This document describes the handling of failures in Ray.
Machine and Process Failures¶
Each raylet (the scheduler process) sends heartbeats to a monitor process. If the monitor does not receive any heartbeats from a given raylet for some period of time (about ten seconds), then it will mark that process as dead.
If an object is needed but is lost or was never created, then the task that created the object will be re-executed to create the object. If necessary, tasks needed to create the input arguments to the task being re-executed will also be re-executed. This is the standard lineage-based fault tolerance strategy used by other systems like Spark.
When an actor dies (either because the actor process crashed or because the node that the actor was on died), by default any attempt to get an object from that actor that cannot be created will raise an exception. Subsequent releases will include an option for automatically restarting actors.
At the moment, Ray does not handle all failure scenarios. We are working on addressing these known problems.
- Ray does not recover from the failure of any of the following processes: a Redis server and the monitor process.
- If a driver fails, that driver will not be restarted and the job will not complete.
- If an object is constructed by a call to
ray.puton the driver, is then evicted, and is later needed, Ray will not reconstruct this object.
- If an object is constructed by an actor method, is then evicted, and is later needed, Ray will not reconstruct this object.