Product

Mesos 0.22.1 release improves task recovery in containers

May 22, 2015

D2iQ

D2iQ

2 min read

 
The Apache Mesos community recently released version 0.22.1 of the popular project, which serves as the kernel of Mesosphere's Datacenter Operating System and manages infrastructure at places such as Twitter and Yelp.
 
This release includes several new features, as well as fixes to task recovery, sandbox permissions, and the Java and Python bindings. Among the improvements are:
 
  • A bug fix in the Docker containerizer where a Docker task that failed immediately after launch might get stuck in a staging mode, and the Mesos slave couldn't recover or terminate the task.
  • A bug fix for task-recovery in slave nodes, particularly where both the Mesos containerizer and the Docker containerizer attempted to recover all the running tasks on recovery, regardless which one it was launched from. This caused a lot of errors in the log and, worse, could cause tasks to fail or stick.
  • A new capability for frameworks to update information such as name, hostname, failover_timeout and webui_url in the Mesos master when they re-register on another machine.
 
See the Changelog for the list of all the fixes and improvements in Mesos 0.22.1.
 
Mesosphere has built and tested Mesos 0.22.1 against all of our supported frameworks, services and tools, and we have made the Mesos 0.22.1 packages available as part of our official package repository. (Go to our downloads page for more information.)
 
In fact, our latest DCOS build already includes Mesos 0.22.1, and will soon include Marathon 0.8.2, as well. That forthcoming version of Marathon will take advantage of the latest fixes in Mesos 0.22.1, particularly the Java binding fixes for state abstraction.
 
Sign up now for the DCOS public beta and you can spin up a cluster with Mesos 0.22.1 and the rest of the Mesosphere ecosystem.

Ready to get started?