aurora-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From David McLaughlin <da...@dmclaughlin.com>
Subject Re: Review Request 63536: Give jobs the ability to determine how to handle partitions by integrating with new Mesos Partition-Aware APIs
Date Wed, 15 Nov 2017 00:06:02 GMT

-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/63536/
-----------------------------------------------------------

(Updated Nov. 15, 2017, 12:05 a.m.)


Review request for Aurora, Jordan Ly, Santhosh Kumar Shanmugham, and Bill Farner.


Changes
-------

Support:

    Job(
     ...
      partition_policy=PartitionPolicy(delay_secs=30)
      ...
    )
    
 and 
 
    Job(
     ...
      partition_policy=Disable()
      ...
    )


Repository: aurora


Description
-------

This is my prototype code for adding partition-awareness to Aurora. There is a proposal document
to accompany this here: https://docs.google.com/document/d/1E3GlsVTJLEMAkDWk2_PTxzkRZcapb8nF_5q5AADQI7g/edit#

I'd like feedback on the high-level approach before adding unit tests, metrics, logging, etc.


Diffs (updated)
-----

  api/src/main/thrift/org/apache/aurora/gen/api.thrift 1d369263d779b549b9304018437f535ddc855966

  examples/vagrant/upstart/aurora-scheduler.conf 5ca3caef03b6632cd4dbf47711b1ef183f6a6449

  src/main/java/org/apache/aurora/scheduler/base/Conversions.java 33cc012a2cad929b0dd1ce236597b870cfc5aba0

  src/main/java/org/apache/aurora/scheduler/base/Jobs.java 8d5f4e57c6b4f847cb74471f246fd0b7dd0cbc36

  src/main/java/org/apache/aurora/scheduler/base/TaskTestUtil.java 7c223eae69503fe1d5bf34c430438637abcbcb9b

  src/main/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModule.java 5e83b2acdde7198d16427d4031e9772f78612554

  src/main/java/org/apache/aurora/scheduler/state/PartitionManager.java PRE-CREATION 
  src/main/java/org/apache/aurora/scheduler/state/SideEffect.java b91a0852e968b4aa9d74801601671cb61af3648a

  src/main/java/org/apache/aurora/scheduler/state/StateManagerImpl.java 9989ed441cd9bc442e6472768880ce7924c3bdd9

  src/main/java/org/apache/aurora/scheduler/state/StateModule.java c03fff11ea3a4086f9daaa8b07315006c1b481e4

  src/main/java/org/apache/aurora/scheduler/state/TaskStateMachine.java eb4fe7d78ad1e6ec430c428df527bd0cf3a053c1

  src/main/python/apache/aurora/client/cli/jobs.py b79ae56bee0e5692cacf1e66f4a4126b06aaffdc

  src/main/python/apache/aurora/config/schema/base.py 18ce826363009e1cc0beac5cce4edf42610d487e

  src/main/python/apache/aurora/config/thrift.py bedf8cd6641e1b1a930602791b758d584af4891c

  src/test/java/org/apache/aurora/scheduler/base/JobsTest.java 13f656f241a8a9a3d339f4053f165070c2669ef3

  src/test/java/org/apache/aurora/scheduler/config/CommandLineTest.java c2d875bb5c393dd95d75251fe86dc649ceba7bd9

  src/test/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModuleTest.java
7b0429109e9a7795e559db264e7737fc55ff0169 
  src/test/java/org/apache/aurora/scheduler/state/PartitionManagerTest.java PRE-CREATION 
  src/test/java/org/apache/aurora/scheduler/state/StateManagerImplTest.java 0366cd6e9ddba0c3b9c88ffb50738767a352a17a

  src/test/java/org/apache/aurora/scheduler/state/TaskStateMachineTest.java 8d6c3fff0af2df39bb929f760b862a2edf5d6fca

  src/test/python/apache/aurora/client/cli/test_task.py 186cb2737ba8e169819b7d54f86a7344a669b6cb

  src/test/python/apache/aurora/config/test_thrift.py 7a1567a9b67917072bb0ba3eea5857e968375f4d



Diff: https://reviews.apache.org/r/63536/diff/5/

Changes: https://reviews.apache.org/r/63536/diff/4-5/


Testing
-------

Manual testing in Vagrant by stopping and starting the Mesos agent. With three jobs:

1) No PartitionPolicy (verified existing behavior of moving from PARTITIONED directly to LOST)
2) PartitionPolicy with custom delay_secs (verified sat in PARTITIONED for a while before
moving to LOST)
3) PartitionPolicy with reschedule=False (verified sat in PARTITIONED indefinitely)

I also verified tasks are able to transition to various states (back to RUNNING or moving
to FAILED, etc.) when you turn the agent back on.


File Attachments
----------------

Task in PARTITIONED state
  https://reviews.apache.org/media/uploaded/files/2017/11/07/02c7fc72-b11d-4ef9-a86b-914e748cad99__Screen_Shot_2017-11-07_at_11.23.41_AM.png
Task back into running when partition resolved
  https://reviews.apache.org/media/uploaded/files/2017/11/07/a0413f54-1572-4410-a386-0a22e78fab13__Screen_Shot_2017-11-07_at_11.26.02_AM.png
Compaction of PARTITIONED cycles (note timestamps)
  https://reviews.apache.org/media/uploaded/files/2017/11/07/edec32e5-b3ec-4fdc-b93f-5449519805ae__Screen_Shot_2017-11-07_at_11.27.47_AM.png


Thanks,

David McLaughlin


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message