rocksdb/db/compaction_iterator_test.cc

466 lines
16 KiB
C++
Raw Normal View History

// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
// 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.
// This source code is also licensed under the GPLv2 license found in the
// COPYING file in the root directory of this source tree.
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
#include "db/compaction_iterator.h"
Compaction Support for Range Deletion Summary: This diff introduces RangeDelAggregator, which takes ownership of iterators provided to it via AddTombstones(). The tombstones are organized in a two-level map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data copy by holding Slices returned by the iterator, which remain valid thanks to pinning. For compaction, we create a hierarchical range tombstone iterator with structure matching the iterator over compaction input data. An aggregator based on that iterator is used by CompactionIterator to determine which keys are covered by range tombstones. In case of merge operand, the same aggregator is used by MergeHelper. Upon finishing each file in the compaction, relevant range tombstones are added to the output file's range tombstone metablock and file boundaries are updated accordingly. To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete() considers tombstones in the key's snapshot stripe. When this function is used outside of compaction, it also checks newer stripes, which can contain covering tombstones. Currently the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges within a stripe such that binary search can be used. RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range to a new table's range tombstone meta-block. Since range tombstones may fall in the gap between files, we may need to extend some files' key-ranges. The strategy is (1) first file extends as far left as possible and other files do not extend left, (2) all files extend right until either the start of the next file or the end of the last range tombstone in the gap, whichever comes first. One other notable change is adding release/move semantics to ScopedArenaIterator such that it can be used to transfer ownership of an arena-allocated iterator, similar to how unique_ptr is used for malloc'd data. Depends on D61473 Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927 Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D62205
2016-10-18 21:04:56 +02:00
#include <string>
#include <vector>
#include "port/port.h"
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
#include "util/testharness.h"
#include "util/testutil.h"
namespace rocksdb {
// Expects no merging attempts.
class NoMergingMergeOp : public MergeOperator {
public:
bool FullMergeV2(const MergeOperationInput& merge_in,
MergeOperationOutput* merge_out) const override {
ADD_FAILURE();
return false;
}
bool PartialMergeMulti(const Slice& key,
const std::deque<Slice>& operand_list,
std::string* new_value,
Logger* logger) const override {
ADD_FAILURE();
return false;
}
const char* Name() const override {
return "CompactionIteratorTest NoMergingMergeOp";
}
};
// Compaction filter that gets stuck when it sees a particular key,
// then gets unstuck when told to.
// Always returns Decition::kRemove.
class StallingFilter : public CompactionFilter {
public:
virtual Decision FilterV2(int level, const Slice& key, ValueType t,
const Slice& existing_value, std::string* new_value,
std::string* skip_until) const override {
int k = std::atoi(key.ToString().c_str());
last_seen.store(k);
while (k >= stall_at.load()) {
std::this_thread::yield();
}
return Decision::kRemove;
}
const char* Name() const override {
return "CompactionIteratorTest StallingFilter";
}
// Wait until the filter sees a key >= k and stalls at that key.
// If `exact`, asserts that the seen key is equal to k.
void WaitForStall(int k, bool exact = true) {
stall_at.store(k);
while (last_seen.load() < k) {
std::this_thread::yield();
}
if (exact) {
EXPECT_EQ(k, last_seen.load());
}
}
// Filter will stall on key >= stall_at. Advance stall_at to unstall.
mutable std::atomic<int> stall_at{0};
// Last key the filter was called with.
mutable std::atomic<int> last_seen{0};
};
class LoggingForwardVectorIterator : public InternalIterator {
public:
struct Action {
enum class Type {
SEEK_TO_FIRST,
SEEK,
NEXT,
};
Type type;
std::string arg;
explicit Action(Type _type, std::string _arg = "")
: type(_type), arg(_arg) {}
bool operator==(const Action& rhs) const {
return std::tie(type, arg) == std::tie(rhs.type, rhs.arg);
}
};
LoggingForwardVectorIterator(const std::vector<std::string>& keys,
const std::vector<std::string>& values)
: keys_(keys), values_(values), current_(keys.size()) {
assert(keys_.size() == values_.size());
}
virtual bool Valid() const override { return current_ < keys_.size(); }
virtual void SeekToFirst() override {
log.emplace_back(Action::Type::SEEK_TO_FIRST);
current_ = 0;
}
virtual void SeekToLast() override { assert(false); }
virtual void Seek(const Slice& target) override {
log.emplace_back(Action::Type::SEEK, target.ToString());
current_ = std::lower_bound(keys_.begin(), keys_.end(), target.ToString()) -
keys_.begin();
}
virtual void SeekForPrev(const Slice& target) override { assert(false); }
virtual void Next() override {
assert(Valid());
log.emplace_back(Action::Type::NEXT);
current_++;
}
virtual void Prev() override { assert(false); }
virtual Slice key() const override {
assert(Valid());
return Slice(keys_[current_]);
}
virtual Slice value() const override {
assert(Valid());
return Slice(values_[current_]);
}
virtual Status status() const override { return Status::OK(); }
std::vector<Action> log;
private:
std::vector<std::string> keys_;
std::vector<std::string> values_;
size_t current_;
};
class FakeCompaction : public CompactionIterator::CompactionProxy {
public:
FakeCompaction() = default;
virtual int level(size_t compaction_input_level) const { return 0; }
virtual bool KeyNotExistsBeyondOutputLevel(
const Slice& user_key, std::vector<size_t>* level_ptrs) const {
return key_not_exists_beyond_output_level;
}
virtual bool bottommost_level() const { return false; }
virtual int number_levels() const { return 1; }
virtual Slice GetLargestUserKey() const {
return "\xff\xff\xff\xff\xff\xff\xff\xff\xff";
}
virtual bool allow_ingest_behind() const { return false; }
bool key_not_exists_beyond_output_level = false;
};
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
class CompactionIteratorTest : public testing::Test {
public:
Compaction Support for Range Deletion Summary: This diff introduces RangeDelAggregator, which takes ownership of iterators provided to it via AddTombstones(). The tombstones are organized in a two-level map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data copy by holding Slices returned by the iterator, which remain valid thanks to pinning. For compaction, we create a hierarchical range tombstone iterator with structure matching the iterator over compaction input data. An aggregator based on that iterator is used by CompactionIterator to determine which keys are covered by range tombstones. In case of merge operand, the same aggregator is used by MergeHelper. Upon finishing each file in the compaction, relevant range tombstones are added to the output file's range tombstone metablock and file boundaries are updated accordingly. To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete() considers tombstones in the key's snapshot stripe. When this function is used outside of compaction, it also checks newer stripes, which can contain covering tombstones. Currently the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges within a stripe such that binary search can be used. RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range to a new table's range tombstone meta-block. Since range tombstones may fall in the gap between files, we may need to extend some files' key-ranges. The strategy is (1) first file extends as far left as possible and other files do not extend left, (2) all files extend right until either the start of the next file or the end of the last range tombstone in the gap, whichever comes first. One other notable change is adding release/move semantics to ScopedArenaIterator such that it can be used to transfer ownership of an arena-allocated iterator, similar to how unique_ptr is used for malloc'd data. Depends on D61473 Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927 Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D62205
2016-10-18 21:04:56 +02:00
CompactionIteratorTest()
: cmp_(BytewiseComparator()), icmp_(cmp_), snapshots_({}) {}
void InitIterators(const std::vector<std::string>& ks,
const std::vector<std::string>& vs,
const std::vector<std::string>& range_del_ks,
const std::vector<std::string>& range_del_vs,
SequenceNumber last_sequence,
MergeOperator* merge_op = nullptr,
CompactionFilter* filter = nullptr) {
Compaction Support for Range Deletion Summary: This diff introduces RangeDelAggregator, which takes ownership of iterators provided to it via AddTombstones(). The tombstones are organized in a two-level map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data copy by holding Slices returned by the iterator, which remain valid thanks to pinning. For compaction, we create a hierarchical range tombstone iterator with structure matching the iterator over compaction input data. An aggregator based on that iterator is used by CompactionIterator to determine which keys are covered by range tombstones. In case of merge operand, the same aggregator is used by MergeHelper. Upon finishing each file in the compaction, relevant range tombstones are added to the output file's range tombstone metablock and file boundaries are updated accordingly. To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete() considers tombstones in the key's snapshot stripe. When this function is used outside of compaction, it also checks newer stripes, which can contain covering tombstones. Currently the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges within a stripe such that binary search can be used. RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range to a new table's range tombstone meta-block. Since range tombstones may fall in the gap between files, we may need to extend some files' key-ranges. The strategy is (1) first file extends as far left as possible and other files do not extend left, (2) all files extend right until either the start of the next file or the end of the last range tombstone in the gap, whichever comes first. One other notable change is adding release/move semantics to ScopedArenaIterator such that it can be used to transfer ownership of an arena-allocated iterator, similar to how unique_ptr is used for malloc'd data. Depends on D61473 Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927 Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D62205
2016-10-18 21:04:56 +02:00
std::unique_ptr<InternalIterator> range_del_iter(
new test::VectorIterator(range_del_ks, range_del_vs));
range_del_agg_.reset(new RangeDelAggregator(icmp_, snapshots_));
ASSERT_OK(range_del_agg_->AddTombstones(std::move(range_del_iter)));
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
std::unique_ptr<CompactionIterator::CompactionProxy> compaction;
if (filter) {
compaction_proxy_ = new FakeCompaction();
compaction.reset(compaction_proxy_);
}
merge_helper_.reset(new MergeHelper(Env::Default(), cmp_, merge_op, filter,
nullptr, false, 0, 0, nullptr,
&shutting_down_));
iter_.reset(new LoggingForwardVectorIterator(ks, vs));
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
iter_->SeekToFirst();
c_iter_.reset(new CompactionIterator(
iter_.get(), cmp_, merge_helper_.get(), last_sequence, &snapshots_,
kMaxSequenceNumber, Env::Default(), false, range_del_agg_.get(),
std::move(compaction), filter, nullptr, &shutting_down_));
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
}
Compaction Support for Range Deletion Summary: This diff introduces RangeDelAggregator, which takes ownership of iterators provided to it via AddTombstones(). The tombstones are organized in a two-level map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data copy by holding Slices returned by the iterator, which remain valid thanks to pinning. For compaction, we create a hierarchical range tombstone iterator with structure matching the iterator over compaction input data. An aggregator based on that iterator is used by CompactionIterator to determine which keys are covered by range tombstones. In case of merge operand, the same aggregator is used by MergeHelper. Upon finishing each file in the compaction, relevant range tombstones are added to the output file's range tombstone metablock and file boundaries are updated accordingly. To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete() considers tombstones in the key's snapshot stripe. When this function is used outside of compaction, it also checks newer stripes, which can contain covering tombstones. Currently the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges within a stripe such that binary search can be used. RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range to a new table's range tombstone meta-block. Since range tombstones may fall in the gap between files, we may need to extend some files' key-ranges. The strategy is (1) first file extends as far left as possible and other files do not extend left, (2) all files extend right until either the start of the next file or the end of the last range tombstone in the gap, whichever comes first. One other notable change is adding release/move semantics to ScopedArenaIterator such that it can be used to transfer ownership of an arena-allocated iterator, similar to how unique_ptr is used for malloc'd data. Depends on D61473 Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927 Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D62205
2016-10-18 21:04:56 +02:00
void AddSnapshot(SequenceNumber snapshot) { snapshots_.push_back(snapshot); }
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
const Comparator* cmp_;
Compaction Support for Range Deletion Summary: This diff introduces RangeDelAggregator, which takes ownership of iterators provided to it via AddTombstones(). The tombstones are organized in a two-level map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data copy by holding Slices returned by the iterator, which remain valid thanks to pinning. For compaction, we create a hierarchical range tombstone iterator with structure matching the iterator over compaction input data. An aggregator based on that iterator is used by CompactionIterator to determine which keys are covered by range tombstones. In case of merge operand, the same aggregator is used by MergeHelper. Upon finishing each file in the compaction, relevant range tombstones are added to the output file's range tombstone metablock and file boundaries are updated accordingly. To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete() considers tombstones in the key's snapshot stripe. When this function is used outside of compaction, it also checks newer stripes, which can contain covering tombstones. Currently the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges within a stripe such that binary search can be used. RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range to a new table's range tombstone meta-block. Since range tombstones may fall in the gap between files, we may need to extend some files' key-ranges. The strategy is (1) first file extends as far left as possible and other files do not extend left, (2) all files extend right until either the start of the next file or the end of the last range tombstone in the gap, whichever comes first. One other notable change is adding release/move semantics to ScopedArenaIterator such that it can be used to transfer ownership of an arena-allocated iterator, similar to how unique_ptr is used for malloc'd data. Depends on D61473 Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927 Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D62205
2016-10-18 21:04:56 +02:00
const InternalKeyComparator icmp_;
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
std::vector<SequenceNumber> snapshots_;
std::unique_ptr<MergeHelper> merge_helper_;
std::unique_ptr<LoggingForwardVectorIterator> iter_;
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
std::unique_ptr<CompactionIterator> c_iter_;
Compaction Support for Range Deletion Summary: This diff introduces RangeDelAggregator, which takes ownership of iterators provided to it via AddTombstones(). The tombstones are organized in a two-level map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data copy by holding Slices returned by the iterator, which remain valid thanks to pinning. For compaction, we create a hierarchical range tombstone iterator with structure matching the iterator over compaction input data. An aggregator based on that iterator is used by CompactionIterator to determine which keys are covered by range tombstones. In case of merge operand, the same aggregator is used by MergeHelper. Upon finishing each file in the compaction, relevant range tombstones are added to the output file's range tombstone metablock and file boundaries are updated accordingly. To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete() considers tombstones in the key's snapshot stripe. When this function is used outside of compaction, it also checks newer stripes, which can contain covering tombstones. Currently the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges within a stripe such that binary search can be used. RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range to a new table's range tombstone meta-block. Since range tombstones may fall in the gap between files, we may need to extend some files' key-ranges. The strategy is (1) first file extends as far left as possible and other files do not extend left, (2) all files extend right until either the start of the next file or the end of the last range tombstone in the gap, whichever comes first. One other notable change is adding release/move semantics to ScopedArenaIterator such that it can be used to transfer ownership of an arena-allocated iterator, similar to how unique_ptr is used for malloc'd data. Depends on D61473 Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927 Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D62205
2016-10-18 21:04:56 +02:00
std::unique_ptr<RangeDelAggregator> range_del_agg_;
std::atomic<bool> shutting_down_{false};
FakeCompaction* compaction_proxy_;
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
};
// It is possible that the output of the compaction iterator is empty even if
// the input is not.
TEST_F(CompactionIteratorTest, EmptyResult) {
Compaction Support for Range Deletion Summary: This diff introduces RangeDelAggregator, which takes ownership of iterators provided to it via AddTombstones(). The tombstones are organized in a two-level map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data copy by holding Slices returned by the iterator, which remain valid thanks to pinning. For compaction, we create a hierarchical range tombstone iterator with structure matching the iterator over compaction input data. An aggregator based on that iterator is used by CompactionIterator to determine which keys are covered by range tombstones. In case of merge operand, the same aggregator is used by MergeHelper. Upon finishing each file in the compaction, relevant range tombstones are added to the output file's range tombstone metablock and file boundaries are updated accordingly. To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete() considers tombstones in the key's snapshot stripe. When this function is used outside of compaction, it also checks newer stripes, which can contain covering tombstones. Currently the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges within a stripe such that binary search can be used. RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range to a new table's range tombstone meta-block. Since range tombstones may fall in the gap between files, we may need to extend some files' key-ranges. The strategy is (1) first file extends as far left as possible and other files do not extend left, (2) all files extend right until either the start of the next file or the end of the last range tombstone in the gap, whichever comes first. One other notable change is adding release/move semantics to ScopedArenaIterator such that it can be used to transfer ownership of an arena-allocated iterator, similar to how unique_ptr is used for malloc'd data. Depends on D61473 Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927 Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D62205
2016-10-18 21:04:56 +02:00
InitIterators({test::KeyStr("a", 5, kTypeSingleDeletion),
test::KeyStr("a", 3, kTypeValue)},
{"", "val"}, {}, {}, 5);
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
c_iter_->SeekToFirst();
ASSERT_FALSE(c_iter_->Valid());
}
// If there is a corruption after a single deletion, the corrupted key should
// be preserved.
TEST_F(CompactionIteratorTest, CorruptionAfterSingleDeletion) {
Compaction Support for Range Deletion Summary: This diff introduces RangeDelAggregator, which takes ownership of iterators provided to it via AddTombstones(). The tombstones are organized in a two-level map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data copy by holding Slices returned by the iterator, which remain valid thanks to pinning. For compaction, we create a hierarchical range tombstone iterator with structure matching the iterator over compaction input data. An aggregator based on that iterator is used by CompactionIterator to determine which keys are covered by range tombstones. In case of merge operand, the same aggregator is used by MergeHelper. Upon finishing each file in the compaction, relevant range tombstones are added to the output file's range tombstone metablock and file boundaries are updated accordingly. To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete() considers tombstones in the key's snapshot stripe. When this function is used outside of compaction, it also checks newer stripes, which can contain covering tombstones. Currently the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges within a stripe such that binary search can be used. RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range to a new table's range tombstone meta-block. Since range tombstones may fall in the gap between files, we may need to extend some files' key-ranges. The strategy is (1) first file extends as far left as possible and other files do not extend left, (2) all files extend right until either the start of the next file or the end of the last range tombstone in the gap, whichever comes first. One other notable change is adding release/move semantics to ScopedArenaIterator such that it can be used to transfer ownership of an arena-allocated iterator, similar to how unique_ptr is used for malloc'd data. Depends on D61473 Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927 Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D62205
2016-10-18 21:04:56 +02:00
InitIterators({test::KeyStr("a", 5, kTypeSingleDeletion),
test::KeyStr("a", 3, kTypeValue, true),
test::KeyStr("b", 10, kTypeValue)},
{"", "val", "val2"}, {}, {}, 10);
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
c_iter_->SeekToFirst();
ASSERT_TRUE(c_iter_->Valid());
ASSERT_EQ(test::KeyStr("a", 5, kTypeSingleDeletion),
c_iter_->key().ToString());
c_iter_->Next();
ASSERT_TRUE(c_iter_->Valid());
ASSERT_EQ(test::KeyStr("a", 3, kTypeValue, true), c_iter_->key().ToString());
c_iter_->Next();
ASSERT_TRUE(c_iter_->Valid());
ASSERT_EQ(test::KeyStr("b", 10, kTypeValue), c_iter_->key().ToString());
c_iter_->Next();
ASSERT_FALSE(c_iter_->Valid());
}
Compaction Support for Range Deletion Summary: This diff introduces RangeDelAggregator, which takes ownership of iterators provided to it via AddTombstones(). The tombstones are organized in a two-level map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data copy by holding Slices returned by the iterator, which remain valid thanks to pinning. For compaction, we create a hierarchical range tombstone iterator with structure matching the iterator over compaction input data. An aggregator based on that iterator is used by CompactionIterator to determine which keys are covered by range tombstones. In case of merge operand, the same aggregator is used by MergeHelper. Upon finishing each file in the compaction, relevant range tombstones are added to the output file's range tombstone metablock and file boundaries are updated accordingly. To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete() considers tombstones in the key's snapshot stripe. When this function is used outside of compaction, it also checks newer stripes, which can contain covering tombstones. Currently the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges within a stripe such that binary search can be used. RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range to a new table's range tombstone meta-block. Since range tombstones may fall in the gap between files, we may need to extend some files' key-ranges. The strategy is (1) first file extends as far left as possible and other files do not extend left, (2) all files extend right until either the start of the next file or the end of the last range tombstone in the gap, whichever comes first. One other notable change is adding release/move semantics to ScopedArenaIterator such that it can be used to transfer ownership of an arena-allocated iterator, similar to how unique_ptr is used for malloc'd data. Depends on D61473 Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927 Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D62205
2016-10-18 21:04:56 +02:00
TEST_F(CompactionIteratorTest, SimpleRangeDeletion) {
InitIterators({test::KeyStr("morning", 5, kTypeValue),
test::KeyStr("morning", 2, kTypeValue),
test::KeyStr("night", 3, kTypeValue)},
{"zao", "zao", "wan"},
{test::KeyStr("ma", 4, kTypeRangeDeletion)}, {"mz"}, 5);
c_iter_->SeekToFirst();
ASSERT_TRUE(c_iter_->Valid());
ASSERT_EQ(test::KeyStr("morning", 5, kTypeValue), c_iter_->key().ToString());
c_iter_->Next();
ASSERT_TRUE(c_iter_->Valid());
ASSERT_EQ(test::KeyStr("night", 3, kTypeValue), c_iter_->key().ToString());
c_iter_->Next();
ASSERT_FALSE(c_iter_->Valid());
}
TEST_F(CompactionIteratorTest, RangeDeletionWithSnapshots) {
AddSnapshot(10);
std::vector<std::string> ks1;
ks1.push_back(test::KeyStr("ma", 28, kTypeRangeDeletion));
std::vector<std::string> vs1{"mz"};
std::vector<std::string> ks2{test::KeyStr("morning", 15, kTypeValue),
test::KeyStr("morning", 5, kTypeValue),
test::KeyStr("night", 40, kTypeValue),
test::KeyStr("night", 20, kTypeValue)};
std::vector<std::string> vs2{"zao 15", "zao 5", "wan 40", "wan 20"};
InitIterators(ks2, vs2, ks1, vs1, 40);
c_iter_->SeekToFirst();
ASSERT_TRUE(c_iter_->Valid());
ASSERT_EQ(test::KeyStr("morning", 5, kTypeValue), c_iter_->key().ToString());
c_iter_->Next();
ASSERT_TRUE(c_iter_->Valid());
ASSERT_EQ(test::KeyStr("night", 40, kTypeValue), c_iter_->key().ToString());
c_iter_->Next();
ASSERT_FALSE(c_iter_->Valid());
}
TEST_F(CompactionIteratorTest, CompactionFilterSkipUntil) {
class Filter : public CompactionFilter {
virtual Decision FilterV2(int level, const Slice& key, ValueType t,
const Slice& existing_value,
std::string* new_value,
std::string* skip_until) const override {
std::string k = key.ToString();
std::string v = existing_value.ToString();
// See InitIterators() call below for the sequence of keys and their
// filtering decisions. Here we closely assert that compaction filter is
// called with the expected keys and only them, and with the right values.
if (k == "a") {
EXPECT_EQ(ValueType::kValue, t);
EXPECT_EQ("av50", v);
return Decision::kKeep;
}
if (k == "b") {
EXPECT_EQ(ValueType::kValue, t);
EXPECT_EQ("bv60", v);
*skip_until = "d+";
return Decision::kRemoveAndSkipUntil;
}
if (k == "e") {
EXPECT_EQ(ValueType::kMergeOperand, t);
EXPECT_EQ("em71", v);
return Decision::kKeep;
}
if (k == "f") {
if (v == "fm65") {
EXPECT_EQ(ValueType::kMergeOperand, t);
*skip_until = "f";
} else {
EXPECT_EQ("fm30", v);
EXPECT_EQ(ValueType::kMergeOperand, t);
*skip_until = "g+";
}
return Decision::kRemoveAndSkipUntil;
}
if (k == "h") {
EXPECT_EQ(ValueType::kValue, t);
EXPECT_EQ("hv91", v);
return Decision::kKeep;
}
if (k == "i") {
EXPECT_EQ(ValueType::kMergeOperand, t);
EXPECT_EQ("im95", v);
*skip_until = "z";
return Decision::kRemoveAndSkipUntil;
}
ADD_FAILURE();
return Decision::kKeep;
}
const char* Name() const override {
return "CompactionIteratorTest.CompactionFilterSkipUntil::Filter";
}
};
NoMergingMergeOp merge_op;
Filter filter;
InitIterators(
{test::KeyStr("a", 50, kTypeValue), // keep
test::KeyStr("a", 45, kTypeMerge),
test::KeyStr("b", 60, kTypeValue), // skip to "d+"
test::KeyStr("b", 40, kTypeValue), test::KeyStr("c", 35, kTypeValue),
test::KeyStr("d", 70, kTypeMerge),
test::KeyStr("e", 71, kTypeMerge), // keep
test::KeyStr("f", 65, kTypeMerge), // skip to "f", aka keep
test::KeyStr("f", 30, kTypeMerge), // skip to "g+"
test::KeyStr("f", 25, kTypeValue), test::KeyStr("g", 90, kTypeValue),
test::KeyStr("h", 91, kTypeValue), // keep
test::KeyStr("i", 95, kTypeMerge), // skip to "z"
test::KeyStr("j", 99, kTypeValue)},
{"av50", "am45", "bv60", "bv40", "cv35", "dm70", "em71", "fm65", "fm30",
"fv25", "gv90", "hv91", "im95", "jv99"},
{}, {}, kMaxSequenceNumber, &merge_op, &filter);
// Compaction should output just "a", "e" and "h" keys.
c_iter_->SeekToFirst();
ASSERT_TRUE(c_iter_->Valid());
ASSERT_EQ(test::KeyStr("a", 50, kTypeValue), c_iter_->key().ToString());
ASSERT_EQ("av50", c_iter_->value().ToString());
c_iter_->Next();
ASSERT_TRUE(c_iter_->Valid());
ASSERT_EQ(test::KeyStr("e", 71, kTypeMerge), c_iter_->key().ToString());
ASSERT_EQ("em71", c_iter_->value().ToString());
c_iter_->Next();
ASSERT_TRUE(c_iter_->Valid());
ASSERT_EQ(test::KeyStr("h", 91, kTypeValue), c_iter_->key().ToString());
ASSERT_EQ("hv91", c_iter_->value().ToString());
c_iter_->Next();
ASSERT_FALSE(c_iter_->Valid());
// Check that the compaction iterator did the correct sequence of calls on
// the underlying iterator.
using A = LoggingForwardVectorIterator::Action;
using T = A::Type;
std::vector<A> expected_actions = {
A(T::SEEK_TO_FIRST),
A(T::NEXT),
A(T::NEXT),
A(T::SEEK, test::KeyStr("d+", kMaxSequenceNumber, kValueTypeForSeek)),
A(T::NEXT),
A(T::NEXT),
A(T::SEEK, test::KeyStr("g+", kMaxSequenceNumber, kValueTypeForSeek)),
A(T::NEXT),
A(T::SEEK, test::KeyStr("z", kMaxSequenceNumber, kValueTypeForSeek))};
ASSERT_EQ(expected_actions, iter_->log);
}
TEST_F(CompactionIteratorTest, ShuttingDownInFilter) {
NoMergingMergeOp merge_op;
StallingFilter filter;
InitIterators(
{test::KeyStr("1", 1, kTypeValue), test::KeyStr("2", 2, kTypeValue),
test::KeyStr("3", 3, kTypeValue), test::KeyStr("4", 4, kTypeValue)},
{"v1", "v2", "v3", "v4"}, {}, {}, kMaxSequenceNumber, &merge_op, &filter);
// Don't leave tombstones (kTypeDeletion) for filtered keys.
compaction_proxy_->key_not_exists_beyond_output_level = true;
std::atomic<bool> seek_done{false};
rocksdb::port::Thread compaction_thread([&] {
c_iter_->SeekToFirst();
EXPECT_FALSE(c_iter_->Valid());
EXPECT_TRUE(c_iter_->status().IsShutdownInProgress());
seek_done.store(true);
});
// Let key 1 through.
filter.WaitForStall(1);
// Shutdown during compaction filter call for key 2.
filter.WaitForStall(2);
shutting_down_.store(true);
EXPECT_FALSE(seek_done.load());
// Unstall filter and wait for SeekToFirst() to return.
filter.stall_at.store(3);
compaction_thread.join();
assert(seek_done.load());
// Check that filter was never called again.
EXPECT_EQ(2, filter.last_seen.load());
}
// Same as ShuttingDownInFilter, but shutdown happens during filter call for
// a merge operand, not for a value.
TEST_F(CompactionIteratorTest, ShuttingDownInMerge) {
NoMergingMergeOp merge_op;
StallingFilter filter;
InitIterators(
{test::KeyStr("1", 1, kTypeValue), test::KeyStr("2", 2, kTypeMerge),
test::KeyStr("3", 3, kTypeMerge), test::KeyStr("4", 4, kTypeValue)},
{"v1", "v2", "v3", "v4"}, {}, {}, kMaxSequenceNumber, &merge_op, &filter);
compaction_proxy_->key_not_exists_beyond_output_level = true;
std::atomic<bool> seek_done{false};
rocksdb::port::Thread compaction_thread([&] {
c_iter_->SeekToFirst();
ASSERT_FALSE(c_iter_->Valid());
ASSERT_TRUE(c_iter_->status().IsShutdownInProgress());
seek_done.store(true);
});
// Let key 1 through.
filter.WaitForStall(1);
// Shutdown during compaction filter call for key 2.
filter.WaitForStall(2);
shutting_down_.store(true);
EXPECT_FALSE(seek_done.load());
// Unstall filter and wait for SeekToFirst() to return.
filter.stall_at.store(3);
compaction_thread.join();
assert(seek_done.load());
// Check that filter was never called again.
EXPECT_EQ(2, filter.last_seen.load());
}
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
} // namespace rocksdb
int main(int argc, char** argv) {
::testing::InitGoogleTest(&argc, argv);
return RUN_ALL_TESTS();
}