Hello Kudu Jenkins,
I'd like you to reexamine a change. Please visit
http://gerrit.cloudera.org:8080/7988
to look at the new patch set (#3).
Change subject: KUDU-2123. Auto-vivify cmeta on tombstoned replicas if doesn't
exist at startup
......................................................................
KUDU-2123. Auto-vivify cmeta on tombstoned replicas if doesn't exist at startup
It is possible for tombstoned replicas to legitimately not have a cmeta
file as a result of crashing during a first tablet copy, or failing a
tablet copy operation in an older version of Kudu. Not having a cmeta
file results in those tombstoned replicas being unable to vote in Raft
leader elections. We remedy this by creating a cmeta object (with an
empty config) at startup time. The empty config is safe for a tombstoned
replica, because the config doesn't affect a replica's ability to vote
in a leader election. Additionally, if the tombstoned replica were ever
to be overwritten by a tablet copy operation, that would also result in
overwriting the config stored in the local cmeta with a valid Raft
config. Finally, all of this assumes that the nonexistence of a cmeta
file guarantees that the replica has never voted in a leader election.
As an optimization, the cmeta is created with the NO_FLUSH_ON_CREATE
flag, meaning that it will only be flushed to disk if the replica ever
votes.
The following changes had to be made to ConsensusMetadata and the
ConsensusMetadataManager to support the above functionality:
* Enable deferred flush on Create() by defining a flag called
NO_FLUSH_ON_CREATE
* Simplify the interface controlling whether a Flush() is allowed to
overwrite (clobber) an existing file by encapsulating that logic in
the implementation, instead of the public interface to
ConsensusMetadata. When a cmeta is instantiated via
ConsensusMetadata::Create(), the next Flush() is not allowed to
overwrite an existing file. In every other case, Flush() is allowed to
overwrite an existing file.
The following tests have been added:
* A unit test for NO_FLUSH_ON_CREATE.
* A test that crashes the target of a tablet copy after writing the
superblock and before writing the cmeta file. The tablet server is
restarted and the replica is expected to be able to vote while
tombstoned.
Previously-written tests that verify ConsensusMetadata::Create() will
not clobber an existing file still pass.
Change-Id: I8ff6255b1fcbb12417b82853bcde9b239291492b
---
M src/kudu/consensus/consensus_meta-test.cc
M src/kudu/consensus/consensus_meta.cc
M src/kudu/consensus/consensus_meta.h
M src/kudu/consensus/consensus_meta_manager-stress-test.cc
M src/kudu/consensus/consensus_meta_manager-test.cc
M src/kudu/consensus/consensus_meta_manager.cc
M src/kudu/consensus/consensus_meta_manager.h
M src/kudu/consensus/raft_consensus_quorum-test.cc
M src/kudu/integration-tests/CMakeLists.txt
A src/kudu/integration-tests/tombstoned_voting-itest.cc
M src/kudu/master/sys_catalog.cc
M src/kudu/tablet/tablet_bootstrap-test.cc
M src/kudu/tablet/tablet_replica-test.cc
M src/kudu/tserver/tablet_copy_client.cc
M src/kudu/tserver/tablet_copy_source_session-test.cc
M src/kudu/tserver/ts_tablet_manager.cc
16 files changed, 357 insertions(+), 38 deletions(-)
git pull ssh://gerrit.cloudera.org:29418/kudu refs/changes/88/7988/3
--
To view, visit http://gerrit.cloudera.org:8080/7988
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings
Gerrit-MessageType: newpatchset
Gerrit-Change-Id: I8ff6255b1fcbb12417b82853bcde9b239291492b
Gerrit-PatchSet: 3
Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-Owner: Mike Percy <[email protected]>
Gerrit-Reviewer: Adar Dembo <[email protected]>
Gerrit-Reviewer: Alexey Serbin <[email protected]>
Gerrit-Reviewer: Kudu Jenkins
Gerrit-Reviewer: Mike Percy <[email protected]>
Gerrit-Reviewer: Todd Lipcon <[email protected]>
