Open ccicconetti opened 2 months ago
It would be best if all the remote methods (implemented with gRPC) had a timeout.
Currently, if the peer does not answer (wrong end-point, peer terminated abruptly, etc.) this causes the originator to hang indefinitely.
(this is only a problem with the ε-ORC)
Currently,
Orchestrator::main_task
handles one request at a time.However, the handling time of a request may be non-negligible. For instance, the start/stop of a function may take several seconds with container functions. During that time, the ε-ORC will react to further requests and not perform internal procedures (e.g., node management).