Summary:
Consider BlockReadAmpBitmap with bytes_per_bit = 32. Suppose bytes [a, b) were used, while bytes [a-32, a)
and [b+1, b+33) weren't used; more formally, the union of ranges passed to BlockReadAmpBitmap::Mark() contains [a, b) and doesn't intersect with [a-32, a) and [b+1, b+33). Then bits [floor(a/32), ceil(b/32)] will be set, and so the number of useful bytes will be estimated as (ceil(b/32) - floor(a/32)) * 32, which is on average equal to b-a+31.
An extreme example: if we use 1 byte from each block, it'll be counted as 32 bytes from each block.
It's easy to remove this bias by slightly changing the semantics of the bitmap. Currently each bit represents a byte range [i*32, (i+1)*32).
This diff makes each bit represent a single byte: i*32 + X, where X is a random number in [0, 31] generated when bitmap is created. So, e.g., if you read a single byte at random, with probability 31/32 it won't be counted at all, and with probability 1/32 it will be counted as 32 bytes; so, on average it's counted as 1 byte.
*But there is one exception: the last bit will always set with the old way.*
(*) - assuming read_amp_bytes_per_bit = 32.
Closes https://github.com/facebook/rocksdb/pull/2259
Differential Revision: D5035652
Pulled By: lightmark
fbshipit-source-id: bd98b1b9b49fbe61f9e3781d07f624e3cbd92356
Summary:
Now if we have iterate_upper_bound set, we continue read until get a key >= upper_bound. For a lot of cases that neighboring data blocks have a user key gap between them, our index key will be a user key in the middle to get a shorter size. For example, if we have blocks:
[a b c d][f g h]
Then the index key for the first block will be 'e'.
then if upper bound is any key between 'd' and 'e', for example, d1, d2, ..., d99999999999, we don't have to read the second block and also know that we have done our iteration by reaching the last key that smaller the upper bound already.
This diff can reduce RA in most cases.
Closes https://github.com/facebook/rocksdb/pull/2239
Differential Revision: D4990693
Pulled By: lightmark
fbshipit-source-id: ab30ea2e3c6edf3fddd5efed3c34fcf7739827ff
Summary:
we align the buffer with logical sector size and should not test it with page size, which is usually 4k.
Closes https://github.com/facebook/rocksdb/pull/2245
Differential Revision: D5001842
Pulled By: lightmark
fbshipit-source-id: a7135fcf6351c6db363e8908956b1e193a4a6291
Summary:
When user doesn't set a limit on compaction output file size, let's use the sum of the input files' sizes. This will avoid passing UINT64_MAX as fallocate()'s length. Reported in #2249.
Test setup:
- command: `TEST_TMPDIR=/data/rocksdb-test/ strace -e fallocate ./db_compaction_test --gtest_filter=DBCompactionTest.ManualCompactionUnknownOutputSize`
- filesystem: xfs
before this diff:
`fallocate(10, 01, 0, 1844674407370955160) = -1 ENOSPC (No space left on device)`
after this diff:
`fallocate(10, 01, 0, 1977) = 0`
Closes https://github.com/facebook/rocksdb/pull/2252
Differential Revision: D5007275
Pulled By: ajkr
fbshipit-source-id: 4491404a6ae8a41328aede2e2d6f4d9ac3e38880
Summary:
With row cache being enabled, table cache is doing a short circuit for reading data. This path needs to be updated to take advantage of pinnable slice. In the meanwhile we disabling pinning in this path.
Closes https://github.com/facebook/rocksdb/pull/2237
Differential Revision: D4982389
Pulled By: maysamyabandeh
fbshipit-source-id: 542630d0cf23cfb1f0c397da82e7053df7966591
Summary:
Every time after a compaction/flush finish, we issue user reads to put the table into block cache which includes a couple of IO that read footer, index blocks, meta block, etc. So we implement Prefetch here to reduce IO.
Closes https://github.com/facebook/rocksdb/pull/2196
Differential Revision: D4931782
Pulled By: lightmark
fbshipit-source-id: 5a13d58dcab209964352322217193bbf7ff78149
Summary:
�fix the buffer size in case of ppl use buffer size as their block_size.
Closes https://github.com/facebook/rocksdb/pull/2198
Differential Revision: D4956878
Pulled By: lightmark
fbshipit-source-id: 8bb0dc9c133887aadcd625d5261a3d1110b71473
Summary:
The goal is to avoid the problem of small number of L0 files triggering compaction to base level (which increased write-amp), while still allowing L0 compaction-by-size (so intra-L0 compactions cause score to increase).
Closes https://github.com/facebook/rocksdb/pull/2172
Differential Revision: D4908552
Pulled By: ajkr
fbshipit-source-id: 4b170142b2b368e24bd7948b2a6f24c69fabf73d
Summary:
index_per_partition should have deprecated deprecated instead of being removed. It is causing backward compatibility issues.
Closes https://github.com/facebook/rocksdb/pull/2173
Differential Revision: D4910947
Pulled By: maysamyabandeh
fbshipit-source-id: 5c52939381847d232ede6866606f67f2b4b857ae
Summary:
Add a function to allow users to reset internal stats without restarting the DB.
Closes https://github.com/facebook/rocksdb/pull/2167
Differential Revision: D4907939
Pulled By: siying
fbshipit-source-id: ab2dd85b88aabe9380da7485320a1d460d3e1f68
Summary:
st_blocks shows 16 though the right value is 8. This happens occasionally which seems a bug.
Closes https://github.com/facebook/rocksdb/pull/2160
Differential Revision: D4893542
Pulled By: lightmark
fbshipit-source-id: 68e832586b58bbc6162efbe83ce273f1570d5be3
Summary:
prefetch some data from the end of the file for each compaction to reduce IO.
Closes https://github.com/facebook/rocksdb/pull/2149
Differential Revision: D4880576
Pulled By: lightmark
fbshipit-source-id: aa767cd1afc84c541837fbf1ad6c0d45b34d3932
Summary:
Replace Options::use_direct_writes with Options::use_direct_io_for_flush_and_compaction
Now if Options::use_direct_io_for_flush_and_compaction = true, we will enable direct io for both reads and writes for flush and compaction job. Whereas Options::use_direct_reads controls user reads like iterator and Get().
Closes https://github.com/facebook/rocksdb/pull/2117
Differential Revision: D4860912
Pulled By: lightmark
fbshipit-source-id: d93575a8a5e780cf7e40797287edc425ee648c19
Summary:
To correct a build process where the JAVA_TEST_LIBDIR is a symlink to a cache directory.
Test -s (size 0) on symlinks returns true, resulting in a mkdir over the top of the symlink resulting in failure.
As a solution -d checks if it is a directory (or the symlink refers to a directory), which works in the case of real directories and symlinks to directories.
Trivial I know but it was really easy for me to use a symlink here to prevent frequent downloads in a CI environment.
Thanks for your consideration.
Closes https://github.com/facebook/rocksdb/pull/1917
Differential Revision: D4612263
Pulled By: siying
fbshipit-source-id: 4d458f8e1760068cdd6b5eae4bce6e12c400df41
Summary: Build Java and RocksDB LITE as a customized unit test under internal_repo_rocksdb. One thing I'm not sure is that whether these two tests are triggered in every flavor.
Reviewed By: IslamAbdelRahman
Differential Revision: D4855868
fbshipit-source-id: 82a1628b458744d7692bbd29ef7424cca1294031
Summary:
Users usually set readahead buffer to a multiple of 4k, more than that, usually a multiple of blocks.
So previously we set real buffer size 512 * n + 4k, which may introduce an additional block reading.
Closes https://github.com/facebook/rocksdb/pull/2138
Differential Revision: D4871504
Pulled By: lightmark
fbshipit-source-id: b070faa51d92e976e8e8468c00692699e585e243
Summary:
filter the warning out and only print it once.
Closes https://github.com/facebook/rocksdb/pull/2137
Differential Revision: D4870925
Pulled By: lightmark
fbshipit-source-id: 91b363ce7f70bce88b0780337f408fc4649139b8
Summary:
In some CI test environment, compression libraries can't be successfully built. It still helps to build RocksDB there. Provide such an option to skip to download and build compression libraries.
Closes https://github.com/facebook/rocksdb/pull/2135
Differential Revision: D4872617
Pulled By: siying
fbshipit-source-id: bb21ac373bc62a2528cdf1ca4547e05fcae86214
Summary:
Moved MergeOperatorPinning tests from db_test2.cc to db_merge_operator_test.cc.
[This is the same code as PR #2104 , which has already been reviewed, but I am creating a new PR as I cannot import from #2104 onto phabricator anymore even after rebasing. I'll close and discard #2104.]
Closes https://github.com/facebook/rocksdb/pull/2125
Differential Revision: D4863312
Pulled By: sagar0
fbshipit-source-id: 0f71a7690aa09c1d03ee85ce2bc1d2d89e4f4399
Summary:
Currently level histogram is only printed out for DB stats and for default CF. This is confusing. Change to print for every CF instead.
Closes https://github.com/facebook/rocksdb/pull/2126
Differential Revision: D4865373
Pulled By: siying
fbshipit-source-id: 1c853e0ac66e00120ee931cabc9daf69ccc2d577
Summary:
Upgrading a shared lock was silently succeeding because the actual locking code was skipped. This is because if the keys are tracked, it is assumed that they are already locked and do not require locking. Fix this by recording in tracked keys whether the key was locked exclusively or not.
Note that lock downgrades are impossible, which is the behaviour we expect.
This fixesfacebook/mysql-5.6#587.
Closes https://github.com/facebook/rocksdb/pull/2122
Differential Revision: D4861489
Pulled By: IslamAbdelRahman
fbshipit-source-id: 58c7ebe7af098bf01b9774b666d3e9867747d8fd
Summary:
Extend TransactionOptions to include max_write_batch_size which determines the maximum size of the writebatch representation. If memory limit is exceeded, the operation will abort with subcode kMemoryLimit.
Closes https://github.com/facebook/rocksdb/pull/2124
Differential Revision: D4861842
Pulled By: lth
fbshipit-source-id: 46fd172ea67cc90bbba829bf0d70cfab2261c161
Summary:
Run the time command before regression tests, parse the output, and add the numbers to the report.
Closes https://github.com/facebook/rocksdb/pull/2101
Differential Revision: D4862781
Pulled By: maysamyabandeh
fbshipit-source-id: 4a81caa5d14187d67093aad154c8f0ad56aba901
Summary:
also did minor refactoring
Closes https://github.com/facebook/rocksdb/pull/2115
Differential Revision: D4855818
Pulled By: maysamyabandeh
fbshipit-source-id: fbca6ac57e5c6677fffe8354f7291e596a50cb77
Summary:
DBIter, and in-turn NewDBIterator and NewArenaWrappedDBIterator, take a bunch of params. They can be reduced by passing in ReadOptions directly instead of passing in every new param separately. It also seems much cleaner as a bunch of the params towards the end seem to be optional.
(Recently I introduced max_skippable_internal_keys, which added one more to the already huge count).
Idea courtesy IslamAbdelRahman
Closes https://github.com/facebook/rocksdb/pull/2116
Differential Revision: D4857128
Pulled By: sagar0
fbshipit-source-id: 7d239df094b94bd9ea79d145cdf825478ac037a8
Summary:
The compiler error:
```
/home/jenkins/workspace/ceph-master/src/rocksdb/db/db_impl.cc:20:10: fatal error: 'jemalloc/jemalloc.h' file not found
^
1 error generated.
```
But is does compile with the `WITH_JEMALLOC` set.
So ignore all the other settings.
Closes https://github.com/facebook/rocksdb/pull/2118
Differential Revision: D4858387
Pulled By: yiwu-arbug
fbshipit-source-id: 05b982969dcab53669a73a903641e71641c714e7
Summary:
Check the result of the benchmark againt a specified truth_db, which is
expected to be produced using the same benchmark but perhaps on a
different commit or with different configs.
The verification is simple and assumes that key/values are generated
deterministically. This assumption would break if db_bench using rand
variable differently from the benchmark that produced truth_db.
Currently it is checked to work on fillrandom and readwhilewriting.
A param finish_after_writes is added to ensure that the background
writing thread will write the same number of entries between two
benchmarks.
Example:
$ TEST_TMPDIR=/dev/shm/truth_db ./db_bench
--benchmarks="fillrandom,readwhilewriting" --num=200000
--finish_after_writes=true
$ TEST_TMPDIR=/dev/shm/tmpdb ./db_bench
--benchmarks="fillrandom,readwhilewriting,verify" --truth_db
/dev/shm/truth_db/dbbench --num=200000 --finish_after_writes=true
Verifying db <= truth_db...
Verifying db >= truth_db...
...Verified
Closes https://github.com/facebook/rocksdb/pull/2098
Differential Revision: D4839233
Pulled By: maysamyabandeh
fbshipit-source-id: 2f4ed31
Summary:
Currently the examples fail, having a job building them would
prevent that from happening.
Closes https://github.com/facebook/rocksdb/pull/2007
Differential Revision: D4831570
Pulled By: yiwu-arbug
fbshipit-source-id: e38adfa
Summary:
OpenBSD doesn't have `O_DIRECT`, so avoid it. (RocksDB compiles successfully on
OpenBSD with this patch.)
Closes https://github.com/facebook/rocksdb/pull/2106
Differential Revision: D4847833
Pulled By: siying
fbshipit-source-id: 214b785
Summary:
1. Move universal compaction picker to separate files compaction_picker_universal.cc and compaction_picker_universal.h.
2. Rename some functions to make the code easier to understand.
3. Move leveled compaction picking code to a dedicated class, so that we we don't need to pass some common variable around when calling functions. It also allowed us to break down LevelCompactionPicker::PickCompaction() to smaller functions.
Closes https://github.com/facebook/rocksdb/pull/2100
Differential Revision: D4845948
Pulled By: siying
fbshipit-source-id: efa0ab4
Summary:
In RocksDB, we sometimes preallocate the estimated space for a file to have better perf with fallocate (if supported). Usually it is a little bit bigger than the real resulting file size. At this time, we have to let the Filesystem reclaim the space not used.
Ideally, calling ftruncate to truncate the file to its real size should be enough. HOWEVER, it isn't on tmpfs, which we witness in our case, with some buggy kernel version. ftruncate a file with preallocated space doesn't change number of the blocks used by the file, which means the space not used by the file is not returned to the filesystems. So in this case we need fallocate with FALLOC_FL_PUNCH_HOLE to explicitly reclaim the used blocks. It is a hack to cope with the kernel bug and usually we should not need it.
Closes https://github.com/facebook/rocksdb/pull/2102
Differential Revision: D4848934
Pulled By: lightmark
fbshipit-source-id: f1b40b5
Summary:
This is an effort to club all string related utility functions into one common place, in string_util, so that it is easier for everyone to know what string processing functions are available. Right now they seem to be spread out across multiple modules, like logging and options_helper.
Check the sub-commits for easier reviewing.
Closes https://github.com/facebook/rocksdb/pull/2094
Differential Revision: D4837730
Pulled By: sagar0
fbshipit-source-id: 344278a
Summary:
While running `make jtest` using IBM Java, it fails at compactRangeToLevel with the below error.
```
Run: org.rocksdb.RocksDBTest testing now -> compactRangeToLevel
JVMJNCK056E JNI error in ReleaseByteArrayElements: Got memory 0x00003FFF94AA8908 from object 0x00000000000C7F78, releasing from 0x00000000000C7F68
JVMJNCK077E Error detected in org/rocksdb/RocksDB.compactRange0(J[BI[BIZII)V
JVMJNCK024E JNI error detected. Aborting.
JVMJNCK025I Use -Xcheck:jni:nonfatal to continue running when errors are detected.
Fatal error: JNI error
Makefile:205: recipe for target 'run_test' failed
make[1]: *** [run_test] Error 87
make[1]: Leaving directory '/home/ubuntu/rocksdb/java'
Makefile:1542: recipe for target 'jtest' failed
make: *** [jtest] Error 2
```
After checking the code, it is vivid that we are messing up the `ReleaseByteArrayElements` args in `rocksdb_compactrange_helper`.
```
.................
1959 s = db->CompactRange(compact_options, &begin_slice, &end_slice);
1960 }
Closes https://github.com/facebook/rocksdb/pull/2060
Differential Revision: D4831427
Pulled By: yiwu-arbug
fbshipit-source-id: dd02037
