We recently released Mesos v0.16.0 on our downloads page. It includes major refactoring work of the leading master election and detection process. This improves the reliability and flexibility of running multiple masters in your cluster, which provides Mesos with high availability.
In high availability mode, if a leading master machine fails, Mesos holds elections to determine a new leader. Slave machines and schedulers detect the new leading master and connect to it, without disrupting services running on Mesos. Leader election implementation details, including how it works with Zookeeper, are detailed in the high availablity documentation.
Aside from the refactoring, v0.16.0 includes fixes for bugs which caused incorrect termination of Mesos masters and slaves:
Click to read the full release notes.
To upgrade a live cluster, please refer to the Upgrades document.