Jie Yu created MESOS-3814:
-----------------------------

             Summary: Add checks to make sure isolators and the launcher are 
compatible.
                 Key: MESOS-3814
                 URL: https://issues.apache.org/jira/browse/MESOS-3814
             Project: Mesos
          Issue Type: Bug
    Affects Versions: 0.26.0
            Reporter: Jie Yu


There's a recent change regarding the picking of which launcher (Linux or 
Posix) to use
https://reviews.apache.org/r/39604

In our environment, cgroups are not auto-mounted after reboot. We rely on Mesos 
itself to mount all relevant cgroups hierachies.

After the reboot, the above patch detects that 'freezer' hierarchy is not 
mounted, therefore, decided to use the Posix launcher (if --launcher is not 
specified explictly).

Port mapping isolator requires network namespace to be created for each 
container (thus requires Linux launcher). But we don't have a check to verify 
that launcher and isolators are compatible.

Slave thus starts fine and task failed with weird error like:
{noformat}
Collect failed: Failed to create the ingress qdisc on mesos61099: Link 
'mesos61099' is not found.
{noformat}

It does take us quite a few time to figure out the root cause.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to