Binary caches can now specify a priority in their nix-cache-info file.
The binary cache substituter checks caches in order of priority. This
is to ensure that fast, static caches like nixos.org/binary-cache are
processed before slow, dynamic caches like hydra.nixos.org.
This allows disabling the use of binary caches, e.g.
$ nix-build ... --option use-binary-caches false
Note that
$ nix-build ... --option binary-caches ''
does not disable all binary caches, since the caches defined by
channels will still be used.
This operation allows fixing corrupted or accidentally deleted store
paths by redownloading them using substituters, if available.
Since the corrupted path cannot be replaced atomically, there is a
very small time window (one system call) during which neither the old
(corrupted) nor the new (repaired) contents are available. So
repairing should be used with some care on critical packages like
Glibc.
Negative lookups are purged from the DB after a day, at most once per
day. However, for non-"have" lookups (e.g. all except "nix-env
-qas"), negative lookups are ignored after one hour. This is to
ensure that you don't have to wait a day for an operation like
"nix-env -i" to start using new binaries in the cache.
Should probably make this configurable.
Channels can now advertise a binary cache by creating a file
<channel-url>/binary-cache-url. The channel unpacker puts these in
its "binary-caches" subdirectory. Thus, the URLS of the binary caches
for the channels added by root appear in
/nix/var/nix/profiles/per-user/eelco/channels/binary-caches/*. The
binary cache substituter reads these and adds them to the list of
binary caches.
Querying all substitutable paths via "nix-env -qas" is potentially
hard on a server, since it involves sending thousands of HEAD
requests. So a binary cache must now have a meta-info file named
"nix-cache-info" that specifies whether the server wants this. It
also specifies the store prefix so that we don't send useless queries
to a binary cache for a different store prefix.
Since SubstitutionGoal::finished() in build.cc computes the hash
anyway, we can prevent the inefficiency of computing the hash twice by
letting the substituter tell Nix about the expected hash, which can
then verify it.
In "nix-env -qas", we don't need the substitute info, we just need to
know if it exists. This can be done using a HTTP HEAD request, which
saves bandwidth.
Note however that curl currently has a bug that prevents it from
reusing HTTP connections if HEAD requests return a 404:
https://sourceforge.net/tracker/?func=detail&aid=3542731&group_id=976&atid=100976
Without the patch attached to the issue, using HEAD is actually quite
a bit slower than GET.
Getting substitute information using the binary cache substituter has
non-trivial latency overhead. A package or NixOS system configuration
can have hundreds of dependencies, and in the worst case (when the
local info cache is empty) we have to do a separate HTTP request for
each of these. If the ping time to the server is t, getting N info
files will take tN seconds; e.g., with a ping time of 0.1s to
nixos.org, sequentially downloading 1000 info files (a typical NixOS
config) will take at least 100 seconds.
To fix this problem, the binary cache substituter can now perform
requests in parallel. This required changing the substituter
interface to support a function querySubstitutablePathInfos() that
queries multiple paths at the same time, and rewriting queryMissing()
to take advantage of parallelism. (Due to local caching,
parallelising queryMissing() is sufficient for most use cases, since
it's almost always called before building a derivation and thus fills
the local info cache.)
For example, parallelism speeds up querying all 1056 paths in a
particular NixOS system configuration from 116s to 2.6s. It works so
well because the eccentricity of the top-level derivation in the
dependency graph is only 9. So we only need 10 round-trips (when
using an unlimited number of parallel connections) to get everything.
Currently we do a maximum of 150 parallel connections to the server.
Thus it's important that the binary cache server (e.g. nixos.org) has
a high connection limit. Alternatively we could use HTTP pipelining,
but WWW::Curl doesn't support it and libcurl has a hard-coded limit of
5 requests per pipeline.
Using WWW::Curl rather than running an external curl process for every
NAR info file halves the time it takes to get info thanks to libcurl's
support for persistent HTTP connections. (We save a roundtrip per
file.) But the real gain will come from using parallel and/or
pipelined requests.