Summary:
Finalize re-sorts (read: mutates) the files_ in Version* and it is called by CompactionPicker during normal runtime. At the same time, this same Version* lives in the SuperVersion* and is accessed without the mutex in GetImpl() code path.
Mutating the files_ in one thread and reading the same files_ in another thread is a bad idea. It caused this issue: http://ci-builds.fb.com/job/rocksdb_crashtest/285/console
Long-term, we need to be more careful with method contracts and clearly document what state can be mutated when. Now that we are much faster because we don't lock in GetImpl(), we keep running into data races that were not a problem before when we were slower. db_stress has been very helpful in detecting those.
Short-term, I removed Finalize() from CompactionPicker.
Note: I believe this is an issue in current 2.7 version running in production.
Test Plan:
make check
Will also run db_stress to see if issue is gone
Reviewers: sdong, ljin, dhruba, haobo
Reviewed By: sdong
CC: leveldb
Differential Revision: https://reviews.facebook.net/D16983
Summary:
When the manifest is getting rolled the following happens:
1) manifest_file_number_ is assigned to a new manifest number (even though the old one is still current)
2) mutex is unlocked
3) SetCurrentFile() creates temporary file manifest_file_number_.dbtmp
4) SetCurrentFile() renames manifest_file_number_.dbtmp to CURRENT
5) mutex is locked
If FindObsoleteFiles happens between (3) and (4) it will:
1) Delete manifest_file_number_.dbtmp (because it's not in pending_outputs_)
2) Delete old manifest (because the manifest_file_number_ already points to a new one)
I introduce the concept of prev_manifest_file_number_ that will avoid the race condition.
However, we should discuss the future of MANIFEST file rolling. We found some race conditions with it last week and who knows how many more are there. Nobody is using it in production because we don't trust the implementation. Should we even support it?
Test Plan: make check
Reviewers: ljin, dhruba, haobo, sdong
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D16929
Summary: Moved LogBuffer class to an internal header. Removed some unneccesary indirection. Enabled log buffer for BackgroundCallFlush. Forced log buffer flush right after Unlock to improve time ordering of info log.
Test Plan: make check; db_bench compare LOG output
Reviewers: sdong
Reviewed By: sdong
CC: leveldb, igor
Differential Revision: https://reviews.facebook.net/D16707
Summary: Now while the background thread is picking compactions, it writes out multiple info_logs, especially for universal compaction, which introduces a chance of waiting log writing in mutex, which is bad. To remove this risk, write all those info logs to a buffer and flush it after releasing the mutex.
Test Plan:
make all check
check the log lines while running some tests that trigger compactions.
Reviewers: haobo, igor, dhruba
Reviewed By: dhruba
CC: i.am.jin.lei, dhruba, yhchiang, leveldb, nkg-
Differential Revision: https://reviews.facebook.net/D16515
Summary: Provide a public API for users to access the table properties for each SSTable.
Test Plan: Added a unit tests to test the function correctness under differnet conditions.
Reviewers: haobo, dhruba, sdong
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D16083
Summary:
This patch optimized Get() code paths by avoiding malloc of iterators. Iterator creation is moved to mem table rep implementations, where a callback is called when any key is found. This is the same practice as what we do in (SST) table readers.
db_bench result for readrandom following a writeseq, with no compression, single thread and tmpfs, we see throughput improved to 144958 from 139027, about 3%.
Test Plan: make all check
Reviewers: dhruba, haobo, igor
Reviewed By: haobo
CC: leveldb, yhchiang
Differential Revision: https://reviews.facebook.net/D14685
Summary:
@dhruba, I'm not sure where we need to sync the directory. I implemented the function in Env() and added the dir sync just after we close the newly created file in the builder.
Should I also add FsyncDir() to new files that get created by a compaction?
Test Plan: Confirmed that FsyncDir is returning Status::OK()
Reviewers: dhruba, haobo
Reviewed By: dhruba
CC: leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D14751
Summary: There is no reason to have functions NeedCompaction(), MaxCompactionScore() and MaxCompactionScoreLevel() in VersionSet, since they don't access any data in VersionSet.
Test Plan: make check
Reviewers: kailiu, haobo, sdong
Reviewed By: kailiu
CC: leveldb
Differential Revision: https://reviews.facebook.net/D15333
Summary:
A lot of our code implicitly assumes number_levels to be static. ReduceNumberOfLevels() breaks that assumption. For example, after calling ReduceNumberOfLevels(), DBImpl::NumberLevels() will be different from VersionSet::NumberLevels(). This is dangerous. Thankfully, it's not in public headers and is only used from LDB cmd tool. LDB tool is only using it statically, i.e. it never calls it with running DB instance. With this diff, we make it explicitly static. This way, we can assume number_levels to be immutable and not break assumption that lot of our code is relying upon. LDB tool can still use the method.
Also, I removed the method from a separate file since it breaks filename completition. version_se<TAB> now completes to "version_set." instead of "version_set" (without the dot). I don't see a big reason that the function should be in a different file.
Test Plan: reduce_levels_test
Reviewers: dhruba, haobo, kailiu, sdong
Reviewed By: kailiu
CC: leveldb
Differential Revision: https://reviews.facebook.net/D15303
Summary: The only thing we do with compaction pointers is set them to some values, we never actually read them. I don't know what we used them for, but it doesn't look like we use them anymore.
Test Plan: make check
Reviewers: dhruba, haobo, kailiu, sdong
Reviewed By: kailiu
CC: leveldb
Differential Revision: https://reviews.facebook.net/D15225
Summary:
This is a big one. This diff moves all the code related to picking compactions from VersionSet to new class CompactionPicker. Column families' compactions will be completely separate processes, so we need to have multiple CompactionPickers.
To make this easier to review, most of the code change is just copy/paste. There is also a small change not to use VersionSet::current_, but rather to take `Version* version` as a parameter. Most of the other code is exactly the same.
In future diffs, I will also make some improvements to CompactionPickers. I think the most important part will be encapsulating it better. Currently Version, VersionSet, Compaction and CompactionPicker are all friend classes, which makes it harder to change the implementation.
This diff depends on D15171, D15183, D15189 and D15201
Test Plan: `make check`
Reviewers: kailiu, sdong, dhruba, haobo
Reviewed By: kailiu
CC: leveldb
Differential Revision: https://reviews.facebook.net/D15207
Summary:
I'm sure we'll all agree that version_set.cc needs simplifying. This diff moves Compaction class to a separate file.
The diff depends on D15171 and D15183
Test Plan: make check
Reviewers: dhruba, haobo, kailiu, sdong
Reviewed By: kailiu
CC: leveldb
Differential Revision: https://reviews.facebook.net/D15189
Summary:
There were some functions in VersionSet that had no reason to be there instead of Version. Moving them to Version will make column families implementation easier.
The functions moved are:
* NumLevelBytes
* LevelSummary
* LevelFileSummary
* MaxNextLevelOverlappingBytes
* AddLiveFiles (previously AddLiveFilesCurrentVersion())
* NeedSlowdownForNumLevel0Files
The diff continues on (and depends on) D15171
Test Plan: make check
Reviewers: dhruba, haobo, kailiu, sdong, emayanke
Reviewed By: sdong
CC: leveldb
Differential Revision: https://reviews.facebook.net/D15183
Summary:
With column families VersionSet will not have a constant number of levels (each CF can have different options), so we'll need to eliminate call to VersionSet::NumberLevels()
This diff decreases number of callsites, but we're not there yet. It associates number of levels with Version (each version is associated with single CF) instead of VersionSet.
I have also slightly changed how VersionSet keeps track of manifest size.
This diff also modifies constructor of Compaction such that it takes input_version and automatically Ref()s it. Before this was done outside of constructor.
In next diffs I will continue to decrease number of callsites of VersionSet::NumberLevels() and also references to current_
Test Plan: make check
Reviewers: haobo, dhruba, kailiu, sdong
Reviewed By: sdong
Differential Revision: https://reviews.facebook.net/D15171
Summary:
When doing CompactRange(), we should first flush the memtable and then calculate max_level_with_files. Also, we want to compact all the levels that have files, including level `max_level_with_files`.
This patch fixed the unit test.
Test Plan: Added a failing unit test and a fix, so it's not failing anymore.
Reviewers: dhruba, haobo, sdong
Reviewed By: haobo
CC: leveldb, xjin
Differential Revision: https://reviews.facebook.net/D14421
Summary: Currently in DBImpl::MakeRoomForWrite(), we do "versions_->NumLevelFiles(0) >= options_.level0_slowdown_writes_trigger" to check whether the writer thread needs to slow down. However, versions_->NumLevelFiles(0) is slightly more expensive than we expected. By caching the result of the comparison when installing a new version, we can avoid this function call every time.
Test Plan:
make all check
Manually trigger this behavior by applying universal compaction style and make sure inserts are made slow after there are certain number of files.
Reviewers: haobo, kailiu, igor
Reviewed By: kailiu
CC: nkg-, leveldb
Differential Revision: https://reviews.facebook.net/D15141
Summary:
In some use cases, table readers for all live files should always be cached. In that case, there will be an opportunity to avoid the table cache look-up while Get() and NewIterator().
We define options.max_open_files = -1 to be the mode that table readers for live files will always be kept. In that mode, table readers are cached in FileMetaData (with a reference count hold in table cache). So that when executing table_cache.Get() and table_cache.newInterator(), LRU cache checking can be by-passed, to reduce latency.
Test Plan: add a test case in db_test
Reviewers: haobo, kailiu
Reviewed By: haobo
CC: dhruba, igor, leveldb
Differential Revision: https://reviews.facebook.net/D15039
Summary: I made some cleanup while reading the source code in `db`. Most changes are about style, naming or C++ 11 new features.
Test Plan: ran `make check`
Reviewers: haobo, dhruba, sdong
CC: leveldb
Differential Revision: https://reviews.facebook.net/D15009
Summary:
Instead of locking and saving a DB state, we can cache a DB state and update it only when it changes. This change reduces lock contention and speeds up read operations on the DB.
Performance improvements are substantial, although there is some cost in no-read workloads. I ran the regression tests on my devserver and here are the numbers:
overwrite 56345 -> 63001
fillseq 193730 -> 185296
readrandom 771301 -> 1219803 (58% improvement!)
readrandom_smallblockcache 677609 -> 862850
readrandom_memtable_sst 710440 -> 1109223
readrandom_fillunique_random 221589 -> 247869
memtablefillrandom 105286 -> 92643
memtablereadrandom 763033 -> 1288862
Test Plan:
make asan_check
I am also running db_stress
Reviewers: dhruba, haobo, sdong, kailiu
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D14679
Summary: When deconstructing an iterator, no need to check obsolete file if it doesn't hold last reference of any version.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, kailiu
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D14595
Summary: In get operations, merge_operands is only used in few cases. Lazily initialize it can reduce average latency in some cases
Test Plan: make all check
Reviewers: haobo, kailiu, dhruba
Reviewed By: haobo
CC: igor, nkg-, leveldb
Differential Revision: https://reviews.facebook.net/D14415
Conflicts:
db/db_impl.cc
db/memtable.cc
Summary: In get operations, merge_operands is only used in few cases. Lazily initialize it can reduce average latency in some cases
Test Plan: make all check
Reviewers: haobo, kailiu, dhruba
Reviewed By: haobo
CC: igor, nkg-, leveldb
Differential Revision: https://reviews.facebook.net/D14415
Summary:
@haobo's suggestions from https://reviews.facebook.net/D13827
Renaming some variables, deprecating purge_log_after_flush, changing for loop into auto for loop.
I have not implemented deleting objects outside of mutex yet because it would require a big code change - we would delete object in db_impl, which currently does not know anything about object because it's defined in version_edit.h (FileMetaData). We should do it at some point, though.
Test Plan: Ran deletefile_test
Reviewers: haobo
Reviewed By: haobo
CC: leveldb, haobo
Differential Revision: https://reviews.facebook.net/D14025
Summary:
Here's one solution we discussed on speeding up FindObsoleteFiles. Keep a set of all files in DBImpl and update the set every time we create a file. I probably missed few other spots where we create a file.
It might speed things up a bit, but makes code uglier. I don't really like it.
Much better approach would be to abstract all file handling to a separate class. Think of it as layer between DBImpl and Env. Having a separate class deal with file namings and deletion would benefit both code cleanliness (especially with huge DBImpl) and speed things up. It will take a huge effort to do this, though.
Let's discuss offline today.
Test Plan: Ran ./db_stress, verified that files are getting deleted
Reviewers: dhruba, haobo, kailiu, emayanke
Reviewed By: dhruba
Differential Revision: https://reviews.facebook.net/D13827
Summary: This is to give application compaction filter a chance to access context information of a specific compaction run. For example, depending on whether a compaction goes through all data files, the application could do things differently.
Test Plan: make check
Reviewers: dhruba, kailiu, sdong
Reviewed By: dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D13683
Summary:
This patch adds a option for universal compaction to allow us to only compress output files if the files compacted previously did not yet reach a specified ratio, to save CPU costs in some cases.
Compression is always skipped for flushing. This is because the size information is not easy to evaluate for flushing case. We can improve it later.
Test Plan:
add test
DBTest.UniversalCompactionCompressRatio1 and DBTest.UniversalCompactionCompressRatio12
Reviewers: dhruba, haobo
Reviewed By: dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D13467
Summary:
Enable background flush thread in this patch and fix unit tests with:
(1) After background flush, schedule a background compaction if condition satisfied;
(2) Fix a bug that if universal compaction is enabled and number of levels are set to be 0, compaction will not be automatically triggered
(3) Fix unit tests to wait for compaction to finish instead of flush, before checking the compaction results.
Test Plan: pass all unit tests
Reviewers: haobo, xjin, dhruba
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D13461
Summary:
So far we only have key/value pairs as well as bloom filter stored in the
sst file. It will be great if we are able to store more metadata about
this table itself, for example, the entry size, bloom filter name, etc.
This diff is the first step of this effort. It allows table to keep the
basic statistics mentioned in http://fburl.com/14995441, as well as
allowing writing user-collected stats to stats block.
After this diff, we will figure out the interface of how to allow user to collect their interested statistics.
Test Plan:
1. Added several unit tests.
2. Ran `make check` to ensure it doesn't break other tests.
Reviewers: dhruba, haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D13419
Summary:
Change namespace from leveldb to rocksdb. This allows a single
application to link in open-source leveldb code as well as
rocksdb code into the same process.
Test Plan: compile rocksdb
Reviewers: emayanke
Reviewed By: emayanke
CC: leveldb
Differential Revision: https://reviews.facebook.net/D13287
Summary:
Added a new field called max_size_amplification_ratio in the
CompactionOptionsUniversal structure. This determines the maximum
percentage overhead of space amplification.
The size amplification is defined to be the ratio between the size of
the oldest file to the sum of the sizes of all other files. If the
size amplification exceeds the specified value, then min_merge_width
and max_merge_width are ignored and a full compaction of all files is done.
A value of 10 means that the size a database that stores 100 bytes
of user data could occupy 110 bytes of physical storage.
Test Plan: Unit test DBTest.UniversalCompactionSpaceAmplification added.
Reviewers: haobo, emayanke, xjin
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12825
Summary:
If ReadOptions.non_blocking_io is set to true, then KeyMayExists
and Iterators will return data that is cached in RAM.
If the Iterator needs to do IO from storage to serve the data,
then the Iterator.status() will return Status::IsRetry().
Test Plan:
Enhanced unit test DBTest.KeyMayExist to detect if there were are IOs
issues from storage. Added DBTest.NonBlockingIteration to verify
nonblocking Iterations.
Reviewers: emayanke, haobo
Reviewed By: haobo
CC: leveldb
Maniphest Tasks: T63
Differential Revision: https://reviews.facebook.net/D12531
Summary: Fix code so that the filter_block layer only assumes keys are internal when prefix_extractor is set.
Test Plan: ./filter_block_test
Reviewers: dhruba, haobo
Reviewed By: dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D12501
Summary: If use_prefix_filters is set and read_range>1, then the random seeks will set a the prefix filter to be the prefix of the key which was randomly selected as the target. Still need to add statistics (perhaps in a separate diff).
Test Plan: ./db_bench --benchmarks=fillseq,prefixscanrandom --num=10000000 --statistics=1 --use_prefix_blooms=1 --use_prefix_api=1 --bloom_bits=10
Reviewers: dhruba
Reviewed By: dhruba
CC: leveldb, haobo
Differential Revision: https://reviews.facebook.net/D12273
Summary: An api to query the level, key ranges, size etc for each SST file and an api to delete a specific file from the db and all associated state in the bookkeeping datastructures.
Notes: Editing the manifest version does not release the obsolete files right away. However deleting the file directly will mess up the iterator. We may need a more aggressive/timely file deletion api.
I have used std::unique_ptr - will switch to boost:: since this is external. thoughts?
Unit test is fragile right now as it expects the compaction at certain levels.
Test Plan: unittest
Reviewers: dhruba, vamsi, emayanke
CC: zshao, leveldb, haobo
Task ID: #
Blame Rev:
Summary:
The pre-existing code was purging a DeleteMarker if thay key did not
exist in deeper levels. But in the Universal Compaction Style, all
files are in Level0. For compaction runs that did not include the
earliest file, we were erroneously purging the DeleteMarkers.
The fix is to purge DeleteMarkers only if the compaction includes
the earlist file.
Test Plan: DBTest.Randomized triggers this code path.
Differential Revision: https://reviews.facebook.net/D12081
Summary:
Here are the major changes to the Merge Interface. It has been expanded
to handle cases where the MergeOperator is not associative. It does so by stacking
up merge operations while scanning through the key history (i.e.: during Get() or
Compaction), until a valid Put/Delete/end-of-history is encountered; it then
applies all of the merge operations in the correct sequence starting with the
base/sentinel value.
I have also introduced an "AssociativeMerge" function which allows the user to
take advantage of associative merge operations (such as in the case of counters).
The implementation will always attempt to merge the operations/operands themselves
together when they are encountered, and will resort to the "stacking" method if
and only if the "associative-merge" fails.
This implementation is conjectured to allow MergeOperator to handle the general
case, while still providing the user with the ability to take advantage of certain
efficiencies in their own merge-operator / data-structure.
NOTE: This is a preliminary diff. This must still go through a lot of review,
revision, and testing. Feedback welcome!
Test Plan:
-This is a preliminary diff. I have only just begun testing/debugging it.
-I will be testing this with the existing MergeOperator use-cases and unit-tests
(counters, string-append, and redis-lists)
-I will be "desk-checking" and walking through the code with the help gdb.
-I will find a way of stress-testing the new interface / implementation using
db_bench, db_test, merge_test, and/or db_stress.
-I will ensure that my tests cover all cases: Get-Memtable,
Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0,
Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found,
end-of-history, end-of-file, etc.
-A lot of feedback from the reviewers.
Reviewers: haobo, dhruba, zshao, emayanke
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D11499
Summary: Removed KeyMayExistImpl because KeyMayExist demanded Get like semantics now. Removed no_io from memtable and imm because we need the proper value now and shouldn't just stop when we see Merge in memtable. Added checks to block_cache. Updated documentation and unit-test
Test Plan: make all check;db_stress for 1 hour
Reviewers: dhruba, haobo
Reviewed By: dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D11853
Summary:
Introduced KeyMayExist checking during writebatch-delete and removed from Outer Delete API because it uses writebatch-delete.
Added code to skip getting Table from disk if not already present in table_cache.
Some renaming of variables.
Introduced KeyMayExistImpl which allows checking since specified sequence number in GetImpl useful to check partially written writebatch.
Changed KeyMayExist to not be pure virtual and provided a default implementation.
Expanded unit-tests in db_test to check appropriately.
Ran db_stress for 1 hour with ./db_stress --max_key=100000 --ops_per_thread=10000000 --delpercent=50 --filter_deletes=1 --statistics=1.
Test Plan: db_stress;make check
Reviewers: dhruba, haobo
Reviewed By: dhruba
CC: leveldb, xjin
Differential Revision: https://reviews.facebook.net/D11745
