2013-10-16 23:59:46 +02:00
|
|
|
// 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.
|
|
|
|
//
|
2011-03-18 23:37:00 +01:00
|
|
|
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
|
|
|
|
// Use of this source code is governed by a BSD-style license that can be
|
|
|
|
// found in the LICENSE file. See the AUTHORS file for names of contributors.
|
|
|
|
|
|
|
|
#include "db/memtable.h"
|
2013-07-23 23:42:27 +02:00
|
|
|
|
|
|
|
#include <memory>
|
|
|
|
|
2011-03-18 23:37:00 +01:00
|
|
|
#include "db/dbformat.h"
|
2013-12-03 03:34:05 +01:00
|
|
|
#include "db/merge_context.h"
|
2013-08-23 17:38:13 +02:00
|
|
|
#include "rocksdb/comparator.h"
|
|
|
|
#include "rocksdb/env.h"
|
|
|
|
#include "rocksdb/iterator.h"
|
|
|
|
#include "rocksdb/merge_operator.h"
|
2011-03-18 23:37:00 +01:00
|
|
|
#include "util/coding.h"
|
2013-08-23 08:10:02 +02:00
|
|
|
#include "util/murmurhash.h"
|
2013-11-27 20:47:40 +01:00
|
|
|
#include "util/mutexlock.h"
|
2013-11-18 20:32:54 +01:00
|
|
|
#include "util/perf_context_imp.h"
|
2013-11-22 23:14:05 +01:00
|
|
|
#include "util/statistics_imp.h"
|
2013-11-18 20:32:54 +01:00
|
|
|
#include "util/stop_watch.h"
|
2011-03-18 23:37:00 +01:00
|
|
|
|
In-place updates for equal keys and similar sized values
Summary:
Currently for each put, a fresh memory is allocated, and a new entry is added to the memtable with a new sequence number irrespective of whether the key already exists in the memtable. This diff is an attempt to update the value inplace for existing keys. It currently handles a very simple case:
1. Key already exists in the current memtable. Does not inplace update values in immutable memtable or snapshot
2. Latest value type is a 'put' ie kTypeValue
3. New value size is less than existing value, to avoid reallocating memory
TODO: For a put of an existing key, deallocate memory take by values, for other value types till a kTypeValue is found, ie. remove kTypeMerge.
TODO: Update the transaction log, to allow consistent reload of the memtable.
Test Plan: Added a unit test verifying the inplace update. But some other unit tests broken due to invalid sequence number checks. WIll fix them next.
Reviewers: xinyaohu, sumeet, haobo, dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12423
Automatic commit by arc
2013-08-19 23:12:47 +02:00
|
|
|
namespace std {
|
|
|
|
template <>
|
|
|
|
struct hash<rocksdb::Slice> {
|
|
|
|
size_t operator()(const rocksdb::Slice& slice) const {
|
|
|
|
return MurmurHash(slice.data(), slice.size(), 0);
|
|
|
|
}
|
|
|
|
};
|
|
|
|
}
|
|
|
|
|
2013-10-04 06:49:15 +02:00
|
|
|
namespace rocksdb {
|
2011-03-18 23:37:00 +01:00
|
|
|
|
2013-07-23 23:42:27 +02:00
|
|
|
MemTable::MemTable(const InternalKeyComparator& cmp,
|
2013-12-03 20:17:58 +01:00
|
|
|
MemTableRepFactory* table_factory,
|
Make arena block size configurable
Summary:
Add an option for arena block size, default value 4096 bytes. Arena will allocate blocks with such size.
I am not sure about passing parameter to skiplist in the new virtualized framework, though I talked to Jim a bit. So add Jim as reviewer.
Test Plan:
new unit test, I am running db_test.
For passing paramter from configured option to Arena, I tried tests like:
TEST(DBTest, Arena_Option) {
std::string dbname = test::TmpDir() + "/db_arena_option_test";
DestroyDB(dbname, Options());
DB* db = nullptr;
Options opts;
opts.create_if_missing = true;
opts.arena_block_size = 1000000; // tested 99, 999999
Status s = DB::Open(opts, dbname, &db);
db->Put(WriteOptions(), "a", "123");
}
and printed some debug info. The results look good. Any suggestion for such a unit-test?
Reviewers: haobo, dhruba, emayanke, jpaton
Reviewed By: dhruba
CC: leveldb, zshao
Differential Revision: https://reviews.facebook.net/D11799
2013-07-31 21:42:23 +02:00
|
|
|
int numlevel,
|
|
|
|
const Options& options)
|
2011-03-18 23:37:00 +01:00
|
|
|
: comparator_(cmp),
|
2011-05-21 04:17:43 +02:00
|
|
|
refs_(0),
|
Make arena block size configurable
Summary:
Add an option for arena block size, default value 4096 bytes. Arena will allocate blocks with such size.
I am not sure about passing parameter to skiplist in the new virtualized framework, though I talked to Jim a bit. So add Jim as reviewer.
Test Plan:
new unit test, I am running db_test.
For passing paramter from configured option to Arena, I tried tests like:
TEST(DBTest, Arena_Option) {
std::string dbname = test::TmpDir() + "/db_arena_option_test";
DestroyDB(dbname, Options());
DB* db = nullptr;
Options opts;
opts.create_if_missing = true;
opts.arena_block_size = 1000000; // tested 99, 999999
Status s = DB::Open(opts, dbname, &db);
db->Put(WriteOptions(), "a", "123");
}
and printed some debug info. The results look good. Any suggestion for such a unit-test?
Reviewers: haobo, dhruba, emayanke, jpaton
Reviewed By: dhruba
CC: leveldb, zshao
Differential Revision: https://reviews.facebook.net/D11799
2013-07-31 21:42:23 +02:00
|
|
|
arena_impl_(options.arena_block_size),
|
|
|
|
table_(table_factory->CreateMemTableRep(comparator_, &arena_impl_)),
|
2012-10-19 23:00:53 +02:00
|
|
|
flush_in_progress_(false),
|
|
|
|
flush_completed_(false),
|
2012-11-29 01:42:36 +01:00
|
|
|
file_number_(0),
|
2013-02-28 23:09:30 +01:00
|
|
|
edit_(numlevel),
|
2013-06-11 23:23:58 +02:00
|
|
|
first_seqno_(0),
|
2013-07-16 20:56:46 +02:00
|
|
|
mem_next_logfile_number_(0),
|
In-place updates for equal keys and similar sized values
Summary:
Currently for each put, a fresh memory is allocated, and a new entry is added to the memtable with a new sequence number irrespective of whether the key already exists in the memtable. This diff is an attempt to update the value inplace for existing keys. It currently handles a very simple case:
1. Key already exists in the current memtable. Does not inplace update values in immutable memtable or snapshot
2. Latest value type is a 'put' ie kTypeValue
3. New value size is less than existing value, to avoid reallocating memory
TODO: For a put of an existing key, deallocate memory take by values, for other value types till a kTypeValue is found, ie. remove kTypeMerge.
TODO: Update the transaction log, to allow consistent reload of the memtable.
Test Plan: Added a unit test verifying the inplace update. But some other unit tests broken due to invalid sequence number checks. WIll fix them next.
Reviewers: xinyaohu, sumeet, haobo, dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12423
Automatic commit by arc
2013-08-19 23:12:47 +02:00
|
|
|
mem_logfile_number_(0),
|
|
|
|
locks_(options.inplace_update_support
|
|
|
|
? options.inplace_update_num_locks
|
2013-11-27 23:27:02 +01:00
|
|
|
: 0),
|
|
|
|
prefix_extractor_(options.prefix_extractor) {
|
|
|
|
|
|
|
|
if (prefix_extractor_ && options.memtable_prefix_bloom_bits > 0) {
|
|
|
|
prefix_bloom_.reset(new DynamicBloom(options.memtable_prefix_bloom_bits,
|
|
|
|
options.memtable_prefix_bloom_probes));
|
|
|
|
}
|
|
|
|
}
|
2011-03-18 23:37:00 +01:00
|
|
|
|
|
|
|
MemTable::~MemTable() {
|
2011-05-21 04:17:43 +02:00
|
|
|
assert(refs_ == 0);
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|
|
|
|
|
2013-07-23 23:42:27 +02:00
|
|
|
size_t MemTable::ApproximateMemoryUsage() {
|
2013-08-23 08:10:02 +02:00
|
|
|
return arena_impl_.ApproximateMemoryUsage() +
|
|
|
|
table_->ApproximateMemoryUsage();
|
2013-07-23 23:42:27 +02:00
|
|
|
}
|
2011-03-18 23:37:00 +01:00
|
|
|
|
|
|
|
int MemTable::KeyComparator::operator()(const char* aptr, const char* bptr)
|
|
|
|
const {
|
|
|
|
// Internal keys are encoded as length-prefixed strings.
|
|
|
|
Slice a = GetLengthPrefixedSlice(aptr);
|
|
|
|
Slice b = GetLengthPrefixedSlice(bptr);
|
|
|
|
return comparator.Compare(a, b);
|
|
|
|
}
|
|
|
|
|
2013-08-23 08:10:02 +02:00
|
|
|
Slice MemTableRep::UserKey(const char* key) const {
|
|
|
|
Slice slice = GetLengthPrefixedSlice(key);
|
|
|
|
return Slice(slice.data(), slice.size() - 8);
|
|
|
|
}
|
|
|
|
|
2011-03-18 23:37:00 +01:00
|
|
|
// Encode a suitable internal key target for "target" and return it.
|
|
|
|
// Uses *scratch as scratch space, and the returned pointer will point
|
|
|
|
// into this scratch space.
|
2013-11-21 04:49:27 +01:00
|
|
|
const char* EncodeKey(std::string* scratch, const Slice& target) {
|
2011-03-18 23:37:00 +01:00
|
|
|
scratch->clear();
|
|
|
|
PutVarint32(scratch, target.size());
|
|
|
|
scratch->append(target.data(), target.size());
|
|
|
|
return scratch->data();
|
|
|
|
}
|
|
|
|
|
|
|
|
class MemTableIterator: public Iterator {
|
|
|
|
public:
|
2013-11-27 23:27:02 +01:00
|
|
|
MemTableIterator(const MemTable& mem, const ReadOptions& options)
|
|
|
|
: mem_(mem), iter_(), dynamic_prefix_seek_(false), valid_(false) {
|
2013-11-04 01:32:46 +01:00
|
|
|
if (options.prefix) {
|
2013-11-27 23:27:02 +01:00
|
|
|
iter_ = mem_.table_->GetPrefixIterator(*options.prefix);
|
2013-11-04 01:32:46 +01:00
|
|
|
} else if (options.prefix_seek) {
|
2013-11-27 23:27:02 +01:00
|
|
|
dynamic_prefix_seek_ = true;
|
|
|
|
iter_ = mem_.table_->GetDynamicPrefixIterator();
|
2013-11-04 01:32:46 +01:00
|
|
|
} else {
|
2013-11-27 23:27:02 +01:00
|
|
|
iter_ = mem_.table_->GetIterator();
|
2013-11-04 01:32:46 +01:00
|
|
|
}
|
|
|
|
}
|
2013-08-23 08:10:02 +02:00
|
|
|
|
2013-11-27 23:27:02 +01:00
|
|
|
virtual bool Valid() const { return valid_; }
|
|
|
|
virtual void Seek(const Slice& k) {
|
|
|
|
if (dynamic_prefix_seek_ && mem_.prefix_bloom_ &&
|
|
|
|
!mem_.prefix_bloom_->MayContain(
|
|
|
|
mem_.prefix_extractor_->Transform(ExtractUserKey(k)))) {
|
|
|
|
valid_ = false;
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
iter_->Seek(k, nullptr);
|
|
|
|
valid_ = iter_->Valid();
|
|
|
|
}
|
|
|
|
virtual void SeekToFirst() {
|
|
|
|
iter_->SeekToFirst();
|
|
|
|
valid_ = iter_->Valid();
|
|
|
|
}
|
|
|
|
virtual void SeekToLast() {
|
|
|
|
iter_->SeekToLast();
|
|
|
|
valid_ = iter_->Valid();
|
|
|
|
}
|
|
|
|
virtual void Next() {
|
|
|
|
assert(Valid());
|
|
|
|
iter_->Next();
|
|
|
|
valid_ = iter_->Valid();
|
|
|
|
}
|
|
|
|
virtual void Prev() {
|
|
|
|
assert(Valid());
|
|
|
|
iter_->Prev();
|
|
|
|
valid_ = iter_->Valid();
|
|
|
|
}
|
2013-07-23 23:42:27 +02:00
|
|
|
virtual Slice key() const {
|
2013-11-27 23:27:02 +01:00
|
|
|
assert(Valid());
|
2013-07-23 23:42:27 +02:00
|
|
|
return GetLengthPrefixedSlice(iter_->key());
|
|
|
|
}
|
2011-03-18 23:37:00 +01:00
|
|
|
virtual Slice value() const {
|
2013-11-27 23:27:02 +01:00
|
|
|
assert(Valid());
|
2013-07-23 23:42:27 +02:00
|
|
|
Slice key_slice = GetLengthPrefixedSlice(iter_->key());
|
2011-03-18 23:37:00 +01:00
|
|
|
return GetLengthPrefixedSlice(key_slice.data() + key_slice.size());
|
|
|
|
}
|
|
|
|
|
|
|
|
virtual Status status() const { return Status::OK(); }
|
|
|
|
|
|
|
|
private:
|
2013-11-27 23:27:02 +01:00
|
|
|
const MemTable& mem_;
|
2013-07-23 23:42:27 +02:00
|
|
|
std::shared_ptr<MemTableRep::Iterator> iter_;
|
2013-11-27 23:27:02 +01:00
|
|
|
bool dynamic_prefix_seek_;
|
|
|
|
bool valid_;
|
2011-03-18 23:37:00 +01:00
|
|
|
|
|
|
|
// No copying allowed
|
|
|
|
MemTableIterator(const MemTableIterator&);
|
|
|
|
void operator=(const MemTableIterator&);
|
|
|
|
};
|
|
|
|
|
2013-11-04 01:32:46 +01:00
|
|
|
Iterator* MemTable::NewIterator(const ReadOptions& options) {
|
2013-11-27 23:27:02 +01:00
|
|
|
return new MemTableIterator(*this, options);
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|
|
|
|
|
In-place updates for equal keys and similar sized values
Summary:
Currently for each put, a fresh memory is allocated, and a new entry is added to the memtable with a new sequence number irrespective of whether the key already exists in the memtable. This diff is an attempt to update the value inplace for existing keys. It currently handles a very simple case:
1. Key already exists in the current memtable. Does not inplace update values in immutable memtable or snapshot
2. Latest value type is a 'put' ie kTypeValue
3. New value size is less than existing value, to avoid reallocating memory
TODO: For a put of an existing key, deallocate memory take by values, for other value types till a kTypeValue is found, ie. remove kTypeMerge.
TODO: Update the transaction log, to allow consistent reload of the memtable.
Test Plan: Added a unit test verifying the inplace update. But some other unit tests broken due to invalid sequence number checks. WIll fix them next.
Reviewers: xinyaohu, sumeet, haobo, dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12423
Automatic commit by arc
2013-08-19 23:12:47 +02:00
|
|
|
port::RWMutex* MemTable::GetLock(const Slice& key) {
|
|
|
|
return &locks_[std::hash<Slice>()(key) % locks_.size()];
|
|
|
|
}
|
|
|
|
|
2011-03-18 23:37:00 +01:00
|
|
|
void MemTable::Add(SequenceNumber s, ValueType type,
|
2013-11-27 23:27:02 +01:00
|
|
|
const Slice& key, /* user key */
|
2011-03-18 23:37:00 +01:00
|
|
|
const Slice& value) {
|
|
|
|
// Format of an entry is concatenation of:
|
|
|
|
// key_size : varint32 of internal_key.size()
|
|
|
|
// key bytes : char[internal_key.size()]
|
|
|
|
// value_size : varint32 of value.size()
|
|
|
|
// value bytes : char[value.size()]
|
|
|
|
size_t key_size = key.size();
|
|
|
|
size_t val_size = value.size();
|
|
|
|
size_t internal_key_size = key_size + 8;
|
|
|
|
const size_t encoded_len =
|
|
|
|
VarintLength(internal_key_size) + internal_key_size +
|
|
|
|
VarintLength(val_size) + val_size;
|
Make arena block size configurable
Summary:
Add an option for arena block size, default value 4096 bytes. Arena will allocate blocks with such size.
I am not sure about passing parameter to skiplist in the new virtualized framework, though I talked to Jim a bit. So add Jim as reviewer.
Test Plan:
new unit test, I am running db_test.
For passing paramter from configured option to Arena, I tried tests like:
TEST(DBTest, Arena_Option) {
std::string dbname = test::TmpDir() + "/db_arena_option_test";
DestroyDB(dbname, Options());
DB* db = nullptr;
Options opts;
opts.create_if_missing = true;
opts.arena_block_size = 1000000; // tested 99, 999999
Status s = DB::Open(opts, dbname, &db);
db->Put(WriteOptions(), "a", "123");
}
and printed some debug info. The results look good. Any suggestion for such a unit-test?
Reviewers: haobo, dhruba, emayanke, jpaton
Reviewed By: dhruba
CC: leveldb, zshao
Differential Revision: https://reviews.facebook.net/D11799
2013-07-31 21:42:23 +02:00
|
|
|
char* buf = arena_impl_.Allocate(encoded_len);
|
2011-03-18 23:37:00 +01:00
|
|
|
char* p = EncodeVarint32(buf, internal_key_size);
|
|
|
|
memcpy(p, key.data(), key_size);
|
|
|
|
p += key_size;
|
|
|
|
EncodeFixed64(p, (s << 8) | type);
|
|
|
|
p += 8;
|
|
|
|
p = EncodeVarint32(p, val_size);
|
|
|
|
memcpy(p, value.data(), val_size);
|
2012-10-19 23:00:53 +02:00
|
|
|
assert((p + val_size) - buf == (unsigned)encoded_len);
|
2013-07-23 23:42:27 +02:00
|
|
|
table_->Insert(buf);
|
2013-02-28 23:09:30 +01:00
|
|
|
|
2013-11-27 23:27:02 +01:00
|
|
|
if (prefix_bloom_) {
|
|
|
|
assert(prefix_extractor_);
|
|
|
|
prefix_bloom_->Add(prefix_extractor_->Transform(key));
|
|
|
|
}
|
|
|
|
|
2013-02-28 23:09:30 +01:00
|
|
|
// The first sequence number inserted into the memtable
|
|
|
|
assert(first_seqno_ == 0 || s > first_seqno_);
|
|
|
|
if (first_seqno_ == 0) {
|
|
|
|
first_seqno_ = s;
|
|
|
|
}
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|
|
|
|
|
2013-03-21 23:59:47 +01:00
|
|
|
bool MemTable::Get(const LookupKey& key, std::string* value, Status* s,
|
2013-12-03 03:34:05 +01:00
|
|
|
MergeContext& merge_context, const Options& options) {
|
2013-11-18 20:32:54 +01:00
|
|
|
StopWatchNano memtable_get_timer(options.env, false);
|
|
|
|
StartPerfTimer(&memtable_get_timer);
|
|
|
|
|
2013-11-27 23:27:02 +01:00
|
|
|
Slice mem_key = key.memtable_key();
|
|
|
|
Slice user_key = key.user_key();
|
|
|
|
|
|
|
|
std::shared_ptr<MemTableRep::Iterator> iter;
|
|
|
|
if (prefix_bloom_ &&
|
|
|
|
!prefix_bloom_->MayContain(prefix_extractor_->Transform(user_key))) {
|
|
|
|
// iter is null if prefix bloom says the key does not exist
|
|
|
|
} else {
|
|
|
|
iter = table_->GetIterator(user_key);
|
|
|
|
iter->Seek(user_key, mem_key.data());
|
|
|
|
}
|
2013-03-21 23:59:47 +01:00
|
|
|
|
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences.
Summary:
Here are the major changes to the Merge Interface. It has been expanded
to handle cases where the MergeOperator is not associative. It does so by stacking
up merge operations while scanning through the key history (i.e.: during Get() or
Compaction), until a valid Put/Delete/end-of-history is encountered; it then
applies all of the merge operations in the correct sequence starting with the
base/sentinel value.
I have also introduced an "AssociativeMerge" function which allows the user to
take advantage of associative merge operations (such as in the case of counters).
The implementation will always attempt to merge the operations/operands themselves
together when they are encountered, and will resort to the "stacking" method if
and only if the "associative-merge" fails.
This implementation is conjectured to allow MergeOperator to handle the general
case, while still providing the user with the ability to take advantage of certain
efficiencies in their own merge-operator / data-structure.
NOTE: This is a preliminary diff. This must still go through a lot of review,
revision, and testing. Feedback welcome!
Test Plan:
-This is a preliminary diff. I have only just begun testing/debugging it.
-I will be testing this with the existing MergeOperator use-cases and unit-tests
(counters, string-append, and redis-lists)
-I will be "desk-checking" and walking through the code with the help gdb.
-I will find a way of stress-testing the new interface / implementation using
db_bench, db_test, merge_test, and/or db_stress.
-I will ensure that my tests cover all cases: Get-Memtable,
Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0,
Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found,
end-of-history, end-of-file, etc.
-A lot of feedback from the reviewers.
Reviewers: haobo, dhruba, zshao, emayanke
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
|
|
|
bool merge_in_progress = s->IsMergeInProgress();
|
2013-08-20 22:35:28 +02:00
|
|
|
auto merge_operator = options.merge_operator.get();
|
2013-03-21 23:59:47 +01:00
|
|
|
auto logger = options.info_log;
|
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences.
Summary:
Here are the major changes to the Merge Interface. It has been expanded
to handle cases where the MergeOperator is not associative. It does so by stacking
up merge operations while scanning through the key history (i.e.: during Get() or
Compaction), until a valid Put/Delete/end-of-history is encountered; it then
applies all of the merge operations in the correct sequence starting with the
base/sentinel value.
I have also introduced an "AssociativeMerge" function which allows the user to
take advantage of associative merge operations (such as in the case of counters).
The implementation will always attempt to merge the operations/operands themselves
together when they are encountered, and will resort to the "stacking" method if
and only if the "associative-merge" fails.
This implementation is conjectured to allow MergeOperator to handle the general
case, while still providing the user with the ability to take advantage of certain
efficiencies in their own merge-operator / data-structure.
NOTE: This is a preliminary diff. This must still go through a lot of review,
revision, and testing. Feedback welcome!
Test Plan:
-This is a preliminary diff. I have only just begun testing/debugging it.
-I will be testing this with the existing MergeOperator use-cases and unit-tests
(counters, string-append, and redis-lists)
-I will be "desk-checking" and walking through the code with the help gdb.
-I will find a way of stress-testing the new interface / implementation using
db_bench, db_test, merge_test, and/or db_stress.
-I will ensure that my tests cover all cases: Get-Memtable,
Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0,
Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found,
end-of-history, end-of-file, etc.
-A lot of feedback from the reviewers.
Reviewers: haobo, dhruba, zshao, emayanke
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
|
|
|
std::string merge_result;
|
|
|
|
|
2013-11-18 20:32:54 +01:00
|
|
|
bool found_final_value = false;
|
2013-11-27 23:27:02 +01:00
|
|
|
for (; !found_final_value && iter && iter->Valid(); iter->Next()) {
|
2011-06-22 04:36:45 +02:00
|
|
|
// entry format is:
|
|
|
|
// klength varint32
|
2013-01-04 02:13:56 +01:00
|
|
|
// userkey char[klength-8]
|
2011-06-22 04:36:45 +02:00
|
|
|
// tag uint64
|
|
|
|
// vlength varint32
|
|
|
|
// value char[vlength]
|
|
|
|
// Check that it belongs to same user key. We do not check the
|
|
|
|
// sequence number since the Seek() call above should have skipped
|
|
|
|
// all entries with overly large sequence numbers.
|
2013-07-23 23:42:27 +02:00
|
|
|
const char* entry = iter->key();
|
2011-06-22 04:36:45 +02:00
|
|
|
uint32_t key_length;
|
In-place updates for equal keys and similar sized values
Summary:
Currently for each put, a fresh memory is allocated, and a new entry is added to the memtable with a new sequence number irrespective of whether the key already exists in the memtable. This diff is an attempt to update the value inplace for existing keys. It currently handles a very simple case:
1. Key already exists in the current memtable. Does not inplace update values in immutable memtable or snapshot
2. Latest value type is a 'put' ie kTypeValue
3. New value size is less than existing value, to avoid reallocating memory
TODO: For a put of an existing key, deallocate memory take by values, for other value types till a kTypeValue is found, ie. remove kTypeMerge.
TODO: Update the transaction log, to allow consistent reload of the memtable.
Test Plan: Added a unit test verifying the inplace update. But some other unit tests broken due to invalid sequence number checks. WIll fix them next.
Reviewers: xinyaohu, sumeet, haobo, dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12423
Automatic commit by arc
2013-08-19 23:12:47 +02:00
|
|
|
const char* key_ptr = GetVarint32Ptr(entry, entry + 5, &key_length);
|
2011-06-22 04:36:45 +02:00
|
|
|
if (comparator_.comparator.user_comparator()->Compare(
|
In-place updates for equal keys and similar sized values
Summary:
Currently for each put, a fresh memory is allocated, and a new entry is added to the memtable with a new sequence number irrespective of whether the key already exists in the memtable. This diff is an attempt to update the value inplace for existing keys. It currently handles a very simple case:
1. Key already exists in the current memtable. Does not inplace update values in immutable memtable or snapshot
2. Latest value type is a 'put' ie kTypeValue
3. New value size is less than existing value, to avoid reallocating memory
TODO: For a put of an existing key, deallocate memory take by values, for other value types till a kTypeValue is found, ie. remove kTypeMerge.
TODO: Update the transaction log, to allow consistent reload of the memtable.
Test Plan: Added a unit test verifying the inplace update. But some other unit tests broken due to invalid sequence number checks. WIll fix them next.
Reviewers: xinyaohu, sumeet, haobo, dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12423
Automatic commit by arc
2013-08-19 23:12:47 +02:00
|
|
|
Slice(key_ptr, key_length - 8), key.user_key()) == 0) {
|
2011-06-22 04:36:45 +02:00
|
|
|
// Correct user key
|
|
|
|
const uint64_t tag = DecodeFixed64(key_ptr + key_length - 8);
|
|
|
|
switch (static_cast<ValueType>(tag & 0xff)) {
|
|
|
|
case kTypeValue: {
|
In-place updates for equal keys and similar sized values
Summary:
Currently for each put, a fresh memory is allocated, and a new entry is added to the memtable with a new sequence number irrespective of whether the key already exists in the memtable. This diff is an attempt to update the value inplace for existing keys. It currently handles a very simple case:
1. Key already exists in the current memtable. Does not inplace update values in immutable memtable or snapshot
2. Latest value type is a 'put' ie kTypeValue
3. New value size is less than existing value, to avoid reallocating memory
TODO: For a put of an existing key, deallocate memory take by values, for other value types till a kTypeValue is found, ie. remove kTypeMerge.
TODO: Update the transaction log, to allow consistent reload of the memtable.
Test Plan: Added a unit test verifying the inplace update. But some other unit tests broken due to invalid sequence number checks. WIll fix them next.
Reviewers: xinyaohu, sumeet, haobo, dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12423
Automatic commit by arc
2013-08-19 23:12:47 +02:00
|
|
|
if (options.inplace_update_support) {
|
|
|
|
GetLock(key.user_key())->ReadLock();
|
|
|
|
}
|
2011-06-22 04:36:45 +02:00
|
|
|
Slice v = GetLengthPrefixedSlice(key_ptr + key_length);
|
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences.
Summary:
Here are the major changes to the Merge Interface. It has been expanded
to handle cases where the MergeOperator is not associative. It does so by stacking
up merge operations while scanning through the key history (i.e.: during Get() or
Compaction), until a valid Put/Delete/end-of-history is encountered; it then
applies all of the merge operations in the correct sequence starting with the
base/sentinel value.
I have also introduced an "AssociativeMerge" function which allows the user to
take advantage of associative merge operations (such as in the case of counters).
The implementation will always attempt to merge the operations/operands themselves
together when they are encountered, and will resort to the "stacking" method if
and only if the "associative-merge" fails.
This implementation is conjectured to allow MergeOperator to handle the general
case, while still providing the user with the ability to take advantage of certain
efficiencies in their own merge-operator / data-structure.
NOTE: This is a preliminary diff. This must still go through a lot of review,
revision, and testing. Feedback welcome!
Test Plan:
-This is a preliminary diff. I have only just begun testing/debugging it.
-I will be testing this with the existing MergeOperator use-cases and unit-tests
(counters, string-append, and redis-lists)
-I will be "desk-checking" and walking through the code with the help gdb.
-I will find a way of stress-testing the new interface / implementation using
db_bench, db_test, merge_test, and/or db_stress.
-I will ensure that my tests cover all cases: Get-Memtable,
Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0,
Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found,
end-of-history, end-of-file, etc.
-A lot of feedback from the reviewers.
Reviewers: haobo, dhruba, zshao, emayanke
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
|
|
|
*s = Status::OK();
|
2013-03-21 23:59:47 +01:00
|
|
|
if (merge_in_progress) {
|
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences.
Summary:
Here are the major changes to the Merge Interface. It has been expanded
to handle cases where the MergeOperator is not associative. It does so by stacking
up merge operations while scanning through the key history (i.e.: during Get() or
Compaction), until a valid Put/Delete/end-of-history is encountered; it then
applies all of the merge operations in the correct sequence starting with the
base/sentinel value.
I have also introduced an "AssociativeMerge" function which allows the user to
take advantage of associative merge operations (such as in the case of counters).
The implementation will always attempt to merge the operations/operands themselves
together when they are encountered, and will resort to the "stacking" method if
and only if the "associative-merge" fails.
This implementation is conjectured to allow MergeOperator to handle the general
case, while still providing the user with the ability to take advantage of certain
efficiencies in their own merge-operator / data-structure.
NOTE: This is a preliminary diff. This must still go through a lot of review,
revision, and testing. Feedback welcome!
Test Plan:
-This is a preliminary diff. I have only just begun testing/debugging it.
-I will be testing this with the existing MergeOperator use-cases and unit-tests
(counters, string-append, and redis-lists)
-I will be "desk-checking" and walking through the code with the help gdb.
-I will find a way of stress-testing the new interface / implementation using
db_bench, db_test, merge_test, and/or db_stress.
-I will ensure that my tests cover all cases: Get-Memtable,
Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0,
Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found,
end-of-history, end-of-file, etc.
-A lot of feedback from the reviewers.
Reviewers: haobo, dhruba, zshao, emayanke
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
|
|
|
assert(merge_operator);
|
2013-12-03 03:34:05 +01:00
|
|
|
if (!merge_operator->FullMerge(key.user_key(), &v,
|
|
|
|
merge_context.GetOperands(), value,
|
|
|
|
logger.get())) {
|
2013-11-22 23:14:05 +01:00
|
|
|
RecordTick(options.statistics.get(), NUMBER_MERGE_FAILURES);
|
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences.
Summary:
Here are the major changes to the Merge Interface. It has been expanded
to handle cases where the MergeOperator is not associative. It does so by stacking
up merge operations while scanning through the key history (i.e.: during Get() or
Compaction), until a valid Put/Delete/end-of-history is encountered; it then
applies all of the merge operations in the correct sequence starting with the
base/sentinel value.
I have also introduced an "AssociativeMerge" function which allows the user to
take advantage of associative merge operations (such as in the case of counters).
The implementation will always attempt to merge the operations/operands themselves
together when they are encountered, and will resort to the "stacking" method if
and only if the "associative-merge" fails.
This implementation is conjectured to allow MergeOperator to handle the general
case, while still providing the user with the ability to take advantage of certain
efficiencies in their own merge-operator / data-structure.
NOTE: This is a preliminary diff. This must still go through a lot of review,
revision, and testing. Feedback welcome!
Test Plan:
-This is a preliminary diff. I have only just begun testing/debugging it.
-I will be testing this with the existing MergeOperator use-cases and unit-tests
(counters, string-append, and redis-lists)
-I will be "desk-checking" and walking through the code with the help gdb.
-I will find a way of stress-testing the new interface / implementation using
db_bench, db_test, merge_test, and/or db_stress.
-I will ensure that my tests cover all cases: Get-Memtable,
Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0,
Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found,
end-of-history, end-of-file, etc.
-A lot of feedback from the reviewers.
Reviewers: haobo, dhruba, zshao, emayanke
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
|
|
|
*s = Status::Corruption("Error: Could not perform merge.");
|
|
|
|
}
|
2013-03-21 23:59:47 +01:00
|
|
|
} else {
|
|
|
|
value->assign(v.data(), v.size());
|
|
|
|
}
|
In-place updates for equal keys and similar sized values
Summary:
Currently for each put, a fresh memory is allocated, and a new entry is added to the memtable with a new sequence number irrespective of whether the key already exists in the memtable. This diff is an attempt to update the value inplace for existing keys. It currently handles a very simple case:
1. Key already exists in the current memtable. Does not inplace update values in immutable memtable or snapshot
2. Latest value type is a 'put' ie kTypeValue
3. New value size is less than existing value, to avoid reallocating memory
TODO: For a put of an existing key, deallocate memory take by values, for other value types till a kTypeValue is found, ie. remove kTypeMerge.
TODO: Update the transaction log, to allow consistent reload of the memtable.
Test Plan: Added a unit test verifying the inplace update. But some other unit tests broken due to invalid sequence number checks. WIll fix them next.
Reviewers: xinyaohu, sumeet, haobo, dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12423
Automatic commit by arc
2013-08-19 23:12:47 +02:00
|
|
|
if (options.inplace_update_support) {
|
|
|
|
GetLock(key.user_key())->Unlock();
|
|
|
|
}
|
2013-11-18 20:32:54 +01:00
|
|
|
found_final_value = true;
|
|
|
|
break;
|
2011-06-22 04:36:45 +02:00
|
|
|
}
|
2013-03-21 23:59:47 +01:00
|
|
|
case kTypeDeletion: {
|
|
|
|
if (merge_in_progress) {
|
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences.
Summary:
Here are the major changes to the Merge Interface. It has been expanded
to handle cases where the MergeOperator is not associative. It does so by stacking
up merge operations while scanning through the key history (i.e.: during Get() or
Compaction), until a valid Put/Delete/end-of-history is encountered; it then
applies all of the merge operations in the correct sequence starting with the
base/sentinel value.
I have also introduced an "AssociativeMerge" function which allows the user to
take advantage of associative merge operations (such as in the case of counters).
The implementation will always attempt to merge the operations/operands themselves
together when they are encountered, and will resort to the "stacking" method if
and only if the "associative-merge" fails.
This implementation is conjectured to allow MergeOperator to handle the general
case, while still providing the user with the ability to take advantage of certain
efficiencies in their own merge-operator / data-structure.
NOTE: This is a preliminary diff. This must still go through a lot of review,
revision, and testing. Feedback welcome!
Test Plan:
-This is a preliminary diff. I have only just begun testing/debugging it.
-I will be testing this with the existing MergeOperator use-cases and unit-tests
(counters, string-append, and redis-lists)
-I will be "desk-checking" and walking through the code with the help gdb.
-I will find a way of stress-testing the new interface / implementation using
db_bench, db_test, merge_test, and/or db_stress.
-I will ensure that my tests cover all cases: Get-Memtable,
Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0,
Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found,
end-of-history, end-of-file, etc.
-A lot of feedback from the reviewers.
Reviewers: haobo, dhruba, zshao, emayanke
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
|
|
|
assert(merge_operator);
|
|
|
|
*s = Status::OK();
|
2013-12-03 03:34:05 +01:00
|
|
|
if (!merge_operator->FullMerge(key.user_key(), nullptr,
|
|
|
|
merge_context.GetOperands(), value,
|
|
|
|
logger.get())) {
|
2013-11-22 23:14:05 +01:00
|
|
|
RecordTick(options.statistics.get(), NUMBER_MERGE_FAILURES);
|
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences.
Summary:
Here are the major changes to the Merge Interface. It has been expanded
to handle cases where the MergeOperator is not associative. It does so by stacking
up merge operations while scanning through the key history (i.e.: during Get() or
Compaction), until a valid Put/Delete/end-of-history is encountered; it then
applies all of the merge operations in the correct sequence starting with the
base/sentinel value.
I have also introduced an "AssociativeMerge" function which allows the user to
take advantage of associative merge operations (such as in the case of counters).
The implementation will always attempt to merge the operations/operands themselves
together when they are encountered, and will resort to the "stacking" method if
and only if the "associative-merge" fails.
This implementation is conjectured to allow MergeOperator to handle the general
case, while still providing the user with the ability to take advantage of certain
efficiencies in their own merge-operator / data-structure.
NOTE: This is a preliminary diff. This must still go through a lot of review,
revision, and testing. Feedback welcome!
Test Plan:
-This is a preliminary diff. I have only just begun testing/debugging it.
-I will be testing this with the existing MergeOperator use-cases and unit-tests
(counters, string-append, and redis-lists)
-I will be "desk-checking" and walking through the code with the help gdb.
-I will find a way of stress-testing the new interface / implementation using
db_bench, db_test, merge_test, and/or db_stress.
-I will ensure that my tests cover all cases: Get-Memtable,
Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0,
Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found,
end-of-history, end-of-file, etc.
-A lot of feedback from the reviewers.
Reviewers: haobo, dhruba, zshao, emayanke
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
|
|
|
*s = Status::Corruption("Error: Could not perform merge.");
|
|
|
|
}
|
2013-03-21 23:59:47 +01:00
|
|
|
} else {
|
2013-12-26 22:49:04 +01:00
|
|
|
*s = Status::NotFound();
|
2013-03-21 23:59:47 +01:00
|
|
|
}
|
2013-11-18 20:32:54 +01:00
|
|
|
found_final_value = true;
|
|
|
|
break;
|
2013-03-21 23:59:47 +01:00
|
|
|
}
|
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences.
Summary:
Here are the major changes to the Merge Interface. It has been expanded
to handle cases where the MergeOperator is not associative. It does so by stacking
up merge operations while scanning through the key history (i.e.: during Get() or
Compaction), until a valid Put/Delete/end-of-history is encountered; it then
applies all of the merge operations in the correct sequence starting with the
base/sentinel value.
I have also introduced an "AssociativeMerge" function which allows the user to
take advantage of associative merge operations (such as in the case of counters).
The implementation will always attempt to merge the operations/operands themselves
together when they are encountered, and will resort to the "stacking" method if
and only if the "associative-merge" fails.
This implementation is conjectured to allow MergeOperator to handle the general
case, while still providing the user with the ability to take advantage of certain
efficiencies in their own merge-operator / data-structure.
NOTE: This is a preliminary diff. This must still go through a lot of review,
revision, and testing. Feedback welcome!
Test Plan:
-This is a preliminary diff. I have only just begun testing/debugging it.
-I will be testing this with the existing MergeOperator use-cases and unit-tests
(counters, string-append, and redis-lists)
-I will be "desk-checking" and walking through the code with the help gdb.
-I will find a way of stress-testing the new interface / implementation using
db_bench, db_test, merge_test, and/or db_stress.
-I will ensure that my tests cover all cases: Get-Memtable,
Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0,
Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found,
end-of-history, end-of-file, etc.
-A lot of feedback from the reviewers.
Reviewers: haobo, dhruba, zshao, emayanke
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
|
|
|
case kTypeMerge: {
|
|
|
|
Slice v = GetLengthPrefixedSlice(key_ptr + key_length);
|
|
|
|
merge_in_progress = true;
|
2013-12-03 03:34:05 +01:00
|
|
|
merge_context.PushOperand(v);
|
|
|
|
while(merge_context.GetNumOperands() >= 2) {
|
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences.
Summary:
Here are the major changes to the Merge Interface. It has been expanded
to handle cases where the MergeOperator is not associative. It does so by stacking
up merge operations while scanning through the key history (i.e.: during Get() or
Compaction), until a valid Put/Delete/end-of-history is encountered; it then
applies all of the merge operations in the correct sequence starting with the
base/sentinel value.
I have also introduced an "AssociativeMerge" function which allows the user to
take advantage of associative merge operations (such as in the case of counters).
The implementation will always attempt to merge the operations/operands themselves
together when they are encountered, and will resort to the "stacking" method if
and only if the "associative-merge" fails.
This implementation is conjectured to allow MergeOperator to handle the general
case, while still providing the user with the ability to take advantage of certain
efficiencies in their own merge-operator / data-structure.
NOTE: This is a preliminary diff. This must still go through a lot of review,
revision, and testing. Feedback welcome!
Test Plan:
-This is a preliminary diff. I have only just begun testing/debugging it.
-I will be testing this with the existing MergeOperator use-cases and unit-tests
(counters, string-append, and redis-lists)
-I will be "desk-checking" and walking through the code with the help gdb.
-I will find a way of stress-testing the new interface / implementation using
db_bench, db_test, merge_test, and/or db_stress.
-I will ensure that my tests cover all cases: Get-Memtable,
Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0,
Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found,
end-of-history, end-of-file, etc.
-A lot of feedback from the reviewers.
Reviewers: haobo, dhruba, zshao, emayanke
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
|
|
|
// Attempt to associative merge. (Returns true if successful)
|
2013-12-03 03:34:05 +01:00
|
|
|
if (merge_operator->PartialMerge(key.user_key(),
|
|
|
|
merge_context.GetOperand(0),
|
|
|
|
merge_context.GetOperand(1),
|
|
|
|
&merge_result, logger.get())) {
|
|
|
|
merge_context.PushPartialMergeResult(merge_result);
|
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences.
Summary:
Here are the major changes to the Merge Interface. It has been expanded
to handle cases where the MergeOperator is not associative. It does so by stacking
up merge operations while scanning through the key history (i.e.: during Get() or
Compaction), until a valid Put/Delete/end-of-history is encountered; it then
applies all of the merge operations in the correct sequence starting with the
base/sentinel value.
I have also introduced an "AssociativeMerge" function which allows the user to
take advantage of associative merge operations (such as in the case of counters).
The implementation will always attempt to merge the operations/operands themselves
together when they are encountered, and will resort to the "stacking" method if
and only if the "associative-merge" fails.
This implementation is conjectured to allow MergeOperator to handle the general
case, while still providing the user with the ability to take advantage of certain
efficiencies in their own merge-operator / data-structure.
NOTE: This is a preliminary diff. This must still go through a lot of review,
revision, and testing. Feedback welcome!
Test Plan:
-This is a preliminary diff. I have only just begun testing/debugging it.
-I will be testing this with the existing MergeOperator use-cases and unit-tests
(counters, string-append, and redis-lists)
-I will be "desk-checking" and walking through the code with the help gdb.
-I will find a way of stress-testing the new interface / implementation using
db_bench, db_test, merge_test, and/or db_stress.
-I will ensure that my tests cover all cases: Get-Memtable,
Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0,
Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found,
end-of-history, end-of-file, etc.
-A lot of feedback from the reviewers.
Reviewers: haobo, dhruba, zshao, emayanke
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
|
|
|
} else {
|
|
|
|
// Stack them because user can't associative merge
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
2014-01-14 16:55:16 +01:00
|
|
|
default:
|
2013-08-15 01:32:46 +02:00
|
|
|
assert(false);
|
|
|
|
break;
|
2011-06-22 04:36:45 +02:00
|
|
|
}
|
2013-05-21 06:28:09 +02:00
|
|
|
} else {
|
|
|
|
// exit loop if user key does not match
|
|
|
|
break;
|
2011-06-22 04:36:45 +02:00
|
|
|
}
|
|
|
|
}
|
2013-03-21 23:59:47 +01:00
|
|
|
|
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences.
Summary:
Here are the major changes to the Merge Interface. It has been expanded
to handle cases where the MergeOperator is not associative. It does so by stacking
up merge operations while scanning through the key history (i.e.: during Get() or
Compaction), until a valid Put/Delete/end-of-history is encountered; it then
applies all of the merge operations in the correct sequence starting with the
base/sentinel value.
I have also introduced an "AssociativeMerge" function which allows the user to
take advantage of associative merge operations (such as in the case of counters).
The implementation will always attempt to merge the operations/operands themselves
together when they are encountered, and will resort to the "stacking" method if
and only if the "associative-merge" fails.
This implementation is conjectured to allow MergeOperator to handle the general
case, while still providing the user with the ability to take advantage of certain
efficiencies in their own merge-operator / data-structure.
NOTE: This is a preliminary diff. This must still go through a lot of review,
revision, and testing. Feedback welcome!
Test Plan:
-This is a preliminary diff. I have only just begun testing/debugging it.
-I will be testing this with the existing MergeOperator use-cases and unit-tests
(counters, string-append, and redis-lists)
-I will be "desk-checking" and walking through the code with the help gdb.
-I will find a way of stress-testing the new interface / implementation using
db_bench, db_test, merge_test, and/or db_stress.
-I will ensure that my tests cover all cases: Get-Memtable,
Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0,
Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found,
end-of-history, end-of-file, etc.
-A lot of feedback from the reviewers.
Reviewers: haobo, dhruba, zshao, emayanke
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
|
|
|
// No change to value, since we have not yet found a Put/Delete
|
|
|
|
|
2013-11-18 20:32:54 +01:00
|
|
|
if (!found_final_value && merge_in_progress) {
|
2013-03-21 23:59:47 +01:00
|
|
|
*s = Status::MergeInProgress("");
|
|
|
|
}
|
2013-11-18 20:32:54 +01:00
|
|
|
BumpPerfTime(&perf_context.get_from_memtable_time, &memtable_get_timer);
|
|
|
|
BumpPerfCount(&perf_context.get_from_memtable_count);
|
|
|
|
return found_final_value;
|
2011-06-22 04:36:45 +02:00
|
|
|
}
|
|
|
|
|
2014-01-14 16:55:16 +01:00
|
|
|
void MemTable::Update(SequenceNumber seq,
|
In-place updates for equal keys and similar sized values
Summary:
Currently for each put, a fresh memory is allocated, and a new entry is added to the memtable with a new sequence number irrespective of whether the key already exists in the memtable. This diff is an attempt to update the value inplace for existing keys. It currently handles a very simple case:
1. Key already exists in the current memtable. Does not inplace update values in immutable memtable or snapshot
2. Latest value type is a 'put' ie kTypeValue
3. New value size is less than existing value, to avoid reallocating memory
TODO: For a put of an existing key, deallocate memory take by values, for other value types till a kTypeValue is found, ie. remove kTypeMerge.
TODO: Update the transaction log, to allow consistent reload of the memtable.
Test Plan: Added a unit test verifying the inplace update. But some other unit tests broken due to invalid sequence number checks. WIll fix them next.
Reviewers: xinyaohu, sumeet, haobo, dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12423
Automatic commit by arc
2013-08-19 23:12:47 +02:00
|
|
|
const Slice& key,
|
|
|
|
const Slice& value) {
|
|
|
|
LookupKey lkey(key, seq);
|
2013-11-27 23:27:02 +01:00
|
|
|
Slice mem_key = lkey.memtable_key();
|
In-place updates for equal keys and similar sized values
Summary:
Currently for each put, a fresh memory is allocated, and a new entry is added to the memtable with a new sequence number irrespective of whether the key already exists in the memtable. This diff is an attempt to update the value inplace for existing keys. It currently handles a very simple case:
1. Key already exists in the current memtable. Does not inplace update values in immutable memtable or snapshot
2. Latest value type is a 'put' ie kTypeValue
3. New value size is less than existing value, to avoid reallocating memory
TODO: For a put of an existing key, deallocate memory take by values, for other value types till a kTypeValue is found, ie. remove kTypeMerge.
TODO: Update the transaction log, to allow consistent reload of the memtable.
Test Plan: Added a unit test verifying the inplace update. But some other unit tests broken due to invalid sequence number checks. WIll fix them next.
Reviewers: xinyaohu, sumeet, haobo, dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12423
Automatic commit by arc
2013-08-19 23:12:47 +02:00
|
|
|
|
|
|
|
std::shared_ptr<MemTableRep::Iterator> iter(
|
2013-12-03 20:17:58 +01:00
|
|
|
table_->GetIterator(lkey.user_key()));
|
2013-11-27 23:27:02 +01:00
|
|
|
iter->Seek(key, mem_key.data());
|
|
|
|
|
In-place updates for equal keys and similar sized values
Summary:
Currently for each put, a fresh memory is allocated, and a new entry is added to the memtable with a new sequence number irrespective of whether the key already exists in the memtable. This diff is an attempt to update the value inplace for existing keys. It currently handles a very simple case:
1. Key already exists in the current memtable. Does not inplace update values in immutable memtable or snapshot
2. Latest value type is a 'put' ie kTypeValue
3. New value size is less than existing value, to avoid reallocating memory
TODO: For a put of an existing key, deallocate memory take by values, for other value types till a kTypeValue is found, ie. remove kTypeMerge.
TODO: Update the transaction log, to allow consistent reload of the memtable.
Test Plan: Added a unit test verifying the inplace update. But some other unit tests broken due to invalid sequence number checks. WIll fix them next.
Reviewers: xinyaohu, sumeet, haobo, dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12423
Automatic commit by arc
2013-08-19 23:12:47 +02:00
|
|
|
|
|
|
|
if (iter->Valid()) {
|
|
|
|
// entry format is:
|
2014-01-14 16:55:16 +01:00
|
|
|
// key_length varint32
|
In-place updates for equal keys and similar sized values
Summary:
Currently for each put, a fresh memory is allocated, and a new entry is added to the memtable with a new sequence number irrespective of whether the key already exists in the memtable. This diff is an attempt to update the value inplace for existing keys. It currently handles a very simple case:
1. Key already exists in the current memtable. Does not inplace update values in immutable memtable or snapshot
2. Latest value type is a 'put' ie kTypeValue
3. New value size is less than existing value, to avoid reallocating memory
TODO: For a put of an existing key, deallocate memory take by values, for other value types till a kTypeValue is found, ie. remove kTypeMerge.
TODO: Update the transaction log, to allow consistent reload of the memtable.
Test Plan: Added a unit test verifying the inplace update. But some other unit tests broken due to invalid sequence number checks. WIll fix them next.
Reviewers: xinyaohu, sumeet, haobo, dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12423
Automatic commit by arc
2013-08-19 23:12:47 +02:00
|
|
|
// userkey char[klength-8]
|
|
|
|
// tag uint64
|
|
|
|
// vlength varint32
|
|
|
|
// value char[vlength]
|
|
|
|
// Check that it belongs to same user key. We do not check the
|
|
|
|
// sequence number since the Seek() call above should have skipped
|
|
|
|
// all entries with overly large sequence numbers.
|
|
|
|
const char* entry = iter->key();
|
|
|
|
uint32_t key_length;
|
|
|
|
const char* key_ptr = GetVarint32Ptr(entry, entry + 5, &key_length);
|
|
|
|
if (comparator_.comparator.user_comparator()->Compare(
|
|
|
|
Slice(key_ptr, key_length - 8), lkey.user_key()) == 0) {
|
|
|
|
// Correct user key
|
|
|
|
const uint64_t tag = DecodeFixed64(key_ptr + key_length - 8);
|
|
|
|
switch (static_cast<ValueType>(tag & 0xff)) {
|
|
|
|
case kTypeValue: {
|
2014-01-14 16:55:16 +01:00
|
|
|
Slice prev_value = GetLengthPrefixedSlice(key_ptr + key_length);
|
|
|
|
uint32_t prev_value_size = prev_value.size();
|
|
|
|
uint32_t new_value_size = value.size();
|
|
|
|
|
In-place updates for equal keys and similar sized values
Summary:
Currently for each put, a fresh memory is allocated, and a new entry is added to the memtable with a new sequence number irrespective of whether the key already exists in the memtable. This diff is an attempt to update the value inplace for existing keys. It currently handles a very simple case:
1. Key already exists in the current memtable. Does not inplace update values in immutable memtable or snapshot
2. Latest value type is a 'put' ie kTypeValue
3. New value size is less than existing value, to avoid reallocating memory
TODO: For a put of an existing key, deallocate memory take by values, for other value types till a kTypeValue is found, ie. remove kTypeMerge.
TODO: Update the transaction log, to allow consistent reload of the memtable.
Test Plan: Added a unit test verifying the inplace update. But some other unit tests broken due to invalid sequence number checks. WIll fix them next.
Reviewers: xinyaohu, sumeet, haobo, dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12423
Automatic commit by arc
2013-08-19 23:12:47 +02:00
|
|
|
// Update value, if newValue size <= curValue size
|
2014-01-14 16:55:16 +01:00
|
|
|
if (new_value_size <= prev_value_size ) {
|
In-place updates for equal keys and similar sized values
Summary:
Currently for each put, a fresh memory is allocated, and a new entry is added to the memtable with a new sequence number irrespective of whether the key already exists in the memtable. This diff is an attempt to update the value inplace for existing keys. It currently handles a very simple case:
1. Key already exists in the current memtable. Does not inplace update values in immutable memtable or snapshot
2. Latest value type is a 'put' ie kTypeValue
3. New value size is less than existing value, to avoid reallocating memory
TODO: For a put of an existing key, deallocate memory take by values, for other value types till a kTypeValue is found, ie. remove kTypeMerge.
TODO: Update the transaction log, to allow consistent reload of the memtable.
Test Plan: Added a unit test verifying the inplace update. But some other unit tests broken due to invalid sequence number checks. WIll fix them next.
Reviewers: xinyaohu, sumeet, haobo, dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12423
Automatic commit by arc
2013-08-19 23:12:47 +02:00
|
|
|
char* p = EncodeVarint32(const_cast<char*>(key_ptr) + key_length,
|
2014-01-14 16:55:16 +01:00
|
|
|
new_value_size);
|
In-place updates for equal keys and similar sized values
Summary:
Currently for each put, a fresh memory is allocated, and a new entry is added to the memtable with a new sequence number irrespective of whether the key already exists in the memtable. This diff is an attempt to update the value inplace for existing keys. It currently handles a very simple case:
1. Key already exists in the current memtable. Does not inplace update values in immutable memtable or snapshot
2. Latest value type is a 'put' ie kTypeValue
3. New value size is less than existing value, to avoid reallocating memory
TODO: For a put of an existing key, deallocate memory take by values, for other value types till a kTypeValue is found, ie. remove kTypeMerge.
TODO: Update the transaction log, to allow consistent reload of the memtable.
Test Plan: Added a unit test verifying the inplace update. But some other unit tests broken due to invalid sequence number checks. WIll fix them next.
Reviewers: xinyaohu, sumeet, haobo, dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12423
Automatic commit by arc
2013-08-19 23:12:47 +02:00
|
|
|
WriteLock wl(GetLock(lkey.user_key()));
|
2014-01-14 16:55:16 +01:00
|
|
|
memcpy(p, value.data(), new_value_size);
|
In-place updates for equal keys and similar sized values
Summary:
Currently for each put, a fresh memory is allocated, and a new entry is added to the memtable with a new sequence number irrespective of whether the key already exists in the memtable. This diff is an attempt to update the value inplace for existing keys. It currently handles a very simple case:
1. Key already exists in the current memtable. Does not inplace update values in immutable memtable or snapshot
2. Latest value type is a 'put' ie kTypeValue
3. New value size is less than existing value, to avoid reallocating memory
TODO: For a put of an existing key, deallocate memory take by values, for other value types till a kTypeValue is found, ie. remove kTypeMerge.
TODO: Update the transaction log, to allow consistent reload of the memtable.
Test Plan: Added a unit test verifying the inplace update. But some other unit tests broken due to invalid sequence number checks. WIll fix them next.
Reviewers: xinyaohu, sumeet, haobo, dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12423
Automatic commit by arc
2013-08-19 23:12:47 +02:00
|
|
|
assert(
|
2014-01-14 16:55:16 +01:00
|
|
|
(p + new_value_size) - entry ==
|
In-place updates for equal keys and similar sized values
Summary:
Currently for each put, a fresh memory is allocated, and a new entry is added to the memtable with a new sequence number irrespective of whether the key already exists in the memtable. This diff is an attempt to update the value inplace for existing keys. It currently handles a very simple case:
1. Key already exists in the current memtable. Does not inplace update values in immutable memtable or snapshot
2. Latest value type is a 'put' ie kTypeValue
3. New value size is less than existing value, to avoid reallocating memory
TODO: For a put of an existing key, deallocate memory take by values, for other value types till a kTypeValue is found, ie. remove kTypeMerge.
TODO: Update the transaction log, to allow consistent reload of the memtable.
Test Plan: Added a unit test verifying the inplace update. But some other unit tests broken due to invalid sequence number checks. WIll fix them next.
Reviewers: xinyaohu, sumeet, haobo, dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12423
Automatic commit by arc
2013-08-19 23:12:47 +02:00
|
|
|
(unsigned) (VarintLength(key_length) +
|
|
|
|
key_length +
|
2014-01-14 16:55:16 +01:00
|
|
|
VarintLength(new_value_size) +
|
|
|
|
new_value_size)
|
In-place updates for equal keys and similar sized values
Summary:
Currently for each put, a fresh memory is allocated, and a new entry is added to the memtable with a new sequence number irrespective of whether the key already exists in the memtable. This diff is an attempt to update the value inplace for existing keys. It currently handles a very simple case:
1. Key already exists in the current memtable. Does not inplace update values in immutable memtable or snapshot
2. Latest value type is a 'put' ie kTypeValue
3. New value size is less than existing value, to avoid reallocating memory
TODO: For a put of an existing key, deallocate memory take by values, for other value types till a kTypeValue is found, ie. remove kTypeMerge.
TODO: Update the transaction log, to allow consistent reload of the memtable.
Test Plan: Added a unit test verifying the inplace update. But some other unit tests broken due to invalid sequence number checks. WIll fix them next.
Reviewers: xinyaohu, sumeet, haobo, dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12423
Automatic commit by arc
2013-08-19 23:12:47 +02:00
|
|
|
);
|
2013-11-27 23:27:02 +01:00
|
|
|
// no need to update bloom, as user key does not change.
|
2014-01-14 16:55:16 +01:00
|
|
|
return;
|
In-place updates for equal keys and similar sized values
Summary:
Currently for each put, a fresh memory is allocated, and a new entry is added to the memtable with a new sequence number irrespective of whether the key already exists in the memtable. This diff is an attempt to update the value inplace for existing keys. It currently handles a very simple case:
1. Key already exists in the current memtable. Does not inplace update values in immutable memtable or snapshot
2. Latest value type is a 'put' ie kTypeValue
3. New value size is less than existing value, to avoid reallocating memory
TODO: For a put of an existing key, deallocate memory take by values, for other value types till a kTypeValue is found, ie. remove kTypeMerge.
TODO: Update the transaction log, to allow consistent reload of the memtable.
Test Plan: Added a unit test verifying the inplace update. But some other unit tests broken due to invalid sequence number checks. WIll fix them next.
Reviewers: xinyaohu, sumeet, haobo, dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12423
Automatic commit by arc
2013-08-19 23:12:47 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
default:
|
|
|
|
// If the latest value is kTypeDeletion, kTypeMerge or kTypeLogData
|
2014-01-14 16:55:16 +01:00
|
|
|
// we don't have enough space for update inplace
|
|
|
|
Add(seq, kTypeValue, key, value);
|
|
|
|
return;
|
In-place updates for equal keys and similar sized values
Summary:
Currently for each put, a fresh memory is allocated, and a new entry is added to the memtable with a new sequence number irrespective of whether the key already exists in the memtable. This diff is an attempt to update the value inplace for existing keys. It currently handles a very simple case:
1. Key already exists in the current memtable. Does not inplace update values in immutable memtable or snapshot
2. Latest value type is a 'put' ie kTypeValue
3. New value size is less than existing value, to avoid reallocating memory
TODO: For a put of an existing key, deallocate memory take by values, for other value types till a kTypeValue is found, ie. remove kTypeMerge.
TODO: Update the transaction log, to allow consistent reload of the memtable.
Test Plan: Added a unit test verifying the inplace update. But some other unit tests broken due to invalid sequence number checks. WIll fix them next.
Reviewers: xinyaohu, sumeet, haobo, dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12423
Automatic commit by arc
2013-08-19 23:12:47 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2014-01-14 16:55:16 +01:00
|
|
|
// key doesn't exist
|
|
|
|
Add(seq, kTypeValue, key, value);
|
|
|
|
}
|
|
|
|
|
|
|
|
bool MemTable::UpdateCallback(SequenceNumber seq,
|
|
|
|
const Slice& key,
|
|
|
|
const Slice& delta,
|
|
|
|
const Options& options) {
|
|
|
|
LookupKey lkey(key, seq);
|
|
|
|
Slice memkey = lkey.memtable_key();
|
|
|
|
|
|
|
|
std::shared_ptr<MemTableRep::Iterator> iter(
|
|
|
|
table_->GetIterator(lkey.user_key()));
|
|
|
|
iter->Seek(key, memkey.data());
|
|
|
|
|
|
|
|
if (iter->Valid()) {
|
|
|
|
// entry format is:
|
|
|
|
// key_length varint32
|
|
|
|
// userkey char[klength-8]
|
|
|
|
// tag uint64
|
|
|
|
// vlength varint32
|
|
|
|
// value char[vlength]
|
|
|
|
// Check that it belongs to same user key. We do not check the
|
|
|
|
// sequence number since the Seek() call above should have skipped
|
|
|
|
// all entries with overly large sequence numbers.
|
|
|
|
const char* entry = iter->key();
|
|
|
|
uint32_t key_length;
|
|
|
|
const char* key_ptr = GetVarint32Ptr(entry, entry + 5, &key_length);
|
|
|
|
if (comparator_.comparator.user_comparator()->Compare(
|
|
|
|
Slice(key_ptr, key_length - 8), lkey.user_key()) == 0) {
|
|
|
|
// Correct user key
|
|
|
|
const uint64_t tag = DecodeFixed64(key_ptr + key_length - 8);
|
|
|
|
switch (static_cast<ValueType>(tag & 0xff)) {
|
|
|
|
case kTypeValue: {
|
|
|
|
Slice prev_value = GetLengthPrefixedSlice(key_ptr + key_length);
|
|
|
|
uint32_t prev_value_size = prev_value.size();
|
|
|
|
|
|
|
|
WriteLock wl(GetLock(lkey.user_key()));
|
|
|
|
std::string str_value;
|
|
|
|
if (options.inplace_callback(const_cast<char*>(prev_value.data()),
|
|
|
|
prev_value_size, delta, &str_value)) {
|
|
|
|
// Value already updated by callback.
|
|
|
|
// TODO: Change size of value in memtable slice.
|
|
|
|
// This works for leaf, since size is already encoded in the
|
|
|
|
// value. It doesn't depend on rocksdb buffer size.
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
Slice slice_value = Slice(str_value);
|
|
|
|
uint32_t new_value_size = slice_value.size();
|
|
|
|
|
|
|
|
// Update value, if newValue size <= curValue size
|
|
|
|
if (new_value_size <= prev_value_size ) {
|
|
|
|
char* p = EncodeVarint32(const_cast<char*>(key_ptr) + key_length,
|
|
|
|
new_value_size);
|
|
|
|
|
|
|
|
memcpy(p, slice_value.data(), new_value_size);
|
|
|
|
assert(
|
|
|
|
(p + new_value_size) - entry ==
|
|
|
|
(unsigned) (VarintLength(key_length) +
|
|
|
|
key_length +
|
|
|
|
VarintLength(new_value_size) +
|
|
|
|
new_value_size)
|
|
|
|
);
|
|
|
|
return true;
|
|
|
|
} else {
|
|
|
|
// If we don't have enough space to update in-place
|
|
|
|
// Return as NotUpdatable, and do normal Add()
|
|
|
|
Add(seq, kTypeValue, key, slice_value);
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
// If the latest value is not kTypeValue
|
|
|
|
// or key doesn't exist
|
In-place updates for equal keys and similar sized values
Summary:
Currently for each put, a fresh memory is allocated, and a new entry is added to the memtable with a new sequence number irrespective of whether the key already exists in the memtable. This diff is an attempt to update the value inplace for existing keys. It currently handles a very simple case:
1. Key already exists in the current memtable. Does not inplace update values in immutable memtable or snapshot
2. Latest value type is a 'put' ie kTypeValue
3. New value size is less than existing value, to avoid reallocating memory
TODO: For a put of an existing key, deallocate memory take by values, for other value types till a kTypeValue is found, ie. remove kTypeMerge.
TODO: Update the transaction log, to allow consistent reload of the memtable.
Test Plan: Added a unit test verifying the inplace update. But some other unit tests broken due to invalid sequence number checks. WIll fix them next.
Reviewers: xinyaohu, sumeet, haobo, dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12423
Automatic commit by arc
2013-08-19 23:12:47 +02:00
|
|
|
return false;
|
|
|
|
}
|
2013-10-04 06:49:15 +02:00
|
|
|
} // namespace rocksdb
|