git: async batch interface

This is a transitionary interface which does NOT require an
event loop.  It can be plugged into in current synchronous code
without major surgery.

It allows HTTP/1.1 pipelining-like functionality by taking
advantage of predictable and well-specified POSIX pipe semantics
by stuffing multiple git cat-file requests into the --batch pipe

With xt/git_async_cmp.t and GIANT_GIT_DIR=git.git, the async
interface is 10-25% faster than the synchronous interface since
it can keep the "git cat-file" process busier.

This is expected to improve performance on systems with slower
storage (but multiple cores).

diff --git a/MANIFEST b/MANIFEST
index 098e656d39502f7f326336bb0ed46ba4005d3d48..ee7dd3e83b706e13c39d44e50c5c876e2933748d 100644 (file)
--- a/MANIFEST
+++ b/MANIFEST
@@ -290,6 +290,7 @@ t/watch_maildir_v2.t
 t/www_listing.t
 t/xcpdb-reshard.t
 xt/git-http-backend.t
+xt/git_async_cmp.t
 xt/nntpd-validate.t
 xt/perf-msgview.t
 xt/perf-nntpd.t

diff --git a/lib/PublicInbox/Git.pm b/lib/PublicInbox/Git.pm
index 218846f3d1d25d42699a4a4be55351394c6a5341..9059584086ca57ecfb3411cb2cae12b87cc9496b 100644 (file)
--- a/lib/PublicInbox/Git.pm
+++ b/lib/PublicInbox/Git.pm
@@ -16,6 +16,11 @@ use PublicInbox::Tmpfile;
 use base qw(Exporter);
 our @EXPORT_OK = qw(git_unquote git_quote);
 
+use constant MAX_INFLIGHT =>
+       ($^O eq 'linux' ? 4096 : POSIX::_POSIX_PIPE_BUF())
+       /
+       65; # SHA-256 hex size + "\n" in preparation for git using non-SHA1
+
 my %GIT_ESC = (
        a => "\a",
        b => "\b",
@@ -124,10 +129,50 @@ sub _bidi_pipe {
        $self->{$in} = $in_r;
 }
 
+sub read_cat_in_full ($$$) {
+       my ($self, $in, $left) = @_;
+       my $offset = 0;
+       my $buf = '';
+       while ($left > 0) {
+               my $r = read($in, $buf, $left, $offset);
+               defined($r) or fail($self, "read failed: $!");
+               $r == 0 and fail($self, 'exited unexpectedly');
+               $left -= $r;
+               $offset += $r;
+       }
+       my $r = read($in, my $lf, 1);
+       defined($r) or fail($self, "read failed: $!");
+       fail($self, 'newline missing after blob') if ($r != 1 || $lf ne "\n");
+       \$buf;
+}
+
+sub _cat_async_step ($$$) {
+       my ($self, $inflight, $in) = @_;
+       my $cb = shift @$inflight or die 'BUG: inflight empty';
+       local $/ = "\n";
+       my $head = $in->getline;
+       return eval { $cb->(undef) } if $head =~ / missing$/;
+
+       $head =~ /^([0-9a-f]{40}) (\S+) ([0-9]+)$/ or
+               fail($self, "Unexpected result from async git cat-file: $head");
+       my ($oid_hex, $type, $size) = ($1, $2, $3 + 0);
+       my $bref = read_cat_in_full($self, $in, $size);
+       eval { $cb->($bref, $oid_hex, $type, $size) };
+}
+
+sub cat_async_wait ($) {
+       my ($self) = @_;
+       my $inflight = delete $self->{inflight} or return;
+       my $in = $self->{in};
+       while (scalar(@$inflight)) {
+               _cat_async_step($self, $inflight, $in);
+       }
+}
+
 sub cat_file {
        my ($self, $obj, $ref) = @_;
        my ($retried, $in, $head);
-
+       cat_async_wait($self);
 again:
        batch_prepare($self);
        $self->{out}->print($obj, "\n") or fail($self, "write error: $!");
@@ -147,26 +192,8 @@ again:
                fail($self, "Unexpected result from git cat-file: $head");
 
        my $size = $1;
-       my $rv;
-       my $left = $size;
        $$ref = $size if $ref;
-
-       my $offset = 0;
-       my $buf = '';
-       while ($left > 0) {
-               my $r = read($in, $buf, $left, $offset);
-               defined($r) or fail($self, "read failed: $!");
-               $r == 0 and fail($self, 'exited unexpectedly');
-               $left -= $r;
-               $offset += $r;
-       }
-       $rv = \$buf;
-
-       my $r = read($in, my $lf, 1);
-       defined($r) or fail($self, "read failed: $!");
-       fail($self, 'newline missing after blob') if ($r != 1 || $lf ne "\n");
-
-       $rv;
+       read_cat_in_full($self, $in, $size);
 }
 
 sub batch_prepare ($) { _bidi_pipe($_[0], qw(--batch in out pid)) }
@@ -206,9 +233,15 @@ sub _destroy {
        waitpid($p, 0) if $@; # wait synchronously if not in event loop
 }
 
+sub cat_async_abort ($) {
+       my ($self) = @_;
+       my $inflight = delete $self->{inflight} or die 'BUG: not in async';
+       cleanup($self);
+}
+
 sub fail {
        my ($self, $msg) = @_;
-       cleanup($self);
+       $self->{inflight} ? cat_async_abort($self) : cleanup($self);
        die $msg;
 }
 
@@ -278,6 +311,25 @@ sub pub_urls {
        local_nick($self);
 }
 
+sub cat_async_begin {
+       my ($self) = @_;
+       cleanup($self) if alternates_changed($self);
+       batch_prepare($self);
+       die 'BUG: already in async' if $self->{inflight};
+       $self->{inflight} = [];
+}
+
+sub cat_async ($$$) {
+       my ($self, $oid, $cb) = @_;
+       my $inflight = $self->{inflight} or die 'BUG: not in async';
+       if (scalar(@$inflight) >= MAX_INFLIGHT) {
+               _cat_async_step($self, $inflight, $self->{in});
+       }
+
+       $self->{out}->print($oid, "\n") or fail($self, "write error: $!");
+       push @$inflight, $cb;
+}
+
 sub commit_title ($$) {
        my ($self, $oid) = @_; # PublicInbox::Git, $sha1hex
        my $buf = cat_file($self, $oid) or return;

diff --git a/xt/git_async_cmp.t b/xt/git_async_cmp.t
new file mode 100644 (file)
index 0000000..f8ffe3d
--- /dev/null
+++ b/xt/git_async_cmp.t
@@ -0,0 +1,61 @@
+#!perl -w
+# Copyright (C) 2019 all contributors <meta@public-inbox.org>
+# License: AGPL-3.0+ <https://www.gnu.org/licenses/agpl-3.0.txt>
+use strict;
+use Test::More;
+use Benchmark qw(:all);
+use Digest::SHA;
+use PublicInbox::TestCommon;
+my $git_dir = $ENV{GIANT_GIT_DIR};
+plan 'skip_all' => "GIANT_GIT_DIR not defined for $0" unless defined($git_dir);
+use_ok 'PublicInbox::Git';
+my $git = PublicInbox::Git->new($git_dir);
+my @cat = qw(cat-file --buffer --batch-check --batch-all-objects);
+if (require_git(2.19, 1)) {
+       push @cat, '--unordered';
+} else {
+       warn "git <2.19, cat-file lacks --unordered, locality suffers\n";
+}
+my @dig;
+my $nr = $ENV{NR} || 1;
+my $async = timeit($nr, sub {
+       my $dig = Digest::SHA->new(1);
+       my $cb = sub {
+               my ($bref) = @_;
+               $dig->add($$bref);
+       };
+       my $cat = $git->popen(@cat);
+       $git->cat_async_begin;
+
+       foreach (<$cat>) {
+               my ($oid, undef, undef) = split(/ /);
+               $git->cat_async($oid, $cb);
+       }
+       close $cat or die "cat: $?";
+       $git->cat_async_wait;
+       push @dig, ['async', $dig->hexdigest ];
+});
+
+my $sync = timeit($nr, sub {
+       my $dig = Digest::SHA->new(1);
+       my $cat = $git->popen(@cat);
+       foreach (<$cat>) {
+               my ($oid, undef, undef) = split(/ /);
+               my $bref = $git->cat_file($oid);
+               $dig->add($$bref);
+       }
+       close $cat or die "cat: $?";
+       push @dig, ['sync', $dig->hexdigest ];
+});
+
+ok(scalar(@dig) >= 2, 'got some digests');
+my $ref = shift @dig;
+my $exp = $ref->[1];
+isnt($exp, Digest::SHA->new(1)->hexdigest, 'not empty');
+foreach (@dig) {
+       is($_->[1], $exp, "digest matches $_->[0] <=> $ref->[0]");
+}
+diag "sync=".timestr($sync);
+diag "async=".timestr($async);
+done_testing;
+1;