[PATCH 0/2] www: per-inbox manifest.js.gz improvements

[PATCH 0/2] www: per-inbox manifest.js.gz improvements

[Eric Wong]

Speed things up for people wanting to clone a single inbox and
significantly reduce server overheads.  [1/2] should be
noticeable to -httpd admins if users are cloning single inboxes.

Eric Wong (2):
  manifest.js.gz: avoid long-lived per-epoch cat-file processes
  www: use ->ALL for per-inbox manifest.js.gz, too

 lib/PublicInbox/Git.pm          | 25 +++++--------------------
 lib/PublicInbox/ManifestJsGz.pm | 16 ++++++++--------
 t/extindex-psgi.t               | 20 ++++++++++++++++++--
 3 files changed, 31 insertions(+), 30 deletions(-)

[PATCH 1/2] manifest.js.gz: avoid long-lived per-epoch cat-file processes

[Eric Wong]

When generating per-inbox manifests, we were forgetting to
cleanup per-epoch "git cat-file --batch" processes.  Our
previous method of generating modified times was also stupidly
inefficient, so replace the pipeline with a single
"git for-each-ref" invocation.
---
 lib/PublicInbox/Git.pm          | 25 +++++--------------------
 lib/PublicInbox/ManifestJsGz.pm |  2 +-
 2 files changed, 6 insertions(+), 21 deletions(-)

diff --git a/lib/PublicInbox/Git.pm b/lib/PublicInbox/Git.pm
index e557f6d6..219a1732 100644
--- a/lib/PublicInbox/Git.pm
+++ b/lib/PublicInbox/Git.pm
@@ -16,7 +16,7 @@ use Errno qw(EINTR EAGAIN);
 use File::Glob qw(bsd_glob GLOB_NOSORT);
 use File::Spec ();
 use Time::HiRes qw(stat);
-use PublicInbox::Spawn qw(popen_rd);
+use PublicInbox::Spawn qw(popen_rd spawn);
 use PublicInbox::Tmpfile;
 use IO::Poll qw(POLLIN);
 use Carp qw(croak);
@@ -466,28 +466,13 @@ sub cat_async ($$$;$) {
 	push(@$inflight, $oid, $cb, $arg);
 }
 
-sub extract_cmt_time {
-	my ($bref, undef, undef, undef, $modified) = @_;
-
-	if ($$bref =~ /^committer .*?> ([0-9]+) [\+\-]?[0-9]+/sm) {
-		my $cmt_time = $1 + 0;
-		$$modified = $cmt_time if $cmt_time > $$modified;
-	}
-}
-
 # returns the modified time of a git repo, same as the "modified" field
 # of a grokmirror manifest
 sub modified ($) {
-	my ($self) = @_;
-	my $modified = 0;
-	my $fh = popen($self, qw(rev-parse --branches));
-	local $/ = "\n";
-	while (my $oid = <$fh>) {
-		chomp $oid;
-		cat_async($self, $oid, \&extract_cmt_time, \$modified);
-	}
-	cat_async_wait($self);
-	$modified || time;
+	# committerdate:unix is git 2.9.4+ (2017-05-05), so using raw instead
+	my $fh = popen($_[0], qw[for-each-ref --sort=-committerdate
+				--format=%(committerdate:raw) --count=1]);
+	(split(/ /, <$fh> // time))[0] + 0; # integerize for JSON
 }
 
 # for grokmirror, which doesn't read gitweb.description
diff --git a/lib/PublicInbox/ManifestJsGz.pm b/lib/PublicInbox/ManifestJsGz.pm
index 69d81fa1..cde60245 100644
--- a/lib/PublicInbox/ManifestJsGz.pm
+++ b/lib/PublicInbox/ManifestJsGz.pm
@@ -27,7 +27,7 @@ sub inject_entry ($$$;$) {
 	$ctx->{manifest}->{$url_path} = $ent;
 }
 
-sub manifest_add ($$;$$) { # slow path w/o extindex "all"
+sub manifest_add ($$;$$) { # slow path w/o extindex "all" (or per-inbox)
 	my ($ctx, $ibx, $epoch, $default_desc) = @_;
 	my $url_path = "/$ibx->{name}";
 	my $git;

[PATCH 2/2] www: use ->ALL for per-inbox manifest.js.gz, too

[Eric Wong]

With 11 epochs on LKML, it's the 11 current epoch time goes from
around 60ms to around 10ms, so it's a significant improvement.
And improve test coverage while we're at it.
---
 lib/PublicInbox/ManifestJsGz.pm | 14 +++++++-------
 t/extindex-psgi.t               | 20 ++++++++++++++++++--
 2 files changed, 25 insertions(+), 9 deletions(-)

diff --git a/lib/PublicInbox/ManifestJsGz.pm b/lib/PublicInbox/ManifestJsGz.pm
index cde60245..d5048a96 100644
--- a/lib/PublicInbox/ManifestJsGz.pm
+++ b/lib/PublicInbox/ManifestJsGz.pm
@@ -53,6 +53,7 @@ sub slow_manifest_add ($$) {
 			manifest_add($ctx, $ibx);
 		}
 	};
+	warn "E: $@" if $@;
 }
 
 sub eidx_manifest_add ($$$) {
@@ -65,6 +66,10 @@ sub eidx_manifest_add ($$$) {
 		}
 	} else {
 		warn "E: `${\$ibx->eidx_key}' not indexed by $ALL->{topdir}\n";
+		# do not use slow path for global manifest since
+		# it can become catastrophically slow.  per-inbox manifest
+		# is not too bad with dozens of epochs, so never fail that:
+		slow_manifest_add($ctx, $ibx) if $ibx == $ctx->{ibx};
 	}
 }
 
@@ -85,12 +90,8 @@ sub response {
 sub ibx_entry {
 	my ($ctx, $ibx) = @_;
 	my $ALL = $ctx->{www}->{pi_cfg}->ALL;
-	if ($ALL) { # FIXME: test this in t/
-		eidx_manifest_add($ctx, $ALL, $ibx);
-	} else {
+	$ALL ? eidx_manifest_add($ctx, $ALL, $ibx) :
 		slow_manifest_add($ctx, $ibx);
-		warn "E: $@" if $@;
-	}
 }
 
 sub hide_key { 'manifest' } # for WwwListing->list_match_i
@@ -112,8 +113,7 @@ sub psgi_triple {
 
 sub per_inbox {
 	my ($ctx) = @_;
-	# only one inbox, slow is probably OK
-	slow_manifest_add($ctx, $ctx->{ibx});
+	ibx_entry($ctx, $ctx->{ibx});
 	psgi_triple($ctx);
 }
 
diff --git a/t/extindex-psgi.t b/t/extindex-psgi.t
index 4e26962e..98dc2e48 100644
--- a/t/extindex-psgi.t
+++ b/t/extindex-psgi.t
@@ -11,6 +11,7 @@ require_git(2.6);
 require_mods(qw(json DBD::SQLite Search::Xapian
 		HTTP::Request::Common Plack::Test URI::Escape Plack::Builder));
 use_ok($_) for (qw(HTTP::Request::Common Plack::Test));
+use IO::Uncompress::Gunzip qw(gunzip);
 require PublicInbox::WWW;
 my ($ro_home, $cfg_path) = setup_public_inboxes;
 my ($tmpdir, $for_destroy) = tmpdir;
@@ -18,11 +19,11 @@ my $home = "$tmpdir/home";
 mkdir $home or BAIL_OUT $!;
 mkdir "$home/.public-inbox" or BAIL_OUT $!;
 my $pi_config = "$home/.public-inbox/config";
-cp("$ro_home/.public-inbox/config", $pi_config) or BAIL_OUT;
+cp($cfg_path, $pi_config) or BAIL_OUT;
 my $env = { HOME => $home };
 run_script([qw(-extindex --all), "$tmpdir/eidx"], $env) or BAIL_OUT;
 {
-	open my $cfgfh, '>', $pi_config or BAIL_OUT;
+	open my $cfgfh, '>>', $pi_config or BAIL_OUT;
 	$cfgfh->autoflush(1);
 	print $cfgfh <<EOM or BAIL_OUT;
 [extindex "all"]
@@ -30,6 +31,7 @@ run_script([qw(-extindex --all), "$tmpdir/eidx"], $env) or BAIL_OUT;
 	url = http://bogus.example.com/all
 [publicinbox]
 	wwwlisting = all
+	grokManifest = all
 EOM
 }
 my $www = PublicInbox::WWW->new(PublicInbox::Config->new($pi_config));
@@ -67,6 +69,20 @@ my $client = sub {
 	is($res->code, 404, 'no inboxes matched');
 	unlike($res->content, qr!no inboxes, yet!,
 		'we have inboxes, just no matches');
+
+	my $m = {};
+	for my $pfx (qw(/t1 /t2), '') {
+		$res = $cb->(GET($pfx.'/manifest.js.gz'));
+		gunzip(\($res->content) => \(my $js));
+		$m->{$pfx} = json_utf8->decode($js);
+	}
+	is_deeply([sort keys %{$m->{''}}],
+		[ sort(keys %{$m->{'/t1'}}, keys %{$m->{'/t2'}}) ],
+		't1 + t2 = all');
+	is_deeply([ sort keys %{$m->{'/t2'}} ], [ '/t2/git/0.git' ],
+		't2 manifest');
+	is_deeply([ sort keys %{$m->{'/t1'}} ], [ '/t1' ],
+		't2 manifest');
 };
 test_psgi(sub { $www->call(@_) }, $client);
 %$env = (%$env, TMPDIR => $tmpdir, PI_CONFIG => $pi_config);