Summary:
There is a new option called hybrid_mode which, when switched on,
causes HBase style compactions. Files from L0 are
compacted back into L0. This meat of this compaction algorithm
is in PickCompactionHybrid().
All files reside in L0. That means all files have overlapping
keys. Each file has a time-bound, i.e. each file contains a
range of keys that were inserted around the same time. The
start-seqno and the end-seqno refers to the timeframe when
these keys were inserted. Files that have contiguous seqno
are compacted together into a larger file. All files are
ordered from most recent to the oldest.
The current compaction algorithm starts to look for
candidate files starting from the most recent file. It continues to
add more files to the same compaction run as long as the
sum of the files chosen till now is smaller than the next
candidate file size. This logic needs to be debated
and validated.
The above logic should reduce write amplification to a
large extent... will publish numbers shortly.
Test Plan: dbstress runs for 6 hours with no data corruption (tested so far).
Differential Revision: https://reviews.facebook.net/D11289
Summary:
There is a new option called hybrid_mode which, when switched on,
causes HBase style compactions. Files from L0 are
compacted back into L0. This meat of this compaction algorithm
is in PickCompactionHybrid().
All files reside in L0. That means all files have overlapping
keys. Each file has a time-bound, i.e. each file contains a
range of keys that were inserted around the same time. The
start-seqno and the end-seqno refers to the timeframe when
these keys were inserted. Files that have contiguous seqno
are compacted together into a larger file. All files are
ordered from most recent to the oldest.
The current compaction algorithm starts to look for
candidate files starting from the most recent file. It continues to
add more files to the same compaction run as long as the
sum of the files chosen till now is smaller than the next
candidate file size. This logic needs to be debated
and validated.
The above logic should reduce write amplification to a
large extent... will publish numbers shortly.
Test Plan: dbstress runs for 6 hours with no data corruption (tested so far).
Differential Revision: https://reviews.facebook.net/D11289
Summary:
Merge multiple multiple memtables in memory before writing it
out to a file in L0.
There is a new config parameter min_write_buffer_number_to_merge
that specifies the number of write buffers that should be merged
together to a single file in storage. The system will not flush
wrte buffers to storage unless at least these many buffers have
accumulated in memory.
The default value of this new parameter is 1, which means that
a write buffer will be immediately flushed to disk as soon it is
ready.
Test Plan: make check
Differential Revision: https://reviews.facebook.net/D11241
Summary: This hopefully gives the right semantics to compaction filter. Will write a small wiki to explain the ideas.
Test Plan: make check; db_stress
Reviewers: dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D11121
Summary:
Preliminary! Introduced the --use_multiget=1 and --keys_per_multiget=n
flags for db_bench. Also updated and tested the ReadRandom() method
to include an option to use multiget. By default,
keys_per_multiget=100.
Preliminary tests imply that multiget is at least 1.25x faster per
key than regular get.
Will continue adding Multiget for ReadMissing, ReadHot,
RandomWithVerify, ReadRandomWriteRandom; soon. Will also think
about ways to better verify benchmarks.
Test Plan:
1. make db_bench
2. ./db_bench --benchmarks=fillrandom
3. ./db_bench --benchmarks=readrandom --use_existing_db=1
--use_multiget=1 --threads=4 --keys_per_multiget=100
4. ./db_bench --benchmarks=readrandom --use_existing_db=1
--threads=4
5. Verify ops/sec (and 1000000 of 1000000 keys found)
Reviewers: haobo, MarkCallaghan, dhruba
Reviewed By: MarkCallaghan
CC: leveldb
Differential Revision: https://reviews.facebook.net/D11127
Summary:
This diff simplifies EnvOptions by treating it as POD, similar to Options.
- virtual functions are removed and member fields are accessed directly.
- StorageOptions is removed.
- Options.allow_readahead and Options.allow_readahead_compactions are deprecated.
- Unused global variables are removed: useOsBuffer, useFsReadAhead, useMmapRead, useMmapWrite
Test Plan: make check; db_stress
Reviewers: dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D11175
Summary:
Without this files could be written out to a level greater than the maximum level possible and is the source of the segfaults that wormhole awas getting. The sequence of steps that was followed:
1. WriteLevel0Table was called when memtable was to be flushed for a file.
2. PickLevelForMemTableOutput was called to determine the level to which this file should be pushed.
3. PickLevelForMemTableOutput returned a wrong result because max_mem_compaction_level was equal to 2 even when num_levels was equal to 0.
The fix to re-initialize max_mem_compaction_level based on num_levels passed seems correct.
Test Plan: make all check; Also made a dummy file to mimic the wormhole-file behaviour which was causing the segfaults and found that the same segfault occurs without this change and not with this.
Reviewers: dhruba, haobo
Reviewed By: dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D11157
Summary:
Implemented the MultiGet operator which takes in a list of keys
and returns their associated values. Currently uses std::vector as its
container data structure. Otherwise, it works identically to "Get".
Test Plan:
1. make db_test ; compile it
2. ./db_test ; test it
3. make all check ; regress / run all tests
4. make release ; (optional) compile with release settings
Reviewers: haobo, MarkCallaghan, dhruba
Reviewed By: dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D10875
Summary: Added the 'score' column to the compaction stats output, which shows the level total size devided by level target size. Could be useful when monitoring compaction decisions...
Test Plan: make check; db_bench
Reviewers: dhruba
CC: leveldb, MarkCallaghan
Differential Revision: https://reviews.facebook.net/D11025
Summary:
This diff adds an option to specify whether PTHREAD_MUTEX_ADAPTIVE_NP will be enabled for the rocksdb single big kernel lock. db_bench also have this option now.
Quickly tested 8 thread cpu bound 100 byte random read.
No fast mutex: ~750k/s ops
With fast mutex: ~880k/s ops
Test Plan: make check; db_bench; db_stress
Reviewers: dhruba
CC: MarkCallaghan, leveldb
Differential Revision: https://reviews.facebook.net/D11031
Summary: a new option block_size_deviation is added.
Test Plan: run db_test and db_bench
Reviewers: dhruba, haobo
Reviewed By: haobo
Differential Revision: https://reviews.facebook.net/D10821
Summary: MaybeDumpStats was causing lock problem
Test Plan: make check; db_stress
Reviewers: dhruba
Reviewed By: dhruba
Differential Revision: https://reviews.facebook.net/D10935
Summary:
Added an option stats_dump_period_sec to dump leveldb.stats to LOG periodically for diagnosis.
By defauly, it's set to a very big number 3600 (1 hour).
Test Plan: make check;
Reviewers: dhruba
Reviewed By: dhruba
CC: leveldb, zshao
Differential Revision: https://reviews.facebook.net/D10761
Summary: There was an artifical limit on the size of the write buffer size.
Test Plan: make check
Reviewers: haobo
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D10911
Summary:
Currently, with paranoid_check on, DB::Open will fail on any log read error on recovery.
If client is ok with losing most recent updates, we could simply skip those errors.
However, it's important to introduce an additional flag, so that paranoid_check can
still guard against more serious problems.
Test Plan: make check; db_stress
Reviewers: dhruba, emayanke
Reviewed By: emayanke
CC: leveldb, emayanke
Differential Revision: https://reviews.facebook.net/D10869
Summary:
Make stop watch a simple implementation, instead of subclass of a virtual class
Allocate stop watches off the stack instead of heap.
Code is more terse now.
Test Plan: make all check, db_bench with --statistics=1
Reviewers: haobo, dhruba
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D10809
Summary:
This diff replaces compaction_filter_args and CompactionFilter with a single compaction_filter parameter. It gives CompactionFilter better encapsulation and a similar look to Comparator and MergeOpertor, which improves consistency of the overall interface.
The change is not backward compatible. Nevertheless, the two references in fbcode are not in production yet.
Test Plan: make check
Reviewers: dhruba
Reviewed By: dhruba
CC: leveldb, zshao
Differential Revision: https://reviews.facebook.net/D10773
Summary:
Currently, compaction filter is run on internal key older than the oldest snapshot, which is incorrect.
Compaction filter should really be run on the most recent internal key when there is no external snapshot.
Test Plan: make check; db_stress
Reviewers: dhruba
Reviewed By: dhruba
Differential Revision: https://reviews.facebook.net/D10641
Summary:
WAL files are moved to archive directory and clear only at DB::Open.
Can lead to a lot of space consumption in a Database. Added logic to periodically clear Archive Directory too.
Test Plan: make all check + add unit test
Reviewers: dhruba, heyongqiang
Reviewed By: heyongqiang
CC: leveldb
Differential Revision: https://reviews.facebook.net/D10617
Summary:
This diff introduces a new Merge operation into rocksdb.
The purpose of this review is mostly getting feedback from the team (everyone please) on the design.
Please focus on the four files under include/leveldb/, as they spell the client visible interface change.
include/leveldb/db.h
include/leveldb/merge_operator.h
include/leveldb/options.h
include/leveldb/write_batch.h
Please go over local/my_test.cc carefully, as it is a concerete use case.
Please also review the impelmentation files to see if the straw man implementation makes sense.
Note that, the diff does pass all make check and truly supports forward iterator over db and a version
of Get that's based on iterator.
Future work:
- Integration with compaction
- A raw Get implementation
I am working on a wiki that explains the design and implementation choices, but coding comes
just naturally and I think it might be a good idea to share the code earlier. The code is
heavily commented.
Test Plan: run all local tests
Reviewers: dhruba, heyongqiang
Reviewed By: dhruba
CC: leveldb, zshao, sheki, emayanke, MarkCallaghan
Differential Revision: https://reviews.facebook.net/D9651
Summary:
- don't see a point exposing table.h to the public.
- fixed make clean to remove also *.d files.
Test Plan: make check; db_stress
Reviewers: dhruba, heyongqiang
Reviewed By: dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D10479
Summary:
- removed the compaction_filter_value from the callback interface. Restrict compaction filter to purging values.
- modify some comments to reflect curent status.
Test Plan: make check
Reviewers: dhruba
Reviewed By: dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D10335
Summary: A better error message. A local change. Did not look at other places where this could be done.
Test Plan: compile
Reviewers: dhruba, MarkCallaghan
Reviewed By: dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D10251
Summary:
FindObsoleteFiles was slow, holding the single big lock, resulted in bad p99 behavior.
Didn't profile anything, but several things could be improved:
1. VersionSet::AddLiveFiles works with std::set, which is by itself slow (a tree).
You also don't know how many dynamic allocations occur just for building up this tree.
switched to std::vector, also added logic to pre-calculate total size and do just one allocation
2. Don't see why env_->GetChildren() needs to be mutex proteced, moved to PurgeObsoleteFiles where
mutex could be unlocked.
3. switched std::set to std:unordered_set, the conversion from vector is also inside PurgeObsoleteFiles
I have a feeling this should pretty much fix it.
Test Plan: make check; db_stress
Reviewers: dhruba, heyongqiang, MarkCallaghan
Reviewed By: dhruba
CC: leveldb, zshao
Differential Revision: https://reviews.facebook.net/D10197
Summary: using unique_ptr to have automatic delete for probableWALfiles in db_impl.cc
Test Plan: make
Reviewers: sheki, dhruba
Reviewed By: sheki
CC: leveldb
Differential Revision: https://reviews.facebook.net/D10083
Summary:
The segfault was happening because the program was unable to open a new
sst file (as part of the compaction) because the process ran out of
file descriptors.
The fix is to check the return status of the file creation before taking
any other action.
Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7fabf03f9700 (LWP 29904)]
leveldb::DBImpl::OpenCompactionOutputFile (this=this@entry=0x7fabf9011400, compact=compact@entry=0x7fabf741a2b0) at db/db_impl.cc:1399
1399 db/db_impl.cc: No such file or directory.
(gdb) where
Test Plan: make check
Reviewers: MarkCallaghan, sheki
Reviewed By: MarkCallaghan
CC: leveldb
Differential Revision: https://reviews.facebook.net/D10101
Summary:
Transaction Log Iterator did not move to the next file in the series if there was a write batch at the end of the currentFile.
The solution is if the last seq no. of the current file is < RequestedSeqNo. Assume the first seqNo. of the next file has to satisfy the request.
Also major refactoring around the code. Moved opening the logreader to a seperate function, got rid of goto.
Test Plan: added a unit test for it.
Reviewers: dhruba, heyongqiang
Reviewed By: heyongqiang
CC: leveldb, emayanke
Differential Revision: https://reviews.facebook.net/D10029
Summary:
During recovery, last_updated_manifest number was not set if there were no records in the Write-ahead log.
Now check for the recovered manifest also and set last_updated_manifest file to the max value.
Test Plan: unit test
Reviewers: heyongqiang
Reviewed By: heyongqiang
CC: leveldb
Differential Revision: https://reviews.facebook.net/D9891
Summary:
If a class owns an object:
- If the object can be null => use a unique_ptr. no delete
- If the object can not be null => don't even need new, let alone delete
- for runtime sized array => use vector, no delete.
Test Plan: make check
Reviewers: dhruba, heyongqiang
Reviewed By: heyongqiang
CC: leveldb, zshao, sheki, emayanke, MarkCallaghan
Differential Revision: https://reviews.facebook.net/D9783
Summary:
RocksDB does a binary search to look at the files which might contain the requested sequence number at the call GetUpdatesSince.
There was a bug in the binary search => when the file pointed by the middle index of bsearch was empty/corrupt it needst to resize the vector and update indexes.
This now fixes that.
Test Plan: existing unit tests pass.
Reviewers: heyongqiang, dhruba
Reviewed By: heyongqiang
CC: leveldb
Differential Revision: https://reviews.facebook.net/D9777
Summary:
If the vector returned by GetUpdatesSince is empty, it is still returned to the
user. This causes it throw an std::range error.
The probable file list is checked and it returns an IOError status instead of OK now.
Test Plan: added a unit test.
Reviewers: dhruba, heyongqiang
Reviewed By: heyongqiang
CC: leveldb
Differential Revision: https://reviews.facebook.net/D9771
Summary:
Use non mmapd files for Write-Ahead log.
Earlier use of MMaped files. made the log iterator read ahead and miss records.
Now the reader and writer will point to the same physical location.
There is no perf regression :
./db_bench --benchmarks=fillseq --db=/dev/shm/mmap_test --num=$(million 20) --use_existing_db=0 --threads=2
with This diff :
fillseq : 10.756 micros/op 185281 ops/sec; 20.5 MB/s
without this dif :
fillseq : 11.085 micros/op 179676 ops/sec; 19.9 MB/s
Test Plan: unit test included
Reviewers: dhruba, heyongqiang
Reviewed By: heyongqiang
CC: leveldb
Differential Revision: https://reviews.facebook.net/D9741
Summary: rocksdb uses a single global lock to protect in memory metadata. We should minimize the mutex protected code section to increase the effective parallelism of the program. See https://our.intern.facebook.com/intern/tasks/?t=2218928
Test Plan:
make check
db_bench
Reviewers: dhruba, heyongqiang
CC: zshao, leveldb
Differential Revision: https://reviews.facebook.net/D9705
Summary:
The events that trigger compaction:
* opening the database
* Get -> only if seek compaction is not disabled and other checks are true
* MakeRoomForWrite -> when memtable is full
* BackgroundCall ->
If the background thread is about to do a compaction run, it schedules
a new background task to trigger a possible compaction. This will cause
additional background threads to find and process other compactions that
can run concurrently.
Test Plan: ran db_bench with overwrite and readonly alternatively.
Reviewers: sheki, MarkCallaghan
Reviewed By: sheki
CC: leveldb
Differential Revision: https://reviews.facebook.net/D9579
Summary:
This patch allows an application to specify whether to use bufferedio,
reads-via-mmaps and writes-via-mmaps per database. Earlier, there
was a global static variable that was used to configure this functionality.
The default setting remains the same (and is backward compatible):
1. use bufferedio
2. do not use mmaps for reads
3. use mmap for writes
4. use readaheads for reads needed for compaction
I also added a parameter to db_bench to be able to explicitly specify
whether to do readaheads for compactions or not.
Test Plan: make check
Reviewers: sheki, heyongqiang, MarkCallaghan
Reviewed By: sheki
CC: leveldb
Differential Revision: https://reviews.facebook.net/D9429
Summary: Makefile had options to ignore sign-comparisons and unused-parameters, which should be there. Also fixed the specific errors in the code-base
Test Plan: make
Reviewers: chip, dhruba
Reviewed By: dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D9531
Summary:
Add --benchmarks=levelstats option to report per-level stats (#files, #bytes)
Change readwhilewriting test to report response time for writes but exclude
them from the stats merged by all threads.
Prevent "NaN" in stats output by preventing division by 0.
Remove "o" file I committed by mistake.
Task ID: #
Blame Rev:
Test Plan:
make check
Revert Plan:
Database Impact:
Memcache Impact:
Other Notes:
EImportant:
- begin *PUBLIC* platform impact section -
Bugzilla: #
- end platform impact -
Reviewers: dhruba
Reviewed By: dhruba
Differential Revision: https://reviews.facebook.net/D9513
Summary:
Rocksdb can create 0 sized log files when it is opened and closed without any operations.
The GetUpdatesSince fails currently if there is a log file of size zero.
This diff fixes this. If there is a log file is 0, it is removed form the probable_file_list
Test Plan: unit test
Reviewers: dhruba, heyongqiang
Reviewed By: heyongqiang
CC: leveldb
Differential Revision: https://reviews.facebook.net/D9507
Summary:
If there is an error while writing an edit to the manifest file, the manifest
file is closed and reopened to check if the edit made it in. However, if the
re-opening of the manifest is unsuccessful and options.paranoid_checks is set
t true, then the db refuses to accept new puts, effectively putting the db
in readonly mode.
In a future diff, I would like to make the default value of paranoid_check
to true.
Test Plan: make check
Reviewers: sheki
Reviewed By: sheki
CC: leveldb
Differential Revision: https://reviews.facebook.net/D9201
Summary:
Store the last flushed, seq no. in db_impl. Check against it in
transaction Log iterator. Do not attempt to read ahead if we do not know
if the data is flushed completely.
Does not work if flush is disabled. Any ideas on fixing that?
* Minor change, iter->Next is called the first time automatically for
* the first time.
Test Plan:
existing test pass.
More ideas on testing this?
Planning to run some stress test.
Reviewers: dhruba, heyongqiang
CC: leveldb
Differential Revision: https://reviews.facebook.net/D9087
Summary:
The compaction process zeros out sequence numbers if the output is
part of the bottommost level.
The Slice is supposed to refer to an immutable data buffer. The
merger that implements the priority queue while reading kvs as
the input of a compaction run reies on this fact. The bug was that
were updating the sequence number of a record in-place and that was
causing suceeding invocations of the merger to return kvs in
arbitrary order of sequence numbers.
The fix is to copy the key to a local memory buffer before setting
its seqno to 0.
Test Plan:
Set Options.purge_redundant_kvs_while_flush = false and then run
db_stress --ops_per_thread=1000 --max_key=320
Reviewers: emayanke, sheki
Reviewed By: emayanke
CC: leveldb
Differential Revision: https://reviews.facebook.net/D9147
Summary:
This adds the rate_delay_limit_milliseconds option to make the delay
configurable in MakeRoomForWrite when the max compaction score is too high.
This delay is called the Ln slowdown. This change also counts the Ln slowdown
per level to make it possible to see where the stalls occur.
From IO-bound performance testing, the Level N stalls occur:
* with compression -> at the largest uncompressed level. This makes sense
because compaction for compressed levels is much
slower. When Lx is uncompressed and Lx+1 is compressed
then files pile up at Lx because the (Lx,Lx+1)->Lx+1
compaction process is the first to be slowed by
compression.
* without compression -> at level 1
Task ID: #1832108
Blame Rev:
Test Plan:
run with real data, added test
Revert Plan:
Database Impact:
Memcache Impact:
Other Notes:
EImportant:
- begin *PUBLIC* platform impact section -
Bugzilla: #
- end platform impact -
Reviewers: dhruba
Reviewed By: dhruba
Differential Revision: https://reviews.facebook.net/D9045
Summary:
Rocks accumulates recent writes and deletes in the in-memory memtable.
When the memtable is full, it writes the contents on the memtable to
a file in L0.
This patch removes redundant records at the time of the flush. If there
are multiple versions of the same key in the memtable, then only the
most recent one is dumped into the output file. The purging of
redundant records occur only if the most recent snapshot is earlier
than the earliest record in the memtable.
Should we switch on this feature by default or should we keep this feature
turned off in the default settings?
Test Plan: Added test case to db_test.cc
Reviewers: sheki, vamsi, emayanke, heyongqiang
Reviewed By: sheki
CC: leveldb
Differential Revision: https://reviews.facebook.net/D8991
Summary:
There was an artifical limit of 50K files per database. This is
insifficient if the database is 1 TB in size and each file is 2 MB.
Test Plan: make check
Reviewers: sheki, emayanke
Reviewed By: emayanke
CC: leveldb
Differential Revision: https://reviews.facebook.net/D8919
Dhruba Borthakur