[
https://issues.apache.org/jira/browse/MESOS-878?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14004154#comment-14004154
]
Yifan Gu commented on MESOS-878:
--------------------------------
Hey guys,
I found that when a slave detects a new master, it will check its state before
sending (re)registration message, if it's in TERMINATING, then it will skip the
registration.
So in such case, this problem appears when the slave enters TERMINATING right
after it sends out the (re)registraton message:
(detected() ==> call doReliableRegistration() ==> enter TERMINATING by some
reason )
I am trying to simulate such process in the test case by changing the slave's
state to TERMINATING before it gets registered. And I am trying to do this by
using DROP_PROTOBUFS to block the slaveReregisteredMessage, but I can't find a
way to unblock it.
Any ideas? Thanks in advance!
> Slave should not register with the master when in TERMINATING.
> --------------------------------------------------------------
>
> Key: MESOS-878
> URL: https://issues.apache.org/jira/browse/MESOS-878
> Project: Mesos
> Issue Type: Bug
> Reporter: Benjamin Mahler
> Assignee: Yifan Gu
>
> If the slave takes a long time to shut down, doReliableRegistration in the
> slave will continue to re-register (causing the master to re-add the slave)
> and the slave will continue to drop the re-registration acknowledgement
> messages since it is in the TERMINATING state.
> The slave should not attempt to register when it is in a TERMINATING state.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
