Fixed Issues

Version 14.0.0

Fixed a network deadlock that could occur on the Compute Server’s internal command channel between the server and its workers while a solve is running. If a response to a worker command was lost in transit, the server had no safe way to retry the request without risking the worker re-executing the command or the connection stalling. The channel now uses a call-index based idempotent retry protocol, active only for the duration of a solve: each request carries a monotonically increasing call index, and repeating a request with the same index replays the previously stored response instead of re-issuing the command to the worker. This resolves the deadlock and is most noticeable in clusters running large numbers of small, short-running jobs.