Stream: t-compiler/cargo-bisect-rustc

Topic: slow downloads


pnkfelix (Feb 20 2020 at 20:20, on Zulip):

@simulacrum as an experiment, or really, to keep myself occupied, I tried doing some manual bisection while cargo-bisect-rustc ran

pnkfelix (Feb 20 2020 at 20:20, on Zulip):

and in the process of doing my own bisection, I had to run rustup update ...

pnkfelix (Feb 20 2020 at 20:21, on Zulip):

and I was a little surprised when it seemed like I was able to get the nightlies for my own bisection faster that way than cargo-bisect-rustc was doing itself.

pnkfelix (Feb 20 2020 at 20:21, on Zulip):

now to be fair, this was a totally not scientific test. I wasn't even doing the race on different computers, so the two ongoing processes were of course stealing time and network bandwidth from eachother

pnkfelix (Feb 20 2020 at 20:22, on Zulip):

but I am still curious: Would you expect cargo-bisect-rustc's download system to be significantly slower than that of rustp? Does the manner in which cargo-bisect-rustc is downloading its CI artifacts have some bottleneck I'm not aware of?

pnkfelix (Feb 20 2020 at 20:28, on Zulip):

(hmm I just thought of one potential reason that my setup could be slowing things down... I was running cargo-bisect-rustc inside of an Emacs shell, but my own bisection in a normal terminal. I know the Emacs shell can slow down interactive terminal stuff. I really should re-do the experiment outside of Emacs. That, and/or add a way to tell cargo-bisect-rustc not to include the progress bars and whatnot.)

simulacrum (Feb 20 2020 at 20:50, on Zulip):

We use reqwest 0.9 (should probably update to async/await 0.10, I guess).

simulacrum (Feb 20 2020 at 20:51, on Zulip):

I think rustup uses curl, which might have better support for some features (e.g., maybe HTTP 3 or something crazy like that?)

simulacrum (Feb 20 2020 at 20:51, on Zulip):

but the terminal thing seems more likely

simulacrum (Feb 20 2020 at 20:51, on Zulip):

at least in theory bisect and rustup should be about on par

pnkfelix (Feb 20 2020 at 20:52, on Zulip):

yeah I wont file an issue until i've gathered actual data

pnkfelix (Feb 20 2020 at 20:52, on Zulip):

in a more scientific manner

simulacrum (Feb 20 2020 at 20:52, on Zulip):

one difference (though I hesitate to suggest it) is that rustup presumably has libcurl ... as the User-Agent

simulacrum (Feb 20 2020 at 20:52, on Zulip):

whereas we have something adhoc or nothing at all

simulacrum (Feb 20 2020 at 20:52, on Zulip):

so maybe Amazon (CloudFront) rate limits bisect as a result?

simulacrum (Feb 20 2020 at 20:52, on Zulip):

that'd be pretty sad

Last update: Apr 03 2020 at 18:30UTC