[
https://issues.apache.org/jira/browse/MESOS-8987?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16522505#comment-16522505
]
Gastón Kleiman commented on MESOS-8987:
---------------------------------------
I agree that the agent should be allowed to retry the (re)registration if the
autenticator/authorizer failed to reach a concrete answer.
However I don't agree that the master should send a {{ShutdownMessage}} if it
the authenticator/authorizer clearly thinks the agent is not allowed to
register. A {{ShutdownMessage} makes an agent kill all the running tasks, which
I think is very drastic should only happen with explicit approval from an
operator. Furthermore this situation can be triggered by transient state as a
result from an admin fat-fingering permissions for a brief period of time.
I propose to allow in both cases the agent to try to (re)register for an long
(or indefinite) period of time instead of asking it to shut down.
> Master asks agent to shutdown upon auth errors
> ----------------------------------------------
>
> Key: MESOS-8987
> URL: https://issues.apache.org/jira/browse/MESOS-8987
> Project: Mesos
> Issue Type: Bug
> Components: master, security
> Affects Versions: 1.4.1, 1.5.1, 1.6.0, 1.7.0
> Reporter: Gastón Kleiman
> Assignee: Gastón Kleiman
> Priority: Blocker
> Labels: mesosphere
>
> The Mesos master sends a {{ShutdownMessage}} to an agent if there is an
> [authentication|https://github.com/apache/mesos/blob/d733b1031350e03bce443aa287044eb4eee1053a/src/master/master.cpp#L6532-L6543]
> or an
> [authorization|https://github.com/apache/mesos/blob/d733b1031350e03bce443aa287044eb4eee1053a/src/master/master.cpp#L6622-L6633]
> error during agent registration.
>
> Upon receipt of this message, the agent kills alls its tasks and commits
> suicide. This means that transient auth errors can lead to whole agents being
> killed along with it's tasks.
> I think the master should stop sending the {{ShutdownMessage}}s in these
> cases, or at least let the agent retry the registration a few times before
> asking it to shutdown.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
