|
|
|
// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
|
|
|
|
// This source code is licensed under both the GPLv2 (found in the
|
|
|
|
// COPYING file in the root directory) and Apache 2.0 License
|
|
|
|
// (found in the LICENSE.Apache file in the root 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.
|
|
|
|
//
|
|
|
|
// Decodes the blocks generated by block_builder.cc.
|
|
|
|
|
|
|
|
#include "table/block.h"
|
|
|
|
#include <algorithm>
|
|
|
|
#include <string>
|
|
|
|
#include <unordered_map>
|
|
|
|
#include <vector>
|
|
|
|
|
|
|
|
#include "monitoring/perf_context_imp.h"
|
|
|
|
#include "port/port.h"
|
|
|
|
#include "port/stack_trace.h"
|
|
|
|
#include "rocksdb/comparator.h"
|
|
|
|
#include "table/block_prefix_index.h"
|
|
|
|
#include "table/format.h"
|
|
|
|
#include "util/coding.h"
|
|
|
|
#include "util/logging.h"
|
|
|
|
|
|
|
|
namespace rocksdb {
|
|
|
|
|
|
|
|
// Helper routine: decode the next block entry starting at "p",
|
|
|
|
// storing the number of shared key bytes, non_shared key bytes,
|
|
|
|
// and the length of the value in "*shared", "*non_shared", and
|
|
|
|
// "*value_length", respectively. Will not derefence past "limit".
|
|
|
|
//
|
|
|
|
// If any errors are detected, returns nullptr. Otherwise, returns a
|
|
|
|
// pointer to the key delta (just past the three decoded values).
|
|
|
|
static inline const char* DecodeEntry(const char* p, const char* limit,
|
|
|
|
uint32_t* shared,
|
|
|
|
uint32_t* non_shared,
|
|
|
|
uint32_t* value_length) {
|
|
|
|
if (limit - p < 3) return nullptr;
|
|
|
|
*shared = reinterpret_cast<const unsigned char*>(p)[0];
|
|
|
|
*non_shared = reinterpret_cast<const unsigned char*>(p)[1];
|
|
|
|
*value_length = reinterpret_cast<const unsigned char*>(p)[2];
|
|
|
|
if ((*shared | *non_shared | *value_length) < 128) {
|
|
|
|
// Fast path: all three values are encoded in one byte each
|
|
|
|
p += 3;
|
|
|
|
} else {
|
|
|
|
if ((p = GetVarint32Ptr(p, limit, shared)) == nullptr) return nullptr;
|
|
|
|
if ((p = GetVarint32Ptr(p, limit, non_shared)) == nullptr) return nullptr;
|
|
|
|
if ((p = GetVarint32Ptr(p, limit, value_length)) == nullptr) return nullptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (static_cast<uint32_t>(limit - p) < (*non_shared + *value_length)) {
|
|
|
|
return nullptr;
|
|
|
|
}
|
|
|
|
return p;
|
|
|
|
}
|
|
|
|
|
|
|
|
void BlockIter::Next() {
|
|
|
|
assert(Valid());
|
|
|
|
ParseNextKey();
|
|
|
|
}
|
|
|
|
|
|
|
|
void BlockIter::Prev() {
|
|
|
|
assert(Valid());
|
|
|
|
|
|
|
|
assert(prev_entries_idx_ == -1 ||
|
|
|
|
static_cast<size_t>(prev_entries_idx_) < prev_entries_.size());
|
|
|
|
// Check if we can use cached prev_entries_
|
|
|
|
if (prev_entries_idx_ > 0 &&
|
|
|
|
prev_entries_[prev_entries_idx_].offset == current_) {
|
|
|
|
// Read cached CachedPrevEntry
|
|
|
|
prev_entries_idx_--;
|
|
|
|
const CachedPrevEntry& current_prev_entry =
|
|
|
|
prev_entries_[prev_entries_idx_];
|
|
|
|
|
|
|
|
const char* key_ptr = nullptr;
|
|
|
|
if (current_prev_entry.key_ptr != nullptr) {
|
|
|
|
// The key is not delta encoded and stored in the data block
|
|
|
|
key_ptr = current_prev_entry.key_ptr;
|
|
|
|
key_pinned_ = true;
|
|
|
|
} else {
|
|
|
|
// The key is delta encoded and stored in prev_entries_keys_buff_
|
|
|
|
key_ptr = prev_entries_keys_buff_.data() + current_prev_entry.key_offset;
|
|
|
|
key_pinned_ = false;
|
|
|
|
}
|
|
|
|
const Slice current_key(key_ptr, current_prev_entry.key_size);
|
|
|
|
|
|
|
|
current_ = current_prev_entry.offset;
|
|
|
|
if (key_includes_seq_) {
|
|
|
|
key_.SetInternalKey(current_key, false /* copy */);
|
|
|
|
} else {
|
|
|
|
key_.SetUserKey(current_key, false /* copy */);
|
|
|
|
}
|
|
|
|
value_ = current_prev_entry.value;
|
|
|
|
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Clear prev entries cache
|
|
|
|
prev_entries_idx_ = -1;
|
|
|
|
prev_entries_.clear();
|
|
|
|
prev_entries_keys_buff_.clear();
|
|
|
|
|
|
|
|
// Scan backwards to a restart point before current_
|
|
|
|
const uint32_t original = current_;
|
|
|
|
while (GetRestartPoint(restart_index_) >= original) {
|
|
|
|
if (restart_index_ == 0) {
|
|
|
|
// No more entries
|
|
|
|
current_ = restarts_;
|
|
|
|
restart_index_ = num_restarts_;
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
restart_index_--;
|
|
|
|
}
|
|
|
|
|
|
|
|
SeekToRestartPoint(restart_index_);
|
|
|
|
|
|
|
|
do {
|
|
|
|
if (!ParseNextKey()) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
Slice current_key = key();
|
|
|
|
|
|
|
|
if (key_.IsKeyPinned()) {
|
|
|
|
// The key is not delta encoded
|
|
|
|
prev_entries_.emplace_back(current_, current_key.data(), 0,
|
|
|
|
current_key.size(), value());
|
|
|
|
} else {
|
|
|
|
// The key is delta encoded, cache decoded key in buffer
|
|
|
|
size_t new_key_offset = prev_entries_keys_buff_.size();
|
|
|
|
prev_entries_keys_buff_.append(current_key.data(), current_key.size());
|
|
|
|
|
|
|
|
prev_entries_.emplace_back(current_, nullptr, new_key_offset,
|
|
|
|
current_key.size(), value());
|
|
|
|
}
|
|
|
|
// Loop until end of current entry hits the start of original entry
|
|
|
|
} while (NextEntryOffset() < original);
|
|
|
|
prev_entries_idx_ = static_cast<int32_t>(prev_entries_.size()) - 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
void BlockIter::Seek(const Slice& target) {
|
|
|
|
Slice seek_key = target;
|
|
|
|
if (!key_includes_seq_) {
|
|
|
|
seek_key = ExtractUserKey(target);
|
|
|
|
}
|
|
|
|
PERF_TIMER_GUARD(block_seek_nanos);
|
|
|
|
if (data_ == nullptr) { // Not init yet
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
uint32_t index = 0;
|
|
|
|
bool ok = false;
|
|
|
|
if (prefix_index_) {
|
|
|
|
ok = PrefixSeek(target, &index);
|
|
|
|
} else {
|
|
|
|
ok = BinarySeek(seek_key, 0, num_restarts_ - 1, &index);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!ok) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
SeekToRestartPoint(index);
|
|
|
|
// Linear search (within restart block) for first key >= target
|
|
|
|
|
|
|
|
while (true) {
|
|
|
|
if (!ParseNextKey() || Compare(key_, seek_key) >= 0) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void BlockIter::SeekForPrev(const Slice& target) {
|
|
|
|
PERF_TIMER_GUARD(block_seek_nanos);
|
|
|
|
Slice seek_key = target;
|
|
|
|
if (!key_includes_seq_) {
|
|
|
|
seek_key = ExtractUserKey(target);
|
|
|
|
}
|
|
|
|
if (data_ == nullptr) { // Not init yet
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
uint32_t index = 0;
|
|
|
|
bool ok = BinarySeek(seek_key, 0, num_restarts_ - 1, &index);
|
|
|
|
|
|
|
|
if (!ok) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
SeekToRestartPoint(index);
|
|
|
|
// Linear search (within restart block) for first key >= seek_key
|
|
|
|
|
|
|
|
while (ParseNextKey() && Compare(key_, seek_key) < 0) {
|
|
|
|
}
|
|
|
|
if (!Valid()) {
|
|
|
|
SeekToLast();
|
|
|
|
} else {
|
|
|
|
while (Valid() && Compare(key_, seek_key) > 0) {
|
|
|
|
Prev();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void BlockIter::SeekToFirst() {
|
|
|
|
if (data_ == nullptr) { // Not init yet
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
SeekToRestartPoint(0);
|
|
|
|
ParseNextKey();
|
|
|
|
}
|
|
|
|
|
|
|
|
void BlockIter::SeekToLast() {
|
|
|
|
if (data_ == nullptr) { // Not init yet
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
SeekToRestartPoint(num_restarts_ - 1);
|
|
|
|
while (ParseNextKey() && NextEntryOffset() < restarts_) {
|
|
|
|
// Keep skipping
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void BlockIter::CorruptionError() {
|
|
|
|
current_ = restarts_;
|
|
|
|
restart_index_ = num_restarts_;
|
|
|
|
status_ = Status::Corruption("bad entry in block");
|
|
|
|
key_.Clear();
|
|
|
|
value_.clear();
|
|
|
|
}
|
|
|
|
|
|
|
|
bool BlockIter::ParseNextKey() {
|
|
|
|
current_ = NextEntryOffset();
|
|
|
|
const char* p = data_ + current_;
|
|
|
|
const char* limit = data_ + restarts_; // Restarts come right after data
|
|
|
|
if (p >= limit) {
|
|
|
|
// No more entries to return. Mark as invalid.
|
|
|
|
current_ = restarts_;
|
|
|
|
restart_index_ = num_restarts_;
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Decode next entry
|
|
|
|
uint32_t shared, non_shared, value_length;
|
|
|
|
p = DecodeEntry(p, limit, &shared, &non_shared, &value_length);
|
|
|
|
if (p == nullptr || key_.Size() < shared) {
|
|
|
|
CorruptionError();
|
|
|
|
return false;
|
|
|
|
} else {
|
|
|
|
if (shared == 0) {
|
|
|
|
// If this key dont share any bytes with prev key then we dont need
|
|
|
|
// to decode it and can use it's address in the block directly.
|
|
|
|
if (key_includes_seq_) {
|
|
|
|
key_.SetInternalKey(Slice(p, non_shared), false /* copy */);
|
|
|
|
} else {
|
|
|
|
key_.SetUserKey(Slice(p, non_shared), false /* copy */);
|
|
|
|
}
|
|
|
|
key_pinned_ = true;
|
|
|
|
} else {
|
|
|
|
// This key share `shared` bytes with prev key, we need to decode it
|
|
|
|
key_.TrimAppend(shared, p, non_shared);
|
|
|
|
key_pinned_ = false;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (global_seqno_ != kDisableGlobalSequenceNumber) {
|
|
|
|
// If we are reading a file with a global sequence number we should
|
|
|
|
// expect that all encoded sequence numbers are zeros and any value
|
|
|
|
// type is kTypeValue, kTypeMerge or kTypeDeletion
|
|
|
|
assert(GetInternalKeySeqno(key_.GetInternalKey()) == 0);
|
|
|
|
|
|
|
|
ValueType value_type = ExtractValueType(key_.GetInternalKey());
|
|
|
|
assert(value_type == ValueType::kTypeValue ||
|
|
|
|
value_type == ValueType::kTypeMerge ||
|
|
|
|
value_type == ValueType::kTypeDeletion);
|
|
|
|
|
|
|
|
if (key_pinned_) {
|
|
|
|
// TODO(tec): Investigate updating the seqno in the loaded block
|
|
|
|
// directly instead of doing a copy and update.
|
|
|
|
|
|
|
|
// We cannot use the key address in the block directly because
|
|
|
|
// we have a global_seqno_ that will overwrite the encoded one.
|
|
|
|
key_.OwnKey();
|
|
|
|
key_pinned_ = false;
|
|
|
|
}
|
|
|
|
|
|
|
|
key_.UpdateInternalKey(global_seqno_, value_type);
|
|
|
|
}
|
|
|
|
|
|
|
|
value_ = Slice(p + non_shared, value_length);
|
|
|
|
while (restart_index_ + 1 < num_restarts_ &&
|
|
|
|
GetRestartPoint(restart_index_ + 1) < current_) {
|
|
|
|
++restart_index_;
|
|
|
|
}
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Binary search in restart array to find the first restart point that
|
|
|
|
// is either the last restart point with a key less than target,
|
|
|
|
// which means the key of next restart point is larger than target, or
|
|
|
|
// the first restart point with a key = target
|
|
|
|
bool BlockIter::BinarySeek(const Slice& target, uint32_t left, uint32_t right,
|
|
|
|
uint32_t* index) {
|
|
|
|
assert(left <= right);
|
|
|
|
|
|
|
|
while (left < right) {
|
|
|
|
uint32_t mid = (left + right + 1) / 2;
|
|
|
|
uint32_t region_offset = GetRestartPoint(mid);
|
Block::Iter::PrefixSeek() to have an extra check to filter out some false matches
Summary:
In block based table's hash index checking, when looking for a key that doesn't exist, there is a high chance that a false block is returned because of hash bucket conflicts. In this revision, another check is done to filter out some of those cases: comparing previous key of the block boundary to see whether the target block is what we are looking for.
In a favored test setting (bloom filter disabled, 8 L0 files), I saw about 80% improvements. In a non-favored test setting (bloom filter enabled, files are all in L1, files are all cached), I see the performance penalty is less than 3%.
Test Plan: make all check
Reviewers: haobo, ljin
Reviewed By: ljin
Subscribers: wuj, leveldb, zagfox, yhchiang
Differential Revision: https://reviews.facebook.net/D20595
10 years ago
|
|
|
uint32_t shared, non_shared, value_length;
|
|
|
|
const char* key_ptr = DecodeEntry(data_ + region_offset, data_ + restarts_,
|
|
|
|
&shared, &non_shared, &value_length);
|
Block::Iter::PrefixSeek() to have an extra check to filter out some false matches
Summary:
In block based table's hash index checking, when looking for a key that doesn't exist, there is a high chance that a false block is returned because of hash bucket conflicts. In this revision, another check is done to filter out some of those cases: comparing previous key of the block boundary to see whether the target block is what we are looking for.
In a favored test setting (bloom filter disabled, 8 L0 files), I saw about 80% improvements. In a non-favored test setting (bloom filter enabled, files are all in L1, files are all cached), I see the performance penalty is less than 3%.
Test Plan: make all check
Reviewers: haobo, ljin
Reviewed By: ljin
Subscribers: wuj, leveldb, zagfox, yhchiang
Differential Revision: https://reviews.facebook.net/D20595
10 years ago
|
|
|
if (key_ptr == nullptr || (shared != 0)) {
|
|
|
|
CorruptionError();
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
Slice mid_key(key_ptr, non_shared);
|
|
|
|
int cmp = Compare(mid_key, target);
|
|
|
|
if (cmp < 0) {
|
|
|
|
// Key at "mid" is smaller than "target". Therefore all
|
|
|
|
// blocks before "mid" are uninteresting.
|
|
|
|
left = mid;
|
|
|
|
} else if (cmp > 0) {
|
|
|
|
// Key at "mid" is >= "target". Therefore all blocks at or
|
|
|
|
// after "mid" are uninteresting.
|
|
|
|
right = mid - 1;
|
|
|
|
} else {
|
|
|
|
left = right = mid;
|
Block::Iter::PrefixSeek() to have an extra check to filter out some false matches
Summary:
In block based table's hash index checking, when looking for a key that doesn't exist, there is a high chance that a false block is returned because of hash bucket conflicts. In this revision, another check is done to filter out some of those cases: comparing previous key of the block boundary to see whether the target block is what we are looking for.
In a favored test setting (bloom filter disabled, 8 L0 files), I saw about 80% improvements. In a non-favored test setting (bloom filter enabled, files are all in L1, files are all cached), I see the performance penalty is less than 3%.
Test Plan: make all check
Reviewers: haobo, ljin
Reviewed By: ljin
Subscribers: wuj, leveldb, zagfox, yhchiang
Differential Revision: https://reviews.facebook.net/D20595
10 years ago
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
*index = left;
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Compare target key and the block key of the block of `block_index`.
|
|
|
|
// Return -1 if error.
|
|
|
|
int BlockIter::CompareBlockKey(uint32_t block_index, const Slice& target) {
|
|
|
|
uint32_t region_offset = GetRestartPoint(block_index);
|
|
|
|
uint32_t shared, non_shared, value_length;
|
|
|
|
const char* key_ptr = DecodeEntry(data_ + region_offset, data_ + restarts_,
|
|
|
|
&shared, &non_shared, &value_length);
|
|
|
|
if (key_ptr == nullptr || (shared != 0)) {
|
|
|
|
CorruptionError();
|
|
|
|
return 1; // Return target is smaller
|
|
|
|
}
|
|
|
|
Slice block_key(key_ptr, non_shared);
|
|
|
|
return Compare(block_key, target);
|
|
|
|
}
|
Block::Iter::PrefixSeek() to have an extra check to filter out some false matches
Summary:
In block based table's hash index checking, when looking for a key that doesn't exist, there is a high chance that a false block is returned because of hash bucket conflicts. In this revision, another check is done to filter out some of those cases: comparing previous key of the block boundary to see whether the target block is what we are looking for.
In a favored test setting (bloom filter disabled, 8 L0 files), I saw about 80% improvements. In a non-favored test setting (bloom filter enabled, files are all in L1, files are all cached), I see the performance penalty is less than 3%.
Test Plan: make all check
Reviewers: haobo, ljin
Reviewed By: ljin
Subscribers: wuj, leveldb, zagfox, yhchiang
Differential Revision: https://reviews.facebook.net/D20595
10 years ago
|
|
|
|
|
|
|
// Binary search in block_ids to find the first block
|
|
|
|
// with a key >= target
|
|
|
|
bool BlockIter::BinaryBlockIndexSeek(const Slice& target, uint32_t* block_ids,
|
|
|
|
uint32_t left, uint32_t right,
|
|
|
|
uint32_t* index) {
|
|
|
|
assert(left <= right);
|
|
|
|
uint32_t left_bound = left;
|
|
|
|
|
|
|
|
while (left <= right) {
|
|
|
|
uint32_t mid = (right + left) / 2;
|
Block::Iter::PrefixSeek() to have an extra check to filter out some false matches
Summary:
In block based table's hash index checking, when looking for a key that doesn't exist, there is a high chance that a false block is returned because of hash bucket conflicts. In this revision, another check is done to filter out some of those cases: comparing previous key of the block boundary to see whether the target block is what we are looking for.
In a favored test setting (bloom filter disabled, 8 L0 files), I saw about 80% improvements. In a non-favored test setting (bloom filter enabled, files are all in L1, files are all cached), I see the performance penalty is less than 3%.
Test Plan: make all check
Reviewers: haobo, ljin
Reviewed By: ljin
Subscribers: wuj, leveldb, zagfox, yhchiang
Differential Revision: https://reviews.facebook.net/D20595
10 years ago
|
|
|
|
|
|
|
int cmp = CompareBlockKey(block_ids[mid], target);
|
|
|
|
if (!status_.ok()) {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
if (cmp < 0) {
|
|
|
|
// Key at "target" is larger than "mid". Therefore all
|
|
|
|
// blocks before or at "mid" are uninteresting.
|
|
|
|
left = mid + 1;
|
|
|
|
} else {
|
|
|
|
// Key at "target" is <= "mid". Therefore all blocks
|
|
|
|
// after "mid" are uninteresting.
|
|
|
|
// If there is only one block left, we found it.
|
|
|
|
if (left == right) break;
|
|
|
|
right = mid;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (left == right) {
|
|
|
|
// In one of the two following cases:
|
|
|
|
// (1) left is the first one of block_ids
|
|
|
|
// (2) there is a gap of blocks between block of `left` and `left-1`.
|
|
|
|
// we can further distinguish the case of key in the block or key not
|
|
|
|
// existing, by comparing the target key and the key of the previous
|
|
|
|
// block to the left of the block found.
|
|
|
|
if (block_ids[left] > 0 &&
|
|
|
|
(left == left_bound || block_ids[left - 1] != block_ids[left] - 1) &&
|
|
|
|
CompareBlockKey(block_ids[left] - 1, target) > 0) {
|
|
|
|
current_ = restarts_;
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
*index = block_ids[left];
|
|
|
|
return true;
|
|
|
|
} else {
|
|
|
|
assert(left > right);
|
|
|
|
// Mark iterator invalid
|
|
|
|
current_ = restarts_;
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
bool BlockIter::PrefixSeek(const Slice& target, uint32_t* index) {
|
|
|
|
assert(prefix_index_);
|
|
|
|
Slice seek_key = target;
|
|
|
|
if (!key_includes_seq_) {
|
|
|
|
seek_key = ExtractUserKey(target);
|
|
|
|
}
|
|
|
|
uint32_t* block_ids = nullptr;
|
|
|
|
uint32_t num_blocks = prefix_index_->GetBlocks(target, &block_ids);
|
|
|
|
|
|
|
|
if (num_blocks == 0) {
|
|
|
|
current_ = restarts_;
|
|
|
|
return false;
|
|
|
|
} else {
|
|
|
|
return BinaryBlockIndexSeek(seek_key, block_ids, 0, num_blocks - 1, index);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
uint32_t Block::NumRestarts() const {
|
|
|
|
assert(size_ >= 2*sizeof(uint32_t));
|
|
|
|
return DecodeFixed32(data_ + size_ - sizeof(uint32_t));
|
|
|
|
}
|
|
|
|
|
|
|
|
Block::Block(BlockContents&& contents, SequenceNumber _global_seqno,
|
|
|
|
size_t read_amp_bytes_per_bit, Statistics* statistics)
|
|
|
|
: contents_(std::move(contents)),
|
|
|
|
data_(contents_.data.data()),
|
|
|
|
size_(contents_.data.size()),
|
|
|
|
restart_offset_(0),
|
|
|
|
num_restarts_(0),
|
|
|
|
global_seqno_(_global_seqno) {
|
|
|
|
if (size_ < sizeof(uint32_t)) {
|
|
|
|
size_ = 0; // Error marker
|
|
|
|
} else {
|
|
|
|
// Should only decode restart points for uncompressed blocks
|
|
|
|
if (compression_type() == kNoCompression) {
|
|
|
|
num_restarts_ = NumRestarts();
|
|
|
|
restart_offset_ =
|
|
|
|
static_cast<uint32_t>(size_) - (1 + num_restarts_) * sizeof(uint32_t);
|
|
|
|
if (restart_offset_ > size_ - sizeof(uint32_t)) {
|
|
|
|
// The size is too small for NumRestarts() and therefore
|
|
|
|
// restart_offset_ wrapped around.
|
|
|
|
size_ = 0;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (read_amp_bytes_per_bit != 0 && statistics && size_ != 0) {
|
|
|
|
read_amp_bitmap_.reset(new BlockReadAmpBitmap(
|
|
|
|
restart_offset_, read_amp_bytes_per_bit, statistics));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
BlockIter* Block::NewIterator(const Comparator* cmp, const Comparator* ucmp,
|
|
|
|
BlockIter* iter, bool total_order_seek,
|
|
|
|
Statistics* stats, bool key_includes_seq) {
|
|
|
|
BlockIter* ret_iter;
|
|
|
|
if (iter != nullptr) {
|
|
|
|
ret_iter = iter;
|
|
|
|
} else {
|
|
|
|
ret_iter = new BlockIter;
|
|
|
|
}
|
|
|
|
if (size_ < 2*sizeof(uint32_t)) {
|
Change and clarify the relationship between Valid(), status() and Seek*() for all iterators. Also fix some bugs
Summary:
Before this PR, Iterator/InternalIterator may simultaneously have non-ok status() and Valid() = true. That state means that the last operation failed, but the iterator is nevertheless positioned on some unspecified record. Likely intended uses of that are:
* If some sst files are corrupted, a normal iterator can be used to read the data from files that are not corrupted.
* When using read_tier = kBlockCacheTier, read the data that's in block cache, skipping over the data that is not.
However, this behavior wasn't documented well (and until recently the wiki on github had misleading incorrect information). In the code there's a lot of confusion about the relationship between status() and Valid(), and about whether Seek()/SeekToLast()/etc reset the status or not. There were a number of bugs caused by this confusion, both inside rocksdb and in the code that uses rocksdb (including ours).
This PR changes the convention to:
* If status() is not ok, Valid() always returns false.
* Any seek operation resets status. (Before the PR, it depended on iterator type and on particular error.)
This does sacrifice the two use cases listed above, but siying said it's ok.
Overview of the changes:
* A commit that adds missing status checks in MergingIterator. This fixes a bug that actually affects us, and we need it fixed. `DBIteratorTest.NonBlockingIterationBugRepro` explains the scenario.
* Changes to lots of iterator types to make all of them conform to the new convention. Some bug fixes along the way. By far the biggest changes are in DBIter, which is a big messy piece of code; I tried to make it less big and messy but mostly failed.
* A stress-test for DBIter, to gain some confidence that I didn't break it. It does a few million random operations on the iterator, while occasionally modifying the underlying data (like ForwardIterator does) and occasionally returning non-ok status from internal iterator.
To find the iterator types that needed changes I searched for "public .*Iterator" in the code. Here's an overview of all 27 iterator types:
Iterators that didn't need changes:
* status() is always ok(), or Valid() is always false: MemTableIterator, ModelIter, TestIterator, KVIter (2 classes with this name anonymous namespaces), LoggingForwardVectorIterator, VectorIterator, MockTableIterator, EmptyIterator, EmptyInternalIterator.
* Thin wrappers that always pass through Valid() and status(): ArenaWrappedDBIter, TtlIterator, InternalIteratorFromIterator.
Iterators with changes (see inline comments for details):
* DBIter - an overhaul:
- It used to silently skip corrupted keys (`FindParseableKey()`), which seems dangerous. This PR makes it just stop immediately after encountering a corrupted key, just like it would for other kinds of corruption. Let me know if there was actually some deeper meaning in this behavior and I should put it back.
- It had a few code paths silently discarding subiterator's status. The stress test caught a few.
- The backwards iteration code path was expecting the internal iterator's set of keys to be immutable. It's probably always true in practice at the moment, since ForwardIterator doesn't support backwards iteration, but this PR fixes it anyway. See added DBIteratorTest.ReverseToForwardBug for an example.
- Some parts of backwards iteration code path even did things like `assert(iter_->Valid())` after a seek, which is never a safe assumption.
- It used to not reset status on seek for some types of errors.
- Some simplifications and better comments.
- Some things got more complicated from the added error handling. I'm open to ideas for how to make it nicer.
* MergingIterator - check status after every operation on every subiterator, and in some places assert that valid subiterators have ok status.
* ForwardIterator - changed to the new convention, also slightly simplified.
* ForwardLevelIterator - fixed some bugs and simplified.
* LevelIterator - simplified.
* TwoLevelIterator - changed to the new convention. Also fixed a bug that would make SeekForPrev() sometimes silently ignore errors from first_level_iter_.
* BlockBasedTableIterator - minor changes.
* BlockIter - replaced `SetStatus()` with `Invalidate()` to make sure non-ok BlockIter is always invalid.
* PlainTableIterator - some seeks used to not reset status.
* CuckooTableIterator - tiny code cleanup.
* ManagedIterator - fixed some bugs.
* BaseDeltaIterator - changed to the new convention and fixed a bug.
* BlobDBIterator - seeks used to not reset status.
* KeyConvertingIterator - some small change.
Closes https://github.com/facebook/rocksdb/pull/3810
Differential Revision: D7888019
Pulled By: al13n321
fbshipit-source-id: 4aaf6d3421c545d16722a815b2fa2e7912bc851d
7 years ago
|
|
|
ret_iter->Invalidate(Status::Corruption("bad block contents"));
|
|
|
|
return ret_iter;
|
|
|
|
}
|
|
|
|
if (num_restarts_ == 0) {
|
Change and clarify the relationship between Valid(), status() and Seek*() for all iterators. Also fix some bugs
Summary:
Before this PR, Iterator/InternalIterator may simultaneously have non-ok status() and Valid() = true. That state means that the last operation failed, but the iterator is nevertheless positioned on some unspecified record. Likely intended uses of that are:
* If some sst files are corrupted, a normal iterator can be used to read the data from files that are not corrupted.
* When using read_tier = kBlockCacheTier, read the data that's in block cache, skipping over the data that is not.
However, this behavior wasn't documented well (and until recently the wiki on github had misleading incorrect information). In the code there's a lot of confusion about the relationship between status() and Valid(), and about whether Seek()/SeekToLast()/etc reset the status or not. There were a number of bugs caused by this confusion, both inside rocksdb and in the code that uses rocksdb (including ours).
This PR changes the convention to:
* If status() is not ok, Valid() always returns false.
* Any seek operation resets status. (Before the PR, it depended on iterator type and on particular error.)
This does sacrifice the two use cases listed above, but siying said it's ok.
Overview of the changes:
* A commit that adds missing status checks in MergingIterator. This fixes a bug that actually affects us, and we need it fixed. `DBIteratorTest.NonBlockingIterationBugRepro` explains the scenario.
* Changes to lots of iterator types to make all of them conform to the new convention. Some bug fixes along the way. By far the biggest changes are in DBIter, which is a big messy piece of code; I tried to make it less big and messy but mostly failed.
* A stress-test for DBIter, to gain some confidence that I didn't break it. It does a few million random operations on the iterator, while occasionally modifying the underlying data (like ForwardIterator does) and occasionally returning non-ok status from internal iterator.
To find the iterator types that needed changes I searched for "public .*Iterator" in the code. Here's an overview of all 27 iterator types:
Iterators that didn't need changes:
* status() is always ok(), or Valid() is always false: MemTableIterator, ModelIter, TestIterator, KVIter (2 classes with this name anonymous namespaces), LoggingForwardVectorIterator, VectorIterator, MockTableIterator, EmptyIterator, EmptyInternalIterator.
* Thin wrappers that always pass through Valid() and status(): ArenaWrappedDBIter, TtlIterator, InternalIteratorFromIterator.
Iterators with changes (see inline comments for details):
* DBIter - an overhaul:
- It used to silently skip corrupted keys (`FindParseableKey()`), which seems dangerous. This PR makes it just stop immediately after encountering a corrupted key, just like it would for other kinds of corruption. Let me know if there was actually some deeper meaning in this behavior and I should put it back.
- It had a few code paths silently discarding subiterator's status. The stress test caught a few.
- The backwards iteration code path was expecting the internal iterator's set of keys to be immutable. It's probably always true in practice at the moment, since ForwardIterator doesn't support backwards iteration, but this PR fixes it anyway. See added DBIteratorTest.ReverseToForwardBug for an example.
- Some parts of backwards iteration code path even did things like `assert(iter_->Valid())` after a seek, which is never a safe assumption.
- It used to not reset status on seek for some types of errors.
- Some simplifications and better comments.
- Some things got more complicated from the added error handling. I'm open to ideas for how to make it nicer.
* MergingIterator - check status after every operation on every subiterator, and in some places assert that valid subiterators have ok status.
* ForwardIterator - changed to the new convention, also slightly simplified.
* ForwardLevelIterator - fixed some bugs and simplified.
* LevelIterator - simplified.
* TwoLevelIterator - changed to the new convention. Also fixed a bug that would make SeekForPrev() sometimes silently ignore errors from first_level_iter_.
* BlockBasedTableIterator - minor changes.
* BlockIter - replaced `SetStatus()` with `Invalidate()` to make sure non-ok BlockIter is always invalid.
* PlainTableIterator - some seeks used to not reset status.
* CuckooTableIterator - tiny code cleanup.
* ManagedIterator - fixed some bugs.
* BaseDeltaIterator - changed to the new convention and fixed a bug.
* BlobDBIterator - seeks used to not reset status.
* KeyConvertingIterator - some small change.
Closes https://github.com/facebook/rocksdb/pull/3810
Differential Revision: D7888019
Pulled By: al13n321
fbshipit-source-id: 4aaf6d3421c545d16722a815b2fa2e7912bc851d
7 years ago
|
|
|
// Empty block.
|
|
|
|
ret_iter->Invalidate(Status::OK());
|
|
|
|
return ret_iter;
|
|
|
|
} else {
|
|
|
|
BlockPrefixIndex* prefix_index_ptr =
|
|
|
|
total_order_seek ? nullptr : prefix_index_.get();
|
|
|
|
ret_iter->Initialize(cmp, ucmp, data_, restart_offset_, num_restarts_,
|
|
|
|
prefix_index_ptr, global_seqno_,
|
Copy Get() result when file reads use mmap
Summary:
For iterator reads, a `SuperVersion` is pinned to preserve a snapshot of SST files, and `Block`s are pinned to allow `key()` and `value()` to return pointers directly into a RocksDB memory region. This works for both non-mmap reads, where the block owns the memory region, and mmap reads, where the file owns the memory region.
For point reads with `PinnableSlice`, only the `Block` object is pinned. This works for non-mmap reads because the block owns the memory region, so even if the file is deleted after compaction, the memory region survives. However, for mmap reads, file deletion causes the memory region to which the `PinnableSlice` refers to be unmapped. The result is usually a segfault upon accessing the `PinnableSlice`, although sometimes it returned wrong results (I repro'd this a bunch of times with `db_stress`).
This PR copies the value into the `PinnableSlice` when it comes from mmap'd memory. We can tell whether the `Block` owns its memory using `Block::cachable()`, which is unset when reads do not use the provided buffer as is the case with mmap file reads. When that is false we ensure the result of `Get()` is copied.
This feels like a short-term solution as ideally we'd have the `PinnableSlice` pin the mmap'd memory so we can do zero-copy reads. It seemed hard so I chose this approach to fix correctness in the meantime.
Closes https://github.com/facebook/rocksdb/pull/3881
Differential Revision: D8076288
Pulled By: ajkr
fbshipit-source-id: 31d78ec010198723522323dbc6ea325122a46b08
7 years ago
|
|
|
read_amp_bitmap_.get(), key_includes_seq, cachable());
|
|
|
|
|
|
|
|
if (read_amp_bitmap_) {
|
|
|
|
if (read_amp_bitmap_->GetStatistics() != stats) {
|
|
|
|
// DB changed the Statistics pointer, we need to notify read_amp_bitmap_
|
|
|
|
read_amp_bitmap_->SetStatistics(stats);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret_iter;
|
|
|
|
}
|
|
|
|
|
|
|
|
void Block::SetBlockPrefixIndex(BlockPrefixIndex* prefix_index) {
|
|
|
|
prefix_index_.reset(prefix_index);
|
|
|
|
}
|
|
|
|
|
|
|
|
size_t Block::ApproximateMemoryUsage() const {
|
Use malloc_usable_size() for accounting block cache size
Summary:
Currently, when we insert something into block cache, we say that the block cache capacity decreased by the size of the block. However, size of the block might be less than the actual memory used by this object. For example, 4.5KB block will actually use 8KB of memory. So even if we configure block cache to 10GB, our actually memory usage of block cache will be 20GB!
This problem showed up a lot in testing and just recently also showed up in MongoRocks production where we were using 30GB more memory than expected.
This diff will fix the problem. Instead of counting the block size, we will count memory used by the block. That way, a block cache configured to be 10GB will actually use only 10GB of memory.
I'm using non-portable function and I couldn't find info on portability on Google. However, it seems to work on Linux, which will cover majority of our use-cases.
Test Plan:
1. fill up mongo instance with 80GB of data
2. restart mongo with block cache size configured to 10GB
3. do a table scan in mongo
4. memory usage before the diff: 12GB. memory usage after the diff: 10.5GB
Reviewers: sdong, MarkCallaghan, rven, yhchiang
Reviewed By: yhchiang
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D40635
10 years ago
|
|
|
size_t usage = usable_size();
|
|
|
|
#ifdef ROCKSDB_MALLOC_USABLE_SIZE
|
|
|
|
usage += malloc_usable_size((void*)this);
|
|
|
|
#else
|
|
|
|
usage += sizeof(*this);
|
|
|
|
#endif // ROCKSDB_MALLOC_USABLE_SIZE
|
|
|
|
if (prefix_index_) {
|
|
|
|
usage += prefix_index_->ApproximateMemoryUsage();
|
|
|
|
}
|
|
|
|
if (read_amp_bitmap_) {
|
|
|
|
usage += read_amp_bitmap_->ApproximateMemoryUsage();
|
|
|
|
}
|
|
|
|
return usage;
|
|
|
|
}
|
|
|
|
|
|
|
|
} // namespace rocksdb
|