The fact that queryPathInfo() is synchronous meant that we needed a
thread for every concurrent binary cache lookup, even though they end
up being handled by the same download thread. Requiring hundreds of
threads is not a good idea. So now there is an asynchronous version of
queryPathInfo() that takes a callback function to process the
result. Similarly, enqueueDownload() now takes a callback rather than
returning a future.
Thus, a command like
nix path-info --store https://cache.nixos.org/ -r /nix/store/slljrzwmpygy1daay14kjszsr9xix063-nixos-16.09beta231.dccf8c5
that returns 4941 paths now takes 1.87s using only 2 threads (the main
thread and the downloader thread). (This is with a prewarmed
CloudFront.)
The binary cache store can now use HTTP/2 to do lookups. This is much
more efficient than HTTP/1.1 due to multiplexing: we can issue many
requests in parallel over a single TCP connection. Thus it's no longer
necessary to use a bunch of concurrent TCP connections (25 by
default).
For example, downloading 802 .narinfo files from
https://cache.nixos.org/, using a single TCP connection, takes 11.8s
with HTTP/1.1, but only 0.61s with HTTP/2.
This did require a fairly substantial rewrite of the Downloader class
to use the curl multi interface, because otherwise curl wouldn't be
able to do multiplexing for us. As a bonus, we get connection reuse
even with HTTP/1.1. All downloads are now handled by a single worker
thread. Clients call Downloader::enqueueDownload() to tell the worker
thread to start the download, getting a std::future to the result.
