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_io_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_io_do when your work can be done on any idle worker
10 package PublicInbox::IPC;
13 use parent qw(Exporter);
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 $MY_MAX_ARG_STRLEN = 4096 * 33; # extra 4K for serialization
21 my $SEQPACKET = eval { Socket::SOCK_SEQPACKET() }; # portable enough?
22 our @EXPORT_OK = qw(ipc_freeze ipc_thaw);
24 # ->imports at BEGIN turns sereal_*_with_object into custom ops on 5.14+
25 # and eliminate method call overhead
28 require Sereal::Encoder;
29 require Sereal::Decoder;
30 Sereal::Encoder->import('sereal_encode_with_object');
31 Sereal::Decoder->import('sereal_decode_with_object');
32 ($enc, $dec) = (Sereal::Encoder->new, Sereal::Decoder->new);
36 if ($enc && $dec) { # should be custom ops
37 *ipc_freeze = sub ($) { sereal_encode_with_object $enc, $_[0] };
38 *ipc_thaw = sub ($) { sereal_decode_with_object $dec, $_[0], my $ret };
41 *ipc_freeze = \&Storable::freeze;
42 *ipc_thaw = \&Storable::thaw;
45 my $recv_cmd = PublicInbox::Spawn->can('recv_cmd4');
46 my $send_cmd = PublicInbox::Spawn->can('send_cmd4') // do {
47 require PublicInbox::CmdIPC4;
48 $recv_cmd //= PublicInbox::CmdIPC4->can('recv_cmd4');
49 PublicInbox::CmdIPC4->can('send_cmd4');
51 require PublicInbox::Syscall;
52 $recv_cmd //= PublicInbox::Syscall->can('recv_cmd4');
53 PublicInbox::Syscall->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 print $w length($buf), "\n", $buf or croak "print: $!";
71 sub ipc_return ($$$) {
72 my ($w, $ret, $exc) = @_;
73 _send_rec($w, $exc ? bless(\$exc, 'PublicInbox::IPC::Die') : $ret);
76 sub ipc_worker_loop ($$$) {
77 my ($self, $r_req, $w_res) = @_;
78 my ($rec, $wantarray, $sub, @args);
80 while ($rec = _get_rec($r_req)) {
81 ($wantarray, $sub, @args) = @$rec;
82 # no waiting if client doesn't care,
83 # this is the overwhelmingly likely case
84 if (!defined($wantarray)) {
85 eval { $self->$sub(@args) };
86 warn "$$ die: $@ (from nowait $sub)\n" if $@;
87 } elsif ($wantarray) {
88 my @ret = eval { $self->$sub(@args) };
89 ipc_return($w_res, \@ret, $@);
90 } else { # '' => wantscalar
91 my $ret = eval { $self->$sub(@args) };
92 ipc_return($w_res, \$ret, $@);
97 # starts a worker if Sereal or Storable is installed
98 sub ipc_worker_spawn {
99 my ($self, $ident, $oldset, $fields) = @_;
100 return if ($self->{-ipc_ppid} // -1) == $$; # idempotent
101 delete(@$self{qw(-ipc_req -ipc_res -ipc_ppid -ipc_pid)});
102 pipe(my ($r_req, $w_req)) or die "pipe: $!";
103 pipe(my ($r_res, $w_res)) or die "pipe: $!";
104 my $sigset = $oldset // PublicInbox::DS::block_signals();
105 $self->ipc_atfork_prepare;
106 my $seed = rand(0xffffffff);
107 my $pid = fork // die "fork: $!";
110 eval { Net::SSLeay::randomize() };
111 eval { PublicInbox::DS->Reset };
112 delete @$self{qw(-wq_s1 -wq_s2 -wq_workers -wq_ppid)};
113 $w_req = $r_res = undef;
114 $w_res->autoflush(1);
115 $SIG{$_} = 'IGNORE' for (qw(TERM INT QUIT));
117 # ensure we properly exit even if warn() dies:
118 my $end = PublicInbox::OnDestroy->new($$, sub { exit(!!$@) });
121 local @$self{keys %$fields} = values(%$fields);
122 my $on_destroy = $self->ipc_atfork_child;
123 local @SIG{keys %SIG} = values %SIG;
124 PublicInbox::DS::sig_setmask($sigset);
125 ipc_worker_loop($self, $r_req, $w_res);
127 warn "worker $ident PID:$$ died: $@\n" if $@;
128 undef $end; # trigger exit
130 PublicInbox::DS::sig_setmask($sigset) unless $oldset;
131 $r_req = $w_res = undef;
132 $w_req->autoflush(1);
133 $self->{-ipc_req} = $w_req;
134 $self->{-ipc_res} = $r_res;
135 $self->{-ipc_ppid} = $$;
136 $self->{-ipc_pid} = $pid;
139 sub ipc_worker_reap { # dwaitpid callback
140 my ($args, $pid) = @_;
141 my ($self, @uargs) = @$args;
142 delete $self->{-wq_workers}->{$pid};
143 return $self->{-reap_do}->($args, $pid) if $self->{-reap_do};
146 # TERM(15) is our default exit signal, PIPE(13) is likely w/ pager
147 warn "$self->{-wq_ident} PID:$pid died \$?=$?\n" if $s != 15 && $s != 13
151 my ($self, $cb, @uargs) = @_;
152 local $PublicInbox::DS::in_loop = 1;
153 $self->{-reap_async} = 1;
154 $self->{-reap_do} = $cb;
155 my @pids = keys %{$self->{-wq_workers}};
156 dwaitpid($_, \&ipc_worker_reap, [ $self, @uargs ]) for @pids;
159 # for base class, override in sub classes
160 sub ipc_atfork_prepare {}
162 sub wq_atexit_child {}
164 sub ipc_atfork_child {
166 my $io = delete($self->{-ipc_atfork_child_close}) or return;
171 # idempotent, can be called regardless of whether worker is active or not
172 sub ipc_worker_stop {
173 my ($self, $args) = @_;
174 my ($pid, $ppid) = delete(@$self{qw(-ipc_pid -ipc_ppid)});
175 my ($w_req, $r_res) = delete(@$self{qw(-ipc_req -ipc_res)});
176 if (!$w_req && !$r_res) {
177 die "unexpected PID:$pid without IPC pipes" if $pid;
180 die 'no PID with IPC pipes' unless $pid;
181 $w_req = $r_res = undef;
183 return if $$ != $ppid;
184 dwaitpid($pid, \&ipc_worker_reap, [$self, $args]);
187 # use this if we have multiple readers reading curl or "pigz -dc"
188 # and writing to the same store
191 $f // die 'BUG: no filename given';
192 require PublicInbox::Lock;
193 $self->{-ipc_lock} //= bless { lock_path => $f }, 'PublicInbox::Lock'
196 sub _wait_return ($$) {
197 my ($r_res, $sub) = @_;
198 my $ret = _get_rec($r_res) // die "no response on $sub";
199 die $$ret if ref($ret) eq 'PublicInbox::IPC::Die';
200 wantarray ? @$ret : $$ret;
203 # call $self->$sub(@args), on a worker if ipc_worker_spawn was used
205 my ($self, $sub, @args) = @_;
206 if (my $w_req = $self->{-ipc_req}) { # run in worker
207 my $ipc_lock = $self->{-ipc_lock};
208 my $lock = $ipc_lock ? $ipc_lock->lock_for_scope : undef;
209 if (defined(wantarray)) {
210 my $r_res = $self->{-ipc_res} or die 'no ipc_res';
211 _send_rec($w_req, [ wantarray, $sub, @args ]);
212 _wait_return($r_res, $sub);
213 } else { # likely, fire-and-forget into pipe
214 _send_rec($w_req, [ undef , $sub, @args ]);
216 } else { # run locally
221 # needed when there's multiple IPC workers and the parent forking
222 # causes newer siblings to inherit older siblings sockets
223 sub ipc_sibling_atfork_child {
225 my ($pid, undef) = delete(@$self{qw(-ipc_pid -ipc_ppid)});
226 delete(@$self{qw(-ipc_req -ipc_res)});
227 $pid == $$ and die "BUG: $$ ipc_atfork_child called on itself";
231 my ($self, $s2, $len, $full_stream) = @_;
232 my @fds = $recv_cmd->($s2, my $buf, $len // $MY_MAX_ARG_STRLEN);
233 return if scalar(@fds) && !defined($fds[0]);
234 my $n = length($buf) or return 0;
237 if (open(my $cmdfh, '+<&=', $fd)) {
238 $self->{$nfd++} = $cmdfh;
239 $cmdfh->autoflush(1);
241 die "$$ open(+<&=$fd) (FD:$nfd): $!";
244 while ($full_stream && $n < $len) {
245 my $r = sysread($s2, $buf, $len - $n, $n) // croak "read: $!";
246 croak "read EOF after $n/$len bytes" if $r == 0;
249 # Sereal dies on truncated data, Storable returns undef
250 my $args = ipc_thaw($buf) // die "thaw error on buffer of size: $n";
252 my $sub = shift @$args;
253 eval { $self->$sub(@$args) };
254 warn "$$ $0 wq_worker: $sub: $@" if $@;
255 delete @$self{0..($nfd-1)};
259 sub wq_worker_loop ($$) {
260 my ($self, $bcast2) = @_;
261 my $wqw = PublicInbox::WQWorker->new($self, $self->{-wq_s2});
262 PublicInbox::WQWorker->new($self, $bcast2) if $bcast2;
263 PublicInbox::DS->SetPostLoopCallback(sub { $wqw->{sock} });
264 PublicInbox::DS::event_loop();
265 PublicInbox::DS->Reset;
268 sub do_sock_stream { # via wq_io_do, for big requests
269 my ($self, $len) = @_;
270 recv_and_run($self, my $s2 = delete $self->{0}, $len, 1);
274 my ($self, $sub, @args) = @_;
275 if (my $wkr = $self->{-wq_workers}) {
276 my $buf = ipc_freeze([$sub, @args]);
277 for my $bcast1 (values %$wkr) {
278 my $sock = $bcast1 // $self->{-wq_s1} // next;
279 send($sock, $buf, MSG_EOR) // croak "send: $!";
280 # XXX shouldn't have to deal with EMSGSIZE here...
283 eval { $self->$sub(@args) };
284 warn "wq_broadcast: $@" if $@;
288 sub stream_in_full ($$$) {
289 my ($s1, $fds, $buf) = @_;
290 socketpair(my $r, my $w, AF_UNIX, SOCK_STREAM, 0) or
291 croak "socketpair: $!";
292 my $n = $send_cmd->($s1, [ fileno($r) ],
293 ipc_freeze(['do_sock_stream', length($buf)]),
294 MSG_EOR) // croak "sendmsg: $!";
296 $n = $send_cmd->($w, $fds, $buf, 0) // croak "sendmsg: $!";
297 while ($n < length($buf)) {
298 my $x = syswrite($w, $buf, length($buf) - $n, $n) //
299 croak "syswrite: $!";
300 croak "syswrite wrote 0 bytes" if $x == 0;
305 sub wq_io_do { # always async
306 my ($self, $sub, $ios, @args) = @_;
307 if (my $s1 = $self->{-wq_s1}) { # run in worker
308 my $fds = [ map { fileno($_) } @$ios ];
309 my $buf = ipc_freeze([$sub, @args]);
310 if (length($buf) > $MY_MAX_ARG_STRLEN) {
311 stream_in_full($s1, $fds, $buf);
313 my $n = $send_cmd->($s1, $fds, $buf, MSG_EOR);
314 return if defined($n); # likely
316 croak "sendmsg: $! (check RLIMIT_NOFILE)";
317 $!{EMSGSIZE} ? stream_in_full($s1, $fds, $buf) :
318 croak("sendmsg: $!");
321 @$self{0..$#$ios} = @$ios;
322 eval { $self->$sub(@args) };
323 warn "wq_io_do: $@" if $@;
324 delete @$self{0..$#$ios}; # don't close
329 my ($self, $wantarray, $sub, @args) = @_;
331 my @ret = eval { $self->$sub(@args) };
332 ipc_return($self->{0}, \@ret, $@);
333 } else { # '' => wantscalar
334 my $ret = eval { $self->$sub(@args) };
335 ipc_return($self->{0}, \$ret, $@);
340 my ($self, $sub, @args) = @_;
341 if (defined(wantarray)) {
342 pipe(my ($r, $w)) or die "pipe: $!";
343 wq_io_do($self, 'wq_sync_run', [ $w ], wantarray, $sub, @args);
345 _wait_return($r, $sub);
347 wq_io_do($self, $sub, [], @args);
351 sub _wq_worker_start ($$$$) {
352 my ($self, $oldset, $fields, $one) = @_;
353 my ($bcast1, $bcast2);
354 $one or socketpair($bcast1, $bcast2, AF_UNIX, $SEQPACKET, 0) or
355 die "socketpair: $!";
356 my $seed = rand(0xffffffff);
357 my $pid = fork // die "fork: $!";
360 eval { Net::SSLeay::randomize() };
362 eval { PublicInbox::DS->Reset };
363 delete @$self{qw(-wq_s1 -wq_ppid)};
364 $self->{-wq_worker_nr} =
365 keys %{delete($self->{-wq_workers}) // {}};
366 $SIG{$_} = 'DEFAULT' for (qw(TTOU TTIN TERM QUIT INT CHLD));
367 local $0 = $one ? $self->{-wq_ident} :
368 "$self->{-wq_ident} $self->{-wq_worker_nr}";
369 # ensure we properly exit even if warn() dies:
370 my $end = PublicInbox::OnDestroy->new($$, sub { exit(!!$@) });
373 local @$self{keys %$fields} = values(%$fields);
374 my $on_destroy = $self->ipc_atfork_child;
375 local @SIG{keys %SIG} = values %SIG;
376 PublicInbox::DS::sig_setmask($oldset);
377 wq_worker_loop($self, $bcast2);
379 warn "worker $self->{-wq_ident} PID:$$ died: $@" if $@;
380 undef $end; # trigger exit
382 $self->{-wq_workers}->{$pid} = $bcast1;
386 # starts workqueue workers if Sereal or Storable is installed
387 sub wq_workers_start {
388 my ($self, $ident, $nr_workers, $oldset, $fields) = @_;
389 ($send_cmd && $recv_cmd && defined($SEQPACKET)) or return;
390 return if $self->{-wq_s1}; # idempotent
391 $self->{-wq_s1} = $self->{-wq_s2} = undef;
392 socketpair($self->{-wq_s1}, $self->{-wq_s2}, AF_UNIX, $SEQPACKET, 0) or
393 die "socketpair: $!";
394 $self->ipc_atfork_prepare;
395 $nr_workers //= $self->{-wq_nr_workers}; # was set earlier
396 my $sigset = $oldset // PublicInbox::DS::block_signals();
397 $self->{-wq_workers} = {};
398 $self->{-wq_ident} = $ident;
399 my $one = $nr_workers == 1;
400 $self->{-wq_nr_workers} = $nr_workers;
401 _wq_worker_start($self, $sigset, $fields, $one) for (1..$nr_workers);
402 PublicInbox::DS::sig_setmask($sigset) unless $oldset;
403 $self->{-wq_ppid} = $$;
408 delete @$self{qw(-wq_s1 -wq_s2)} or return;
409 return if $self->{-reap_async};
410 my @pids = keys %{$self->{-wq_workers}};
411 dwaitpid($_, \&ipc_worker_reap, [ $self ]) for @pids;
415 my ($self, $sig) = @_;
416 kill($sig // 'TERM', keys %{$self->{-wq_workers}});
421 my $ppid = $self->{-wq_ppid};
422 wq_kill($self) if $ppid && $ppid == $$;
424 ipc_worker_stop($self);
427 sub detect_nproc () {
428 # _SC_NPROCESSORS_ONLN = 84 on both Linux glibc and musl
429 return POSIX::sysconf(84) if $^O eq 'linux';
430 return POSIX::sysconf(58) if $^O eq 'freebsd';
433 # getconf(1) is POSIX, but *NPROCESSORS* vars are not
434 for (qw(_NPROCESSORS_ONLN NPROCESSORS_ONLN)) {
435 `getconf $_ 2>/dev/null` =~ /^(\d+)$/ and return $1;
437 for my $nproc (qw(nproc gnproc)) { # GNU coreutils nproc
438 `$nproc 2>/dev/null` =~ /^(\d+)$/ and return $1;
441 # should we bother with `sysctl hw.ncpu`? Those only give
442 # us total processor count, not online processor count.