When running a job with a shell step on a remote node, there's a case where the master will lose connectivity with the remote node. All logging statements in trying to determine the functionality are dumped in the job log. After several hours, there is no reconnection for that pipeline, even if other jobs are run on that remote node successfully. The shell step was completed successfully.

An example job with the error can be found in the links section

This might be an issue with the pipeline and the durable-task plugin.