Summary:
When subcompaction is used, all subcompactions share the same Compaction
pointer in CompactionJob while each subcompaction all keeps their mutable
stats in SubcompactionState.  However, there're still some mutable part
that is currently store in the shared Compaction pointer.

This patch makes two changes:

1. Make the shared Compaction pointer const so that it can never be modified
   during the compaction.
2. Move necessary states from Compaction to SubcompactionState.
3. Make functions of Compaction const if the function does not modify
   its internal state.

Test Plan: rocksdb and MyRocks test

Reviewers: sdong, kradhakrishnan, andrewkr, IslamAbdelRahman

Reviewed By: IslamAbdelRahman

Subscribers: andrewkr, dhruba, yoshinorim, gunnarku, leveldb

Differential Revision: https://reviews.facebook.net/D55923

Add support for db_test2 for dev and CI runs

Summary: Document the skipped counters in perf context more clearly.

Test Plan: Comment only.

Reviewers: IslamAbdelRahman, yhchiang, MarkCallaghan

Reviewed By: MarkCallaghan

Subscribers: leveldb, andrewkr, dhruba

Differential Revision: https://reviews.facebook.net/D55833

Summary:
Without this diff, this is what happens to compaction output file if it's a subclass of WritableFileWrapper:
- during compaction, all `PrepareWrite()` calls update `last_preallocated_block_` of the `WritableFileWrapper` itself, not of `target_`, since `PrepareWrite()` is not virtual,
- `PrepareWrite()` calls `Allocate()`, which is virtual; it does `fallocate()` on `target_`,
- after writing data, `target_->Close()` calls `GetPreallocationStatus()` of `target_`; it returns `last_preallocated_block_` of `target_`, which is zero because it was never touched before,
- `target_->Close()` doesn't call `ftruncate()`; file remains big.

This diff fixes it in a straightforward way, by making the methods virtual. `WritableFileWrapper` ends up having the useless fields `last_preallocated_block_` and `preallocation_block_size_`. I think ideally the preallocation logic should be outside `WritableFile`, the same way as `log_writer.h` and `file_reader_writer.h` moved some non-platform-specific logic out of Env, but that's probably not worth the effort now.

Test Plan: `make -j check`; I'm going to deploy it on our test tier and see if it fixes space reclamation problem there

Reviewers: yhchiang, sdong

Reviewed By: sdong

Subscribers: andrewkr, dhruba, sdong

Differential Revision: https://reviews.facebook.net/D54681

Avoid overloaded virtual function

Summary: Something changed and the special charactor seems to be conflict with an exisitng value. Change it to unblock the build.

Test Plan: Run the test and make sure it passes

Reviewers: kradhakrishnan

Reviewed By: kradhakrishnan

Subscribers: leveldb, andrewkr, dhruba

Differential Revision: https://reviews.facebook.net/D55845

Allow per-column-family processing in wal_filter (publish log numbers for column family to wal_filter, and provide log number in the record callback)

This reverts commit 522de4f5.

It has bug of index block cleaning up.

This reverts commit d7ae42b0.

This is reverted as auto buld failure. This commit itself doesn't have any problem. Reverting as it depends on the commit to revert.

tools/check_format_compatible.sh to use consistent version when testing backward and forward compatibility

Summary: Test seems to fail if we don't use consistent version between testing forward and backward compatibility.

Test Plan: Run the script (with some version removed manually to make it shorter)

Reviewers: IslamAbdelRahman

Reviewed By: IslamAbdelRahman

Subscribers: leveldb, andrewkr, dhruba

Differential Revision: https://reviews.facebook.net/D55773

Test Plan: sent diff to sdong, passes :)

Reviewers: sdong

Reviewed By: sdong

Subscribers: leveldb, andrewkr, dhruba

Differential Revision: https://reviews.facebook.net/D55749

Summary:
I ran into this assert when stress testing transactions.  It's pretty easy to repro.

Changing VersionSet::last_sequence_ to start at 1 seems pretty straightforward.  We would just need to change the 4 callers of SetLastSequence(), including recovery code.  I'd make this change myself, but I do not have enough time to test changes to recovery code-paths this week.  But checking in this test case (disabled) for future fixing.

Test Plan: n/a

Reviewers: yhchiang, kradhakrishnan, andrewkr, anthony, IslamAbdelRahman

Reviewed By: IslamAbdelRahman

Subscribers: andrewkr, dhruba, leveldb

Differential Revision: https://reviews.facebook.net/D55311

Summary:
Basic test cases:

- Manifest is lost or corrupt
- Manifest refers to too many or too few SST files
- SST file is corrupt
- Unflushed data is present when RepairDB is called

Depends on D55065 for its CreateFile() function in file_utils

Test Plan: Ran the tests.

Reviewers: IslamAbdelRahman, yhchiang, yoshinorim, sdong

Reviewed By: sdong

Subscribers: leveldb, andrewkr, dhruba

Differential Revision: https://reviews.facebook.net/D55485

Fix in HistogramWindowingImpl

Summary: After introducing a less forward-compatible change, update the backward compatible checking tool.

Test Plan: Run it.

Reviewers: IslamAbdelRahman

Reviewed By: IslamAbdelRahman

Subscribers: IslamAbdelRahman, leveldb, andrewkr, dhruba

Differential Revision: https://reviews.facebook.net/D55695

Test Plan: make -j40 check OPT=-g, on both /tmp and /dev/shm

Reviewers: sdong

Reviewed By: sdong

Subscribers: andrewkr, dhruba

Differential Revision: https://reviews.facebook.net/D55701

Summary:
When a block based table file is opened, if prefetch_index_and_filter is true, it will prefetch the index and filter blocks, putting them into the block cache.
What this feature adds: when a L0 block based table file is opened, if pin_l0_filter_and_index_blocks_in_cache is true in the options (and prefetch_index_and_filter is true), then the filter and index blocks aren't released back to the block cache at the end of BlockBasedTableReader::Open(). Instead the table reader takes ownership of them, hence pinning them, ie. the LRU cache will never push them out. Meanwhile in the table reader, further accesses will not hit the block cache, thus avoiding lock contention.
When the table reader is destroyed, it releases the pinned blocks (if there were any). This has to happen before the cache is destroyed, so I had to introduce a TableReader::Close(), to guarantee the order of destruction.

Test Plan:
Added two unit tests for this. Existing unit tests run fine (default is pin_l0_filter_and_index_blocks_in_cache=false).

DISABLE_JEMALLOC=1 OPT=-g make all valgrind_check -j32
  Mac: OK.
  Linux: with D55287 patched in it's OK.

Reviewers: sdong

Reviewed By: sdong

Subscribers: andrewkr, leveldb, dhruba

Differential Revision: https://reviews.facebook.net/D54801

Latest versions of Jemalloc library do not require je_init()/je_unint()

  calls. #ifdef in the source code and make this a default build option.

Summary: For compactions benchmarks (both level and universal) we'll use `--max_write_buffer_number=4`. For all the other benchmarks which don't customize the value of `--max_background_flushes` we'll continue using `--max_write_buffer_number=8`.

Test Plan:
To validate basic correctness and command-line options:

```
cd ~/rocksdb
NKEYS=10000000 ./tools/run_flash_bench.sh
```

Reviewers: MarkCallaghan

Reviewed By: MarkCallaghan

Subscribers: andrewkr, dhruba

Differential Revision: https://reviews.facebook.net/D55497

Summary:
This fixes a similar issue as D54711: "CURRENT" file can mutate between
GetLiveFiles() and copy to the tmp directory, in which case it would reference
the wrong manifest filename. To fix this, I forge the "CURRENT" file such that
it simply contains the filename for the manifest returned by GetLiveFiles().

- Changed CreateCheckpoint() to forge current file
- Added CreateFile() utility function
- Added test case that rolls manifest during checkpoint creation

Test Plan:
  $ ./checkpoint_test

Reviewers: sdong, IslamAbdelRahman

Reviewed By: IslamAbdelRahman

Subscribers: dhruba, leveldb

Differential Revision: https://reviews.facebook.net/D55065

Fixes warnings and ensure correct int behavior on 32-bit platforms.

Improve documentation of the allow_os_buffer parameter.

Summary:
DBBlockCacheTest.TestWithCompressedBlockCache is depending on compression using snappy, so this test fail when snappy is not available
block this test when we don't have snappy

https://ci-builds.fb.com/view/rocksdb/job/rocksdb_no_compression/833/console

Test Plan: run the test when compression libraries are not avaliable

Reviewers: sdong, yiwu

Reviewed By: yiwu

Subscribers: andrewkr, dhruba

Differential Revision: https://reviews.facebook.net/D55413

Fix FB internal CI build failure

Histogram Concurrency Improvement and Time-Windowing Support

Fix AppVeyor build error

Summary:
Using gcc 5.2.1 to compile RocksDB on Ubuntu 15.10 results in a warning about unused variable. Warning is treated as an error and all of it results in a build break. Fix the issue and make sure that everything compiles with that particular configuration.

P.S. Lint complains about a non-ASCII character used in the source code ("Wagner-Fischer"). Fix this as well.

Test Plan:
  # Everything compiler cleanly on the system which exhibited the problem.
  # `make clean && make -j 16` on CentOS.
  # `make clean && USE_CLANG=1 make -j 16` on CentOS.

Reviewers: sdong, igor

Reviewed By: igor

Subscribers: andrewkr, dhruba

Differential Revision: https://reviews.facebook.net/D55503