]> git-server-git.apps.pok.os.sepia.ceph.com Git - rocksdb.git/log
rocksdb.git
7 years agoFix a backward compatibility problem with table_properties being nullptr
Zhongyi Xie [Tue, 22 May 2018 20:51:21 +0000 (13:51 -0700)]
Fix a backward compatibility problem with table_properties being nullptr

Summary:
Currently when ldb built from master tries to open a DB from version 2.2, there will be a segfault because table_properties didn't exist back then.
Closes https://github.com/facebook/rocksdb/pull/3890

Differential Revision: D8100914

Pulled By: miasantreble

fbshipit-source-id: b255e8aedc54695432be2e704839c857dabdd65a

7 years agoFix Issue #3771: Slice ctor checks for nullptr and creates empty string
Jacquin Mininger [Tue, 22 May 2018 20:34:26 +0000 (13:34 -0700)]
Fix Issue #3771: Slice ctor checks for nullptr and creates empty string

Summary:
Fix Issue #3771   : Check for nullptr in Slice constructor
Slice ctor checks for nullptr and creates empty string if the string does not exist
Closes https://github.com/facebook/rocksdb/pull/3887

Differential Revision: D8098852

Pulled By: ajkr

fbshipit-source-id: 04471077defa9776ce7b8c389a61312ce31002fb

7 years agoAvoid sleep in DBTest.GroupCommitTest to fix flakiness
Andrew Kryczka [Tue, 22 May 2018 18:58:51 +0000 (11:58 -0700)]
Avoid sleep in DBTest.GroupCommitTest to fix flakiness

Summary:
DBTest.GroupCommitTest would often fail when run under valgrind because its sleeps were insufficient to guarantee a group commit had multiple entries. Instead we can use sync point to force a leader to wait until a non-leader thread has enqueued its work, thus guaranteeing a leader can do group commit work for multiple threads.
Closes https://github.com/facebook/rocksdb/pull/3883

Differential Revision: D8079429

Pulled By: ajkr

fbshipit-source-id: 61dc50fad29d2c85547842f681288de60fa29049

7 years agoAvoid single-deleting merge operands in db_stress
Andrew Kryczka [Tue, 22 May 2018 17:50:09 +0000 (10:50 -0700)]
Avoid single-deleting merge operands in db_stress

Summary:
I repro'd some of the "unexpected value" failures showing up in our CI lately and they always happened on keys that have a mix of single deletes and merge operands. The `SingleDelete()` API comment mentions it's incompatible with `Merge()`, so this PR prevents `db_stress` from mixing them.
Closes https://github.com/facebook/rocksdb/pull/3878

Differential Revision: D8097346

Pulled By: ajkr

fbshipit-source-id: 357a48c6a31156f4f8db3ce565638ad924c437a1