IOError(EOFError()

[quinnj]

@samoconnor, for the IOError(EOFError()) ... errors, I wonder if we could include how long the underlying connection had been idle (if applicable); this could be an easy way to detect if an idle timeout should be used (for users who run in to this problem). I'm in a situation right now where I realized I'm running in to these errors every so often, but it's not totally straightforward how to go about figuring out if this is an idle connection timeout error or not; if I had the timeout value in the error message, I could quickly scan all these types of errors in my logs and (possibly) find a pattern where requests failed most consistently w/ connections that were over 30 seconds or whatever.

[samoconnor]

The show(io::IO, c::Connection) method displays inactive seconds if the value is > 5. However, it is probably always < 5 in the case where you get an IOError because startread has probably just be called and will have reset the timestamp.

This change: edb044c#diff-e9e66c505dad95c5784e8813441444afR3
... adds timestamps to all of the debug log messages. I hope that will help with figuring out timing related issues.

[samoconnor]

With this change: c82979e#diff-4f12f5771d582a1c2672dbe6a4346be5R590
... time since last read is always displayed in show(::IO, ::Connection) so the ♻️ Idle log when an idle connection is reused will now show the number of seconds since the last read (46s):

julia> HTTP.get("http://httpbin.org/ip")
DEBUG: 2018-04-16T15:05:32.758 8b0b ♻️  Idle:           ⏸    1↑     1↓   46s httpbin.org:80:61070 ≣16 RawFD(23)
DEBUG: 2018-04-16T15:05:32.76  8b0b 👁  Start write:T1  ⏸    1↑     1↓   46s httpbin.org:80:61070 ≣16 RawFD(23)

[samoconnor]

Also related, this fix JuliaLang/MbedTLS.jl#145 may make this type of EOFError much less common.

[samoconnor]

@quinnj any further action needed here? or are the logging enhancements + #236 / JuliaLang/MbedTLS.jl#145 enough?

[quinnj]

Yeah, we're good. Thanks.