]> git-server-git.apps.pok.os.sepia.ceph.com Git - rocksdb.git/log
rocksdb.git
8 years agoBump version to 5.8.7
Yi Wu [Tue, 28 Nov 2017 20:34:06 +0000 (12:34 -0800)]
Bump version to 5.8.7

8 years agoFix IOError on WAL write doesn't propagate to write group follower
Yi Wu [Tue, 28 Nov 2017 19:40:40 +0000 (11:40 -0800)]
Fix IOError on WAL write doesn't propagate to write group follower

Summary:
This is a simpler version of #3097 by removing all unrelated changes.

Fixing the bug where concurrent writes may get Status::OK while it actually gets IOError on WAL write. This happens when multiple writes form a write batch group, and the leader get an IOError while writing to WAL. The leader failed to pass the error to followers in the group, and the followers end up returning Status::OK() while actually writing nothing. The bug only affect writes in a batch group. Future writes after the batch group will correctly return immediately with the IOError.
Closes https://github.com/facebook/rocksdb/pull/3201

Differential Revision: D6421644

Pulled By: yiwu-arbug

fbshipit-source-id: 1c2a455c5b73f6842423785eb8a9dbfbb191dc0e

8 years agoBump version to 5.8.6 rocksdb-5.8.6 v5.8.6
Yi Wu [Mon, 20 Nov 2017 21:51:43 +0000 (13:51 -0800)]
Bump version to 5.8.6

8 years agoEnable cacheline_aligned_alloc() to allocate from jemalloc if enabled.
Dmitri Smirnov [Fri, 27 Oct 2017 20:14:07 +0000 (13:14 -0700)]
Enable cacheline_aligned_alloc() to allocate from jemalloc if enabled.

Summary:
Reuse WITH_JEMALLOC option in preparation for module search unification.
  Move jemalloc overrides into a separate .cc
  Remote obsolete JEMALLOC_NOINIT option.
Closes https://github.com/facebook/rocksdb/pull/3078

Differential Revision: D6174826

Pulled By: yiwu-arbug

fbshipit-source-id: 9970a0289b4490272d15853920d9d7531af91140

8 years agoAdd -DPORTABLE=1 to MSVC CI build
Dmitri Smirnov [Thu, 31 Aug 2017 23:42:05 +0000 (16:42 -0700)]
Add -DPORTABLE=1 to MSVC CI build

Summary:
Add -DPORTABLE=1
  port::cacheline_aligned_alloc() has arguments swapped which prevents every single test from running.
Closes https://github.com/facebook/rocksdb/pull/2815

Differential Revision: D5751661

Pulled By: siying

fbshipit-source-id: e0857d6e138ec46035b3c23d7c3c751901a0a4a0

8 years agoBump version to 5.8.5
Yi Wu [Tue, 14 Nov 2017 18:38:22 +0000 (10:38 -0800)]
Bump version to 5.8.5

8 years agoBlob DB: not using PinnableSlice move assignment
Yi Wu [Tue, 14 Nov 2017 02:03:57 +0000 (18:03 -0800)]
Blob DB: not using PinnableSlice move assignment

Summary:
The current implementation of PinnableSlice move assignment have an issue #3163. We are moving away from it instead of try to get the move assignment right, since it is too tricky.
Closes https://github.com/facebook/rocksdb/pull/3164

Differential Revision: D6319201

Pulled By: yiwu-arbug

fbshipit-source-id: 8f3279021f3710da4a4caa14fd238ed2df902c48

8 years agoBump version to 5.8.4
Yi Wu [Thu, 9 Nov 2017 05:28:42 +0000 (21:28 -0800)]
Bump version to 5.8.4

8 years agoBlob DB: Fix race condition between flush and write
Yi Wu [Thu, 9 Nov 2017 03:33:12 +0000 (19:33 -0800)]
Blob DB: Fix race condition between flush and write

Summary:
A race condition will happen when:
* a user thread writes a value, but it hits the write stop condition because there are too many un-flushed memtables, while holding blob_db_impl.write_mutex_.
* Flush is triggered and call flush begin listener and try to acquire blob_db_impl.write_mutex_.

Fixing it.
Closes https://github.com/facebook/rocksdb/pull/3149

Differential Revision: D6279805

Pulled By: yiwu-arbug

fbshipit-source-id: 0e3c58afb78795ebe3360a2c69e05651e3908c40

8 years agoBlob DB: Fix release build
Yi Wu [Wed, 8 Nov 2017 21:08:15 +0000 (13:08 -0800)]
Blob DB: Fix release build

Summary:
`compression` shadow the method name in `BlobFile`. Rename it.
Closes https://github.com/facebook/rocksdb/pull/3148

Differential Revision: D6274498

Pulled By: yiwu-arbug

fbshipit-source-id: 7d293596530998b23b6b8a8940f983f9b6343a98

8 years agoBump version to 5.8.3
Yi Wu [Wed, 8 Nov 2017 06:59:14 +0000 (22:59 -0800)]
Bump version to 5.8.3

8 years agoBlob DB: use compression in file header instead of global options
Yi Wu [Wed, 8 Nov 2017 01:40:44 +0000 (17:40 -0800)]
Blob DB: use compression in file header instead of global options

Summary:
To fix the issue of failing to decompress existing value after reopen DB with a different compression settings.
Closes https://github.com/facebook/rocksdb/pull/3142

Differential Revision: D6267260

Pulled By: yiwu-arbug

fbshipit-source-id: c7cf7f3e33b0cd25520abf4771cdf9180cc02a5f