBCommandExecuteResult(EXEC_SUCCEED, msg);
+  }
+
+  static LDBCommandExecuteResult FAILED(std::string msg) {
+    return LDBCommandExecuteResult(EXEC_FAILED, msg);
+  }
+
+private:
+  State state_;
+  std::string message_;
+
+  bool operator==(const LDBCommandExecuteResult&);
+  bool operator!=(const LDBCommandExecuteResult&);
+};
+
+class LDBCommand {
+public:
+
+  /* Constructor */
+  LDBCommand(std::string& db_name, std::vector<std::string>& args) :
+    db_path_(db_name),
+    db_(NULL) {
+  }
+
+  virtual leveldb::Options PrepareOptionsForOpenDB()  {
+    leveldb::Options opt;
+    opt.create_if_missing = false;
+    return opt;
+  }
+
+  virtual ~LDBCommand() {
+    if (db_ != NULL) {
+      delete db_;
+      db_ = NULL;
+    }
+  }
+
+  /* Print the help message */
+  static void Help(std::string& ret) {
+    ret.append("--db=DB_PATH ");
+  }
+
+  /* Run the command, and return the execute result. */
+  void Run() {
+    if (!exec_state_.IsNotStarted()) {
+      return;
+    }
+
+   if (db_ == NULL) {
+      OpenDB();
+    }
+    DoCommand();
+    if (exec_state_.IsNotStarted()) {
+      exec_state_ = LDBCommandExecuteResult::SUCCEED("");
+    }
+    CloseDB ();
+  }
+
+  virtual void DoCommand() = 0;
+
+  LDBCommandExecuteResult GetExecuteState() {
+    return exec_state_;
+  }
+
+  void ClearPreviousRunState() {
+    exec_state_.Reset();
+  }
+
+  static std::string HexToString(const std::string& str) {
+    std::string parsed;
+    for (int i = 0; i < str.length();) {
+      int c;
+      sscanf(str.c_str() + i, "%2X", &c);
+      parsed.push_back(c);
+      i += 2;
+    }
+    return parsed;
+  }
+
+protected:
+
+  void OpenDB() {
+    leveldb::Options opt = PrepareOptionsForOpenDB();
+    // Open the DB.
+    leveldb::Status st = leveldb::DB::Open(opt, db_path_, &db_);
+    if (!st.ok()) {
+      std::string msg = st.ToString();
+      exec_state_ = LDBCommandExecuteResult::FAILED(msg);
+    }
+  }
+
+  void CloseDB () {
+    if (db_ != NULL) {
+      delete db_;
+      db_ = NULL;
+    }
+  }
+
+  static const char* FROM_ARG;
+  static const char* END_ARG;
+  static const char* HEX_ARG;
+  LDBCommandExecuteResult exec_state_;
+  std::string db_path_;
+  leveldb::DB* db_;
+};
+
+class Compactor: public LDBCommand {
+public:
+  Compactor(std::string& db_name, std::vector<std::string>& args);
+
+  virtual ~Compactor() {}
+
+  static void Help(std::string& ret);
+
+  virtual void DoCommand();
+
+private:
+  bool null_from_;
+  std::string from_;
+  bool null_to_;
+  std::string to_;
+  bool hex_;
+};
+
+class DBDumper: public LDBCommand {
+public:
+  DBDumper(std::string& db_name, std::vector<std::string>& args);
+  virtual ~DBDumper() {}
+  static void Help(std::string& ret);
+  virtual void DoCommand();
+private:
+  bool null_from_;
+  std::string from_;
+  bool null_to_;
+  std::string to_;
+  int max_keys_;
+  bool count_only_;
+  bool print_stats_;
+  bool hex_;
+  bool hex_output_;
+
+  static const char* MAX_KEYS_ARG;
+  static const char* COUNT_ONLY_ARG;
+  static const char* STATS_ARG;
+  static const char* HEX_OUTPUT_ARG;
+};
+
+class ReduceDBLevels : public LDBCommand {
+public:
+
+  ReduceDBLevels (std::string& db_name, std::vector<std::string>& args);
+
+  ~ReduceDBLevels() {}
+
+  virtual leveldb::Options PrepareOptionsForOpenDB();
+
+  virtual void DoCommand();
+  static void Help(std::string& msg);
+
+  static std::vector<std::string> PrepareArgs(int new_levels,
+      bool print_old_level = false);
+
+private:
+  int new_levels_;
+  bool print_old_levels_;
+
+  static const char* NEW_LEVLES_ARG;
+  static const char* PRINT_OLD_LEVELS_ARG;
+};
+
+}
+
+#endif