You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@mesos.apache.org by "Benjamin Mahler (JIRA)" <ji...@apache.org> on 2013/08/23 01:35:51 UTC

[jira] [Created] (MESOS-660) Better slave recovery semantics when flags are changed.

Benjamin Mahler created MESOS-660:
-------------------------------------

             Summary: Better slave recovery semantics when flags are changed.
                 Key: MESOS-660
                 URL: https://issues.apache.org/jira/browse/MESOS-660
             Project: Mesos
          Issue Type: Improvement
            Reporter: Benjamin Mahler
            Assignee: Vinod Kone
             Fix For: 0.15.0


Ideally we can have one of the following semantics apply for all of the slave flags:

1. Safety: If the flag is modified and this will have an strange or unintended behavior, do not proceed.

2. "Do The Right Thing": If flags are modified and the slave can take action to ensure the expected behavior occurs. Many of the isolator related flags can DTRT.

We currently have neither of these properties covered for all flags. For example:
  -Any of the slave flags that affect SlaveInfo will have "Safe" semantics. The slave will not proceed with recovery as the master has cached the SlaveInfo.
  -Adding --cgroups_cfs_enabled will DTRT (apply cfs_quota_us to existing executors). But removing it will not DTRT (will leave cfs_quota_us untouched if it is already applied)!

It's not clear what level of support we want to provide for flag changes, but at the very least we need Safety, otherwise things will behave strangely.

Assigning this to Vinod as a placeholder for now.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira