As a Dotscientist, I'd like to get an immediate error message when trying to launch Jupyter if one of the state machine involved is too busy to service my request, rather than seeing an infinite series of inscrutable intercluster transfer not found errors.
Implementation
Rather than letting requests pile up while a transfer is in progress, empty out f.innerRequests by refusing all requests with a "busy" message. Better to fail quickly than hang for hours, especially when things like RPCs will tend to timeout.
User story
As a Dotscientist, I'd like to get an immediate error message when trying to launch Jupyter if one of the state machine involved is too busy to service my request, rather than seeing an infinite series of inscrutable
intercluster transfer not founderrors.Implementation
Rather than letting requests pile up while a transfer is in progress, empty out
f.innerRequestsby refusing all requests with a "busy" message. Better to fail quickly than hang for hours, especially when things like RPCs will tend to timeout.