1 # ----------------------------------------------------------------
2 # Copyright (C) 2010 Equinox Software, Inc.
3 # Bill Erickson <erickson@esilibrary.com>
5 # This program is free software; you can redistribute it and/or
6 # modify it under the terms of the GNU General Public License
7 # as published by the Free Software Foundation; either version 2
8 # of the License, or (at your option) any later version.
10 # This program is distributed in the hope that it will be useful,
11 # but WITHOUT ANY WARRANTY; without even the implied warranty of
12 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 # GNU General Public License for more details.
14 # ----------------------------------------------------------------
15 package OpenSRF::Server;
18 use OpenSRF::Transport;
19 use OpenSRF::Application;
20 use OpenSRF::Utils::Config;
21 use OpenSRF::Transport::PeerHandle;
22 use OpenSRF::Utils::SettingsClient;
23 use OpenSRF::Utils::Logger qw($logger);
24 use OpenSRF::Transport::SlimJabber::Client;
26 use POSIX qw/:sys_wait_h :errno_h/;
27 use Fcntl qw(F_GETFL F_SETFL O_NONBLOCK);
28 use Time::HiRes qw/usleep/;
31 our $chatty = 1; # disable for production
33 use constant STATUS_PIPE_DATA_SIZE => 12;
34 use constant WRITE_PIPE_DATA_SIZE => 12;
37 my($class, $service, %args) = @_;
38 my $self = bless(\%args, $class);
40 $self->{service} = $service; # service name
41 $self->{num_children} = 0; # number of child processes
42 $self->{osrf_handle} = undef; # xmpp handle
43 $self->{routers} = []; # list of registered routers
44 $self->{active_list} = []; # list of active children
45 $self->{idle_list} = []; # list of idle children
46 $self->{sighup_pending} = [];
47 $self->{pid_map} = {}; # map of child pid to child for cleaner access
48 $self->{sig_pipe} = 0; # true if last syswrite failed
50 $self->{stderr_log} = $self->{stderr_log_path} . "/${service}_stderr.log"
51 if $self->{stderr_log_path};
53 $self->{min_spare_children} ||= 0;
55 $self->{max_spare_children} = $self->{min_spare_children} + 1 if
56 $self->{max_spare_children} and
57 $self->{max_spare_children} <= $self->{min_spare_children};
62 # ----------------------------------------------------------------
63 # Disconnects from routers and waits for child processes to exit.
64 # ----------------------------------------------------------------
70 $logger->info("server: shutting down and cleaning up...");
73 $self->unregister_routers;
76 # graceful shutdown waits for all active
77 # children to complete their in-process tasks.
79 while (@{$self->{active_list}}) {
80 $logger->info("server: graceful shutdown with ".
81 @{$self->{active_list}}." active children...");
83 # block until a child is becomes available
84 $self->check_status(1);
86 $logger->info("server: all clear for graceful shutdown");
89 # don't get sidetracked by signals while we're cleaning up.
90 # it could result in unexpected behavior with list traversal
91 $SIG{CHLD} = 'IGNORE';
93 # terminate the child processes
94 $self->kill_child($_) for
95 (@{$self->{idle_list}}, @{$self->{active_list}});
97 $self->{osrf_handle}->disconnect;
99 # clean up our dead children
100 $self->reap_children(1);
102 exit(0) unless $no_exit;
105 # ----------------------------------------------------------------
106 # SIGHUP handler. Kill all idle children. Copy list of active
107 # children into sighup_pending list for later cleanup.
108 # ----------------------------------------------------------------
111 $logger->info("server: caught SIGHUP; reloading children");
113 # reload the opensrf config
114 # note: calling ::Config->load() results in ever-growing
115 # package names, which eventually causes an exception
116 OpenSRF::Utils::Config->current->_load(
118 config_file => OpenSRF::Utils::Config->current->FILE
121 # force-reload the logger config
122 OpenSRF::Utils::Logger::set_config(1);
124 # copy active list into pending list for later cleanup
125 $self->{sighup_pending} = [ @{$self->{active_list}} ];
127 # idle_list will be modified as children are reaped.
128 my @idle = @{$self->{idle_list}};
130 # idle children are the reaper's plaything
131 $self->kill_child($_) for @idle;
134 # ----------------------------------------------------------------
135 # Waits on the jabber socket for inbound data from the router.
136 # Each new message is passed off to a child process for handling.
137 # At regular intervals, wake up for min/max spare child maintenance
138 # ----------------------------------------------------------------
142 $logger->set_service($self->{service});
144 $SIG{$_} = sub { $self->cleanup; } for (qw/INT QUIT/);
145 $SIG{TERM} = sub { $self->cleanup(0, 1); };
146 $SIG{CHLD} = sub { $self->reap_children(); };
147 $SIG{HUP} = sub { $self->handle_sighup(); };
148 $SIG{USR1} = sub { $self->unregister_routers; };
149 $SIG{USR2} = sub { $self->register_routers; };
151 $self->spawn_children;
152 $self->build_osrf_handle;
153 $self->register_routers;
160 $self->{child_died} = 0;
162 my $msg = $self->{osrf_handle}->process($wait_time);
164 # we woke up for any reason, reset the wait time to allow
165 # for idle maintenance as necessary
170 if(my $child = pop(@{$self->{idle_list}})) {
172 # we have an idle child to handle the request
173 $chatty and $logger->internal("server: passing request to idle child $child");
174 push(@{$self->{active_list}}, $child);
175 $self->write_child($child, $msg);
177 } elsif($self->{num_children} < $self->{max_children}) {
179 # spawning a child to handle the request
180 $chatty and $logger->internal("server: spawning child to handle request");
181 $self->write_child($self->spawn_child(1), $msg);
184 $logger->warn("server: no children available, waiting... consider increasing " .
185 "max_children for this application higher than $self->{max_children} ".
186 "in the OpenSRF configuration if this message occurs frequently");
187 $self->check_status(1); # block until child is available
189 my $child = pop(@{$self->{idle_list}});
190 push(@{$self->{active_list}}, $child);
191 $self->write_child($child, $msg);
196 # don't perform idle maint immediately when woken by SIGCHLD
197 unless($self->{child_died}) {
199 # when we hit equilibrium, there's no need for regular
200 # maintenance, so set wait_time to 'forever'
202 !$self->perform_idle_maintenance and # no maintenance performed this time
203 @{$self->{active_list}} == 0; # no active children
209 # ----------------------------------------------------------------
210 # Launch a new spare child or kill an extra spare child. To
211 # prevent large-scale spawning or die-offs, spawn or kill only
212 # 1 process per idle maintenance loop.
213 # Returns true if any idle maintenance occurred, 0 otherwise
214 # ----------------------------------------------------------------
215 sub perform_idle_maintenance {
218 $chatty and $logger->internal(
220 "server: %d idle, %d active, %d min_spare, %d max_spare in idle maintenance",
221 scalar(@{$self->{idle_list}}),
222 scalar(@{$self->{active_list}}),
223 $self->{min_spare_children},
224 $self->{max_spare_children}
228 # spawn 1 spare child per maintenance loop if necessary
229 if( $self->{min_spare_children} and
230 $self->{num_children} < $self->{max_children} and
231 scalar(@{$self->{idle_list}}) < $self->{min_spare_children} ) {
233 $chatty and $logger->internal("server: spawning spare child");
237 # kill 1 excess spare child per maintenance loop if necessary
238 } elsif($self->{max_spare_children} and
239 $self->{num_children} > $self->{min_children} and
240 scalar(@{$self->{idle_list}}) > $self->{max_spare_children} ) {
242 $chatty and $logger->internal("server: killing spare child");
252 my $child = shift || pop(@{$self->{idle_list}}) or return;
253 $chatty and $logger->internal("server: killing child $child");
254 kill('TERM', $child->{pid});
257 # ----------------------------------------------------------------
258 # Jabber connection inbound message arrive on.
259 # ----------------------------------------------------------------
260 sub build_osrf_handle {
263 my $conf = OpenSRF::Utils::Config->current;
264 my $username = $conf->bootstrap->username;
265 my $password = $conf->bootstrap->passwd;
266 my $domain = $conf->bootstrap->domain;
267 my $port = $conf->bootstrap->port;
268 my $resource = $self->{service} . '_listener_' . $conf->env->hostname;
270 $logger->debug("server: inbound connecting as $username\@$domain/$resource on port $port");
272 $self->{osrf_handle} =
273 OpenSRF::Transport::SlimJabber::Client->new(
274 username => $username,
275 resource => $resource,
276 password => $password,
281 $self->{osrf_handle}->initialize;
285 # ----------------------------------------------------------------
286 # Sends request data to a child process
287 # ----------------------------------------------------------------
289 my($self, $child, $msg) = @_;
290 my $xml = encode_utf8(decode_utf8($msg->to_xml));
292 # tell the child how much data to expect, minus the header
294 {use bytes; $write_size = length($xml)}
295 $write_size = sprintf("%*s", WRITE_PIPE_DATA_SIZE, $write_size);
299 $self->{sig_pipe} = 0;
300 local $SIG{'PIPE'} = sub { $self->{sig_pipe} = 1; };
302 # send message to child data pipe
303 syswrite($child->{pipe_to_child}, $write_size . $xml);
305 last unless $self->{sig_pipe};
306 $logger->error("server: got SIGPIPE writing to $child, retrying...");
307 usleep(50000); # 50 msec
310 $logger->error("server: unable to send request message to child $child") if $self->{sig_pipe};
313 # ----------------------------------------------------------------
314 # Checks to see if any child process has reported its availability
315 # In blocking mode, blocks until a child has reported.
316 # ----------------------------------------------------------------
318 my($self, $block) = @_;
320 return unless @{$self->{active_list}};
326 # if can_read or sysread is interrupted while bloking, go back and
327 # wait again until we have at least 1 free child
329 # refresh the read_set handles in case we lost a child in the previous iteration
330 my $read_set = IO::Select->new;
331 $read_set->add($_->{pipe_to_child}) for @{$self->{active_list}};
333 if(my @handles = $read_set->can_read(($block) ? undef : 0)) {
335 for my $pipe (@handles) {
336 sysread($pipe, $pid, STATUS_PIPE_DATA_SIZE) or next;
337 push(@pids, int($pid));
341 last unless $block and !@pids;
346 $chatty and $logger->internal("server: ".scalar(@pids)." children reporting for duty: (@pids)");
351 # move the children from the active list to the idle list
352 for my $proc (@{$self->{active_list}}) {
353 if(grep { $_ == $proc->{pid} } @pids) {
354 push(@{$self->{idle_list}}, $proc);
356 push(@new_actives, $proc);
360 $self->{active_list} = [@new_actives];
362 $chatty and $logger->internal(sprintf(
363 "server: %d idle and %d active children after status update",
364 scalar(@{$self->{idle_list}}), scalar(@{$self->{active_list}})));
366 # some children just went from active to idle. let's see
367 # if any of them need to be killed from a previous sighup.
369 for my $child (@{$self->{sighup_pending}}) {
370 if (grep {$_ == $child->{pid}} @pids) {
372 $chatty and $logger->internal(
373 "server: killing previously-active ".
374 "child after receiving SIGHUP: $child");
376 # remove the pending child
377 $self->{sighup_pending} = [
378 grep {$_->{pid} != $child->{pid}}
379 @{$self->{sighup_pending}}
382 # kill the pending child
383 $self->kill_child($child)
388 # ----------------------------------------------------------------
389 # Cleans up any child processes that have exited.
390 # In shutdown mode, block until all children have washed ashore
391 # ----------------------------------------------------------------
393 my($self, $shutdown) = @_;
394 $self->{child_died} = 1;
398 my $pid = waitpid(-1, ($shutdown) ? 0 : WNOHANG);
401 $chatty and $logger->internal("server: reaping child $pid");
403 my $child = $self->{pid_map}->{$pid};
405 close($child->{pipe_to_parent});
406 close($child->{pipe_to_child});
408 $self->{active_list} = [ grep { $_->{pid} != $pid } @{$self->{active_list}} ];
409 $self->{idle_list} = [ grep { $_->{pid} != $pid } @{$self->{idle_list}} ];
411 $self->{num_children}--;
412 delete $self->{pid_map}->{$pid};
413 delete $child->{$_} for keys %$child; # destroy with a vengeance
416 $self->spawn_children unless $shutdown;
418 $chatty and $logger->internal(sprintf(
419 "server: %d idle and %d active children after reap_children",
420 scalar(@{$self->{idle_list}}), scalar(@{$self->{active_list}})));
423 # ----------------------------------------------------------------
424 # Spawn up to max_children processes
425 # ----------------------------------------------------------------
428 $self->spawn_child while $self->{num_children} < $self->{min_children};
431 # ----------------------------------------------------------------
432 # Spawns a new child. If $active is set, the child goes directly
433 # into the active_list.
434 # ----------------------------------------------------------------
436 my($self, $active) = @_;
438 my $child = OpenSRF::Server::Child->new($self);
440 # socket for sending message data to the child
442 $child->{pipe_to_child},
443 $child->{pipe_to_parent},
444 AF_UNIX, SOCK_STREAM, PF_UNSPEC)) {
445 $logger->error("server: error creating data socketpair: $!");
449 $child->{pipe_to_child}->autoflush(1);
450 $child->{pipe_to_parent}->autoflush(1);
452 $child->{pid} = fork();
454 if($child->{pid}) { # parent process
455 $self->{num_children}++;
456 $self->{pid_map}->{$child->{pid}} = $child;
459 push(@{$self->{active_list}}, $child);
461 push(@{$self->{idle_list}}, $child);
464 $chatty and $logger->internal("server: server spawned child $child with ".$self->{num_children}." total children");
468 } else { # child process
470 # recover default handling for any signal whose handler
471 # may have been adopted from the parent process.
472 $SIG{$_} = 'DEFAULT' for qw/TERM INT QUIT HUP CHLD USR1 USR2/;
474 if($self->{stderr_log}) {
476 $chatty and $logger->internal("server: redirecting STDERR to " . $self->{stderr_log});
479 unless( open(STDERR, '>>' . $self->{stderr_log}) ) {
480 $logger->error("server: unable to open STDERR log file: " . $self->{stderr_log} . " : $@");
481 open STDERR, '>/dev/null'; # send it back to /dev/null
489 OpenSRF::Transport::PeerHandle->retrieve->disconnect;
491 $logger->error("server: child process died: $@") if $@;
496 # ----------------------------------------------------------------
497 # Sends the register command to the configured routers
498 # ----------------------------------------------------------------
499 sub register_routers {
502 my $conf = OpenSRF::Utils::Config->current;
503 my $routers = $conf->bootstrap->routers;
504 my $router_name = $conf->bootstrap->router_name;
507 for my $router (@$routers) {
510 if( !$router->{services} ||
511 !$router->{services}->{service} ||
513 ref($router->{services}->{service}) eq 'ARRAY' and
514 grep { $_ eq $self->{service} } @{$router->{services}->{service}}
515 ) || $router->{services}->{service} eq $self->{service}) {
517 my $name = $router->{name};
518 my $domain = $router->{domain};
519 push(@targets, "$name\@$domain/router");
523 push(@targets, "$router_name\@$router/router");
528 $logger->info("server: registering with router $_");
529 $self->{osrf_handle}->send(
531 body => 'registering',
532 router_command => 'register',
533 router_class => $self->{service}
537 $self->{routers} = \@targets;
540 # ----------------------------------------------------------------
541 # Sends the unregister command to any routers we have registered
543 # ----------------------------------------------------------------
544 sub unregister_routers {
546 return unless $self->{osrf_handle}->tcp_connected;
548 for my $router (@{$self->{routers}}) {
549 $logger->info("server: disconnecting from router $router");
550 $self->{osrf_handle}->send(
552 body => "unregistering",
553 router_command => "unregister",
554 router_class => $self->{service}
560 package OpenSRF::Server::Child;
563 use OpenSRF::Transport;
564 use OpenSRF::Application;
565 use OpenSRF::Transport::PeerHandle;
566 use OpenSRF::Transport::SlimJabber::XMPPMessage;
567 use OpenSRF::Utils::Logger qw($logger);
568 use OpenSRF::DomainObject::oilsResponse qw/:status/;
569 use Fcntl qw(F_GETFL F_SETFL O_NONBLOCK);
570 use Time::HiRes qw(time usleep);
571 use POSIX qw/:sys_wait_h :errno_h/;
573 use overload '""' => sub { return '[' . shift()->{pid} . ']'; };
576 my($class, $parent) = @_;
577 my $self = bless({}, $class);
578 $self->{pid} = 0; # my process ID
579 $self->{parent} = $parent; # Controller parent process
580 $self->{num_requests} = 0; # total serviced requests
581 $self->{sig_pipe} = 0; # true if last syswrite failed
587 my $flags = fcntl($fh, F_GETFL, 0);
588 fcntl($fh, F_SETFL, $flags | O_NONBLOCK);
593 my $flags = fcntl($fh, F_GETFL, 0);
594 $flags &= ~O_NONBLOCK;
595 fcntl($fh, F_SETFL, $flags);
598 # ----------------------------------------------------------------
599 # Connects to Jabber and runs the application child_init
600 # ----------------------------------------------------------------
603 my $service = $self->{parent}->{service};
604 $0 = "OpenSRF Drone [$service]";
605 OpenSRF::Transport::PeerHandle->construct($service);
606 OpenSRF::Application->application_implementation->child_init
607 if (OpenSRF::Application->application_implementation->can('child_init'));
610 # ----------------------------------------------------------------
611 # Waits for messages from the parent process, handles the message,
612 # then goes into the keepalive loop if this is a stateful session.
613 # When max_requests is hit, the process exits.
614 # ----------------------------------------------------------------
617 my $network = OpenSRF::Transport::PeerHandle->retrieve;
619 # main child run loop. Ends when this child hits max requests.
622 my $data = $self->wait_for_request or next;
624 # Update process name to show activity
628 # Discard extraneous data from the jabber socket
629 if(!$network->flush_socket()) {
630 $logger->error("server: network disconnected! child dropping request and exiting: $data");
634 my $session = OpenSRF::Transport->handler(
635 $self->{parent}->{service},
636 OpenSRF::Transport::SlimJabber::XMPPMessage->new(xml => $data)
639 $self->keepalive_loop($session);
641 last if ++$self->{num_requests} == $self->{parent}->{max_requests};
643 # Tell the parent process we are available to process requests
646 # Repair process name
650 $chatty and $logger->internal("server: child process shutting down after reaching max_requests");
652 OpenSRF::Application->application_implementation->child_exit
653 if (OpenSRF::Application->application_implementation->can('child_exit'));
656 # ----------------------------------------------------------------
657 # waits for a request data on the parent pipe and returns it.
658 # ----------------------------------------------------------------
659 sub wait_for_request {
662 my $data = ''; # final request data
663 my $buf_size = 4096; # default linux pipe_buf (atomic window, not total size)
664 my $read_pipe = $self->{pipe_to_parent};
665 my $bytes_needed; # size of the data we are about to receive
666 my $bytes_recvd; # number of bytes read so far
667 my $first_read = 1; # true for first loop iteration
672 # wait for some data to start arriving
673 my $read_set = IO::Select->new;
674 $read_set->add($read_pipe);
677 # if can_read is interrupted while blocking,
678 # go back and wait again until it succeeds.
679 last if $read_set->can_read;
682 # parent started writing, let's start reading
683 $self->set_nonblock($read_pipe);
686 # read all of the available data
689 my $nbytes = sysread($self->{pipe_to_parent}, $buf, $buf_size);
691 unless(defined $nbytes) {
693 $logger->error("server: error reading data from parent: $!. ".
694 "bytes_needed=$bytes_needed; bytes_recvd=$bytes_recvd; data=$data");
700 last if $nbytes <= 0; # no more data available for reading
702 $bytes_recvd += $nbytes;
706 $self->set_block($self->{pipe_to_parent});
707 return undef if $read_error;
709 # extract the data size and remove the header from the final data
711 my $wps_size = OpenSRF::Server::WRITE_PIPE_DATA_SIZE;
712 $bytes_needed = int(substr($data, 0, $wps_size)) + $wps_size;
713 $data = substr($data, $wps_size);
718 if ($bytes_recvd == $bytes_needed) {
719 # we've read all the data. Nothing left to do
723 $logger->info("server: child process read all available pipe data. ".
724 "waiting for more data from parent. bytes_needed=$bytes_needed; bytes_recvd=$bytes_recvd");
731 # ----------------------------------------------------------------
732 # If this is a stateful opensrf session, wait up to $keepalive
733 # seconds for subsequent requests from the client
734 # ----------------------------------------------------------------
736 my($self, $session) = @_;
737 my $keepalive = $self->{parent}->{keepalive};
739 while($session->state and $session->state == $session->CONNECTED) {
741 unless( $session->queue_wait($keepalive) ) {
743 # client failed to disconnect before timeout
744 $logger->info("server: no request was received in $keepalive seconds, exiting stateful session");
746 my $res = OpenSRF::DomainObject::oilsConnectStatus->new(
747 status => "Disconnected on timeout",
748 statusCode => STATUS_TIMEOUT
751 $session->status($res);
752 $session->state($session->DISCONNECTED);
757 $chatty and $logger->internal("server: child done with request(s)");
761 # ----------------------------------------------------------------
762 # Report our availability to our parent process
763 # ----------------------------------------------------------------
769 $self->{sig_pipe} = 0;
770 local $SIG{'PIPE'} = sub { $self->{sig_pipe} = 1; };
773 $self->{pipe_to_parent},
774 sprintf("%*s", OpenSRF::Server::STATUS_PIPE_DATA_SIZE, $self->{pid})
777 last unless $self->{sig_pipe};
778 $logger->error("server: $self got SIGPIPE writing status to parent, retrying...");
779 usleep(50000); # 50 msec
782 $logger->error("server: $self unable to send status to parent") if $self->{sig_pipe};