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
Summary:
There were three implementations of VectorIterator (util/vector_iterator, test_util/testutil.h and LoggingForwardVectorIterator). Merged them into one class to increase code coverage/testing and reduce duplication.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/8901
Reviewed By: pdillinger
Differential Revision: D31022673
Pulled By: mrambacher
fbshipit-source-id: 8e3acbd2dfd60b4df609d02cc72846de2389d531
Summary:
When dynamically linking two binaries together, different builds of RocksDB from two sources might cause errors. To provide a tool for user to solve the problem, the RocksDB namespace is changed to a flag which can be overridden in build time.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6433
Test Plan: Build release, all and jtest. Try to build with ROCKSDB_NAMESPACE with another flag.
Differential Revision: D19977691
fbshipit-source-id: aa7f2d0972e1c31d75339ac48478f34f6cfcfb3e
Summary:
There are too many types of files under util/. Some test related files don't belong to there or just are just loosely related. Mo
ve them to a new directory test_util/, so that util/ is cleaner.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5377
Differential Revision: D15551366
Pulled By: siying
fbshipit-source-id: 0f5c8653832354ef8caa31749c0143815d719e2c
Summary:
Now that v2 is fully functional, the v1 aggregator is removed.
The v2 aggregator has been renamed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4778
Differential Revision: D13495930
Pulled By: abhimadan
fbshipit-source-id: 9d69500a60a283e79b6c4fa938fc68a8aa4d40d6
Summary:
To more accurately truncate range tombstones at SST boundaries,
we now represent them in RangeDelAggregator using InternalKeys, which
are end-key-exclusive as they were before this change.
During compaction, "atomic compaction unit boundaries" (the range of
keys contained in neighbouring and overlaping SSTs) are propagated down
to RangeDelAggregator to truncate range tombstones at those boundariies
instead. See https://github.com/facebook/rocksdb/pull/4432#discussion_r221072219 and https://github.com/facebook/rocksdb/pull/4432#discussion_r221138683
for motivating examples.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4432
Differential Revision: D10263952
Pulled By: abhimadan
fbshipit-source-id: 2fe85ff8a02b3a6a2de2edfe708012797a7bd579
Summary:
The CollapsedRangeDelMap was entirely mishandling tombstones at the same
sequence number when the tombstones did not have identical start and end
keys. Such tombstones are common since 90fc40690, which causes
tombstones to be split during compactions.
For example, if the tombstone [a, c) @ 1 lies across a compaction
boundary at b, it will be split into [a, b) @ 1 and [b, c) @ 1. Without
this patch, the collapsed range deletion map would look like this:
a -> 1
b -> 1
c -> 0
Notice how the b -> 1 entry is redundant. When the tombstones overlap,
the problem is even worse. Consider tombstones [a, c) @ 1 and [b, d) @
1, which produces this map without this patch:
a -> 1
b -> 1
c -> 0
d -> 0
This map is corrupt, as a map can never contain adjacent sentinel (zero)
entries. When the iterator advances from b to c, it will notice that c
is a sentinel enty and skip to d--but d is also a sentinel entry! Asking
what tombstone this iterator points to will trigger an assertion, as it
is not pointing to a valid tombstone.
/cc ajkr
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4424
Differential Revision: D10039248
Pulled By: abhimadan
fbshipit-source-id: 6d737c1e88d60e80cf27286726627ba44463e7f4
Summary:
Add a unit test for range collapsing when non-default comparator is used. This exposes the bug fixed in #4386.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4388
Differential Revision: D9918252
Pulled By: ajkr
fbshipit-source-id: 99501b96b251eab41791a7e33b27055ee36c5c39
Summary:
`RangeDelAggregator::AddTombstones` contained an assertion which stated that, if a range tombstone extended past the largest key in the sstable, then `FileMetaData::largest` must have a sentinel sequence number of `kMaxSequenceNumber`, which implies that the tombstone's end key is safe to truncate. However, `largest` will not be a sentinel key when the next sstable in the level's smallest key is equal to the current sstable's largest key, which caused the assertion to fail.
The assertion must hold for the truncation to be safe, so it has been moved to an additional check on end-key truncation.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4356
Differential Revision: D9760891
Pulled By: abhimadan
fbshipit-source-id: 7c20c3885cd919dcd14f291f88fd27aa33defebc
Summary:
Do not consider the range tombstone sentinel key as causing 2 adjacent
sstables in a level to overlap. When a range tombstone's end key is the
largest key in an sstable, the sstable's end key is so to a "sentinel"
value that is the smallest key in the next sstable with a sequence
number of kMaxSequenceNumber. This "sentinel" is guaranteed to not
overlap in internal-key space with the next sstable. Unfortunately,
GetOverlappingFiles uses user-keys to determine overlap and was thus
considering 2 adjacent sstables in a level to overlap if they were
separated by this sentinel key. This in turn would cause compactions to
be larger than necessary.
Note that this conflicts with
https://github.com/facebook/rocksdb/pull/2769 and cases
`DBRangeDelTest.CompactionTreatsSplitInputLevelDeletionAtomically` to
fail.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4050
Differential Revision: D8844423
Pulled By: ajkr
fbshipit-source-id: df3f9f1db8f4cff2bff77376b98b83c2ae1d155b
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
Summary:
Before we were checking every file in the level which was unnecessary. We can piggyback onto the code for checking point-key overlap, which already opens all the files that could possibly contain overlapping range deletions. This PR makes us check just the range deletions from those files, so no extra ones will be opened.
Closes https://github.com/facebook/rocksdb/pull/3179
Differential Revision: D6358125
Pulled By: ajkr
fbshipit-source-id: 00e200770fdb8f3cc6b1b2da232b755e4ba36279
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
Summary:
Added a tombstone-collapsing mode to RangeDelAggregator, which eliminates overlap in the TombstoneMap. In this mode, we can check whether a tombstone covers a user key using upper_bound() (i.e., binary search). However, the tradeoff is the overhead to add tombstones is now higher, so at first I've only enabled it for range scans (compaction/flush/user iterators), where we expect a high number of calls to ShouldDelete() for the same tombstones. Point queries like Get() will still use the linear scan approach.
Also in this diff I changed RangeDelAggregator's TombstoneMap to use multimap with user keys instead of map with internal keys. Callers sometimes provided ParsedInternalKey directly, from which it would've required string copying to derive an internal key Slice with which we could search the map.
Closes https://github.com/facebook/rocksdb/pull/1614
Differential Revision: D4270397
Pulled By: ajkr
fbshipit-source-id: 93092c7
Changyu Bi