Fault tolerance is an important requirement for a data center OS. The ability to keep functioning in the event of a failure becomes indispensable, when working with large-scale systems. Mesos has no single point of failure in the architecture and can continue to operate in case of faults of various entities. There are three modes of fault tolerance that we have to deal with: machine failures, bugs in Mesos processes, and upgrades. Note that all of the points mentioned earlier can happen with any entity in Mesos. There are mainly three components of Mesos that need to be resilient to these faults: master, slave, and framework.
In case of machine running the Mesos slave fails, the master will notice, and inform the frameworks about the slave failure event. The framework can choose to reschedule the tasks running on that slave to other healthy slaves. Once the machine is fixed and the slave process is restarted in a healthy mode, it will reregister with the master and will, again...