tvl-depot/scripts/download-from-binary-cache.pl.in

397 lines
12 KiB
Perl
Raw Normal View History

#! @perl@ -w @perlFlags@
use DBI;
use File::Basename;
use IO::Select;
use Nix::Config;
use Nix::Store;
use WWW::Curl::Easy;
use WWW::Curl::Multi;
use strict;
my @binaryCacheUrls = map { s/\/+$//; $_ } split(/ /, ($ENV{"NIX_BINARY_CACHES"} || ""));
download-from-binary-cache: parallelise fetching of NAR info files 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.
2012-07-07 01:08:20 +02:00
my $maxParallelRequests = 150;
my ($dbh, $insertNAR, $queryNAR, $insertNegativeNAR, $queryNegativeNAR);
my %cacheIds;
my $curlm = WWW::Curl::Multi->new;
my $activeRequests = 0;
my $curlIdCount = 1;
download-from-binary-cache: parallelise fetching of NAR info files 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.
2012-07-07 01:08:20 +02:00
my %requests;
my %scheduled;
my $caBundle = $ENV{"CURL_CA_BUNDLE"} || $ENV{"OPENSSL_X509_CERT_FILE"};
sub addRequest {
download-from-binary-cache: parallelise fetching of NAR info files Getting substitute information using the binary cache substituter has non-trivial latency overhead. A package or NixOS system configuration can have hundreds of dependencies, and in the worst case (when the local info cache is empty) we have to do a separate HTTP request for each of these. If the ping time to the server is t, getting N info files will take tN seconds; e.g., with a ping time of 0.1s to nixos.org, sequentially downloading 1000 info files (a typical NixOS config) will take at least 100 seconds. To fix this problem, the binary cache substituter can now perform requests in parallel. This required changing the substituter interface to support a function querySubstitutablePathInfos() that queries multiple paths at the same time, and rewriting queryMissing() to take advantage of parallelism. (Due to local caching, parallelising queryMissing() is sufficient for most use cases, since it's almost always called before building a derivation and thus fills the local info cache.) For example, parallelism speeds up querying all 1056 paths in a particular NixOS system configuration from 116s to 2.6s. It works so well because the eccentricity of the top-level derivation in the dependency graph is only 9. So we only need 10 round-trips (when using an unlimited number of parallel connections) to get everything. Currently we do a maximum of 150 parallel connections to the server. Thus it's important that the binary cache server (e.g. nixos.org) has a high connection limit. Alternatively we could use HTTP pipelining, but WWW::Curl doesn't support it and libcurl has a hard-coded limit of 5 requests per pipeline.
2012-07-07 01:08:20 +02:00
my ($storePath, $url) = @_;
my $curl = WWW::Curl::Easy->new;
my $curlId = $curlIdCount++;
download-from-binary-cache: parallelise fetching of NAR info files 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.
2012-07-07 01:08:20 +02:00
$requests{$curlId} = { storePath => $storePath, url => $url, handle => $curl, content => "" };
$curl->setopt(CURLOPT_PRIVATE, $curlId);
$curl->setopt(CURLOPT_URL, $url);
download-from-binary-cache: parallelise fetching of NAR info files 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.
2012-07-07 01:08:20 +02:00
$curl->setopt(CURLOPT_WRITEDATA, \$requests{$curlId}->{content});
$curl->setopt(CURLOPT_FOLLOWLOCATION, 1);
$curl->setopt(CURLOPT_CAINFO, $caBundle) if defined $caBundle;
download-from-binary-cache: parallelise fetching of NAR info files Getting substitute information using the binary cache substituter has non-trivial latency overhead. A package or NixOS system configuration can have hundreds of dependencies, and in the worst case (when the local info cache is empty) we have to do a separate HTTP request for each of these. If the ping time to the server is t, getting N info files will take tN seconds; e.g., with a ping time of 0.1s to nixos.org, sequentially downloading 1000 info files (a typical NixOS config) will take at least 100 seconds. To fix this problem, the binary cache substituter can now perform requests in parallel. This required changing the substituter interface to support a function querySubstitutablePathInfos() that queries multiple paths at the same time, and rewriting queryMissing() to take advantage of parallelism. (Due to local caching, parallelising queryMissing() is sufficient for most use cases, since it's almost always called before building a derivation and thus fills the local info cache.) For example, parallelism speeds up querying all 1056 paths in a particular NixOS system configuration from 116s to 2.6s. It works so well because the eccentricity of the top-level derivation in the dependency graph is only 9. So we only need 10 round-trips (when using an unlimited number of parallel connections) to get everything. Currently we do a maximum of 150 parallel connections to the server. Thus it's important that the binary cache server (e.g. nixos.org) has a high connection limit. Alternatively we could use HTTP pipelining, but WWW::Curl doesn't support it and libcurl has a hard-coded limit of 5 requests per pipeline.
2012-07-07 01:08:20 +02:00
if ($activeRequests >= $maxParallelRequests) {
$scheduled{$curlId} = 1;
} else {
$curlm->add_handle($curl);
$activeRequests++;
}
download-from-binary-cache: parallelise fetching of NAR info files 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.
2012-07-07 01:08:20 +02:00
return $requests{$curlId};
}
sub processRequests {
while ($activeRequests) {
my ($rfds, $wfds, $efds) = $curlm->fdset();
#print STDERR "R = @{$rfds}, W = @{$wfds}, E = @{$efds}\n";
# Sleep until we can read or write some data.
if (scalar @{$rfds} + scalar @{$wfds} + scalar @{$efds} > 0) {
IO::Select->select(IO::Select->new(@{$rfds}), IO::Select->new(@{$wfds}), IO::Select->new(@{$efds}), 0.1);
}
if ($curlm->perform() != $activeRequests) {
while (my ($id, $result) = $curlm->info_read) {
if ($id) {
download-from-binary-cache: parallelise fetching of NAR info files 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.
2012-07-07 01:08:20 +02:00
my $handle = $requests{$id}->{handle};
$requests{$id}->{result} = $result;
$requests{$id}->{httpStatus} = $handle->getinfo(CURLINFO_HTTP_CODE);
#print STDERR "\nRequest completed ($id, $result, $requests{$id}->{httpStatus})\n";
$activeRequests--;
download-from-binary-cache: parallelise fetching of NAR info files 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.
2012-07-07 01:08:20 +02:00
delete $requests{$id}->{handle};
if (scalar(keys %scheduled) > 0) {
my $id2 = (keys %scheduled)[0];
$curlm->add_handle($requests{$id2}->{handle});
$activeRequests++;
delete $scheduled{$id2};
}
}
}
}
}
}
sub initCache {
my $dbPath = "$Nix::Config::stateDir/binary-cache-v1.sqlite";
# Open/create the database.
$dbh = DBI->connect("dbi:SQLite:dbname=$dbPath", "", "")
or die "cannot open database `$dbPath'";
$dbh->{RaiseError} = 1;
$dbh->{PrintError} = 0;
$dbh->do("pragma synchronous = off"); # we can always reproduce the cache
$dbh->do("pragma journal_mode = truncate");
# Initialise the database schema, if necessary.
$dbh->do(<<EOF);
create table if not exists BinaryCaches (
id integer primary key autoincrement not null,
url text unique not null
);
EOF
$dbh->do(<<EOF);
create table if not exists NARs (
cache integer not null,
storePath text not null,
url text not null,
compression text not null,
fileHash text,
fileSize integer,
narHash text,
narSize integer,
refs text,
deriver text,
system text,
timestamp integer not null,
primary key (cache, storePath),
foreign key (cache) references BinaryCaches(id) on delete cascade
);
EOF
$dbh->do(<<EOF);
create table if not exists NegativeNARs (
cache integer not null,
storePath text not null,
timestamp integer not null,
primary key (cache, storePath),
foreign key (cache) references BinaryCaches(id) on delete cascade
);
EOF
$insertNAR = $dbh->prepare(
"insert or replace into NARs(cache, storePath, url, compression, fileHash, fileSize, narHash, " .
"narSize, refs, deriver, system, timestamp) values (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)") or die;
$queryNAR = $dbh->prepare("select * from NARs where cache = ? and storePath = ?") or die;
$insertNegativeNAR = $dbh->prepare(
"insert or replace into NegativeNARs(cache, storePath, timestamp) values (?, ?, ?)") or die;
$queryNegativeNAR = $dbh->prepare("select 1 from NegativeNARs where cache = ? and storePath = ?") or die;
}
download-from-binary-cache: parallelise fetching of NAR info files Getting substitute information using the binary cache substituter has non-trivial latency overhead. A package or NixOS system configuration can have hundreds of dependencies, and in the worst case (when the local info cache is empty) we have to do a separate HTTP request for each of these. If the ping time to the server is t, getting N info files will take tN seconds; e.g., with a ping time of 0.1s to nixos.org, sequentially downloading 1000 info files (a typical NixOS config) will take at least 100 seconds. To fix this problem, the binary cache substituter can now perform requests in parallel. This required changing the substituter interface to support a function querySubstitutablePathInfos() that queries multiple paths at the same time, and rewriting queryMissing() to take advantage of parallelism. (Due to local caching, parallelising queryMissing() is sufficient for most use cases, since it's almost always called before building a derivation and thus fills the local info cache.) For example, parallelism speeds up querying all 1056 paths in a particular NixOS system configuration from 116s to 2.6s. It works so well because the eccentricity of the top-level derivation in the dependency graph is only 9. So we only need 10 round-trips (when using an unlimited number of parallel connections) to get everything. Currently we do a maximum of 150 parallel connections to the server. Thus it's important that the binary cache server (e.g. nixos.org) has a high connection limit. Alternatively we could use HTTP pipelining, but WWW::Curl doesn't support it and libcurl has a hard-coded limit of 5 requests per pipeline.
2012-07-07 01:08:20 +02:00
sub negativeHit {
my ($storePath, $binaryCacheUrl) = @_;
$queryNegativeNAR->execute(getCacheId($binaryCacheUrl), basename($storePath));
return @{$queryNegativeNAR->fetchall_arrayref()} != 0;
}
download-from-binary-cache: parallelise fetching of NAR info files 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.
2012-07-07 01:08:20 +02:00
sub processNARInfo {
my ($storePath, $binaryCacheUrl, $request) = @_;
download-from-binary-cache: parallelise fetching of NAR info files 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.
2012-07-07 01:08:20 +02:00
my $cacheId = getCacheId($binaryCacheUrl);
if ($request->{result} != 0 || $request->{httpStatus} != 200) {
if ($request->{httpStatus} != 404) {
download-from-binary-cache: parallelise fetching of NAR info files 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.
2012-07-07 01:08:20 +02:00
print STDERR "could not download $request->{url} (" .
($request->{result} != 0 ? "Curl error $request->{result}" : "HTTP status $request->{httpStatus}") . ")\n";
} else {
$insertNegativeNAR->execute($cacheId, basename($storePath), time());
}
return undef;
}
my ($storePath2, $url, $fileHash, $fileSize, $narHash, $narSize, $deriver, $system);
my $compression = "bzip2";
my @refs;
foreach my $line (split "\n", $request->{content}) {
$line =~ /^(.*): (.*)$/ or return undef;
if ($1 eq "StorePath") { $storePath2 = $2; }
elsif ($1 eq "URL") { $url = $2; }
2012-07-02 00:46:38 +02:00
elsif ($1 eq "Compression") { $compression = $2; }
elsif ($1 eq "FileHash") { $fileHash = $2; }
elsif ($1 eq "FileSize") { $fileSize = int($2); }
elsif ($1 eq "NarHash") { $narHash = $2; }
elsif ($1 eq "NarSize") { $narSize = int($2); }
elsif ($1 eq "References") { @refs = split / /, $2; }
elsif ($1 eq "Deriver") { $deriver = $2; }
elsif ($1 eq "System") { $system = $2; }
}
return undef if $storePath ne $storePath2;
if ($storePath ne $storePath2 || !defined $url || !defined $narHash) {
download-from-binary-cache: parallelise fetching of NAR info files 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.
2012-07-07 01:08:20 +02:00
print STDERR "bad NAR info file $request->{url}\n";
return undef;
}
# Cache the result.
$insertNAR->execute(
$cacheId, basename($storePath), $url, $compression, $fileHash, $fileSize,
$narHash, $narSize, join(" ", @refs), $deriver, $system, time());
return
{ url => $url
, compression => $compression
, fileHash => $fileHash
, fileSize => $fileSize
, narHash => $narHash
, narSize => $narSize
, refs => [ @refs ]
, deriver => $deriver
, system => $system
};
}
sub getCacheId {
my ($binaryCacheUrl) = @_;
my $cacheId = $cacheIds{$binaryCacheUrl};
return $cacheId if defined $cacheId;
# FIXME: not atomic.
my @res = @{$dbh->selectcol_arrayref("select id from BinaryCaches where url = ?", {}, $binaryCacheUrl)};
if (scalar @res == 1) {
$cacheId = $res[0];
} else {
$dbh->do("insert into BinaryCaches(url) values (?)",
{}, $binaryCacheUrl);
$cacheId = $dbh->last_insert_id("", "", "", "");
}
$cacheIds{$binaryCacheUrl} = $cacheId;
return $cacheId;
}
sub cachedGetInfoFrom {
my ($storePath, $pathHash, $binaryCacheUrl) = @_;
$queryNAR->execute(getCacheId($binaryCacheUrl), basename($storePath));
my $res = $queryNAR->fetchrow_hashref();
return undef unless defined $res;
return
{ url => $res->{url}
, compression => $res->{compression}
, fileHash => $res->{fileHash}
, fileSize => $res->{fileSize}
, narHash => $res->{narHash}
, narSize => $res->{narSize}
, refs => [ split " ", $res->{refs} ]
, deriver => $res->{deriver}
} if defined $res;
}
download-from-binary-cache: parallelise fetching of NAR info files Getting substitute information using the binary cache substituter has non-trivial latency overhead. A package or NixOS system configuration can have hundreds of dependencies, and in the worst case (when the local info cache is empty) we have to do a separate HTTP request for each of these. If the ping time to the server is t, getting N info files will take tN seconds; e.g., with a ping time of 0.1s to nixos.org, sequentially downloading 1000 info files (a typical NixOS config) will take at least 100 seconds. To fix this problem, the binary cache substituter can now perform requests in parallel. This required changing the substituter interface to support a function querySubstitutablePathInfos() that queries multiple paths at the same time, and rewriting queryMissing() to take advantage of parallelism. (Due to local caching, parallelising queryMissing() is sufficient for most use cases, since it's almost always called before building a derivation and thus fills the local info cache.) For example, parallelism speeds up querying all 1056 paths in a particular NixOS system configuration from 116s to 2.6s. It works so well because the eccentricity of the top-level derivation in the dependency graph is only 9. So we only need 10 round-trips (when using an unlimited number of parallel connections) to get everything. Currently we do a maximum of 150 parallel connections to the server. Thus it's important that the binary cache server (e.g. nixos.org) has a high connection limit. Alternatively we could use HTTP pipelining, but WWW::Curl doesn't support it and libcurl has a hard-coded limit of 5 requests per pipeline.
2012-07-07 01:08:20 +02:00
sub printInfo {
my ($storePath, $info) = @_;
print "$storePath\n";
print $info->{deriver} ? "$Nix::Config::storeDir/$info->{deriver}" : "", "\n";
print scalar @{$info->{refs}}, "\n";
print "$Nix::Config::storeDir/$_\n" foreach @{$info->{refs}};
print $info->{fileSize} || 0, "\n";
print $info->{narSize} || 0, "\n";
}
download-from-binary-cache: parallelise fetching of NAR info files 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.
2012-07-07 01:08:20 +02:00
sub printInfoParallel {
my @paths = @_;
# First print all paths for which we have cached info.
my @left;
foreach my $storePath (@paths) {
my $pathHash = substr(basename($storePath), 0, 32);
my $found = 0;
foreach my $binaryCacheUrl (@binaryCacheUrls) {
my $info = cachedGetInfoFrom($storePath, $pathHash, $binaryCacheUrl);
if (defined $info) {
printInfo($storePath, $info);
$found = 1;
last;
}
}
push @left, $storePath if !$found;
}
download-from-binary-cache: parallelise fetching of NAR info files 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.
2012-07-07 01:08:20 +02:00
return if scalar @left == 0;
foreach my $binaryCacheUrl (@binaryCacheUrls) {
download-from-binary-cache: parallelise fetching of NAR info files 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.
2012-07-07 01:08:20 +02:00
my @left2;
%requests = ();
foreach my $storePath (@left) {
my $pathHash = substr(basename($storePath), 0, 32);
if (negativeHit($storePath, $binaryCacheUrl)) {
push @left2, $storePath;
next;
}
my $infoUrl = "$binaryCacheUrl/$pathHash.narinfo";
addRequest($storePath, $infoUrl);
}
processRequests;
foreach my $request (values %requests) {
my $info = processNARInfo($request->{storePath}, $binaryCacheUrl, $request);
if (defined $info) {
printInfo($request->{storePath}, $info);
} else {
push @left2, $request->{storePath};
}
}
@left = @left2;
}
}
sub downloadBinary {
my ($storePath) = @_;
my $pathHash = substr(basename($storePath), 0, 32);
cache: foreach my $binaryCacheUrl (@binaryCacheUrls) {
my $info = cachedGetInfoFrom($storePath, $pathHash, $binaryCacheUrl);
download-from-binary-cache: parallelise fetching of NAR info files 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.
2012-07-07 01:08:20 +02:00
unless (defined $info) {
next if negativeHit($storePath, $binaryCacheUrl);
my $request = addRequest($storePath, "$binaryCacheUrl/$pathHash.narinfo");
processRequests;
$info = processNARInfo($storePath, $binaryCacheUrl, $request);
}
next unless defined $info;
my $decompressor;
if ($info->{compression} eq "bzip2") { $decompressor = "$Nix::Config::bzip2 -d"; }
elsif ($info->{compression} eq "xz") { $decompressor = "$Nix::Config::xz -d"; }
else {
print STDERR "unknown compression method $info->{compression}\n";
next;
}
print STDERR "\n*** Downloading $info->{url} into $storePath...\n";
if (system("$Nix::Config::curl --fail --location $binaryCacheUrl/$info->{url} | $decompressor | $Nix::Config::binDir/nix-store --restore $storePath") != 0) {
die "download of `$info->{url}' failed" . ($! ? ": $!" : "") . "\n" unless $? == 0;
next;
}
download-from-binary-cache: parallelise fetching of NAR info files 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.
2012-07-07 01:08:20 +02:00
# The hash in the manifest can be either in base-16 or
# base-32. Handle both.
$info->{narHash} =~ /^sha256:(.*)$/ or die "invalid hash";
my $hash = $1;
my $hash2 = hashPath("sha256", 1, $storePath);
die "hash mismatch in downloaded path $storePath; expected $hash, got $hash2\n"
if $hash ne $hash2;
print STDERR "\n";
return 1;
}
return 0;
}
initCache();
if ($ARGV[0] eq "--query") {
while (<STDIN>) {
download-from-binary-cache: parallelise fetching of NAR info files 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.
2012-07-07 01:08:20 +02:00
chomp;
my ($cmd, @args) = split " ", $_;
if ($cmd eq "have") {
my $storePath = <STDIN>; chomp $storePath;
# FIXME: want to give correct info here, but it's too slow.
download-from-binary-cache: parallelise fetching of NAR info files 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.
2012-07-07 01:08:20 +02:00
print "0\n";
#my $info = getInfo($storePath);
#if (defined $info) { print "1\n"; } else { print "0\n"; }
}
elsif ($cmd eq "info") {
download-from-binary-cache: parallelise fetching of NAR info files 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.
2012-07-07 01:08:20 +02:00
printInfoParallel(@args);
print "\n";
}
else { die "unknown command `$cmd'"; }
flush STDOUT;
}
}
elsif ($ARGV[0] eq "--substitute") {
my $storePath = $ARGV[1] or die;
if (!downloadBinary($storePath)) {
print STDERR "could not download $storePath from any binary cache\n";
}
}
else {
die;
}