[
https://issues.apache.org/jira/browse/MESOS-8601?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16375270#comment-16375270
]
Greg Mann commented on MESOS-8601:
----------------------------------
{code}
commit b4e210678c04e57c2fa9f277b44f6d011da1846a (HEAD -> master, origin/master,
origin/HEAD, merge)
Author: Chun-Hung Hsiao <[email protected]>
Date: Fri Feb 23 18:37:17 2018 -0800
Added a master API test for agent re-registration after master failover.
This test verifies that subscribing to the 'api/v1' endpoint between a
master failover and an agent re-registration won't cause the master to
crash.
Review: https://reviews.apache.org/r/65775/
{code}
{code}
commit f2ec2b288e823424b2efe71d62ef90101b7a863f
Author: Chun-Hung Hsiao <[email protected]>
Date: Fri Feb 23 18:37:12 2018 -0800
Fixed a master API bug for agent re-registration after master failover.
When the master fails over and a client subscribes to the master before
agent re-registration, the master will crash when sending `TASK_ADDED`
because the framework info might not have been added to the master yet.
This patch fixes this bug.
Review: https://reviews.apache.org/r/65774/
{code}
> Master crashes during slave reregistration after failover.
> ----------------------------------------------------------
>
> Key: MESOS-8601
> URL: https://issues.apache.org/jira/browse/MESOS-8601
> Project: Mesos
> Issue Type: Bug
> Components: master
> Affects Versions: 1.5.0
> Reporter: Chun-Hung Hsiao
> Assignee: Chun-Hung Hsiao
> Priority: Blocker
> Labels: master
>
> The following happened after a master failover.
> During slave reregistration, new tasks were added and the new leading master
> notified all of its subscribers, and triggered the following check failure:
> {noformat}
> F0222 15:53:44.440387 2805 master.cpp:11190] Check failed: 'framework' Must
> be non NULL
> *** Check failure stack trace: ***
> @ 0x7f1357be521d google::LogMessage::Fail()
> @ 0x7f1357be704d google::LogMessage::SendToLog()
> @ 0x7f1357be4e0c google::LogMessage::Flush()
> @ 0x7f1357be7949 google::LogMessageFatal::~LogMessageFatal()
> @ 0x7f1356c80e2d google::CheckNotNull<>()
> @ 0x7f1356ce2666 mesos::internal::master::Master::Subscribers::send()
> @ 0x7f1356cece83 mesos::internal::master::Slave::addTask()
> @ 0x7f1356cf3206 mesos::internal::master::Slave::Slave()
> @ 0x7f1356cf5b90 mesos::internal::master::Master::__reregisterSlave()
> @ 0x7f1356d02cf8 mesos::internal::master::Master::_reregisterSlave()
> @ 0x7f1357b43761 process::ProcessBase::consume()
> @ 0x7f1357b5248c process::ProcessManager::resume()
> @ 0x7f1357b579f6
> _ZNSt6thread5_ImplISt12_Bind_simpleIFZN7process14ProcessManager12init_threadsEvEUlvE_vEEE6_M_runEv
> @ 0x7f1354e6c230 (unknown)
> @ 0x7f135468ae25 start_thread
> @ 0x7f13543b834d __clone
> {noformat}
> This was because the master tried to get the framework info when sending the
> notification:
> https://github.com/apache/mesos/blob/1.5.x/src/master/master.cpp#L11190
> But it added the framework after that:
> https://github.com/apache/mesos/blob/1.5.x/src/master/master.cpp#L6963
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)