You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "ASF GitHub Bot (JIRA)" <ji...@apache.org> on 2018/07/02 12:20:00 UTC
[jira] [Commented] (FLINK-9004) Cluster test: Run general purpose
job with failures with Yarn session
[ https://issues.apache.org/jira/browse/FLINK-9004?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16529796#comment-16529796 ]
ASF GitHub Bot commented on FLINK-9004:
---------------------------------------
GitHub user GJL opened a pull request:
https://github.com/apache/flink/pull/6239
[FLINK-9004][tests] Implement Jepsen tests to test job availability.
## What is the purpose of the change
*Use the Jepsen framework (https://github.com/jepsen-io/jepsen) to implement
tests that verify Flink's HA capabilities under real-world faults, such as
sudden TaskManager/JobManager termination, HDFS NameNode unavailability, network
partitions, etc. The Flink cluster under test is automatically deployed on YARN
(session & job mode) and Mesos.
Provide Dockerfiles for local test development.*
## Brief change log
- *Implement Jepsen tests.*
## Verifying this change
This change added tests and can be verified as follows:
- *The changes themselves are tests.*
- *Run Jepsen tests in docker containers.*
- *Run unit tests with `lein test`*
## Does this pull request potentially affect one of the following parts:
- Dependencies (does it add or upgrade a dependency): (yes / **no** (at least not to Flink))
- The public API, i.e., is any changed class annotated with `@Public(Evolving)`: (yes / **no**)
- The serializers: (yes / **no** / don't know)
- The runtime per-record code paths (performance sensitive): (yes / **no** / don't know)
- Anything that affects deployment or recovery: JobManager (and its components), Checkpointing, Yarn/Mesos, ZooKeeper: (yes / **no** (but it will as soon as test failures appear) / don't know)
- The S3 file system connector: (yes / **no** / don't know)
## Documentation
- Does this pull request introduce a new feature? (yes / **no**)
- If yes, how is the feature documented? (**not applicable** / docs / JavaDocs / not documented)
cc: @tillrohrmann @cewood
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/GJL/flink FLINK-9004
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/flink/pull/6239.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 #6239
----
commit 063e4621a5982b55ee7f7b0935290bbc717a5a45
Author: gyao <ga...@...>
Date: 2018-03-05T21:23:33Z
[FLINK-9004][tests] Implement Jepsen tests to test job availability.
Use the Jepsen framework (https://github.com/jepsen-io/jepsen) to implement
tests that verify Flink's HA capabilities under real-world faults, such as
sudden TaskManager/JobManager termination, HDFS NameNode unavailability, network
partitions, etc. The Flink cluster under test is automatically deployed on YARN
(session & job mode) and Mesos.
Provide Dockerfiles for local test development.
----
> Cluster test: Run general purpose job with failures with Yarn session
> ---------------------------------------------------------------------
>
> Key: FLINK-9004
> URL: https://issues.apache.org/jira/browse/FLINK-9004
> Project: Flink
> Issue Type: Sub-task
> Components: Tests
> Affects Versions: 1.5.0
> Reporter: Till Rohrmann
> Assignee: Gary Yao
> Priority: Blocker
> Labels: pull-request-available
> Fix For: 1.6.0, 1.5.1
>
>
> Similar to FLINK-8973, we should run the general purpose job (FLINK-8971) on a Yarn session cluster and simulate failures.
> The job jar should be ill-packaged, meaning that we include too many dependencies in the user jar. We should include the Scala library, Hadoop and Flink itself to verify that there are no class loading issues.
> The general purpose job should run with misbehavior activated. Additionally, we should simulate at least the following failure scenarios:
> * Kill Flink processes
> * Kill connection to storage system for checkpoints and jobs
> * Simulate network partition
> We should run the test at least with the following state backend: RocksDB incremental async and checkpointing to S3.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)