column_family.cc 23.4 KB
Newer Older
1 2 3 4 5 6 7 8 9
//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
//  This source code is licensed under the BSD-style license found in the
//  LICENSE file in the root directory of this source tree. An additional grant
//  of patent rights can be found in the PATENTS file in the same directory.
//
// 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.

I
Igor Canadi 已提交
10
#include "db/column_family.h"
11

12 13 14 15 16
#ifndef __STDC_FORMAT_MACROS
#define __STDC_FORMAT_MACROS
#endif

#include <inttypes.h>
17 18 19
#include <vector>
#include <string>
#include <algorithm>
I
Igor Canadi 已提交
20
#include <limits>
21

22
#include "db/db_impl.h"
I
Igor Canadi 已提交
23
#include "db/version_set.h"
24
#include "db/internal_stats.h"
25
#include "db/compaction_picker.h"
26
#include "db/table_properties_collector.h"
27
#include "db/write_controller.h"
28
#include "util/autovector.h"
29
#include "util/hash_skiplist_rep.h"
I
Igor Canadi 已提交
30 31 32

namespace rocksdb {

33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62
namespace {
// This function computes the amount of time in microseconds by which a write
// should be delayed based on the number of level-0 files according to the
// following formula:
// if n < bottom, return 0;
// if n >= top, return 1000;
// otherwise, let r = (n - bottom) /
//                    (top - bottom)
//  and return r^2 * 1000.
// The goal of this formula is to gradually increase the rate at which writes
// are slowed. We also tried linear delay (r * 1000), but it seemed to do
// slightly worse. There is no other particular reason for choosing quadratic.
uint64_t SlowdownAmount(int n, double bottom, double top) {
  uint64_t delay;
  if (n >= top) {
    delay = 1000;
  } else if (n < bottom) {
    delay = 0;
  } else {
    // If we are here, we know that:
    //   level0_start_slowdown <= n < level0_slowdown
    // since the previous two conditions are false.
    double how_much = static_cast<double>(n - bottom) / (top - bottom);
    delay = std::max(how_much * how_much * 1000, 100.0);
  }
  assert(delay <= 1000);
  return delay;
}
}  // namespace

63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79
ColumnFamilyHandleImpl::ColumnFamilyHandleImpl(ColumnFamilyData* cfd,
                                               DBImpl* db, port::Mutex* mutex)
    : cfd_(cfd), db_(db), mutex_(mutex) {
  if (cfd_ != nullptr) {
    cfd_->Ref();
  }
}

ColumnFamilyHandleImpl::~ColumnFamilyHandleImpl() {
  if (cfd_ != nullptr) {
    DBImpl::DeletionState deletion_state;
    mutex_->Lock();
    if (cfd_->Unref()) {
      delete cfd_;
    }
    db_->FindObsoleteFiles(deletion_state, false, true);
    mutex_->Unlock();
80 81 82
    if (deletion_state.HaveSomethingToDelete()) {
      db_->PurgeObsoleteFiles(deletion_state);
    }
83 84 85
  }
}

86 87
uint32_t ColumnFamilyHandleImpl::GetID() const { return cfd()->GetID(); }

88 89 90 91
ColumnFamilyOptions SanitizeOptions(const InternalKeyComparator* icmp,
                                    const ColumnFamilyOptions& src) {
  ColumnFamilyOptions result = src;
  result.comparator = icmp;
I
Igor Canadi 已提交
92 93 94 95
#ifdef OS_MACOSX
  // TODO(icanadi) make write_buffer_size uint64_t instead of size_t
  ClipToRange(&result.write_buffer_size, ((size_t)64) << 10, ((size_t)1) << 30);
#else
96 97
  ClipToRange(&result.write_buffer_size,
              ((size_t)64) << 10, ((size_t)64) << 30);
I
Igor Canadi 已提交
98
#endif
99 100 101 102 103 104 105 106 107 108 109 110 111 112 113
  // if user sets arena_block_size, we trust user to use this value. Otherwise,
  // calculate a proper value from writer_buffer_size;
  if (result.arena_block_size <= 0) {
    result.arena_block_size = result.write_buffer_size / 10;
  }
  result.min_write_buffer_number_to_merge =
      std::min(result.min_write_buffer_number_to_merge,
               result.max_write_buffer_number - 1);
  result.compression_per_level = src.compression_per_level;
  if (result.max_mem_compaction_level >= result.num_levels) {
    result.max_mem_compaction_level = result.num_levels - 1;
  }
  if (result.soft_rate_limit > result.hard_rate_limit) {
    result.soft_rate_limit = result.hard_rate_limit;
  }
114 115 116
  if (result.max_write_buffer_number < 2) {
    result.max_write_buffer_number = 2;
  }
117 118 119 120 121
  if (!result.prefix_extractor) {
    assert(result.memtable_factory);
    Slice name = result.memtable_factory->Name();
    if (name.compare("HashSkipListRepFactory") == 0 ||
        name.compare("HashLinkListRepFactory") == 0) {
122 123 124 125 126 127 128 129
      result.memtable_factory = std::make_shared<SkipListFactory>();
    }
  }

  // -- Sanitize the table properties collector
  // All user defined properties collectors will be wrapped by
  // UserKeyTablePropertiesCollector since for them they only have the
  // knowledge of the user keys; internal keys are invisible to them.
I
Igor Canadi 已提交
130 131 132 133 134 135 136
  auto& collector_factories = result.table_properties_collector_factories;
  for (size_t i = 0; i < result.table_properties_collector_factories.size();
       ++i) {
    assert(collector_factories[i]);
    collector_factories[i] =
        std::make_shared<UserKeyTablePropertiesCollectorFactory>(
            collector_factories[i]);
137 138
  }
  // Add collector to collect internal key statistics
I
Igor Canadi 已提交
139 140
  collector_factories.push_back(
      std::make_shared<InternalKeyPropertiesCollectorFactory>());
141

I
Igor Canadi 已提交
142 143 144 145 146 147 148 149 150
  if (result.compaction_style == kCompactionStyleFIFO) {
    result.num_levels = 1;
    // since we delete level0 files in FIFO compaction when there are too many
    // of them, these options don't really mean anything
    result.level0_file_num_compaction_trigger = std::numeric_limits<int>::max();
    result.level0_slowdown_writes_trigger = std::numeric_limits<int>::max();
    result.level0_stop_writes_trigger = std::numeric_limits<int>::max();
  }

151 152 153
  return result;
}

154 155 156
int SuperVersion::dummy = 0;
void* const SuperVersion::kSVInUse = &SuperVersion::dummy;
void* const SuperVersion::kSVObsolete = nullptr;
157

158 159 160 161 162 163 164 165 166 167 168 169 170
SuperVersion::~SuperVersion() {
  for (auto td : to_delete) {
    delete td;
  }
}

SuperVersion* SuperVersion::Ref() {
  refs.fetch_add(1, std::memory_order_relaxed);
  return this;
}

bool SuperVersion::Unref() {
  // fetch_sub returns the previous value of ref
171 172 173
  uint32_t previous_refs = refs.fetch_sub(1, std::memory_order_relaxed);
  assert(previous_refs > 0);
  return previous_refs == 1;
174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196
}

void SuperVersion::Cleanup() {
  assert(refs.load(std::memory_order_relaxed) == 0);
  imm->Unref(&to_delete);
  MemTable* m = mem->Unref();
  if (m != nullptr) {
    to_delete.push_back(m);
  }
  current->Unref();
}

void SuperVersion::Init(MemTable* new_mem, MemTableListVersion* new_imm,
                        Version* new_current) {
  mem = new_mem;
  imm = new_imm;
  current = new_current;
  mem->Ref();
  imm->Ref();
  current->Ref();
  refs.store(1, std::memory_order_relaxed);
}

197 198
namespace {
void SuperVersionUnrefHandle(void* ptr) {
199 200 201 202
  // UnrefHandle is called when a thread exists or a ThreadLocalPtr gets
  // destroyed. When former happens, the thread shouldn't see kSVInUse.
  // When latter happens, we are in ~ColumnFamilyData(), no get should happen as
  // well.
203 204 205 206 207 208 209 210 211 212
  SuperVersion* sv = static_cast<SuperVersion*>(ptr);
  if (sv->Unref()) {
    sv->db_mutex->Lock();
    sv->Cleanup();
    sv->db_mutex->Unlock();
    delete sv;
  }
}
}  // anonymous namespace

213
ColumnFamilyData::ColumnFamilyData(uint32_t id, const std::string& name,
I
Igor Canadi 已提交
214
                                   Version* dummy_versions, Cache* table_cache,
215
                                   const ColumnFamilyOptions& options,
I
Igor Canadi 已提交
216
                                   const DBOptions* db_options,
L
Lei Jin 已提交
217
                                   const EnvOptions& env_options,
218
                                   ColumnFamilySet* column_family_set)
219 220 221 222
    : id_(id),
      name_(name),
      dummy_versions_(dummy_versions),
      current_(nullptr),
223 224
      refs_(0),
      dropped_(false),
225
      internal_comparator_(options.comparator),
226
      options_(*db_options, SanitizeOptions(&internal_comparator_, options)),
L
Lei Jin 已提交
227
      ioptions_(options_),
228
      mem_(nullptr),
I
Igor Canadi 已提交
229
      imm_(options_.min_write_buffer_number_to_merge),
230 231
      super_version_(nullptr),
      super_version_number_(0),
232
      local_sv_(new ThreadLocalPtr(&SuperVersionUnrefHandle)),
233 234
      next_(nullptr),
      prev_(nullptr),
235
      log_number_(0),
236 237 238
      column_family_set_(column_family_set) {
  Ref();

239 240
  // if dummy_versions is nullptr, then this is a dummy column family.
  if (dummy_versions != nullptr) {
241
    internal_stats_.reset(
242
        new InternalStats(options_.num_levels, db_options->env, this));
L
Lei Jin 已提交
243
    table_cache_.reset(new TableCache(ioptions_, env_options, table_cache));
244
    if (options_.compaction_style == kCompactionStyleUniversal) {
I
Igor Canadi 已提交
245 246
      compaction_picker_.reset(
          new UniversalCompactionPicker(&options_, &internal_comparator_));
I
Igor Canadi 已提交
247
    } else if (options_.compaction_style == kCompactionStyleLevel) {
I
Igor Canadi 已提交
248 249
      compaction_picker_.reset(
          new LevelCompactionPicker(&options_, &internal_comparator_));
I
Igor Canadi 已提交
250 251 252 253
    } else {
      assert(options_.compaction_style == kCompactionStyleFIFO);
      compaction_picker_.reset(
          new FIFOCompactionPicker(&options_, &internal_comparator_));
254
    }
255

I
Igor Canadi 已提交
256
    Log(options_.info_log, "Options for column family \"%s\":\n",
257
        name.c_str());
I
Igor Canadi 已提交
258 259
    const ColumnFamilyOptions* cf_options = &options_;
    cf_options->Dump(options_.info_log.get());
260
  }
261 262

  RecalculateWriteStallConditions();
263
}
I
Igor Canadi 已提交
264

265
// DB mutex held
I
Igor Canadi 已提交
266
ColumnFamilyData::~ColumnFamilyData() {
267 268 269 270 271 272 273 274 275 276
  assert(refs_ == 0);
  // remove from linked list
  auto prev = prev_;
  auto next = next_;
  prev->next_ = next;
  next->prev_ = prev;

  // it's nullptr for dummy CFD
  if (column_family_set_ != nullptr) {
    // remove from column_family_set
I
Igor Canadi 已提交
277
    column_family_set_->RemoveColumnFamily(this);
278 279 280 281 282 283
  }

  if (current_ != nullptr) {
    current_->Unref();
  }

I
Igor Canadi 已提交
284 285 286 287 288 289 290 291 292 293 294 295 296 297
  if (super_version_ != nullptr) {
    // Release SuperVersion reference kept in ThreadLocalPtr.
    // This must be done outside of mutex_ since unref handler can lock mutex.
    super_version_->db_mutex->Unlock();
    local_sv_.reset();
    super_version_->db_mutex->Lock();

    bool is_last_reference __attribute__((unused));
    is_last_reference = super_version_->Unref();
    assert(is_last_reference);
    super_version_->Cleanup();
    delete super_version_;
    super_version_ = nullptr;
  }
298

299 300 301 302 303
  if (dummy_versions_ != nullptr) {
    // List must be empty
    assert(dummy_versions_->next_ == dummy_versions_);
    delete dummy_versions_;
  }
304

305 306
  if (mem_ != nullptr) {
    delete mem_->Unref();
307
  }
308
  autovector<MemTable*> to_delete;
309
  imm_.current()->Unref(&to_delete);
310 311 312 313 314
  for (MemTable* m : to_delete) {
    delete m;
  }
}

315 316
void ColumnFamilyData::RecalculateWriteStallConditions() {
  if (current_ != nullptr) {
317 318 319 320 321 322 323 324 325 326 327 328
    const double score = current_->MaxCompactionScore();
    const int max_level = current_->MaxCompactionScoreLevel();

    auto write_controller = column_family_set_->write_controller_;

    if (imm()->size() == options_.max_write_buffer_number) {
      write_controller_token_ = write_controller->GetStopToken();
      internal_stats_->AddCFStats(InternalStats::MEMTABLE_COMPACTION, 1);
      Log(options_.info_log,
          "[%s] Stopping writes because we have %d immutable memtables "
          "(waiting for flush)",
          name_.c_str(), imm()->size());
329 330 331 332 333 334 335
    } else if (current_->NumLevelFiles(0) >=
               options_.level0_stop_writes_trigger) {
      write_controller_token_ = write_controller->GetStopToken();
      internal_stats_->AddCFStats(InternalStats::LEVEL0_NUM_FILES, 1);
      Log(options_.info_log,
          "[%s] Stopping writes because we have %d level-0 files",
          name_.c_str(), current_->NumLevelFiles(0));
336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371
    } else if (options_.level0_slowdown_writes_trigger >= 0 &&
               current_->NumLevelFiles(0) >=
                   options_.level0_slowdown_writes_trigger) {
      uint64_t slowdown = SlowdownAmount(
          current_->NumLevelFiles(0), options_.level0_slowdown_writes_trigger,
          options_.level0_stop_writes_trigger);
      write_controller_token_ = write_controller->GetDelayToken(slowdown);
      internal_stats_->AddCFStats(InternalStats::LEVEL0_SLOWDOWN, slowdown);
      Log(options_.info_log,
          "[%s] Stalling writes because we have %d level-0 files (%" PRIu64
          "us)",
          name_.c_str(), current_->NumLevelFiles(0), slowdown);
    } else if (options_.hard_rate_limit > 1.0 &&
               score > options_.hard_rate_limit) {
      uint64_t kHardLimitSlowdown = 1000;
      write_controller_token_ =
          write_controller->GetDelayToken(kHardLimitSlowdown);
      internal_stats_->RecordLevelNSlowdown(max_level, kHardLimitSlowdown,
                                            false);
      Log(options_.info_log,
          "[%s] Stalling writes because we hit hard limit on level %d. "
          "(%" PRIu64 "us)",
          name_.c_str(), max_level, kHardLimitSlowdown);
    } else if (options_.soft_rate_limit > 0.0 &&
               score > options_.soft_rate_limit) {
      uint64_t slowdown = SlowdownAmount(score, options_.soft_rate_limit,
                                         options_.hard_rate_limit);
      write_controller_token_ = write_controller->GetDelayToken(slowdown);
      internal_stats_->RecordLevelNSlowdown(max_level, slowdown, true);
      Log(options_.info_log,
          "[%s] Stalling writes because we hit soft limit on level %d (%" PRIu64
          "us)",
          name_.c_str(), max_level, slowdown);
    } else {
      write_controller_token_.reset();
    }
372 373 374
  }
}

L
Lei Jin 已提交
375
const EnvOptions* ColumnFamilyData::soptions() const {
L
Lei Jin 已提交
376
  return &(column_family_set_->env_options_);
L
Lei Jin 已提交
377 378
}

379
void ColumnFamilyData::SetCurrent(Version* current) { current_ = current; }
380

381
void ColumnFamilyData::CreateNewMemtable() {
382 383 384
  assert(current_ != nullptr);
  if (mem_ != nullptr) {
    delete mem_->Unref();
385
  }
L
Lei Jin 已提交
386 387
  mem_ = new MemTable(internal_comparator_, ioptions_,
                      MemTableOptions(options_));
388 389 390
  mem_->Ref();
}

391
Compaction* ColumnFamilyData::PickCompaction(LogBuffer* log_buffer) {
392 393
  auto result = compaction_picker_->PickCompaction(current_, log_buffer);
  return result;
394 395 396
}

Compaction* ColumnFamilyData::CompactRange(int input_level, int output_level,
397
                                           uint32_t output_path_id,
398 399 400 401
                                           const InternalKey* begin,
                                           const InternalKey* end,
                                           InternalKey** compaction_end) {
  return compaction_picker_->CompactRange(current_, input_level, output_level,
402 403
                                          output_path_id, begin, end,
                                          compaction_end);
404 405
}

406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486
SuperVersion* ColumnFamilyData::GetReferencedSuperVersion(
    port::Mutex* db_mutex) {
  SuperVersion* sv = nullptr;
  if (LIKELY(column_family_set_->db_options_->allow_thread_local)) {
    sv = GetThreadLocalSuperVersion(db_mutex);
    sv->Ref();
    if (!ReturnThreadLocalSuperVersion(sv)) {
      sv->Unref();
    }
  } else {
    db_mutex->Lock();
    sv = super_version_->Ref();
    db_mutex->Unlock();
  }
  return sv;
}

SuperVersion* ColumnFamilyData::GetThreadLocalSuperVersion(
    port::Mutex* db_mutex) {
  SuperVersion* sv = nullptr;
  // The SuperVersion is cached in thread local storage to avoid acquiring
  // mutex when SuperVersion does not change since the last use. When a new
  // SuperVersion is installed, the compaction or flush thread cleans up
  // cached SuperVersion in all existing thread local storage. To avoid
  // acquiring mutex for this operation, we use atomic Swap() on the thread
  // local pointer to guarantee exclusive access. If the thread local pointer
  // is being used while a new SuperVersion is installed, the cached
  // SuperVersion can become stale. In that case, the background thread would
  // have swapped in kSVObsolete. We re-check the value at when returning
  // SuperVersion back to thread local, with an atomic compare and swap.
  // The superversion will need to be released if detected to be stale.
  void* ptr = local_sv_->Swap(SuperVersion::kSVInUse);
  // Invariant:
  // (1) Scrape (always) installs kSVObsolete in ThreadLocal storage
  // (2) the Swap above (always) installs kSVInUse, ThreadLocal storage
  // should only keep kSVInUse before ReturnThreadLocalSuperVersion call
  // (if no Scrape happens).
  assert(ptr != SuperVersion::kSVInUse);
  sv = static_cast<SuperVersion*>(ptr);
  if (sv == SuperVersion::kSVObsolete ||
      sv->version_number != super_version_number_.load()) {
    RecordTick(options_.statistics.get(), NUMBER_SUPERVERSION_ACQUIRES);
    SuperVersion* sv_to_delete = nullptr;

    if (sv && sv->Unref()) {
      RecordTick(options_.statistics.get(), NUMBER_SUPERVERSION_CLEANUPS);
      db_mutex->Lock();
      // NOTE: underlying resources held by superversion (sst files) might
      // not be released until the next background job.
      sv->Cleanup();
      sv_to_delete = sv;
    } else {
      db_mutex->Lock();
    }
    sv = super_version_->Ref();
    db_mutex->Unlock();

    delete sv_to_delete;
  }
  assert(sv != nullptr);
  return sv;
}

bool ColumnFamilyData::ReturnThreadLocalSuperVersion(SuperVersion* sv) {
  assert(sv != nullptr);
  // Put the SuperVersion back
  void* expected = SuperVersion::kSVInUse;
  if (local_sv_->CompareAndSwap(static_cast<void*>(sv), expected)) {
    // When we see kSVInUse in the ThreadLocal, we are sure ThreadLocal
    // storage has not been altered and no Scrape has happend. The
    // SuperVersion is still current.
    return true;
  } else {
    // ThreadLocal scrape happened in the process of this GetImpl call (after
    // thread local Swap() at the beginning and before CompareAndSwap()).
    // This means the SuperVersion it holds is obsolete.
    assert(expected == SuperVersion::kSVObsolete);
  }
  return false;
}

487
SuperVersion* ColumnFamilyData::InstallSuperVersion(
488
    SuperVersion* new_superversion, port::Mutex* db_mutex) {
489
  new_superversion->db_mutex = db_mutex;
490 491 492 493
  new_superversion->Init(mem_, imm_.current(), current_);
  SuperVersion* old_superversion = super_version_;
  super_version_ = new_superversion;
  ++super_version_number_;
494
  super_version_->version_number = super_version_number_;
495 496 497 498
  // Reset SuperVersions cached in thread local storage
  if (column_family_set_->db_options_->allow_thread_local) {
    ResetThreadLocalSuperVersions();
  }
499 500 501

  RecalculateWriteStallConditions();

502 503 504 505 506
  if (old_superversion != nullptr && old_superversion->Unref()) {
    old_superversion->Cleanup();
    return old_superversion;  // will let caller delete outside of mutex
  }
  return nullptr;
I
Igor Canadi 已提交
507 508
}

509 510
void ColumnFamilyData::ResetThreadLocalSuperVersions() {
  autovector<void*> sv_ptrs;
511
  local_sv_->Scrape(&sv_ptrs, SuperVersion::kSVObsolete);
512 513
  for (auto ptr : sv_ptrs) {
    assert(ptr);
514 515 516
    if (ptr == SuperVersion::kSVInUse) {
      continue;
    }
517 518 519 520 521 522 523 524
    auto sv = static_cast<SuperVersion*>(ptr);
    if (sv->Unref()) {
      sv->Cleanup();
      delete sv;
    }
  }
}

I
Igor Canadi 已提交
525
ColumnFamilySet::ColumnFamilySet(const std::string& dbname,
I
Igor Canadi 已提交
526
                                 const DBOptions* db_options,
L
Lei Jin 已提交
527
                                 const EnvOptions& env_options,
528 529
                                 Cache* table_cache,
                                 WriteController* write_controller)
530
    : max_column_family_(0),
531
      dummy_cfd_(new ColumnFamilyData(0, "", nullptr, nullptr,
532
                                      ColumnFamilyOptions(), db_options,
533
                                      env_options, nullptr)),
I
Igor Canadi 已提交
534
      default_cfd_cache_(nullptr),
I
Igor Canadi 已提交
535 536
      db_name_(dbname),
      db_options_(db_options),
L
Lei Jin 已提交
537
      env_options_(env_options),
538
      table_cache_(table_cache),
539
      write_controller_(write_controller),
540
      spin_lock_(ATOMIC_FLAG_INIT) {
541
  // initialize linked list
542 543
  dummy_cfd_->prev_ = dummy_cfd_;
  dummy_cfd_->next_ = dummy_cfd_;
544
}
I
Igor Canadi 已提交
545 546

ColumnFamilySet::~ColumnFamilySet() {
547 548 549 550
  while (column_family_data_.size() > 0) {
    // cfd destructor will delete itself from column_family_data_
    auto cfd = column_family_data_.begin()->second;
    cfd->Unref();
I
Igor Canadi 已提交
551 552
    delete cfd;
  }
553
  dummy_cfd_->Unref();
554
  delete dummy_cfd_;
I
Igor Canadi 已提交
555 556 557
}

ColumnFamilyData* ColumnFamilySet::GetDefault() const {
I
Igor Canadi 已提交
558 559
  assert(default_cfd_cache_ != nullptr);
  return default_cfd_cache_;
I
Igor Canadi 已提交
560 561 562 563 564 565 566 567 568 569 570
}

ColumnFamilyData* ColumnFamilySet::GetColumnFamily(uint32_t id) const {
  auto cfd_iter = column_family_data_.find(id);
  if (cfd_iter != column_family_data_.end()) {
    return cfd_iter->second;
  } else {
    return nullptr;
  }
}

571 572 573
ColumnFamilyData* ColumnFamilySet::GetColumnFamily(const std::string& name)
    const {
  auto cfd_iter = column_families_.find(name);
I
Igor Canadi 已提交
574 575 576 577 578
  if (cfd_iter != column_families_.end()) {
    auto cfd = GetColumnFamily(cfd_iter->second);
    assert(cfd != nullptr);
    return cfd;
  } else {
579 580
    return nullptr;
  }
I
Igor Canadi 已提交
581 582 583 584 585 586
}

uint32_t ColumnFamilySet::GetNextColumnFamilyID() {
  return ++max_column_family_;
}

587 588 589 590 591 592
uint32_t ColumnFamilySet::GetMaxColumnFamily() { return max_column_family_; }

void ColumnFamilySet::UpdateMaxColumnFamily(uint32_t new_max_column_family) {
  max_column_family_ = std::max(new_max_column_family, max_column_family_);
}

593 594 595 596
size_t ColumnFamilySet::NumberOfColumnFamilies() const {
  return column_families_.size();
}

597
// under a DB mutex
I
Igor Canadi 已提交
598 599 600 601 602
ColumnFamilyData* ColumnFamilySet::CreateColumnFamily(
    const std::string& name, uint32_t id, Version* dummy_versions,
    const ColumnFamilyOptions& options) {
  assert(column_families_.find(name) == column_families_.end());
  ColumnFamilyData* new_cfd =
603
      new ColumnFamilyData(id, name, dummy_versions, table_cache_, options,
L
Lei Jin 已提交
604
                           db_options_, env_options_, this);
605 606
  Lock();
  column_families_.insert({name, id});
I
Igor Canadi 已提交
607
  column_family_data_.insert({id, new_cfd});
608
  Unlock();
I
Igor Canadi 已提交
609
  max_column_family_ = std::max(max_column_family_, id);
610
  // add to linked list
611 612 613 614 615
  new_cfd->next_ = dummy_cfd_;
  auto prev = dummy_cfd_->prev_;
  new_cfd->prev_ = prev;
  prev->next_ = new_cfd;
  dummy_cfd_->prev_ = new_cfd;
I
Igor Canadi 已提交
616 617 618
  if (id == 0) {
    default_cfd_cache_ = new_cfd;
  }
I
Igor Canadi 已提交
619 620 621
  return new_cfd;
}

I
Igor Canadi 已提交
622 623 624 625 626 627 628 629
void ColumnFamilySet::Lock() {
  // spin lock
  while (spin_lock_.test_and_set(std::memory_order_acquire)) {
  }
}

void ColumnFamilySet::Unlock() { spin_lock_.clear(std::memory_order_release); }

630 631 632 633 634 635 636 637 638 639 640 641 642 643
// REQUIRES: DB mutex held
void ColumnFamilySet::FreeDeadColumnFamilies() {
  autovector<ColumnFamilyData*> to_delete;
  for (auto cfd = dummy_cfd_->next_; cfd != dummy_cfd_; cfd = cfd->next_) {
    if (cfd->refs_ == 0) {
      to_delete.push_back(cfd);
    }
  }
  for (auto cfd : to_delete) {
    // this is very rare, so it's not a problem that we do it under a mutex
    delete cfd;
  }
}

644
// under a DB mutex
I
Igor Canadi 已提交
645
void ColumnFamilySet::RemoveColumnFamily(ColumnFamilyData* cfd) {
646
  auto cfd_iter = column_family_data_.find(cfd->GetID());
647
  assert(cfd_iter != column_family_data_.end());
648
  Lock();
649
  column_family_data_.erase(cfd_iter);
650 651
  column_families_.erase(cfd->GetName());
  Unlock();
I
Igor Canadi 已提交
652 653
}

654
bool ColumnFamilyMemTablesImpl::Seek(uint32_t column_family_id) {
I
Igor Canadi 已提交
655 656 657 658 659 660 661 662 663
  if (column_family_id == 0) {
    // optimization for common case
    current_ = column_family_set_->GetDefault();
  } else {
    // maybe outside of db mutex, should lock
    column_family_set_->Lock();
    current_ = column_family_set_->GetColumnFamily(column_family_id);
    column_family_set_->Unlock();
  }
664
  handle_.SetCFD(current_);
665 666
  return current_ != nullptr;
}
667

668 669 670 671 672 673 674 675 676 677
uint64_t ColumnFamilyMemTablesImpl::GetLogNumber() const {
  assert(current_ != nullptr);
  return current_->GetLogNumber();
}

MemTable* ColumnFamilyMemTablesImpl::GetMemTable() const {
  assert(current_ != nullptr);
  return current_->mem();
}

I
Igor Canadi 已提交
678
const Options* ColumnFamilyMemTablesImpl::GetOptions() const {
679
  assert(current_ != nullptr);
I
Igor Canadi 已提交
680
  return current_->options();
681 682
}

683
ColumnFamilyHandle* ColumnFamilyMemTablesImpl::GetColumnFamilyHandle() {
684
  assert(current_ != nullptr);
685
  return &handle_;
686 687
}

688 689 690 691 692 693 694 695 696
uint32_t GetColumnFamilyID(ColumnFamilyHandle* column_family) {
  uint32_t column_family_id = 0;
  if (column_family != nullptr) {
    auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
    column_family_id = cfh->GetID();
  }
  return column_family_id;
}

I
Igor Canadi 已提交
697
}  // namespace rocksdb