As of PR #376, we are using request sessions by default. If a job doesn't emit output for a long time, then tries to emit output, that session object might have been disconnected by the server, at which point the job fails. Allowing any retries fixes this.
Perhaps there is a better way to have both sessions and not have to add a retry adapter.
As of PR #376, we are using request sessions by default. If a job doesn't emit output for a long time, then tries to emit output, that session object might have been disconnected by the server, at which point the job fails. Allowing any retries fixes this.
Perhaps there is a better way to have both sessions and not have to add a retry adapter.