[
https://issues.apache.org/jira/browse/MESOS-2464?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14590734#comment-14590734
]
Till Toenshoff commented on MESOS-2464:
---------------------------------------
IMHO https://reviews.apache.org/r/31838/ still is a valid patch as it in any
case demonstrates a defensive pattern in not assuming that an optional
proto-member was always present.
However, for making sure that the slave's executors wont get shutdown on
authentication failures, it makes total sense to update the _authenticate
continuation to not send a shutdown message but instead simply EXIT -- thus
also not triggering the bug described (but fixed by the suggested RR).
> Authentication failure may lead to slave crash
> ----------------------------------------------
>
> Key: MESOS-2464
> URL: https://issues.apache.org/jira/browse/MESOS-2464
> Project: Mesos
> Issue Type: Bug
> Reporter: Till Toenshoff
> Assignee: Till Toenshoff
> Fix For: 0.23.0
>
>
> When slave authentication fails, the following attempt to transmit a
> {{UnregisterSlaveMessage}} may cause a crash within the slave.
> {noformat}
> E0309 01:08:34.819758 336699392 slave.cpp:740] Master
> [email protected]:5050 refused authentication
> I0309 01:08:34.819787 336699392 slave.cpp:538] Master refused authentication;
> unregistering and shutting down
> [libprotobuf FATAL google/protobuf/message_lite.cc:273] CHECK failed:
> IsInitialized(): Can't serialize message of type
> "mesos.internal.UnregisterSlaveMessage" because it is missing required
> fields: slave_id.value
> libprocess: slave(1)@192.168.178.20:5051 terminating due to CHECK failed:
> IsInitialized(): Can't serialize message of type
> "mesos.internal.UnregisterSlaveMessage" because it is missing required
> fields: slave_id.value
> {noformat}
> The problem here is the following code:
> {noformat}
> UnregisterSlaveMessage message_;
> message_.mutable_slave_id()->MergeFrom(info.id());
> {noformat}
> Authentication happens before registration. {{info.id}} is an optional member
> (of {{SlaveInfo}}) and not known yet. It is set later, while registering. So
> {{slave_id}} will remain unset.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
