ray.rllib.algorithms.algorithm.Algorithm.restore_workers

Algorithm.restore_workers(workers: WorkerSet) → None

Try syncing previously failed and restarted workers with local, if necessary.

Algorithms that use custom EnvRunners may override this method to disable default, and create custom restoration logics. Note that “restoring” does not include the actual restarting process, but merely what should happen after such a restart of a (previously failed) worker.

Parameters:

workers – The WorkerSet to restore. This may be Rollout or Evaluation workers.