Summary:
For iterator reads, a `SuperVersion` is pinned to preserve a snapshot of SST files, and `Block`s are pinned to allow `key()` and `value()` to return pointers directly into a RocksDB memory region. This works for both non-mmap reads, where the block owns the memory region, and mmap reads, where the file owns the memory region.
For point reads with `PinnableSlice`, only the `Block` object is pinned. This works for non-mmap reads because the block owns the memory region, so even if the file is deleted after compaction, the memory region survives. However, for mmap reads, file deletion causes the memory region to which the `PinnableSlice` refers to be unmapped. The result is usually a segfault upon accessing the `PinnableSlice`, although sometimes it returned wrong results (I repro'd this a bunch of times with `db_stress`).
This PR copies the value into the `PinnableSlice` when it comes from mmap'd memory. We can tell whether the `Block` owns its memory using `Block::cachable()`, which is unset when reads do not use the provided buffer as is the case with mmap file reads. When that is false we ensure the result of `Get()` is copied.
This feels like a short-term solution as ideally we'd have the `PinnableSlice` pin the mmap'd memory so we can do zero-copy reads. It seemed hard so I chose this approach to fix correctness in the meantime.
Closes https://github.com/facebook/rocksdb/pull/3881
Differential Revision: D8076288
Pulled By: ajkr
fbshipit-source-id: 31d78ec010198723522323dbc6ea325122a46b08
Summary:
Previously `db_stress` attempted to configure direct I/O dynamically in `SetOptions()` which had multiple problems (ummm must've never been tested):
- It's a DB option so SetDBOptions should've been called instead
- It's not a dynamic option so even SetDBOptions would fail
- It required enabling SyncPoint to mask O_DIRECT since it had no way to detect whether the DB directory was in tmpfs or not. This required locking that consumed ~80% of db_stress CPU.
In this PR I delete the broken dynamic config and instead configure it statically, only enabling it if the DB directory truly supports O_DIRECT.
Closes https://github.com/facebook/rocksdb/pull/3939
Differential Revision: D8238120
Pulled By: ajkr
fbshipit-source-id: 60bb2deebe6c9b54a3f788079261715b4a229279
Summary:
As titled.
I have not extended the Compatibility tests because the new WAL markers are still unimplemented.
Closes https://github.com/facebook/rocksdb/pull/3941
Differential Revision: D8238394
Pulled By: lth
fbshipit-source-id: 980e3d44837bbf2cfa64047f9738f559dfac4b1d
Summary:
This should resolve the performance regression caused by the unnecessary copying of the shared_ptr.
Closes https://github.com/facebook/rocksdb/pull/3937
Differential Revision: D8232330
Pulled By: miasantreble
fbshipit-source-id: 7885bf7cd190b6f87164c52d6edd328298c13f97
Summary:
DBTestBase::OptionConfig includes the scenarios that unit tests could iterate over them by calling ChangeOptions(). Some of the options have been mistakenly put after kEnd which makes them essentially invisible to ChangeOptions() caller. This patch fixes it except for kUniversalSubcompactions which is left as TODO since it would break some unit tests.
Closes https://github.com/facebook/rocksdb/pull/3935
Differential Revision: D8230748
Pulled By: maysamyabandeh
fbshipit-source-id: edddb8fffcd161af1809fef24798ce118f8593db
Summary:
DBImpl::FindObsoleteFiles() may call GetChildren() multiple times if different CFs are on the same path. Fix it.
Closes https://github.com/facebook/rocksdb/pull/3885
Differential Revision: D8084634
Pulled By: siying
fbshipit-source-id: b471fbc251f6a05e9243304dc14c0831060cc0b0
Summary:
Small format error below causes build to fail. I believe that this :
```
fprintf(stderr, "num reads to do %lu\n", reads_);
```
Can be changed to this:
```
fprintf(stderr, "num reads to do %" PRIu64 "\n", reads_);
```
Successful build
```
CC utilities/blob_db/blob_dump_tool.o
AR librocksdb_debug.a
ar: creating archive librocksdb_debug.a
/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/ranlib: file: librocksdb_debug.a(rocks_lua_compaction_filter.o) has no symbols
CC tools/db_bench.o
CC tools/db_bench_tool.o
tools/db_bench_tool.cc:4532:46: error: format specifies type 'unsigned long' but the argument has type 'int64_t' (aka 'long long') [-Werror,-Wformat]
fprintf(stderr, "num reads to do %lu\n", reads_);
~~~ ^~~~~~
%lld
1 error generated.
make: *** [tools/db_bench_tool.o] Error 1
```
```
$ cd rocksdb
$ make all
$ g++ --version
Configured with: --prefix=/Applications/Xcode.app/Contents/Developer/usr --with-gxx-include-dir=/usr/include/c++/4.2.1
Apple LLVM version 9.1.0 (clang-902.0.39.1)
Target: x86_64-apple-darwin17.5.0
Thread model: posix
InstalledDir: /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin
```
Closes https://github.com/facebook/rocksdb/pull/3909
Differential Revision: D8215710
Pulled By: siying
fbshipit-source-id: 15e49fb02a818fec846e9f9b2a50e372b6b67751
Summary:
In order to make valgrind check test to pass in a day, remove some tests that run prohibitively slow under valgrind.
Closes https://github.com/facebook/rocksdb/pull/3924
Differential Revision: D8210184
Pulled By: siying
fbshipit-source-id: 5b06fb08f3cf57571d422d05a0dbddc9f9376f7a
Summary:
This is still WIP, but I'm hoping for early feedback on the overall approach.
This patch implements deletion triggered compaction, which till now only
worked for leveled, for universal style. SST files are marked for
compaction by the CompactOnDeletionCollertor table property. This is
expected to be used when free disk space is low and the user wants to
reclaim space by deleting a bunch of keys. The deletions are expected to
be dense. In such a situation, we want to avoid a full compaction due to
its space overhead.
The strategy used in this case is similar to leveled. We pick one file
from the set of files marked for compaction. We then expand the inputs
to a clean cut on the same level, and then pick overlapping files from
the next non-mepty level. Picking files from the next level can cause
the key range to expand, and we opportunistically expand inputs in the
source level to include files wholly in this key range.
The main side effect of this is that it breaks the property of no time
range overlap between levels. This shouldn't break any functionality.
Closes https://github.com/facebook/rocksdb/pull/3860
Differential Revision: D8124397
Pulled By: anand1976
fbshipit-source-id: bfa2a9dd6817930e991b35d3a8e7e61304ed3dcf
Summary:
The very old sst formats do not have table_properties and rep_->table_properties is thus nullptr. The recent patch in https://github.com/facebook/rocksdb/pull/3894 does not check for nullptr and hence makes it backward incompatible. This patch adds the check.
Closes https://github.com/facebook/rocksdb/pull/3918
Differential Revision: D8188638
Pulled By: maysamyabandeh
fbshipit-source-id: b1d986665ecf0b4d1c442adfa8a193b97707d47b
Summary:
Index blocks have the same format as data blocks. The keys therefore similarly to the keys in the data blocks are internal keys, which means that in addition to the user key it also has 8 bytes that encodes sequence number and value type. This extra 8 bytes however is not necessary in index blocks since the index keys act as an separator between two data blocks. The only exception is when the last key of a block and the first key of the next block share the same user key, in which the sequence number is required to act as a separator.
The patch excludes the sequence from index keys only if the above special case does not happen for any of the index keys. It then records that in the property block. The reader looks at the property block to see if it should expect sequence numbers in the keys of the index block.s
Closes https://github.com/facebook/rocksdb/pull/3894
Differential Revision: D8118775
Pulled By: maysamyabandeh
fbshipit-source-id: 915479f028b5799ca91671d67455ecdefbd873bd
Summary:
This was missed in a refactor of `ReadBlockContents` (2f1a3a4).
Closes https://github.com/facebook/rocksdb/pull/3906
Differential Revision: D8172648
Pulled By: ajkr
fbshipit-source-id: 27e453b19795fea974bfed4721105be6f3a12090
Summary:
Please refer to earlier discussion in [issue 3609](https://github.com/facebook/rocksdb/issues/3609).
There was also an alternative fix in [PR 3888](https://github.com/facebook/rocksdb/pull/3888), but the proposed solution requires complex change.
To summarize the cause of the problem. Upon creation of a column family, a `BlockBasedTableFactory` object is `new`ed and encapsulated by a `std::shared_ptr`. Since there is no other `std::shared_ptr` pointing to this `BlockBasedTableFactory`, when the column family is dropped, the `ColumnFamilyData` is `delete`d, causing the destructor of `std::shared_ptr`. Since there is no other `std::shared_ptr`, the underlying memory is also freed.
Later when the db exits, it releases all the table readers, including the table readers that have been operating on the dropped column family. This needs to access the `table_options` owned by `BlockBasedTableFactory` that has already been deleted. Therefore, a segfault is raised.
Previous workaround is to purge all obsolete files upon `ColumnFamilyData` destruction, which leads to a force release of table readers of the dropped column family. However this does not work when the user disables file deletion.
Our solution in this PR is making a copy of `table_options` in `BlockBasedTable::Rep`. This solution increases memory copy and usage, but is much simpler.
Test plan
```
$ make -j16
$ ./column_family_test --gtest_filter=ColumnFamilyTest.CreateDropAndDestroy:ColumnFamilyTest.CreateDropAndDestroyWithoutFileDeletion
```
Expected behavior:
All tests should pass.
Closes https://github.com/facebook/rocksdb/pull/3898
Differential Revision: D8149421
Pulled By: riversand963
fbshipit-source-id: eaecc2e064057ef607fbdd4cc275874f866c3438
Summary:
```PosixMmapReadableFile::fd_``` is closed after created, but needs to remain open for the lifetime of `PosixMmapReadableFile` since it is used whenever `InvalidateCache` is called.
Closes https://github.com/facebook/rocksdb/pull/2764
Differential Revision: D8152515
Pulled By: ajkr
fbshipit-source-id: b738a6a55ba4e392f9b0f374ff396a1e61c64f65
Summary:
Add flush_before_backup to rocksdb_backup_engine_create_new_backup. make c api able to control the flush before backup behavior.
Closes https://github.com/facebook/rocksdb/pull/3897
Differential Revision: D8157676
Pulled By: ajkr
fbshipit-source-id: 88998c62f89f087bf8672398fd7ddafabbada505
Summary:
Implement midpoint insertion strategy where new blocks will be insert to the middle of LRU list, then move the head on the first hit in cache.
Closes https://github.com/facebook/rocksdb/pull/3877
Differential Revision: D8100895
Pulled By: yiwu-arbug
fbshipit-source-id: f4bd83cb8be469e5d02072cfc8bd66011391f3da
Summary:
Catch up with Posix features
NewWritableRWFile must fail when file does not exists
Implement Env::Truncate()
Adjust Env options optimization functions
Implement MemoryMappedBuffer on Windows.
Closes https://github.com/facebook/rocksdb/pull/3857
Differential Revision: D8053610
Pulled By: ajkr
fbshipit-source-id: ccd0d46c29648a9f6f496873bc1c9d6c5547487e
Summary:
to workaround issue of http://tracker.ceph.com/issues/21422 .
and in tcmalloc aligned_alloc and posix_memalign() are basically the
same thing. the same applies to GNU glibc.
fixes#3175
Signed-off-by: Kefu Chai <tchaikov@gmail.com>
Closes https://github.com/facebook/rocksdb/pull/3862
Differential Revision: D8147930
Pulled By: yiwu-arbug
fbshipit-source-id: 355afe93c4dd0a96a0d711ef190e8b86fbe8d11d
Summary:
Update LRUCacheShard constructor so that adding new params to it don't need to add extra SetXXX() methods.
Closes https://github.com/facebook/rocksdb/pull/3896
Differential Revision: D8128618
Pulled By: yiwu-arbug
fbshipit-source-id: 6afa715de1493a50de413678761a765e3af9b83b
Summary:
Previously we were using -1 as the default for every library, which was legacy from our zlib options. That worked for a while, but after zstd introduced a146ee04ae, it started giving poor compression ratios by default in zstd.
This PR adds a constant to RocksDB public API, `CompressionOptions::kDefaultCompressionLevel`, which will get translated to the default value specific to the compression library being used in "util/compression.h". The constant uses a number that appears to be larger than any library's maximum compression level.
Closes https://github.com/facebook/rocksdb/pull/3895
Differential Revision: D8125780
Pulled By: ajkr
fbshipit-source-id: 2db157a89118cd4f94577c2f4a0a5ff31c8391c6
Summary:
Explicitly specify the underlying type of enums help developers understand the physical storage.
Closes https://github.com/facebook/rocksdb/pull/3892
Differential Revision: D8107027
Pulled By: riversand963
fbshipit-source-id: a00efecbba46df4a3c8eed0994a2d4972ad1a1d3
Summary:
Currently when ldb built from master tries to open a DB from version 2.2, there will be a segfault because table_properties didn't exist back then.
Closes https://github.com/facebook/rocksdb/pull/3890
Differential Revision: D8100914
Pulled By: miasantreble
fbshipit-source-id: b255e8aedc54695432be2e704839c857dabdd65a
Summary:
Fix Issue #3771 : Check for nullptr in Slice constructor
Slice ctor checks for nullptr and creates empty string if the string does not exist
Closes https://github.com/facebook/rocksdb/pull/3887
Differential Revision: D8098852
Pulled By: ajkr
fbshipit-source-id: 04471077defa9776ce7b8c389a61312ce31002fb
Summary:
DBTest.GroupCommitTest would often fail when run under valgrind because its sleeps were insufficient to guarantee a group commit had multiple entries. Instead we can use sync point to force a leader to wait until a non-leader thread has enqueued its work, thus guaranteeing a leader can do group commit work for multiple threads.
Closes https://github.com/facebook/rocksdb/pull/3883
Differential Revision: D8079429
Pulled By: ajkr
fbshipit-source-id: 61dc50fad29d2c85547842f681288de60fa29049
Summary:
I repro'd some of the "unexpected value" failures showing up in our CI lately and they always happened on keys that have a mix of single deletes and merge operands. The `SingleDelete()` API comment mentions it's incompatible with `Merge()`, so this PR prevents `db_stress` from mixing them.
Closes https://github.com/facebook/rocksdb/pull/3878
Differential Revision: D8097346
Pulled By: ajkr
fbshipit-source-id: 357a48c6a31156f4f8db3ce565638ad924c437a1
Summary:
By using WritableFileWriter rather than WritableFile directly, we can buffer multiple Append() calls to one write() file system call, which will be expensive to underlying Env without its own write buffering.
Closes https://github.com/facebook/rocksdb/pull/3882
Differential Revision: D8080673
Pulled By: siying
fbshipit-source-id: e0db900cb3c178166aa738f3985db65e3ae2cf1b
Summary:
Currently it is not possible to change bloom filter config without restart the db, which is causing a lot of operational complexity for users.
This PR aims to make it possible to dynamically change bloom filter config.
Closes https://github.com/facebook/rocksdb/pull/3601
Differential Revision: D7253114
Pulled By: miasantreble
fbshipit-source-id: f22595437d3e0b86c95918c484502de2ceca120c
Summary:
Change `keys_` from `set<string>` to `vector<set<string>>` so that each column
family's keys are stored in one set.
ajkr When you have a chance, can you PTAL? Thanks!
Closes https://github.com/facebook/rocksdb/pull/3871
Differential Revision: D8056447
Pulled By: riversand963
fbshipit-source-id: 650d0f9cad02b1bc005fc329ad76edbf053e6386
Summary:
Previously it only printed percentiles, even though our histogram keeps track of count and sum (and more). There have been many times we want to know more than the percentiles. For example, we currently want sum of "rocksdb.compression.times.nanos" and sum of "rocksdb.decompression.times.nanos", which would allow us to know the relative cost of compression vs decompression.
This PR adds count and sum to the string printed by `StatisticsImpl::ToString`. This is a bit risky as there are definitely parsers assuming the old format. I will mention it in HISTORY.md and hope for the best...
Closes https://github.com/facebook/rocksdb/pull/3863
Differential Revision: D8038831
Pulled By: ajkr
fbshipit-source-id: 0465b72e4b0cbf18ef965f4efe402601d16d5b5c
Summary:
`RangeDelAggregator` holds the pointers returned by `BlockIter::key()` and `BlockIter::value()` so requires the data to which they point is pinned. `BlockIter::key()` points into block memory and is guaranteed to be pinned if and only if prefix encoding is disabled (or, equivalently, restart interval is set to one). I think `BlockIter::value()` is always pinned. Added an assert for these and removed the wrong TODO about increasing restart interval, which would enable key prefix encoding and break the assertion.
Closes https://github.com/facebook/rocksdb/pull/3875
Differential Revision: D8063667
Pulled By: ajkr
fbshipit-source-id: 60b5ebcc0cdd610dd6aad9e74a23378793672c41
Summary:
We only generate the header dependency (".cc.d") files for files mentioned in "src.mk". When we don't generate them, changes to header dependencies do not cause `make` to recompile the dependent ".o". Then it takes a while for developers (or maybe just me) to realize `make clean` is necessary.
Closes https://github.com/facebook/rocksdb/pull/3876
Differential Revision: D8065389
Pulled By: ajkr
fbshipit-source-id: 0f62eee7bcab15b0215791564e6ab3775d46996b
Summary:
Right now, every Block::NewIterator() reads num_restarts_ from the block, which is already read in Block::Block(). This sometimes cause a CPU cache miss. Although fetching this cacheline can usually benefit follow-up block restart offset reading, as they are close to each other, it's almost free to get ride of this read by storing it in the Block class.
Closes https://github.com/facebook/rocksdb/pull/3869
Differential Revision: D8052493
Pulled By: siying
fbshipit-source-id: 9c72360f0c2d7329f3c198ce4eaedd2bc14b87c1
Summary:
In the past, the default value of max_manifest_file_size is uint64_t::MAX,
allowing a long running RocksDB process to grow its MANIFEST file to take up
the entire disk, as reported in [issue 3851](https://github.com/facebook/rocksdb/issues/3851). It is reasonable and common to provide a default non-max value for this option. Therefore, I set the value to 1GB.
siying miasantreble Please let me know whether this looks good to you. Thanks!
Closes https://github.com/facebook/rocksdb/pull/3867
Differential Revision: D8051524
Pulled By: riversand963
fbshipit-source-id: 50251f0804b1fa933a19a30d19d261ea8b9d2b72
Summary:
Right now ReverseBytewiseComparator::FindShortestSeparator() doesn't really shorten key, and ReverseBytewiseComparator::FindShortestSuccessor() seems to return wrong results. The code is confusing too as it uses BytewiseComparatorImpl::FindShortestSeparator() but the function actually won't do anything if the the first key is larger than the second.
Implement ReverseBytewiseComparator::FindShortestSeparator() and override ReverseBytewiseComparator::FindShortestSuccessor() to be empty.
Closes https://github.com/facebook/rocksdb/pull/3836
Differential Revision: D7959762
Pulled By: siying
fbshipit-source-id: 93acb621c16ce6f23e087ae4e19f7d84d1254683
Summary:
this will fix the failing clang_check test
Closes https://github.com/facebook/rocksdb/pull/3868
Differential Revision: D8050880
Pulled By: miasantreble
fbshipit-source-id: 749932e2e4025f835c961c068d601e522a126da6
Summary:
Before this PR, Iterator/InternalIterator may simultaneously have non-ok status() and Valid() = true. That state means that the last operation failed, but the iterator is nevertheless positioned on some unspecified record. Likely intended uses of that are:
* If some sst files are corrupted, a normal iterator can be used to read the data from files that are not corrupted.
* When using read_tier = kBlockCacheTier, read the data that's in block cache, skipping over the data that is not.
However, this behavior wasn't documented well (and until recently the wiki on github had misleading incorrect information). In the code there's a lot of confusion about the relationship between status() and Valid(), and about whether Seek()/SeekToLast()/etc reset the status or not. There were a number of bugs caused by this confusion, both inside rocksdb and in the code that uses rocksdb (including ours).
This PR changes the convention to:
* If status() is not ok, Valid() always returns false.
* Any seek operation resets status. (Before the PR, it depended on iterator type and on particular error.)
This does sacrifice the two use cases listed above, but siying said it's ok.
Overview of the changes:
* A commit that adds missing status checks in MergingIterator. This fixes a bug that actually affects us, and we need it fixed. `DBIteratorTest.NonBlockingIterationBugRepro` explains the scenario.
* Changes to lots of iterator types to make all of them conform to the new convention. Some bug fixes along the way. By far the biggest changes are in DBIter, which is a big messy piece of code; I tried to make it less big and messy but mostly failed.
* A stress-test for DBIter, to gain some confidence that I didn't break it. It does a few million random operations on the iterator, while occasionally modifying the underlying data (like ForwardIterator does) and occasionally returning non-ok status from internal iterator.
To find the iterator types that needed changes I searched for "public .*Iterator" in the code. Here's an overview of all 27 iterator types:
Iterators that didn't need changes:
* status() is always ok(), or Valid() is always false: MemTableIterator, ModelIter, TestIterator, KVIter (2 classes with this name anonymous namespaces), LoggingForwardVectorIterator, VectorIterator, MockTableIterator, EmptyIterator, EmptyInternalIterator.
* Thin wrappers that always pass through Valid() and status(): ArenaWrappedDBIter, TtlIterator, InternalIteratorFromIterator.
Iterators with changes (see inline comments for details):
* DBIter - an overhaul:
- It used to silently skip corrupted keys (`FindParseableKey()`), which seems dangerous. This PR makes it just stop immediately after encountering a corrupted key, just like it would for other kinds of corruption. Let me know if there was actually some deeper meaning in this behavior and I should put it back.
- It had a few code paths silently discarding subiterator's status. The stress test caught a few.
- The backwards iteration code path was expecting the internal iterator's set of keys to be immutable. It's probably always true in practice at the moment, since ForwardIterator doesn't support backwards iteration, but this PR fixes it anyway. See added DBIteratorTest.ReverseToForwardBug for an example.
- Some parts of backwards iteration code path even did things like `assert(iter_->Valid())` after a seek, which is never a safe assumption.
- It used to not reset status on seek for some types of errors.
- Some simplifications and better comments.
- Some things got more complicated from the added error handling. I'm open to ideas for how to make it nicer.
* MergingIterator - check status after every operation on every subiterator, and in some places assert that valid subiterators have ok status.
* ForwardIterator - changed to the new convention, also slightly simplified.
* ForwardLevelIterator - fixed some bugs and simplified.
* LevelIterator - simplified.
* TwoLevelIterator - changed to the new convention. Also fixed a bug that would make SeekForPrev() sometimes silently ignore errors from first_level_iter_.
* BlockBasedTableIterator - minor changes.
* BlockIter - replaced `SetStatus()` with `Invalidate()` to make sure non-ok BlockIter is always invalid.
* PlainTableIterator - some seeks used to not reset status.
* CuckooTableIterator - tiny code cleanup.
* ManagedIterator - fixed some bugs.
* BaseDeltaIterator - changed to the new convention and fixed a bug.
* BlobDBIterator - seeks used to not reset status.
* KeyConvertingIterator - some small change.
Closes https://github.com/facebook/rocksdb/pull/3810
Differential Revision: D7888019
Pulled By: al13n321
fbshipit-source-id: 4aaf6d3421c545d16722a815b2fa2e7912bc851d
Summary:
log_ contract specifies that it should not be modified unless both mutex_ and log_write_mutex_ are held. log_.erase however does that with only holding mutex_. This causes a race condition with two_write_queues since logs_.back is read with holding only log_write_mutex_ (which is correct according to logs_ contract) but logs_.erase is called concurrently. This is probably the cause of logs_.back returning nullptr in https://github.com/facebook/rocksdb/issues/3852 although I could not reproduce it.
Fixes https://github.com/facebook/rocksdb/issues/3852
Closes https://github.com/facebook/rocksdb/pull/3859
Differential Revision: D8026103
Pulled By: maysamyabandeh
fbshipit-source-id: ee394e00fe4aa520d884c5ef87981e9d6b5ccb28
Andrew Kryczka