* Recover after feed fetch failure with exponential backoff
The current implementation stops attempting to fetch a feed if fusion encounters any error fetching it. The only way to continue fetching the feed is if the user manually forces a refresh.
This allows fusion to recover from feed fetch errors by tracking the number of consecutive failures and slowing down requests for consistent failure. If a feed always fails, we eventually slow to only checking it once per week.
Fixes#67
* Add comment
There's some complexity to deciding whether or not to update a feed, and there will be even more if we add support for things like honoring the Retry-After response header. I thought it would be helpful to extract the decision to a dedicated function so that we can unit test it.
Note that this change adjusts the decision logic slightly. Previously, we'd skip a suspended feed even if the force parameter was set to true. Now, we'll check a suspended feed if the force parameter was true. I don't think this actually changes anything because it doesn't seem like the user can force a refresh on a suspended feed, but I just wanted to call attention to the change.
I thought it would be helpful to get unit test coverage for the feed fetch logic.
Currently, our logic is pretty simple, but I'd like fusion to eventually be able to support the If-Modified-Since HTTP header, so we don't force servers to send the same data over and over. At that point, extra testing will be especially helpful, but I thought it would be a good idea to get testing infrastructure in place now.
