-# Copyright (C) 2016-2018 all contributors <meta@public-inbox.org>
+# Copyright (C) 2016-2019 all contributors <meta@public-inbox.org>
# License: AGPL-3.0+ <https://www.gnu.org/licenses/agpl-3.0.txt>
-# Limits the number of processes spawned
-# This does not depend on Danga::Socket or any other external
-# scheduling mechanism, you just need to call start and finish
-# appropriately
+# Like most Perl modules in public-inbox, this is internal and
+# NOT subject to any stability guarantees! It is only documented
+# for other hackers.
+#
+# This is used to limit the number of processes spawned by the
+# PSGI server, so it acts like a semaphore and queues up extra
+# commands to be run if currently at the limit. Multiple "limiters"
+# may be configured which give inboxes different channels to
+# operate in. This can be useful to ensure smaller inboxes can
+# be cloned while cloning of large inboxes is maxed out.
+#
+# This does not depend on PublicInbox::DS or any other external
+# scheduling mechanism, you just need to call start() and finish()
+# appropriately. However, public-inbox-httpd (which uses PublicInbox::DS)
+# will be able to schedule this based on readability of stdout from
+# the spawned process. See GitHTTPBackend.pm and SolverGit.pm for
+# usage examples. It does not depend on any form of threading.
+#
+# This is useful for scheduling CGI execution of both long-lived
+# git-http-backend(1) process (for "git clone") as well as short-lived
+# processes such as git-apply(1).
+
package PublicInbox::Qspawn;
use strict;
use warnings;
use PublicInbox::Spawn qw(popen_rd);
require Plack::Util;
+
my $def_limiter;
+# declares a command to spawn (but does not spawn it).
+# $cmd is the command to spawn
+# $env is the environ for the child process
+# $opt can include redirects and perhaps other process spawning options
sub new ($$$;) {
my ($class, $cmd, $env, $opt) = @_;
bless { args => [ $cmd, $env, $opt ] }, $class;
sub _do_spawn {
my ($self, $cb) = @_;
my $err;
+ my ($cmd, $env, $opts) = @{$self->{args}};
+ my %opts = %{$opts || {}};
+ my $limiter = $self->{limiter};
+ foreach my $k (PublicInbox::Spawn::RLIMITS()) {
+ if (defined(my $rlimit = $limiter->{$k})) {
+ $opts{$k} = $rlimit;
+ }
+ }
- ($self->{rpipe}, $self->{pid}) = popen_rd(@{$self->{args}});
+ ($self->{rpipe}, $self->{pid}) = popen_rd($cmd, $env, \%opts);
if (defined $self->{pid}) {
- $self->{limiter}->{running}++;
+ $limiter->{running}++;
} else {
$self->{err} = $!;
}
$cb->($self->{rpipe});
}
+sub child_err ($) {
+ my ($child_error) = @_; # typically $?
+ my $exitstatus = ($child_error >> 8) or return;
+ my $sig = $child_error & 127;
+ my $msg = "exit status=$exitstatus";
+ $msg .= " signal=$sig" if $sig;
+ $msg;
+}
+
sub finish ($) {
my ($self) = @_;
my $limiter = $self->{limiter};
my $running;
if (delete $self->{rpipe}) {
my $pid = delete $self->{pid};
- $self->{err} = $pid == waitpid($pid, 0) ? $? :
+ $self->{err} = $pid == waitpid($pid, 0) ? child_err($?) :
"PID:$pid still running?";
$running = --$limiter->{running};
}
}
}
+# Similar to `backtick` or "qx" ("perldoc -f qx"), it calls $qx_cb with
+# the stdout of the given command when done; but respects the given limiter
+# $env is the PSGI env. As with ``/qx; only use this when output is small
+# and safe to slurp.
sub psgi_qx {
my ($self, $env, $limiter, $qx_cb) = @_;
my $qx = PublicInbox::Qspawn::Qx->new;
});
}
+# Used for streaming the stdout of one process as a PSGI response.
+#
+# $env is the PSGI env.
+# optional keys in $env:
+# $env->{'qspawn.wcb'} - the write callback from the PSGI server
+# optional, use this if you've already
+# captured it elsewhere. If not given,
+# psgi_return will return an anonymous
+# sub for the PSGI server to call
+#
+# $env->{'qspawn.filter'} - filter callback, receives a string as input,
+# undef on EOF
+#
+# $limiter - the Limiter object to use (uses the def_limiter if not given)
+#
+# $parse_hdr - Initial read function; often for parsing CGI header output.
+# It will be given the return value of sysread from the pipe
+# and a string ref of the current buffer. Returns an arrayref
+# for PSGI responses. 2-element arrays in PSGI mean the
+# body will be streamed, later, via writes (push-based) to
+# psgix.io. 3-element arrays means the body is available
+# immediately (or streamed via ->getline (pull-based)).
sub psgi_return {
my ($self, $env, $limiter, $parse_hdr) = @_;
my ($fh, $rpipe);
return if !defined($r) && ($!{EINTR} || $!{EAGAIN});
$parse_hdr->($r, \$buf);
};
- my $res = delete $env->{'qspawn.response'};
+
+ my $wcb = delete $env->{'qspawn.wcb'};
my $async = $env->{'pi-httpd.async'};
+
my $cb = sub {
my $r = $rd_hdr->() or return;
$rd_hdr = undef;
$rpipe->close;
$end->();
}
- $res->($r);
+ $wcb->($r);
} elsif ($async) {
- $fh = $res->($r); # scalar @$r == 2
+ $fh = $wcb->($r); # scalar @$r == 2
$fh = filter_fh($fh, $filter) if $filter;
$async->async_pass($env->{'psgix.io'}, $fh, \$buf);
} else { # for synchronous PSGI servers
require PublicInbox::GetlineBody;
$r->[2] = PublicInbox::GetlineBody->new($rpipe, $end,
$buf, $filter);
- $res->($r);
+ $wcb->($r);
}
};
$limiter ||= $def_limiter ||= PublicInbox::Qspawn::Limiter->new(32);
}
};
- return $self->start($limiter, $start_cb) if $res;
+ # the caller already captured the PSGI write callback from
+ # the PSGI server, so we can call ->start, here:
+ return $self->start($limiter, $start_cb) if $wcb;
+ # the caller will return this sub to the PSGI server, so
+ # it can set the response callback (that is, for PublicInbox::HTTP,
+ # the chunked_wcb or identity_wcb callback), but other HTTP servers
+ # are supported:
sub {
- ($res) = @_;
+ ($wcb) = @_;
$self->start($limiter, $start_cb);
};
}
max => $max || 32,
running => 0,
run_queue => [],
+ # RLIMIT_CPU => undef,
+ # RLIMIT_DATA => undef,
+ # RLIMIT_CORE => undef,
}, $class;
}
+sub setup_rlimit {
+ my ($self, $name, $config) = @_;
+ foreach my $rlim (PublicInbox::Spawn::RLIMITS()) {
+ my $k = lc($rlim);
+ $k =~ tr/_//d;
+ $k = "publicinboxlimiter.$name.$k";
+ defined(my $v = $config->{$k}) or next;
+ my @rlimit = split(/\s*,\s*/, $v);
+ if (scalar(@rlimit) == 1) {
+ push @rlimit, $rlimit[0];
+ } elsif (scalar(@rlimit) != 2) {
+ warn "could not parse $k: $v\n";
+ }
+ eval { require BSD::Resource };
+ if ($@) {
+ warn "BSD::Resource missing for $rlim";
+ next;
+ }
+ foreach my $i (0..$#rlimit) {
+ next if $rlimit[$i] ne 'INFINITY';
+ $rlimit[$i] = BSD::Resource::RLIM_INFINITY();
+ }
+ $self->{$rlim} = \@rlimit;
+ }
+}
+
# captures everything into a buffer and executes a callback when done
package PublicInbox::Qspawn::Qx;
use strict;