Currently, when debugging haskell-gargantext#161 (closed) I encounter an issue that when some async task is running in the backgroud and a backend error appears, the GET timer for task is not cleared. Hence the frontend can keep asking indefinitely for a broken task. I think this should be solved in some way: maybe implement a backoff (i.e. wait 1 second, if error then 2 seconds, if error then 4, if error then 8 etc) or just keep asking for erronous task a predefined number of times and no more (e.g. 3)?

Easy way to reproduce this is to start search, then kill server: the frontend will keep asking its GET requests.