Network requests sleep for 10 nanoseconds. Is that intentional?

[seh]

Running staticcheck over fargo reveals one potential problem: in file rpc.go's netReq function, when a request fails with a temporary error, we sleep for ten nanoseconds. Note that the magnitude is specified without any unit, allowing its conversion to a time.Duration with the default unit of nanoseconds.

% staticcheck                                    
[...]/github.com/hudl/fargo/rpc.go:111:15: sleeping for 10 nanoseconds is probably a bug. Be explicit if it isn't: time.Sleep(10 * time.Nanosecond) (SA1004)

Error SA1004 has the following description:

Suspiciously small untyped constant in time.Sleep

What was the intended duration there? @ryansb added that statement in commit 4d053d5. Ryan, do you remember?

[ryansb]

Oh, whoops. I think I intended 10 seconds, but that may be too long. A smarter option would probably be to back off, trying first 3 seconds later, then 9, then 30 or so.

[seh]

I agree. There are various backoff libraries out there, if you'd consider introducing a dependency for this. Otherwise, I can implement a simple enough adjustment, so long as we limit how configurable it is. These things tend to devolve into a sampler pack of patterns—strategies, policies, and such.