1 # Copyright (C) 2020-2021 all contributors <meta@public-inbox.org>
2 # License: AGPL-3.0+ <https://www.gnu.org/licenses/agpl-3.0.txt>
4 # base class for remote IPC calls and workqueues, requires Storable or Sereal
5 # - ipc_do and ipc_worker_* is for a single worker/producer and uses pipes
6 # - wq_do and wq_worker* is for a single producer and multiple workers,
7 # using SOCK_SEQPACKET for work distribution
8 # use ipc_do when you need work done on a certain process
9 # use wq_do when your work can be done on any idle worker
10 package PublicInbox::IPC;
13 use parent qw(Exporter);
14 use Carp qw(confess croak);
15 use PublicInbox::DS qw(dwaitpid);
16 use PublicInbox::Spawn;
17 use PublicInbox::OnDestroy;
18 use PublicInbox::WQWorker;
19 use Socket qw(AF_UNIX MSG_EOR SOCK_STREAM);
20 my $SEQPACKET = eval { Socket::SOCK_SEQPACKET() }; # portable enough?
21 use constant PIPE_BUF => $^O eq 'linux' ? 4096 : POSIX::_POSIX_PIPE_BUF();
22 our @EXPORT_OK = qw(ipc_freeze ipc_thaw);
23 my $WQ_MAX_WORKERS = 4096;
25 # ->imports at BEGIN turns sereal_*_with_object into custom ops on 5.14+
26 # and eliminate method call overhead
29 require Sereal::Encoder;
30 require Sereal::Decoder;
31 Sereal::Encoder->import('sereal_encode_with_object');
32 Sereal::Decoder->import('sereal_decode_with_object');
33 ($enc, $dec) = (Sereal::Encoder->new, Sereal::Decoder->new);
37 if ($enc && $dec) { # should be custom ops
38 *ipc_freeze = sub ($) { sereal_encode_with_object $enc, $_[0] };
39 *ipc_thaw = sub ($) { sereal_decode_with_object $dec, $_[0], my $ret };
41 eval { # some distros have Storable as a separate package from Perl
43 *ipc_freeze = \&Storable::freeze;
44 *ipc_thaw = \&Storable::thaw;
46 } // warn("Storable (part of Perl) missing: $@\n");
49 my $recv_cmd = PublicInbox::Spawn->can('recv_cmd4');
50 my $send_cmd = PublicInbox::Spawn->can('send_cmd4') // do {
51 require PublicInbox::CmdIPC4;
52 $recv_cmd //= PublicInbox::CmdIPC4->can('recv_cmd4');
53 PublicInbox::CmdIPC4->can('send_cmd4');
58 defined(my $len = <$r>) or return;
59 chop($len) eq "\n" or croak "no LF byte in $len";
60 defined(my $n = read($r, my $buf, $len)) or croak "read error: $!";
61 $n == $len or croak "short read: $n != $len";
67 my $buf = ipc_freeze($ref);
68 length($buf) . "\n" . $buf;
73 print $w _pack_rec($ref) or croak "print: $!";
76 sub ipc_return ($$$) {
77 my ($w, $ret, $exc) = @_;
78 _send_rec($w, $exc ? bless(\$exc, 'PublicInbox::IPC::Die') : $ret);
81 sub ipc_worker_loop ($$$) {
82 my ($self, $r_req, $w_res) = @_;
83 my ($rec, $wantarray, $sub, @args);
85 while ($rec = _get_rec($r_req)) {
86 ($wantarray, $sub, @args) = @$rec;
87 # no waiting if client doesn't care,
88 # this is the overwhelmingly likely case
89 if (!defined($wantarray)) {
90 eval { $self->$sub(@args) };
91 warn "$$ die: $@ (from nowait $sub)\n" if $@;
92 } elsif ($wantarray) {
93 my @ret = eval { $self->$sub(@args) };
94 ipc_return($w_res, \@ret, $@);
95 } else { # '' => wantscalar
96 my $ret = eval { $self->$sub(@args) };
97 ipc_return($w_res, \$ret, $@);
102 # starts a worker if Sereal or Storable is installed
103 sub ipc_worker_spawn {
104 my ($self, $ident, $oldset, $fields) = @_;
105 return unless $enc; # no Sereal or Storable
106 return if ($self->{-ipc_ppid} // -1) == $$; # idempotent
107 delete(@$self{qw(-ipc_req -ipc_res -ipc_ppid -ipc_pid)});
108 pipe(my ($r_req, $w_req)) or die "pipe: $!";
109 pipe(my ($r_res, $w_res)) or die "pipe: $!";
110 my $sigset = $oldset // PublicInbox::DS::block_signals();
111 $self->ipc_atfork_prepare;
112 my $seed = rand(0xffffffff);
113 my $pid = fork // die "fork: $!";
116 eval { PublicInbox::DS->Reset };
117 delete @$self{qw(-wq_s1 -wq_s2 -wq_workers -wq_ppid)};
118 $w_req = $r_res = undef;
119 $w_res->autoflush(1);
120 $SIG{$_} = 'IGNORE' for (qw(TERM INT QUIT));
122 PublicInbox::DS::sig_setmask($sigset);
123 # ensure we properly exit even if warn() dies:
124 my $end = PublicInbox::OnDestroy->new($$, sub { exit(!!$@) });
127 local @$self{keys %$fields} = values(%$fields);
128 my $on_destroy = $self->ipc_atfork_child;
130 ipc_worker_loop($self, $r_req, $w_res);
132 die "worker $ident PID:$$ died: $@\n" if $@;
133 undef $end; # trigger exit
135 PublicInbox::DS::sig_setmask($sigset) unless $oldset;
136 $r_req = $w_res = undef;
137 $w_req->autoflush(1);
138 $self->{-ipc_req} = $w_req;
139 $self->{-ipc_res} = $r_res;
140 $self->{-ipc_ppid} = $$;
141 $self->{-ipc_pid} = $pid;
144 sub ipc_worker_reap { # dwaitpid callback
145 my ($args, $pid) = @_;
147 # TERM(15) is our default exit signal, PIPE(13) is likely w/ pager
149 warn "PID:$pid died with \$?=$?\n" if $s != 15 && $s != 13;
153 my ($self, $args) = @_;
154 my $pids = delete $self->{"-wq_old_pids.$$"} or return;
155 dwaitpid($_, \&ipc_worker_reap, [$self, $args]) for @$pids;
158 # for base class, override in sub classes
159 sub ipc_atfork_prepare {}
161 sub wq_atexit_child {}
163 sub ipc_atfork_child {
165 my $io = delete($self->{-ipc_atfork_child_close}) or return;
170 # idempotent, can be called regardless of whether worker is active or not
171 sub ipc_worker_stop {
172 my ($self, $args) = @_;
173 my ($pid, $ppid) = delete(@$self{qw(-ipc_pid -ipc_ppid)});
174 my ($w_req, $r_res) = delete(@$self{qw(-ipc_req -ipc_res)});
175 if (!$w_req && !$r_res) {
176 die "unexpected PID:$pid without IPC pipes" if $pid;
179 die 'no PID with IPC pipes' unless $pid;
180 $w_req = $r_res = undef;
182 return if $$ != $ppid;
183 dwaitpid($pid, \&ipc_worker_reap, [$self, $args]);
186 # use this if we have multiple readers reading curl or "pigz -dc"
187 # and writing to the same store
190 require PublicInbox::Lock;
191 $self->{-ipc_lock} //= bless { lock_path => $f }, 'PublicInbox::Lock'
194 sub ipc_async_wait ($$) {
195 my ($self, $max) = @_; # max == -1 to wait for all
196 my $aif = $self->{-async_inflight} or return;
197 my $r_res = $self->{-ipc_res} or die 'BUG: no ipc_res';
198 while (my ($sub, $bytes, $cb, $cb_arg) = splice(@$aif, 0, 4)) {
199 my $ret = _get_rec($r_res) //
200 die "no response on $sub (req.size=$bytes)";
201 $self->{-async_inflight_bytes} -= $bytes;
203 eval { $cb->($cb_arg, $ret) };
204 warn "E: $sub callback error: $@\n" if $@;
205 return if --$max == 0;
209 # call $self->$sub(@args), on a worker if ipc_worker_spawn was used
211 my ($self, $sub, @args) = @_;
212 if (my $w_req = $self->{-ipc_req}) { # run in worker
213 my $ipc_lock = $self->{-ipc_lock};
214 my $lock = $ipc_lock ? $ipc_lock->lock_for_scope : undef;
215 if (defined(wantarray)) {
216 my $r_res = $self->{-ipc_res} or die 'BUG: no ipc_res';
217 ipc_async_wait($self, -1);
218 _send_rec($w_req, [ wantarray, $sub, @args ]);
219 my $ret = _get_rec($r_res) // die "no response on $sub";
220 die $$ret if ref($ret) eq 'PublicInbox::IPC::Die';
221 wantarray ? @$ret : $$ret;
222 } else { # likely, fire-and-forget into pipe
223 _send_rec($w_req, [ undef , $sub, @args ]);
225 } else { # run locally
231 my ($self, $sub, $sub_args, $cb, $cb_arg) = @_;
232 if (my $w_req = $self->{-ipc_req}) { # run in worker
233 my $rec = _pack_rec([ 1, $sub, @$sub_args ]);
234 my $cur_bytes = \($self->{-async_inflight_bytes} //= 0);
235 while (($$cur_bytes + length($rec)) > PIPE_BUF) {
236 ipc_async_wait($self, 1);
238 my $ipc_lock = $self->{-ipc_lock};
239 my $lock = $ipc_lock ? $ipc_lock->lock_for_scope : undef;
240 print $w_req $rec or croak "print: $!";
241 $$cur_bytes += length($rec);
242 push @{$self->{-async_inflight}},
243 $sub, length($rec), $cb, $cb_arg;
245 my $ret = [ eval { $self->$sub(@$sub_args) } ];
247 $ret = ( bless(\$exc, 'PublicInbox::IPC::Die') );
249 eval { $cb->($cb_arg, $ret) };
250 warn "E: $sub callback error: $@\n" if $@;
254 # needed when there's multiple IPC workers and the parent forking
255 # causes newer siblings to inherit older siblings sockets
256 sub ipc_sibling_atfork_child {
258 my ($pid, undef) = delete(@$self{qw(-ipc_pid -ipc_ppid)});
259 delete(@$self{qw(-ipc_req -ipc_res)});
260 $pid == $$ and die "BUG: $$ ipc_atfork_child called on itself";
264 my ($self, $s2, $len, $full_stream) = @_;
265 my @fds = $recv_cmd->($s2, my $buf, $len);
266 return if scalar(@fds) && !defined($fds[0]);
267 my $n = length($buf) or return 0;
270 if (open(my $cmdfh, '+<&=', $fd)) {
271 $self->{$nfd++} = $cmdfh;
272 $cmdfh->autoflush(1);
274 die "$$ open(+<&=$fd) (FD:$nfd): $!";
277 while ($full_stream && $n < $len) {
278 my $r = sysread($s2, $buf, $len - $n, $n) // croak "read: $!";
279 croak "read EOF after $n/$len bytes" if $r == 0;
282 # Sereal dies on truncated data, Storable returns undef
283 my $args = ipc_thaw($buf) // die "thaw error on buffer of size: $n";
285 my $sub = shift @$args;
286 eval { $self->$sub(@$args) };
287 warn "$$ wq_worker: $@" if $@;
288 delete @$self{0..($nfd-1)};
292 sub wq_worker_loop ($) {
294 my $wqw = PublicInbox::WQWorker->new($self);
295 PublicInbox::DS->SetPostLoopCallback(sub { $wqw->{sock} });
296 PublicInbox::DS->EventLoop;
297 PublicInbox::DS->Reset;
300 sub do_sock_stream { # via wq_do, for big requests
301 my ($self, $len) = @_;
302 recv_and_run($self, delete $self->{0}, $len, 1);
305 sub wq_do { # always async
306 my ($self, $sub, $ios, @args) = @_;
307 if (my $s1 = $self->{-wq_s1}) { # run in worker
308 my $fds = [ map { fileno($_) } @$ios ];
309 my $buf = ipc_freeze([$sub, @args]);
310 my $n = $send_cmd->($s1, $fds, $buf, MSG_EOR);
311 return if defined($n); # likely
312 croak "sendmsg: $! (check RLIMIT_NOFILE)" if $!{ETOOMANYREFS};
313 croak "sendmsg: $!" if !$!{EMSGSIZE};
314 socketpair(my $r, my $w, AF_UNIX, SOCK_STREAM, 0) or
315 croak "socketpair: $!";
316 $n = $send_cmd->($s1, [ fileno($r) ],
317 ipc_freeze(['do_sock_stream', length($buf)]),
318 MSG_EOR) // croak "sendmsg: $!";
320 $n = $send_cmd->($w, $fds, $buf, 0) // croak "sendmsg: $!";
321 while ($n < length($buf)) {
322 my $x = syswrite($w, $buf, length($buf) - $n, $n) //
323 croak "syswrite: $!";
324 croak "syswrite wrote 0 bytes" if $x == 0;
328 @$self{0..$#$ios} = @$ios;
329 eval { $self->$sub(@args) };
330 warn "wq_do: $@" if $@;
331 delete @$self{0..$#$ios}; # don't close
335 sub _wq_worker_start ($$$) {
336 my ($self, $oldset, $fields) = @_;
337 my $seed = rand(0xffffffff);
338 my $pid = fork // die "fork: $!";
341 eval { PublicInbox::DS->Reset };
342 delete @$self{qw(-wq_s1 -wq_workers -wq_ppid)};
343 $SIG{$_} = 'IGNORE' for (qw(PIPE));
344 $SIG{$_} = 'DEFAULT' for (qw(TTOU TTIN TERM QUIT INT CHLD));
345 local $0 = $self->{-wq_ident};
346 PublicInbox::DS::sig_setmask($oldset);
347 # ensure we properly exit even if warn() dies:
348 my $end = PublicInbox::OnDestroy->new($$, sub { exit(!!$@) });
351 local @$self{keys %$fields} = values(%$fields);
352 my $on_destroy = $self->ipc_atfork_child;
354 wq_worker_loop($self);
356 warn "worker $self->{-wq_ident} PID:$$ died: $@" if $@;
357 undef $end; # trigger exit
359 $self->{-wq_workers}->{$pid} = \undef;
363 # starts workqueue workers if Sereal or Storable is installed
364 sub wq_workers_start {
365 my ($self, $ident, $nr_workers, $oldset, $fields) = @_;
366 ($enc && $send_cmd && $recv_cmd && defined($SEQPACKET)) or return;
367 return if $self->{-wq_s1}; # idempotent
368 $self->{-wq_s1} = $self->{-wq_s2} = undef;
369 socketpair($self->{-wq_s1}, $self->{-wq_s2}, AF_UNIX, $SEQPACKET, 0) or
370 die "socketpair: $!";
371 $self->ipc_atfork_prepare;
373 $nr_workers = $WQ_MAX_WORKERS if $nr_workers > $WQ_MAX_WORKERS;
374 my $sigset = $oldset // PublicInbox::DS::block_signals();
375 $self->{-wq_workers} = {};
376 $self->{-wq_ident} = $ident;
377 _wq_worker_start($self, $sigset, $fields) for (1..$nr_workers);
378 PublicInbox::DS::sig_setmask($sigset) unless $oldset;
379 $self->{-wq_ppid} = $$;
382 sub wq_worker_incr { # SIGTTIN handler
383 my ($self, $oldset, $fields) = @_;
384 $self->{-wq_s2} or return;
385 return if wq_workers($self) >= $WQ_MAX_WORKERS;
386 $self->ipc_atfork_prepare;
387 my $sigset = $oldset // PublicInbox::DS::block_signals();
388 _wq_worker_start($self, $sigset, $fields);
389 PublicInbox::DS::sig_setmask($sigset) unless $oldset;
392 sub wq_exit { # wakes up wq_worker_decr_wait
393 send($_[0]->{-wq_s2}, $$, MSG_EOR) // die "$$ send: $!";
397 sub wq_worker_decr { # SIGTTOU handler, kills first idle worker
399 return unless wq_workers($self);
400 my $s2 = $self->{-wq_s2} // die 'BUG: no wq_s2';
401 $self->wq_do('wq_exit', [ $s2, $s2, $s2 ]);
402 # caller must call wq_worker_decr_wait in main loop
405 sub wq_worker_decr_wait {
406 my ($self, $timeout) = @_;
407 return if $self->{-wq_ppid} != $$; # can't reap siblings or parents
408 my $s1 = $self->{-wq_s1} // croak 'BUG: no wq_s1';
409 vec(my $rin = '', fileno($s1), 1) = 1;
410 select(my $rout = $rin, undef, undef, $timeout) or
411 croak 'timed out waiting for wq_exit';
412 recv($s1, my $pid, 64, 0) // croak "recv: $!";
413 my $workers = $self->{-wq_workers} // croak 'BUG: no wq_workers';
414 delete $workers->{$pid} // croak "BUG: PID:$pid invalid";
415 dwaitpid($pid, \&ipc_worker_reap, $self);
418 # set or retrieve number of workers
420 my ($self, $nr) = @_;
421 my $cur = $self->{-wq_workers} or return;
423 while (scalar(keys(%$cur)) > $nr) {
424 $self->wq_worker_decr;
425 $self->wq_worker_decr_wait;
427 $self->wq_worker_incr while scalar(keys(%$cur)) < $nr;
433 my ($self, $nohang) = @_;
434 delete @$self{qw(-wq_s1 -wq_s2)} or return;
435 my $ppid = delete $self->{-wq_ppid} or return;
436 my $workers = delete $self->{-wq_workers} // die 'BUG: no wq_workers';
437 return if $ppid != $$; # can't reap siblings or parents
438 my @pids = map { $_ + 0 } keys %$workers;
440 push @{$self->{"-wq_old_pids.$$"}}, @pids;
442 dwaitpid($_, \&ipc_worker_reap, $self) for @pids;
448 my $pids = $self->{"-wq_old_pids.$$"} or return;
453 my ($self, $sig) = @_;
454 my $workers = $self->{-wq_workers} or return;
455 kill($sig // 'TERM', keys %$workers);
458 sub WQ_MAX_WORKERS { $WQ_MAX_WORKERS }
462 my $ppid = $self->{-wq_ppid};
463 wq_kill($self) if $ppid && $ppid == $$;
466 ipc_worker_stop($self);
469 sub detect_nproc () {
470 # _SC_NPROCESSORS_ONLN = 84 on both Linux glibc and musl
471 return POSIX::sysconf(84) if $^O eq 'linux';
472 return POSIX::sysconf(58) if $^O eq 'freebsd';
475 # getconf(1) is POSIX, but *NPROCESSORS* vars are not
476 for (qw(_NPROCESSORS_ONLN NPROCESSORS_ONLN)) {
477 `getconf $_ 2>/dev/null` =~ /^(\d+)$/ and return $1;
479 for my $nproc (qw(nproc gnproc)) { # GNU coreutils nproc
480 `$nproc 2>/dev/null` =~ /^(\d+)$/ and return $1;
483 # should we bother with `sysctl hw.ncpu`? Those only give
484 # us total processor count, not online processor count.