[
https://issues.apache.org/jira/browse/FLINK-4322?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15418952#comment-15418952
]
ASF GitHub Bot commented on FLINK-4322:
---------------------------------------
GitHub user uce opened a pull request:
https://github.com/apache/flink/pull/2366
[FLINK-4322] Unify CheckpointCoordinator and SavepointCoordinator
The CheckpointCoordinator now also takes over the role of the
SavepointCoordinator. Savepoints are just like other checkpoints - they only
store the metadata in addition. Restoring from a savepoint means loading it
into the CheckpointStore at startup.
This simplifies the code quite a bit. We get rid of the savepoint
coordinator and related classes and cumbersome restoring logic in the main
code. For the tests, we can replace some integration tests with unit tests.
`PendingSavepoint` instances are finalized to become a
`CompletedCheckpoint` like regular `PendingCheckpoint` instances, but in
addition store the savepoint meta data and complete a Promise for callbacks.
`PendingSavepoints` cannot be subsumed and a `CompletedCheckpoint` from a
savepoint does not delete its associated state when being disposed.
@StephanEwen did most of the work and I added and fixed some tests.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/uce/flink savepointunify
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/flink/pull/2366.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #2366
----
commit 9d13d3b9c78b5fe6ec436c476492a82b846338aa
Author: Stephan Ewen <[email protected]>
Date: 2016-08-08T17:18:44Z
[FLINK-4322] [checkpointing] Unify CheckpointCoordinator and
SavepointCoordinator
The CheckpointCoordinator now also takes over the role of the
SavepointCoordinator.
Savepoints are just like other checkpoints - they only store the metadata
in addition.
Restoring from a savepoint means loading it into the CheckpointStore at
startup.
commit bcb6cf0b573314449437bc869febfc68f798b0f4
Author: Ufuk Celebi <[email protected]>
Date: 2016-08-11T17:40:07Z
[FLINK-4322] [checkpointing] Add and fix tests
----
> Unify CheckpointCoordinator and SavepointCoordinator
> ----------------------------------------------------
>
> Key: FLINK-4322
> URL: https://issues.apache.org/jira/browse/FLINK-4322
> Project: Flink
> Issue Type: Improvement
> Components: State Backends, Checkpointing
> Affects Versions: 1.1.0
> Reporter: Stephan Ewen
> Fix For: 1.2.0
>
>
> The Checkpoint coordinator should have the functionality of both handling
> checkpoints and savepoints.
> The difference between checkpoints and savepoints is minimal:
> - savepoints always write the root metadata of the checkpoint
> - savepoints are always full (never incremental)
> The commonalities are large
> - jobs should be able to resume from checkpoint or savepoints
> - jobs should fall back to the latest checkpoint or savepoint
> This subsumes issue https://issues.apache.org/jira/browse/FLINK-3397
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
