2018-12-18 01:26:56 +00:00
|
|
|
// Copyright (c) 2018-present, Facebook, Inc. All rights reserved.
|
2017-07-15 23:03:42 +00:00
|
|
|
// This source code is licensed under both the GPLv2 (found in the
|
|
|
|
// COPYING file in the root directory) and Apache 2.0 License
|
|
|
|
// (found in the LICENSE.Apache file in the root directory).
|
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 19:04:56 +00:00
|
|
|
|
|
|
|
#include "db/range_del_aggregator.h"
|
|
|
|
|
2019-05-31 18:52:59 +00:00
|
|
|
#include "db/compaction/compaction_iteration_stats.h"
|
2018-12-18 01:26:56 +00:00
|
|
|
#include "db/dbformat.h"
|
|
|
|
#include "db/pinned_iterators_manager.h"
|
|
|
|
#include "db/range_del_aggregator.h"
|
|
|
|
#include "db/range_tombstone_fragmenter.h"
|
|
|
|
#include "db/version_edit.h"
|
2020-01-22 00:11:08 +00:00
|
|
|
#include "rocksdb/comparator.h"
|
|
|
|
#include "rocksdb/types.h"
|
2018-12-18 01:26:56 +00:00
|
|
|
#include "table/internal_iterator.h"
|
|
|
|
#include "table/scoped_arena_iterator.h"
|
|
|
|
#include "table/table_builder.h"
|
|
|
|
#include "util/heap.h"
|
|
|
|
#include "util/kv_map.h"
|
|
|
|
#include "util/vector_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 19:04:56 +00:00
|
|
|
|
2020-02-20 20:07:53 +00:00
|
|
|
namespace ROCKSDB_NAMESPACE {
|
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 19:04:56 +00:00
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
TruncatedRangeDelIterator::TruncatedRangeDelIterator(
|
|
|
|
std::unique_ptr<FragmentedRangeTombstoneIterator> iter,
|
|
|
|
const InternalKeyComparator* icmp, const InternalKey* smallest,
|
|
|
|
const InternalKey* largest)
|
|
|
|
: iter_(std::move(iter)),
|
|
|
|
icmp_(icmp),
|
|
|
|
smallest_ikey_(smallest),
|
|
|
|
largest_ikey_(largest) {
|
|
|
|
if (smallest != nullptr) {
|
|
|
|
pinned_bounds_.emplace_back();
|
|
|
|
auto& parsed_smallest = pinned_bounds_.back();
|
2020-10-28 17:11:13 +00:00
|
|
|
Status pik_status = ParseInternalKey(smallest->Encode(), &parsed_smallest,
|
|
|
|
false /* log_err_key */); // TODO
|
|
|
|
pik_status.PermitUncheckedError();
|
|
|
|
assert(pik_status.ok());
|
2018-12-18 01:26:56 +00:00
|
|
|
smallest_ = &parsed_smallest;
|
|
|
|
}
|
|
|
|
if (largest != nullptr) {
|
|
|
|
pinned_bounds_.emplace_back();
|
|
|
|
auto& parsed_largest = pinned_bounds_.back();
|
2020-10-01 02:15:42 +00:00
|
|
|
|
2020-10-28 17:11:13 +00:00
|
|
|
Status pik_status = ParseInternalKey(largest->Encode(), &parsed_largest,
|
|
|
|
false /* log_err_key */); // TODO
|
|
|
|
pik_status.PermitUncheckedError();
|
|
|
|
assert(pik_status.ok());
|
2020-10-01 02:15:42 +00:00
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
if (parsed_largest.type == kTypeRangeDeletion &&
|
|
|
|
parsed_largest.sequence == kMaxSequenceNumber) {
|
|
|
|
// The file boundary has been artificially extended by a range tombstone.
|
|
|
|
// We do not need to adjust largest to properly truncate range
|
|
|
|
// tombstones that extend past the boundary.
|
|
|
|
} else if (parsed_largest.sequence == 0) {
|
|
|
|
// The largest key in the sstable has a sequence number of 0. Since we
|
|
|
|
// guarantee that no internal keys with the same user key and sequence
|
|
|
|
// number can exist in a DB, we know that the largest key in this sstable
|
|
|
|
// cannot exist as the smallest key in the next sstable. This further
|
|
|
|
// implies that no range tombstone in this sstable covers largest;
|
|
|
|
// otherwise, the file boundary would have been artificially extended.
|
|
|
|
//
|
|
|
|
// Therefore, we will never truncate a range tombstone at largest, so we
|
|
|
|
// can leave it unchanged.
|
|
|
|
} else {
|
|
|
|
// The same user key may straddle two sstable boundaries. To ensure that
|
|
|
|
// the truncated end key can cover the largest key in this sstable, reduce
|
|
|
|
// its sequence number by 1.
|
|
|
|
parsed_largest.sequence -= 1;
|
Skip swaths of range tombstone covered keys in merging iterator (2022 edition) (#10449)
Summary:
Delete range logic is moved from `DBIter` to `MergingIterator`, and `MergingIterator` will seek to the end of a range deletion if possible instead of scanning through each key and check with `RangeDelAggregator`.
With the invariant that a key in level L (consider memtable as the first level, each immutable and L0 as a separate level) has a larger sequence number than all keys in any level >L, a range tombstone `[start, end)` from level L covers all keys in its range in any level >L. This property motivates optimizations in iterator:
- in `Seek(target)`, if level L has a range tombstone `[start, end)` that covers `target.UserKey`, then for all levels > L, we can do Seek() on `end` instead of `target` to skip some range tombstone covered keys.
- in `Next()/Prev()`, if the current key is covered by a range tombstone `[start, end)` from level L, we can do `Seek` to `end` for all levels > L.
This PR implements the above optimizations in `MergingIterator`. As all range tombstone covered keys are now skipped in `MergingIterator`, the range tombstone logic is removed from `DBIter`. The idea in this PR is similar to https://github.com/facebook/rocksdb/issues/7317, but this PR leaves `InternalIterator` interface mostly unchanged. **Credit**: the cascading seek optimization and the sentinel key (discussed below) are inspired by [Pebble](https://github.com/cockroachdb/pebble/blob/master/merging_iter.go) and suggested by ajkr in https://github.com/facebook/rocksdb/issues/7317. The two optimizations are mostly implemented in `SeekImpl()/SeekForPrevImpl()` and `IsNextDeleted()/IsPrevDeleted()` in `merging_iterator.cc`. See comments for each method for more detail.
One notable change is that the minHeap/maxHeap used by `MergingIterator` now contains range tombstone end keys besides point key iterators. This helps to reduce the number of key comparisons. For example, for a range tombstone `[start, end)`, a `start` and an `end` `HeapItem` are inserted into the heap. When a `HeapItem` for range tombstone start key is popped from the minHeap, we know this range tombstone becomes "active" in the sense that, before the range tombstone's end key is popped from the minHeap, all the keys popped from this heap is covered by the range tombstone's internal key range `[start, end)`.
Another major change, *delete range sentinel key*, is made to `LevelIterator`. Before this PR, when all point keys in an SST file are iterated through in `MergingIterator`, a level iterator would advance to the next SST file in its level. In the case when an SST file has a range tombstone that covers keys beyond the SST file's last point key, advancing to the next SST file would lose this range tombstone. Consequently, `MergingIterator` could return keys that should have been deleted by some range tombstone. We prevent this by pretending that file boundaries in each SST file are sentinel keys. A `LevelIterator` now only advance the file iterator once the sentinel key is processed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10449
Test Plan:
- Added many unit tests in db_range_del_test
- Stress test: `./db_stress --readpercent=5 --prefixpercent=19 --writepercent=20 -delpercent=10 --iterpercent=44 --delrangepercent=2`
- Additional iterator stress test is added to verify against iterators against expected state: https://github.com/facebook/rocksdb/issues/10538. This is based on ajkr's previous attempt https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913.
```
python3 ./tools/db_crashtest.py blackbox --simple --write_buffer_size=524288 --target_file_size_base=524288 --max_bytes_for_level_base=2097152 --compression_type=none --max_background_compactions=8 --value_size_mult=33 --max_key=5000000 --interval=10 --duration=7200 --delrangepercent=3 --delpercent=9 --iterpercent=25 --writepercent=60 --readpercent=3 --prefixpercent=0 --num_iterations=1000 --range_deletion_width=100 --verify_iterator_with_expected_state_one_in=1
```
- Performance benchmark: I used a similar setup as in the blog [post](http://rocksdb.org/blog/2018/11/21/delete-range.html) that introduced DeleteRange, "a database with 5 million data keys, and 10000 range tombstones (ignoring those dropped during compaction) that were written in regular intervals after 4.5 million data keys were written". As expected, the performance with this PR depends on the range tombstone width.
```
# Setup:
TEST_TMPDIR=/dev/shm ./db_bench_main --benchmarks=fillrandom --writes=4500000 --num=5000000
TEST_TMPDIR=/dev/shm ./db_bench_main --benchmarks=overwrite --writes=500000 --num=5000000 --use_existing_db=true --writes_per_range_tombstone=50
# Scan entire DB
TEST_TMPDIR=/dev/shm ./db_bench_main --benchmarks=readseq[-X5] --use_existing_db=true --num=5000000 --disable_auto_compactions=true
# Short range scan (10 Next())
TEST_TMPDIR=/dev/shm/width-100/ ./db_bench_main --benchmarks=seekrandom[-X5] --use_existing_db=true --num=500000 --reads=100000 --seek_nexts=10 --disable_auto_compactions=true
# Long range scan(1000 Next())
TEST_TMPDIR=/dev/shm/width-100/ ./db_bench_main --benchmarks=seekrandom[-X5] --use_existing_db=true --num=500000 --reads=2500 --seek_nexts=1000 --disable_auto_compactions=true
```
Avg over of 10 runs (some slower tests had fews runs):
For the first column (tombstone), 0 means no range tombstone, 100-10000 means width of the 10k range tombstones, and 1 means there is a single range tombstone in the entire DB (width is 1000). The 1 tombstone case is to test regression when there's very few range tombstones in the DB, as no range tombstone is likely to take a different code path than with range tombstones.
- Scan entire DB
| tombstone width | Pre-PR ops/sec | Post-PR ops/sec | ±% |
| ------------- | ------------- | ------------- | ------------- |
| 0 range tombstone |2525600 (± 43564) |2486917 (± 33698) |-1.53% |
| 100 |1853835 (± 24736) |2073884 (± 32176) |+11.87% |
| 1000 |422415 (± 7466) |1115801 (± 22781) |+164.15% |
| 10000 |22384 (± 227) |227919 (± 6647) |+918.22% |
| 1 range tombstone |2176540 (± 39050) |2434954 (± 24563) |+11.87% |
- Short range scan
| tombstone width | Pre-PR ops/sec | Post-PR ops/sec | ±% |
| ------------- | ------------- | ------------- | ------------- |
| 0 range tombstone |35398 (± 533) |35338 (± 569) |-0.17% |
| 100 |28276 (± 664) |31684 (± 331) |+12.05% |
| 1000 |7637 (± 77) |25422 (± 277) |+232.88% |
| 10000 |1367 |28667 |+1997.07% |
| 1 range tombstone |32618 (± 581) |32748 (± 506) |+0.4% |
- Long range scan
| tombstone width | Pre-PR ops/sec | Post-PR ops/sec | ±% |
| ------------- | ------------- | ------------- | ------------- |
| 0 range tombstone |2262 (± 33) |2353 (± 20) |+4.02% |
| 100 |1696 (± 26) |1926 (± 18) |+13.56% |
| 1000 |410 (± 6) |1255 (± 29) |+206.1% |
| 10000 |25 |414 |+1556.0% |
| 1 range tombstone |1957 (± 30) |2185 (± 44) |+11.65% |
- Microbench does not show significant regression: https://gist.github.com/cbi42/59f280f85a59b678e7e5d8561e693b61
Reviewed By: ajkr
Differential Revision: D38450331
Pulled By: cbi42
fbshipit-source-id: b5ef12e8d8c289ed2e163ccdf277f5039b511fca
2022-09-02 16:51:19 +00:00
|
|
|
// This line is not needed for correctness, but it ensures that the
|
|
|
|
// truncated end key is not covering keys from the next SST file.
|
|
|
|
parsed_largest.type = kValueTypeForSeek;
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
}
|
2018-12-18 01:26:56 +00:00
|
|
|
largest_ = &parsed_largest;
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
}
|
2018-12-18 01:26:56 +00:00
|
|
|
}
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
bool TruncatedRangeDelIterator::Valid() const {
|
Skip swaths of range tombstone covered keys in merging iterator (2022 edition) (#10449)
Summary:
Delete range logic is moved from `DBIter` to `MergingIterator`, and `MergingIterator` will seek to the end of a range deletion if possible instead of scanning through each key and check with `RangeDelAggregator`.
With the invariant that a key in level L (consider memtable as the first level, each immutable and L0 as a separate level) has a larger sequence number than all keys in any level >L, a range tombstone `[start, end)` from level L covers all keys in its range in any level >L. This property motivates optimizations in iterator:
- in `Seek(target)`, if level L has a range tombstone `[start, end)` that covers `target.UserKey`, then for all levels > L, we can do Seek() on `end` instead of `target` to skip some range tombstone covered keys.
- in `Next()/Prev()`, if the current key is covered by a range tombstone `[start, end)` from level L, we can do `Seek` to `end` for all levels > L.
This PR implements the above optimizations in `MergingIterator`. As all range tombstone covered keys are now skipped in `MergingIterator`, the range tombstone logic is removed from `DBIter`. The idea in this PR is similar to https://github.com/facebook/rocksdb/issues/7317, but this PR leaves `InternalIterator` interface mostly unchanged. **Credit**: the cascading seek optimization and the sentinel key (discussed below) are inspired by [Pebble](https://github.com/cockroachdb/pebble/blob/master/merging_iter.go) and suggested by ajkr in https://github.com/facebook/rocksdb/issues/7317. The two optimizations are mostly implemented in `SeekImpl()/SeekForPrevImpl()` and `IsNextDeleted()/IsPrevDeleted()` in `merging_iterator.cc`. See comments for each method for more detail.
One notable change is that the minHeap/maxHeap used by `MergingIterator` now contains range tombstone end keys besides point key iterators. This helps to reduce the number of key comparisons. For example, for a range tombstone `[start, end)`, a `start` and an `end` `HeapItem` are inserted into the heap. When a `HeapItem` for range tombstone start key is popped from the minHeap, we know this range tombstone becomes "active" in the sense that, before the range tombstone's end key is popped from the minHeap, all the keys popped from this heap is covered by the range tombstone's internal key range `[start, end)`.
Another major change, *delete range sentinel key*, is made to `LevelIterator`. Before this PR, when all point keys in an SST file are iterated through in `MergingIterator`, a level iterator would advance to the next SST file in its level. In the case when an SST file has a range tombstone that covers keys beyond the SST file's last point key, advancing to the next SST file would lose this range tombstone. Consequently, `MergingIterator` could return keys that should have been deleted by some range tombstone. We prevent this by pretending that file boundaries in each SST file are sentinel keys. A `LevelIterator` now only advance the file iterator once the sentinel key is processed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10449
Test Plan:
- Added many unit tests in db_range_del_test
- Stress test: `./db_stress --readpercent=5 --prefixpercent=19 --writepercent=20 -delpercent=10 --iterpercent=44 --delrangepercent=2`
- Additional iterator stress test is added to verify against iterators against expected state: https://github.com/facebook/rocksdb/issues/10538. This is based on ajkr's previous attempt https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913.
```
python3 ./tools/db_crashtest.py blackbox --simple --write_buffer_size=524288 --target_file_size_base=524288 --max_bytes_for_level_base=2097152 --compression_type=none --max_background_compactions=8 --value_size_mult=33 --max_key=5000000 --interval=10 --duration=7200 --delrangepercent=3 --delpercent=9 --iterpercent=25 --writepercent=60 --readpercent=3 --prefixpercent=0 --num_iterations=1000 --range_deletion_width=100 --verify_iterator_with_expected_state_one_in=1
```
- Performance benchmark: I used a similar setup as in the blog [post](http://rocksdb.org/blog/2018/11/21/delete-range.html) that introduced DeleteRange, "a database with 5 million data keys, and 10000 range tombstones (ignoring those dropped during compaction) that were written in regular intervals after 4.5 million data keys were written". As expected, the performance with this PR depends on the range tombstone width.
```
# Setup:
TEST_TMPDIR=/dev/shm ./db_bench_main --benchmarks=fillrandom --writes=4500000 --num=5000000
TEST_TMPDIR=/dev/shm ./db_bench_main --benchmarks=overwrite --writes=500000 --num=5000000 --use_existing_db=true --writes_per_range_tombstone=50
# Scan entire DB
TEST_TMPDIR=/dev/shm ./db_bench_main --benchmarks=readseq[-X5] --use_existing_db=true --num=5000000 --disable_auto_compactions=true
# Short range scan (10 Next())
TEST_TMPDIR=/dev/shm/width-100/ ./db_bench_main --benchmarks=seekrandom[-X5] --use_existing_db=true --num=500000 --reads=100000 --seek_nexts=10 --disable_auto_compactions=true
# Long range scan(1000 Next())
TEST_TMPDIR=/dev/shm/width-100/ ./db_bench_main --benchmarks=seekrandom[-X5] --use_existing_db=true --num=500000 --reads=2500 --seek_nexts=1000 --disable_auto_compactions=true
```
Avg over of 10 runs (some slower tests had fews runs):
For the first column (tombstone), 0 means no range tombstone, 100-10000 means width of the 10k range tombstones, and 1 means there is a single range tombstone in the entire DB (width is 1000). The 1 tombstone case is to test regression when there's very few range tombstones in the DB, as no range tombstone is likely to take a different code path than with range tombstones.
- Scan entire DB
| tombstone width | Pre-PR ops/sec | Post-PR ops/sec | ±% |
| ------------- | ------------- | ------------- | ------------- |
| 0 range tombstone |2525600 (± 43564) |2486917 (± 33698) |-1.53% |
| 100 |1853835 (± 24736) |2073884 (± 32176) |+11.87% |
| 1000 |422415 (± 7466) |1115801 (± 22781) |+164.15% |
| 10000 |22384 (± 227) |227919 (± 6647) |+918.22% |
| 1 range tombstone |2176540 (± 39050) |2434954 (± 24563) |+11.87% |
- Short range scan
| tombstone width | Pre-PR ops/sec | Post-PR ops/sec | ±% |
| ------------- | ------------- | ------------- | ------------- |
| 0 range tombstone |35398 (± 533) |35338 (± 569) |-0.17% |
| 100 |28276 (± 664) |31684 (± 331) |+12.05% |
| 1000 |7637 (± 77) |25422 (± 277) |+232.88% |
| 10000 |1367 |28667 |+1997.07% |
| 1 range tombstone |32618 (± 581) |32748 (± 506) |+0.4% |
- Long range scan
| tombstone width | Pre-PR ops/sec | Post-PR ops/sec | ±% |
| ------------- | ------------- | ------------- | ------------- |
| 0 range tombstone |2262 (± 33) |2353 (± 20) |+4.02% |
| 100 |1696 (± 26) |1926 (± 18) |+13.56% |
| 1000 |410 (± 6) |1255 (± 29) |+206.1% |
| 10000 |25 |414 |+1556.0% |
| 1 range tombstone |1957 (± 30) |2185 (± 44) |+11.65% |
- Microbench does not show significant regression: https://gist.github.com/cbi42/59f280f85a59b678e7e5d8561e693b61
Reviewed By: ajkr
Differential Revision: D38450331
Pulled By: cbi42
fbshipit-source-id: b5ef12e8d8c289ed2e163ccdf277f5039b511fca
2022-09-02 16:51:19 +00:00
|
|
|
assert(iter_ != nullptr);
|
2018-12-18 01:26:56 +00:00
|
|
|
return iter_->Valid() &&
|
|
|
|
(smallest_ == nullptr ||
|
|
|
|
icmp_->Compare(*smallest_, iter_->parsed_end_key()) < 0) &&
|
|
|
|
(largest_ == nullptr ||
|
|
|
|
icmp_->Compare(iter_->parsed_start_key(), *largest_) < 0);
|
|
|
|
}
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
|
2022-09-30 23:13:03 +00:00
|
|
|
// NOTE: target is a user key, with timestamp if enabled.
|
2018-12-18 01:26:56 +00:00
|
|
|
void TruncatedRangeDelIterator::Seek(const Slice& target) {
|
|
|
|
if (largest_ != nullptr &&
|
|
|
|
icmp_->Compare(*largest_, ParsedInternalKey(target, kMaxSequenceNumber,
|
|
|
|
kTypeRangeDeletion)) <= 0) {
|
|
|
|
iter_->Invalidate();
|
|
|
|
return;
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
}
|
2018-12-18 01:26:56 +00:00
|
|
|
if (smallest_ != nullptr &&
|
|
|
|
icmp_->user_comparator()->Compare(target, smallest_->user_key) < 0) {
|
|
|
|
iter_->Seek(smallest_->user_key);
|
|
|
|
return;
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
}
|
2018-12-18 01:26:56 +00:00
|
|
|
iter_->Seek(target);
|
|
|
|
}
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
|
2022-09-30 23:13:03 +00:00
|
|
|
// NOTE: target is a user key, with timestamp if enabled.
|
2018-12-18 01:26:56 +00:00
|
|
|
void TruncatedRangeDelIterator::SeekForPrev(const Slice& target) {
|
|
|
|
if (smallest_ != nullptr &&
|
|
|
|
icmp_->Compare(ParsedInternalKey(target, 0, kTypeRangeDeletion),
|
|
|
|
*smallest_) < 0) {
|
|
|
|
iter_->Invalidate();
|
|
|
|
return;
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
}
|
2018-12-18 01:26:56 +00:00
|
|
|
if (largest_ != nullptr &&
|
|
|
|
icmp_->user_comparator()->Compare(largest_->user_key, target) < 0) {
|
|
|
|
iter_->SeekForPrev(largest_->user_key);
|
|
|
|
return;
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
}
|
2018-12-18 01:26:56 +00:00
|
|
|
iter_->SeekForPrev(target);
|
|
|
|
}
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
void TruncatedRangeDelIterator::SeekToFirst() {
|
|
|
|
if (smallest_ != nullptr) {
|
|
|
|
iter_->Seek(smallest_->user_key);
|
|
|
|
return;
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
}
|
2018-12-18 01:26:56 +00:00
|
|
|
iter_->SeekToTopFirst();
|
|
|
|
}
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
void TruncatedRangeDelIterator::SeekToLast() {
|
|
|
|
if (largest_ != nullptr) {
|
|
|
|
iter_->SeekForPrev(largest_->user_key);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
iter_->SeekToTopLast();
|
|
|
|
}
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
std::map<SequenceNumber, std::unique_ptr<TruncatedRangeDelIterator>>
|
|
|
|
TruncatedRangeDelIterator::SplitBySnapshot(
|
|
|
|
const std::vector<SequenceNumber>& snapshots) {
|
|
|
|
using FragmentedIterPair =
|
|
|
|
std::pair<const SequenceNumber,
|
|
|
|
std::unique_ptr<FragmentedRangeTombstoneIterator>>;
|
|
|
|
|
|
|
|
auto split_untruncated_iters = iter_->SplitBySnapshot(snapshots);
|
|
|
|
std::map<SequenceNumber, std::unique_ptr<TruncatedRangeDelIterator>>
|
|
|
|
split_truncated_iters;
|
|
|
|
std::for_each(
|
|
|
|
split_untruncated_iters.begin(), split_untruncated_iters.end(),
|
|
|
|
[&](FragmentedIterPair& iter_pair) {
|
2022-08-30 02:09:29 +00:00
|
|
|
auto truncated_iter = std::make_unique<TruncatedRangeDelIterator>(
|
|
|
|
std::move(iter_pair.second), icmp_, smallest_ikey_, largest_ikey_);
|
2018-12-18 01:26:56 +00:00
|
|
|
split_truncated_iters.emplace(iter_pair.first,
|
|
|
|
std::move(truncated_iter));
|
|
|
|
});
|
|
|
|
return split_truncated_iters;
|
|
|
|
}
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
ForwardRangeDelIterator::ForwardRangeDelIterator(
|
2018-12-18 22:10:31 +00:00
|
|
|
const InternalKeyComparator* icmp)
|
2018-12-18 01:26:56 +00:00
|
|
|
: icmp_(icmp),
|
|
|
|
unused_idx_(0),
|
|
|
|
active_seqnums_(SeqMaxComparator()),
|
|
|
|
active_iters_(EndKeyMinComparator(icmp)),
|
|
|
|
inactive_iters_(StartKeyMinComparator(icmp)) {}
|
|
|
|
|
|
|
|
bool ForwardRangeDelIterator::ShouldDelete(const ParsedInternalKey& parsed) {
|
|
|
|
// Move active iterators that end before parsed.
|
|
|
|
while (!active_iters_.empty() &&
|
|
|
|
icmp_->Compare((*active_iters_.top())->end_key(), parsed) <= 0) {
|
|
|
|
TruncatedRangeDelIterator* iter = PopActiveIter();
|
|
|
|
do {
|
|
|
|
iter->Next();
|
|
|
|
} while (iter->Valid() && icmp_->Compare(iter->end_key(), parsed) <= 0);
|
|
|
|
PushIter(iter, parsed);
|
|
|
|
assert(active_iters_.size() == active_seqnums_.size());
|
|
|
|
}
|
|
|
|
|
|
|
|
// Move inactive iterators that start before parsed.
|
|
|
|
while (!inactive_iters_.empty() &&
|
|
|
|
icmp_->Compare(inactive_iters_.top()->start_key(), parsed) <= 0) {
|
|
|
|
TruncatedRangeDelIterator* iter = PopInactiveIter();
|
|
|
|
while (iter->Valid() && icmp_->Compare(iter->end_key(), parsed) <= 0) {
|
|
|
|
iter->Next();
|
|
|
|
}
|
|
|
|
PushIter(iter, parsed);
|
|
|
|
assert(active_iters_.size() == active_seqnums_.size());
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
}
|
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
return active_seqnums_.empty()
|
|
|
|
? false
|
|
|
|
: (*active_seqnums_.begin())->seq() > parsed.sequence;
|
2016-11-19 00:54:09 +00:00
|
|
|
}
|
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
void ForwardRangeDelIterator::Invalidate() {
|
|
|
|
unused_idx_ = 0;
|
|
|
|
active_iters_.clear();
|
|
|
|
active_seqnums_.clear();
|
|
|
|
inactive_iters_.clear();
|
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 19:04:56 +00:00
|
|
|
}
|
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
ReverseRangeDelIterator::ReverseRangeDelIterator(
|
2018-12-18 22:10:31 +00:00
|
|
|
const InternalKeyComparator* icmp)
|
2018-12-18 01:26:56 +00:00
|
|
|
: icmp_(icmp),
|
|
|
|
unused_idx_(0),
|
|
|
|
active_seqnums_(SeqMaxComparator()),
|
|
|
|
active_iters_(StartKeyMaxComparator(icmp)),
|
|
|
|
inactive_iters_(EndKeyMaxComparator(icmp)) {}
|
|
|
|
|
|
|
|
bool ReverseRangeDelIterator::ShouldDelete(const ParsedInternalKey& parsed) {
|
|
|
|
// Move active iterators that start after parsed.
|
|
|
|
while (!active_iters_.empty() &&
|
|
|
|
icmp_->Compare(parsed, (*active_iters_.top())->start_key()) < 0) {
|
|
|
|
TruncatedRangeDelIterator* iter = PopActiveIter();
|
|
|
|
do {
|
|
|
|
iter->Prev();
|
|
|
|
} while (iter->Valid() && icmp_->Compare(parsed, iter->start_key()) < 0);
|
|
|
|
PushIter(iter, parsed);
|
|
|
|
assert(active_iters_.size() == active_seqnums_.size());
|
|
|
|
}
|
|
|
|
|
|
|
|
// Move inactive iterators that end after parsed.
|
|
|
|
while (!inactive_iters_.empty() &&
|
|
|
|
icmp_->Compare(parsed, inactive_iters_.top()->end_key()) < 0) {
|
|
|
|
TruncatedRangeDelIterator* iter = PopInactiveIter();
|
|
|
|
while (iter->Valid() && icmp_->Compare(parsed, iter->start_key()) < 0) {
|
|
|
|
iter->Prev();
|
|
|
|
}
|
|
|
|
PushIter(iter, parsed);
|
|
|
|
assert(active_iters_.size() == active_seqnums_.size());
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
}
|
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
return active_seqnums_.empty()
|
|
|
|
? false
|
|
|
|
: (*active_seqnums_.begin())->seq() > parsed.sequence;
|
2016-11-04 01:40:23 +00:00
|
|
|
}
|
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
void ReverseRangeDelIterator::Invalidate() {
|
|
|
|
unused_idx_ = 0;
|
|
|
|
active_iters_.clear();
|
|
|
|
active_seqnums_.clear();
|
|
|
|
inactive_iters_.clear();
|
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 19:04:56 +00:00
|
|
|
}
|
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
bool RangeDelAggregator::StripeRep::ShouldDelete(
|
|
|
|
const ParsedInternalKey& parsed, RangeDelPositioningMode mode) {
|
|
|
|
if (!InStripe(parsed.sequence) || IsEmpty()) {
|
2017-11-28 19:18:42 +00:00
|
|
|
return false;
|
|
|
|
}
|
2018-12-18 01:26:56 +00:00
|
|
|
switch (mode) {
|
|
|
|
case RangeDelPositioningMode::kForwardTraversal:
|
|
|
|
InvalidateReverseIter();
|
|
|
|
|
|
|
|
// Pick up previously unseen iterators.
|
|
|
|
for (auto it = std::next(iters_.begin(), forward_iter_.UnusedIdx());
|
|
|
|
it != iters_.end(); ++it, forward_iter_.IncUnusedIdx()) {
|
|
|
|
auto& iter = *it;
|
|
|
|
forward_iter_.AddNewIter(iter.get(), parsed);
|
|
|
|
}
|
|
|
|
|
|
|
|
return forward_iter_.ShouldDelete(parsed);
|
|
|
|
case RangeDelPositioningMode::kBackwardTraversal:
|
|
|
|
InvalidateForwardIter();
|
|
|
|
|
|
|
|
// Pick up previously unseen iterators.
|
|
|
|
for (auto it = std::next(iters_.begin(), reverse_iter_.UnusedIdx());
|
|
|
|
it != iters_.end(); ++it, reverse_iter_.IncUnusedIdx()) {
|
|
|
|
auto& iter = *it;
|
|
|
|
reverse_iter_.AddNewIter(iter.get(), parsed);
|
|
|
|
}
|
|
|
|
|
|
|
|
return reverse_iter_.ShouldDelete(parsed);
|
|
|
|
default:
|
|
|
|
assert(false);
|
|
|
|
return false;
|
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 19:04:56 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
bool RangeDelAggregator::StripeRep::IsRangeOverlapped(const Slice& start,
|
|
|
|
const Slice& end) {
|
|
|
|
Invalidate();
|
|
|
|
|
|
|
|
// Set the internal start/end keys so that:
|
|
|
|
// - if start_ikey has the same user key and sequence number as the
|
|
|
|
// current end key, start_ikey will be considered greater; and
|
|
|
|
// - if end_ikey has the same user key and sequence number as the current
|
|
|
|
// start key, end_ikey will be considered greater.
|
|
|
|
ParsedInternalKey start_ikey(start, kMaxSequenceNumber,
|
|
|
|
static_cast<ValueType>(0));
|
|
|
|
ParsedInternalKey end_ikey(end, 0, static_cast<ValueType>(0));
|
|
|
|
for (auto& iter : iters_) {
|
|
|
|
bool checked_candidate_tombstones = false;
|
|
|
|
for (iter->SeekForPrev(start);
|
|
|
|
iter->Valid() && icmp_->Compare(iter->start_key(), end_ikey) <= 0;
|
|
|
|
iter->Next()) {
|
|
|
|
checked_candidate_tombstones = true;
|
|
|
|
if (icmp_->Compare(start_ikey, iter->end_key()) < 0 &&
|
|
|
|
icmp_->Compare(iter->start_key(), end_ikey) <= 0) {
|
|
|
|
return true;
|
2018-07-14 00:34:54 +00:00
|
|
|
}
|
|
|
|
}
|
2018-12-18 01:26:56 +00:00
|
|
|
|
|
|
|
if (!checked_candidate_tombstones) {
|
|
|
|
// Do an additional check for when the end of the range is the begin
|
|
|
|
// key of a tombstone, which we missed earlier since SeekForPrev'ing
|
|
|
|
// to the start was invalid.
|
|
|
|
iter->SeekForPrev(end);
|
|
|
|
if (iter->Valid() && icmp_->Compare(start_ikey, iter->end_key()) < 0 &&
|
|
|
|
icmp_->Compare(iter->start_key(), end_ikey) <= 0) {
|
|
|
|
return true;
|
2018-07-14 00:34:54 +00:00
|
|
|
}
|
|
|
|
}
|
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 19:04:56 +00:00
|
|
|
}
|
2018-12-18 01:26:56 +00:00
|
|
|
return false;
|
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 19:04:56 +00:00
|
|
|
}
|
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
void ReadRangeDelAggregator::AddTombstones(
|
|
|
|
std::unique_ptr<FragmentedRangeTombstoneIterator> input_iter,
|
|
|
|
const InternalKey* smallest, const InternalKey* largest) {
|
|
|
|
if (input_iter == nullptr || input_iter->empty()) {
|
Maintain position in range deletions map
Summary:
When deletion-collapsing mode is enabled (i.e., for DBIter/CompactionIterator), we maintain position in the tombstone maps across calls to ShouldDelete(). Since iterators often access keys sequentially (or reverse-sequentially), scanning forward/backward from the last position can be faster than binary-searching the map for every key.
- When Next() is invoked on an iterator, we use kForwardTraversal to scan forwards, if needed, until arriving at the range deletion containing the next key.
- Similarly for Prev(), we use kBackwardTraversal to scan backwards in the range deletion map.
- When the iterator seeks, we use kBinarySearch for repositioning
- After tombstones are added or before the first ShouldDelete() invocation, the current position is set to invalid, which forces kBinarySearch to be used.
- Non-iterator users (i.e., Get()) use kFullScan, which has the same behavior as before---scan the whole map for every key passed to ShouldDelete().
Closes https://github.com/facebook/rocksdb/pull/1701
Differential Revision: D4350318
Pulled By: ajkr
fbshipit-source-id: 5129b76
2017-01-05 18:22:46 +00:00
|
|
|
return;
|
|
|
|
}
|
2022-08-30 02:09:29 +00:00
|
|
|
rep_.AddTombstones(std::make_unique<TruncatedRangeDelIterator>(
|
|
|
|
std::move(input_iter), icmp_, smallest, largest));
|
2016-12-20 00:44:30 +00:00
|
|
|
}
|
|
|
|
|
2019-04-18 19:22:29 +00:00
|
|
|
bool ReadRangeDelAggregator::ShouldDeleteImpl(const ParsedInternalKey& parsed,
|
|
|
|
RangeDelPositioningMode mode) {
|
2018-12-18 01:26:56 +00:00
|
|
|
return rep_.ShouldDelete(parsed, mode);
|
2018-10-17 18:45:30 +00:00
|
|
|
}
|
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
bool ReadRangeDelAggregator::IsRangeOverlapped(const Slice& start,
|
|
|
|
const Slice& end) {
|
|
|
|
InvalidateRangeDelMapPositions();
|
|
|
|
return rep_.IsRangeOverlapped(start, end);
|
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 19:04:56 +00:00
|
|
|
}
|
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
void CompactionRangeDelAggregator::AddTombstones(
|
|
|
|
std::unique_ptr<FragmentedRangeTombstoneIterator> input_iter,
|
|
|
|
const InternalKey* smallest, const InternalKey* largest) {
|
|
|
|
if (input_iter == nullptr || input_iter->empty()) {
|
|
|
|
return;
|
2016-11-19 00:54:09 +00:00
|
|
|
}
|
2022-09-30 23:13:03 +00:00
|
|
|
// This bounds output of CompactionRangeDelAggregator::NewIterator.
|
|
|
|
if (!trim_ts_.empty()) {
|
|
|
|
assert(icmp_->user_comparator()->timestamp_size() > 0);
|
|
|
|
input_iter->SetTimestampUpperBound(&trim_ts_);
|
|
|
|
}
|
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
assert(input_iter->lower_bound() == 0);
|
|
|
|
assert(input_iter->upper_bound() == kMaxSequenceNumber);
|
|
|
|
parent_iters_.emplace_back(new TruncatedRangeDelIterator(
|
|
|
|
std::move(input_iter), icmp_, smallest, largest));
|
|
|
|
|
2022-09-30 23:13:03 +00:00
|
|
|
Slice* ts_upper_bound = nullptr;
|
|
|
|
if (!ts_upper_bound_.empty()) {
|
|
|
|
assert(icmp_->user_comparator()->timestamp_size() > 0);
|
|
|
|
ts_upper_bound = &ts_upper_bound_;
|
|
|
|
}
|
2018-12-18 01:26:56 +00:00
|
|
|
auto split_iters = parent_iters_.back()->SplitBySnapshot(*snapshots_);
|
|
|
|
for (auto& split_iter : split_iters) {
|
|
|
|
auto it = reps_.find(split_iter.first);
|
|
|
|
if (it == reps_.end()) {
|
|
|
|
bool inserted;
|
|
|
|
SequenceNumber upper_bound = split_iter.second->upper_bound();
|
|
|
|
SequenceNumber lower_bound = split_iter.second->lower_bound();
|
|
|
|
std::tie(it, inserted) = reps_.emplace(
|
|
|
|
split_iter.first, StripeRep(icmp_, upper_bound, lower_bound));
|
|
|
|
assert(inserted);
|
2016-11-04 18:53:38 +00:00
|
|
|
}
|
2018-12-18 01:26:56 +00:00
|
|
|
assert(it != reps_.end());
|
2022-09-30 23:13:03 +00:00
|
|
|
// ts_upper_bound is used to bound ShouldDelete() to only consider
|
|
|
|
// range tombstones under full_history_ts_low_ and trim_ts_. Keys covered by
|
|
|
|
// range tombstones that are above full_history_ts_low_ should not be
|
|
|
|
// dropped prematurely: user may read with a timestamp between the range
|
|
|
|
// tombstone and the covered key. Note that we cannot set timestamp
|
|
|
|
// upperbound on the original `input_iter` since `input_iter`s are later
|
|
|
|
// used in CompactionRangeDelAggregator::NewIterator to output range
|
|
|
|
// tombstones for persistence. We do not want to only persist range
|
|
|
|
// tombstones with timestamp lower than ts_upper_bound.
|
|
|
|
split_iter.second->SetTimestampUpperBound(ts_upper_bound);
|
2018-12-18 01:26:56 +00:00
|
|
|
it->second.AddTombstones(std::move(split_iter.second));
|
2016-11-04 18:53:38 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
bool CompactionRangeDelAggregator::ShouldDelete(const ParsedInternalKey& parsed,
|
|
|
|
RangeDelPositioningMode mode) {
|
|
|
|
auto it = reps_.lower_bound(parsed.sequence);
|
|
|
|
if (it == reps_.end()) {
|
|
|
|
return false;
|
2018-05-04 23:37:39 +00:00
|
|
|
}
|
2018-12-18 01:26:56 +00:00
|
|
|
return it->second.ShouldDelete(parsed, mode);
|
2018-05-04 23:37:39 +00:00
|
|
|
}
|
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
namespace {
|
|
|
|
|
2022-09-30 23:13:03 +00:00
|
|
|
// Produce a sorted (by start internal key) stream of range tombstones from
|
|
|
|
// `children`. lower_bound and upper_bound on user key can be
|
|
|
|
// optionally specified. Range tombstones that ends before lower_bound or starts
|
|
|
|
// after upper_bound are excluded.
|
|
|
|
// If user-defined timestamp is enabled, lower_bound and upper_bound should
|
|
|
|
// contain timestamp, but comparison is done ignoring timestamps.
|
2018-12-18 01:26:56 +00:00
|
|
|
class TruncatedRangeDelMergingIter : public InternalIterator {
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
public:
|
2018-12-18 01:26:56 +00:00
|
|
|
TruncatedRangeDelMergingIter(
|
|
|
|
const InternalKeyComparator* icmp, const Slice* lower_bound,
|
|
|
|
const Slice* upper_bound, bool upper_bound_inclusive,
|
|
|
|
const std::vector<std::unique_ptr<TruncatedRangeDelIterator>>& children)
|
|
|
|
: icmp_(icmp),
|
|
|
|
lower_bound_(lower_bound),
|
|
|
|
upper_bound_(upper_bound),
|
|
|
|
upper_bound_inclusive_(upper_bound_inclusive),
|
2022-09-30 23:13:03 +00:00
|
|
|
heap_(StartKeyMinComparator(icmp)),
|
|
|
|
ts_sz_(icmp_->user_comparator()->timestamp_size()) {
|
2018-12-18 01:26:56 +00:00
|
|
|
for (auto& child : children) {
|
|
|
|
if (child != nullptr) {
|
|
|
|
assert(child->lower_bound() == 0);
|
|
|
|
assert(child->upper_bound() == kMaxSequenceNumber);
|
|
|
|
children_.push_back(child.get());
|
|
|
|
}
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
bool Valid() const override {
|
|
|
|
return !heap_.empty() && BeforeEndKey(heap_.top());
|
|
|
|
}
|
|
|
|
Status status() const override { return Status::OK(); }
|
|
|
|
|
|
|
|
void SeekToFirst() override {
|
|
|
|
heap_.clear();
|
|
|
|
for (auto& child : children_) {
|
|
|
|
if (lower_bound_ != nullptr) {
|
|
|
|
child->Seek(*lower_bound_);
|
|
|
|
} else {
|
|
|
|
child->SeekToFirst();
|
|
|
|
}
|
|
|
|
if (child->Valid()) {
|
|
|
|
heap_.push(child);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
|
|
|
|
void Next() override {
|
2018-12-18 01:26:56 +00:00
|
|
|
auto* top = heap_.top();
|
|
|
|
top->InternalNext();
|
|
|
|
if (top->Valid()) {
|
|
|
|
heap_.replace_top(top);
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
} else {
|
|
|
|
heap_.pop();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
Slice key() const override {
|
|
|
|
auto* top = heap_.top();
|
2022-09-30 23:13:03 +00:00
|
|
|
if (ts_sz_) {
|
|
|
|
cur_start_key_.Set(top->start_key().user_key, top->seq(),
|
|
|
|
kTypeRangeDeletion, top->timestamp());
|
|
|
|
} else {
|
|
|
|
cur_start_key_.Set(top->start_key().user_key, top->seq(),
|
|
|
|
kTypeRangeDeletion);
|
|
|
|
}
|
|
|
|
assert(top->start_key().user_key.size() >= ts_sz_);
|
2018-12-18 01:26:56 +00:00
|
|
|
return cur_start_key_.Encode();
|
2018-10-09 22:15:27 +00:00
|
|
|
}
|
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
Slice value() const override {
|
|
|
|
auto* top = heap_.top();
|
2022-09-30 23:13:03 +00:00
|
|
|
if (!ts_sz_) {
|
|
|
|
return top->end_key().user_key;
|
|
|
|
}
|
|
|
|
assert(top->timestamp().size() == ts_sz_);
|
|
|
|
cur_end_key_.clear();
|
|
|
|
cur_end_key_.append(top->end_key().user_key.data(),
|
|
|
|
top->end_key().user_key.size() - ts_sz_);
|
|
|
|
cur_end_key_.append(top->timestamp().data(), ts_sz_);
|
|
|
|
return cur_end_key_;
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
}
|
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
// Unused InternalIterator methods
|
|
|
|
void Prev() override { assert(false); }
|
|
|
|
void Seek(const Slice& /* target */) override { assert(false); }
|
|
|
|
void SeekForPrev(const Slice& /* target */) override { assert(false); }
|
|
|
|
void SeekToLast() override { assert(false); }
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
|
|
|
|
private:
|
2018-12-18 01:26:56 +00:00
|
|
|
bool BeforeEndKey(const TruncatedRangeDelIterator* iter) const {
|
|
|
|
if (upper_bound_ == nullptr) {
|
|
|
|
return true;
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
}
|
2022-09-30 23:13:03 +00:00
|
|
|
int cmp = icmp_->user_comparator()->CompareWithoutTimestamp(
|
|
|
|
iter->start_key().user_key, *upper_bound_);
|
2018-12-18 01:26:56 +00:00
|
|
|
return upper_bound_inclusive_ ? cmp <= 0 : cmp < 0;
|
|
|
|
}
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
const InternalKeyComparator* icmp_;
|
|
|
|
const Slice* lower_bound_;
|
|
|
|
const Slice* upper_bound_;
|
|
|
|
bool upper_bound_inclusive_;
|
|
|
|
BinaryHeap<TruncatedRangeDelIterator*, StartKeyMinComparator> heap_;
|
|
|
|
std::vector<TruncatedRangeDelIterator*> children_;
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
|
2018-12-18 01:26:56 +00:00
|
|
|
mutable InternalKey cur_start_key_;
|
2022-09-30 23:13:03 +00:00
|
|
|
mutable std::string cur_end_key_;
|
|
|
|
size_t ts_sz_;
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
};
|
|
|
|
|
2022-11-02 21:34:24 +00:00
|
|
|
} // anonymous namespace
|
2018-12-18 01:26:56 +00:00
|
|
|
|
|
|
|
std::unique_ptr<FragmentedRangeTombstoneIterator>
|
|
|
|
CompactionRangeDelAggregator::NewIterator(const Slice* lower_bound,
|
|
|
|
const Slice* upper_bound,
|
|
|
|
bool upper_bound_inclusive) {
|
|
|
|
InvalidateRangeDelMapPositions();
|
2022-08-30 02:09:29 +00:00
|
|
|
auto merging_iter = std::make_unique<TruncatedRangeDelMergingIter>(
|
|
|
|
icmp_, lower_bound, upper_bound, upper_bound_inclusive, parent_iters_);
|
2018-12-18 01:26:56 +00:00
|
|
|
|
|
|
|
auto fragmented_tombstone_list =
|
|
|
|
std::make_shared<FragmentedRangeTombstoneList>(
|
|
|
|
std::move(merging_iter), *icmp_, true /* for_compaction */,
|
|
|
|
*snapshots_);
|
|
|
|
|
2022-08-30 02:09:29 +00:00
|
|
|
return std::make_unique<FragmentedRangeTombstoneIterator>(
|
|
|
|
fragmented_tombstone_list, *icmp_, kMaxSequenceNumber /* upper_bound */);
|
Range deletion performance improvements + cleanup (#4014)
Summary:
This fixes the same performance issue that #3992 fixes but with much more invasive cleanup.
I'm more excited about this PR because it paves the way for fixing another problem we uncovered at Cockroach where range deletion tombstones can cause massive compactions. For example, suppose L4 contains deletions from [a, c) and [x, z) and no other keys, and L5 is entirely empty. L6, however, is full of data. When compacting L4 -> L5, we'll end up with one file that spans, massively, from [a, z). When we go to compact L5 -> L6, we'll have to rewrite all of L6! If, instead of range deletions in L4, we had keys a, b, x, y, and z, RocksDB would have been smart enough to create two files in L5: one for a and b and another for x, y, and z.
With the changes in this PR, it will be possible to adjust the compaction logic to split tombstones/start new output files when they would span too many files in the grandparent level.
ajkr please take a look when you have a minute!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4014
Differential Revision: D8773253
Pulled By: ajkr
fbshipit-source-id: ec62fa85f648fdebe1380b83ed997f9baec35677
2018-07-12 21:28:10 +00:00
|
|
|
}
|
|
|
|
|
2020-02-20 20:07:53 +00:00
|
|
|
} // namespace ROCKSDB_NAMESPACE
|