user/dev discussion of public-inbox itself
 help / color / mirror / code / Atom feed
Search results ordered by [date|relevance]  view[summary|nested|Atom feed]
thread overview below | download mbox.gz: |
* [PATCH 2/2] githttpbackend: clamp to one smart HTTP request at-a-time
  2016-04-28  1:56  5% [PATCH 0/2] githttpbackend: dumb HTTP fallbacks Eric Wong
@ 2016-04-28  1:56  7% ` Eric Wong
  0 siblings, 0 replies; 2+ results
From: Eric Wong @ 2016-04-28  1:56 UTC (permalink / raw)
  To: meta

Server admins may not be able to afford to have too many
git-pack-objects processes running at once.  Since PSGI
HTTP servers should already be configured to use multiple
processes for other requests; limit concurrency of smart
backends to one; and fall back to dumb responses if we're
already generating a pack.
---
 lib/PublicInbox/GitHTTPBackend.pm | 12 ++++++++++++
 t/httpd.t                         |  2 +-
 2 files changed, 13 insertions(+), 1 deletion(-)

diff --git a/lib/PublicInbox/GitHTTPBackend.pm b/lib/PublicInbox/GitHTTPBackend.pm
index c44c67d..a7cac10 100644
--- a/lib/PublicInbox/GitHTTPBackend.pm
+++ b/lib/PublicInbox/GitHTTPBackend.pm
@@ -10,6 +10,14 @@ use Fcntl qw(:seek);
 use IO::File;
 use PublicInbox::Spawn qw(spawn);
 
+# TODO: make configurable, but keep in mind it's better to have
+# multiple -httpd worker processes which are already scaled to
+# the proper number of CPUs and memory.  git-pack-objects(1) may
+# also use threads and bust memory limits, too, so I recommend
+# limiting threads to 1 (via `pack.threads` knob in git) for serving.
+my $LIMIT = 1;
+my $nr_running = 0;
+
 # n.b. serving "description" and "cloneurl" should be innocuous enough to
 # not cause problems.  serving "config" might...
 my @text = qw[HEAD info/refs
@@ -31,6 +39,8 @@ sub r {
 
 sub serve {
 	my ($cgi, $git, $path) = @_;
+	return serve_dumb($cgi, $git, $path) if $nr_running >= $LIMIT;
+
 	my $service = $cgi->param('service') || '';
 	if ($service =~ /\Agit-\w+-pack\z/ || $path =~ /\Agit-\w+-pack\z/) {
 		my $ok = serve_smart($cgi, $git, $path);
@@ -174,6 +184,7 @@ sub serve_smart {
 	$wpipe = $in = undef;
 	$buf = '';
 	my ($vin, $fh, $res);
+	$nr_running++;
 	my $end = sub {
 		if ($fh) {
 			$fh->close;
@@ -182,6 +193,7 @@ sub serve_smart {
 		if ($rpipe) {
 			$rpipe->close; # _may_ be Danga::Socket::close
 			$rpipe = undef;
+			$nr_running--;
 		}
 		if (defined $pid && $pid != waitpid($pid, 0)) {
 			$err->print("git http-backend ($git_dir): $?\n");
diff --git a/t/httpd.t b/t/httpd.t
index 0379031..781fe03 100644
--- a/t/httpd.t
+++ b/t/httpd.t
@@ -104,7 +104,7 @@ EOF
 
 	is(system(qw(git clone -q --mirror),
 			"http://$host:$port/$group", "$tmpdir/clone.git"),
-		0, 'clone successful');
+		0, 'smart clone successful');
 
 	# ensure dumb cloning works, too:
 	is(system('git', "--git-dir=$maindir",
-- 
EW


^ permalink raw reply related	[relevance 7%]

* [PATCH 0/2] githttpbackend: dumb HTTP fallbacks
@ 2016-04-28  1:56  5% Eric Wong
  2016-04-28  1:56  7% ` [PATCH 2/2] githttpbackend: clamp to one smart HTTP request at-a-time Eric Wong
  0 siblings, 1 reply; 2+ results
From: Eric Wong @ 2016-04-28  1:56 UTC (permalink / raw)
  To: meta

Smart HTTP cloning is nice for bandwidth utilization, but it
can eat up memory and CPU on a busy server; so fall back to
dumb HTTP if there is already a running backend or if an admin
decides to set `http.uploadpack' to `false' entirely.

Not that our large static file serving is great when it comes
to slow clients, at the moment; but of course I shall
improve on that...

Eric Wong (2):
      githttpbackend: fall back to dumb if smart HTTP is off
      githttpbackend: clamp to one smart HTTP request at-a-time

 lib/PublicInbox/GitHTTPBackend.pm | 52 ++++++++++++++++++++++++++++-----------
 t/httpd.t                         |  9 +++++++
 2 files changed, 46 insertions(+), 15 deletions(-)


^ permalink raw reply	[relevance 5%]

Results 1-2 of 2 | reverse | options above
-- pct% links below jump to the message on this page, permalinks otherwise --
2016-04-28  1:56  5% [PATCH 0/2] githttpbackend: dumb HTTP fallbacks Eric Wong
2016-04-28  1:56  7% ` [PATCH 2/2] githttpbackend: clamp to one smart HTTP request at-a-time Eric Wong

Code repositories for project(s) associated with this public inbox

	https://80x24.org/public-inbox.git

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).