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 $chatty = 1; # disable for production
33 use constant STATUS_PIPE_DATA_SIZE => 12;
36 my($class, $service, %args) = @_;
37 my $self = bless(\%args, $class);
39 $self->{service} = $service; # service name
40 $self->{num_children} = 0; # number of child processes
41 $self->{osrf_handle} = undef; # xmpp handle
42 $self->{routers} = []; # list of registered routers
43 $self->{active_list} = []; # list of active children
44 $self->{idle_list} = []; # list of idle children
45 $self->{pid_map} = {}; # map of child pid to child for cleaner access
46 $self->{sig_pipe} = 0; # true if last syswrite failed
48 $self->{stderr_log} = $self->{stderr_log_path} . "/${service}_stderr.log"
49 if $self->{stderr_log_path};
51 $self->{min_spare_children} ||= 0;
53 $self->{max_spare_children} = $self->{min_spare_children} + 1 if
54 $self->{max_spare_children} and
55 $self->{max_spare_children} <= $self->{min_spare_children};
60 # ----------------------------------------------------------------
61 # Disconnects from routers and waits for child processes to exit.
62 # ----------------------------------------------------------------
67 $logger->info("server: shutting down and cleaning up...");
69 # don't get sidetracked by signals while we're cleaning up.
70 # it could result in unexpected behavior with list traversal
71 $SIG{CHLD} = 'IGNORE';
73 # terminate the child processes
74 $self->kill_child($_) for
75 (@{$self->{idle_list}}, @{$self->{active_list}});
78 $self->unregister_routers;
80 $self->{osrf_handle}->disconnect;
82 # clean up our dead children
83 $self->reap_children(1);
85 exit(0) unless $no_exit;
89 # ----------------------------------------------------------------
90 # Waits on the jabber socket for inbound data from the router.
91 # Each new message is passed off to a child process for handling.
92 # At regular intervals, wake up for min/max spare child maintenance
93 # ----------------------------------------------------------------
97 $logger->set_service($self->{service});
99 $SIG{$_} = sub { $self->cleanup; } for (qw/INT TERM QUIT/);
100 $SIG{CHLD} = sub { $self->reap_children(); };
102 $self->spawn_children;
103 $self->build_osrf_handle;
104 $self->register_routers;
111 $self->{child_died} = 0;
113 my $msg = $self->{osrf_handle}->process($wait_time);
115 # we woke up for any reason, reset the wait time to allow
116 # for idle maintenance as necessary
121 if(my $child = pop(@{$self->{idle_list}})) {
123 # we have an idle child to handle the request
124 $chatty and $logger->internal("server: passing request to idle child $child");
125 push(@{$self->{active_list}}, $child);
126 $self->write_child($child, $msg);
128 } elsif($self->{num_children} < $self->{max_children}) {
130 # spawning a child to handle the request
131 $chatty and $logger->internal("server: spawning child to handle request");
132 $self->write_child($self->spawn_child(1), $msg);
135 $logger->warn("server: no children available, waiting... consider increasing " .
136 "max_children for this application higher than $self->{max_children} ".
137 "in the OpenSRF configuration if this message occurs frequently");
138 $self->check_status(1); # block until child is available
140 my $child = pop(@{$self->{idle_list}});
141 push(@{$self->{active_list}}, $child);
142 $self->write_child($child, $msg);
147 # don't perform idle maint immediately when woken by SIGCHLD
148 unless($self->{child_died}) {
150 # when we hit equilibrium, there's no need for regular
151 # maintenance, so set wait_time to 'forever'
153 !$self->perform_idle_maintenance and # no maintenance performed this time
154 @{$self->{active_list}} == 0; # no active children
160 # ----------------------------------------------------------------
161 # Launch a new spare child or kill an extra spare child. To
162 # prevent large-scale spawning or die-offs, spawn or kill only
163 # 1 process per idle maintenance loop.
164 # Returns true if any idle maintenance occurred, 0 otherwise
165 # ----------------------------------------------------------------
166 sub perform_idle_maintenance {
169 $chatty and $logger->internal(
171 "server: %d idle, %d active, %d min_spare, %d max_spare in idle maintenance",
172 scalar(@{$self->{idle_list}}),
173 scalar(@{$self->{active_list}}),
174 $self->{min_spare_children},
175 $self->{max_spare_children}
179 # spawn 1 spare child per maintenance loop if necessary
180 if( $self->{min_spare_children} and
181 $self->{num_children} < $self->{max_children} and
182 scalar(@{$self->{idle_list}}) < $self->{min_spare_children} ) {
184 $chatty and $logger->internal("server: spawning spare child");
188 # kill 1 excess spare child per maintenance loop if necessary
189 } elsif($self->{max_spare_children} and
190 $self->{num_children} > $self->{min_children} and
191 scalar(@{$self->{idle_list}}) > $self->{max_spare_children} ) {
193 $chatty and $logger->internal("server: killing spare child");
203 my $child = shift || pop(@{$self->{idle_list}}) or return;
204 $chatty and $logger->internal("server: killing child $child");
205 kill('TERM', $child->{pid});
208 # ----------------------------------------------------------------
209 # Jabber connection inbound message arrive on.
210 # ----------------------------------------------------------------
211 sub build_osrf_handle {
214 my $conf = OpenSRF::Utils::Config->current;
215 my $username = $conf->bootstrap->username;
216 my $password = $conf->bootstrap->passwd;
217 my $domain = $conf->bootstrap->domain;
218 my $port = $conf->bootstrap->port;
219 my $resource = $self->{service} . '_listener_' . $conf->env->hostname;
221 $logger->debug("server: inbound connecting as $username\@$domain/$resource on port $port");
223 $self->{osrf_handle} =
224 OpenSRF::Transport::SlimJabber::Client->new(
225 username => $username,
226 resource => $resource,
227 password => $password,
232 $self->{osrf_handle}->initialize;
236 # ----------------------------------------------------------------
237 # Sends request data to a child process
238 # ----------------------------------------------------------------
240 my($self, $child, $msg) = @_;
241 my $xml = encode_utf8(decode_utf8($msg->to_xml));
245 $self->{sig_pipe} = 0;
246 local $SIG{'PIPE'} = sub { $self->{sig_pipe} = 1; };
248 # send message to child data pipe
249 syswrite($child->{pipe_to_child}, $xml);
251 last unless $self->{sig_pipe};
252 $logger->error("server: got SIGPIPE writing to $child, retrying...");
253 usleep(50000); # 50 msec
256 $logger->error("server: unable to send request message to child $child") if $self->{sig_pipe};
259 # ----------------------------------------------------------------
260 # Checks to see if any child process has reported its availability
261 # In blocking mode, blocks until a child has reported.
262 # ----------------------------------------------------------------
264 my($self, $block) = @_;
266 return unless @{$self->{active_list}};
268 my $read_set = IO::Select->new;
269 $read_set->add($_->{pipe_to_child}) for @{$self->{active_list}};
275 # if can_read or sysread is interrupted while bloking, go back and
276 # wait again until we have at least 1 free child
278 if(my @handles = $read_set->can_read(($block) ? undef : 0)) {
280 for my $pipe (@handles) {
281 sysread($pipe, $pid, STATUS_PIPE_DATA_SIZE) or next;
282 push(@pids, int($pid));
286 last unless $block and !@pids;
291 $chatty and $logger->internal("server: ".scalar(@pids)." children reporting for duty: (@pids)");
296 # move the children from the active list to the idle list
297 for my $proc (@{$self->{active_list}}) {
298 if(grep { $_ == $proc->{pid} } @pids) {
299 push(@{$self->{idle_list}}, $proc);
301 push(@new_actives, $proc);
305 $self->{active_list} = [@new_actives];
307 $chatty and $logger->internal(sprintf(
308 "server: %d idle and %d active children after status update",
309 scalar(@{$self->{idle_list}}), scalar(@{$self->{active_list}})));
312 # ----------------------------------------------------------------
313 # Cleans up any child processes that have exited.
314 # In shutdown mode, block until all children have washed ashore
315 # ----------------------------------------------------------------
317 my($self, $shutdown) = @_;
318 $self->{child_died} = 1;
322 my $pid = waitpid(-1, ($shutdown) ? 0 : WNOHANG);
325 $chatty and $logger->internal("server: reaping child $pid");
327 my $child = $self->{pid_map}->{$pid};
329 close($child->{pipe_to_parent});
330 close($child->{pipe_to_child});
332 $self->{active_list} = [ grep { $_->{pid} != $pid } @{$self->{active_list}} ];
333 $self->{idle_list} = [ grep { $_->{pid} != $pid } @{$self->{idle_list}} ];
335 $self->{num_children}--;
336 delete $self->{pid_map}->{$pid};
337 delete $child->{$_} for keys %$child; # destroy with a vengeance
340 $self->spawn_children unless $shutdown;
342 $chatty and $logger->internal(sprintf(
343 "server: %d idle and %d active children after reap_children",
344 scalar(@{$self->{idle_list}}), scalar(@{$self->{active_list}})));
348 # ----------------------------------------------------------------
349 # Spawn up to max_children processes
350 # ----------------------------------------------------------------
353 $self->spawn_child while $self->{num_children} < $self->{min_children};
356 # ----------------------------------------------------------------
357 # Spawns a new child. If $active is set, the child goes directly
358 # into the active_list.
359 # ----------------------------------------------------------------
361 my($self, $active) = @_;
363 my $child = OpenSRF::Server::Child->new($self);
365 # socket for sending message data to the child
367 $child->{pipe_to_child},
368 $child->{pipe_to_parent},
369 AF_UNIX, SOCK_STREAM, PF_UNSPEC)) {
370 $logger->error("server: error creating data socketpair: $!");
374 $child->{pipe_to_child}->autoflush(1);
375 $child->{pipe_to_parent}->autoflush(1);
377 $child->{pid} = fork();
379 if($child->{pid}) { # parent process
380 $self->{num_children}++;
381 $self->{pid_map}->{$child->{pid}} = $child;
384 push(@{$self->{active_list}}, $child);
386 push(@{$self->{idle_list}}, $child);
389 $chatty and $logger->internal("server: server spawned child $child with ".$self->{num_children}." total children");
393 } else { # child process
395 $SIG{$_} = 'DEFAULT' for (qw/INT TERM QUIT HUP/);
397 if($self->{stderr_log}) {
399 $chatty and $logger->internal("server: redirecting STDERR to " . $self->{stderr_log});
402 unless( open(STDERR, '>>' . $self->{stderr_log}) ) {
403 $logger->error("server: unable to open STDERR log file: " . $self->{stderr_log} . " : $@");
404 open STDERR, '>/dev/null'; # send it back to /dev/null
412 OpenSRF::Transport::PeerHandle->retrieve->disconnect;
414 $logger->error("server: child process died: $@") if $@;
419 # ----------------------------------------------------------------
420 # Sends the register command to the configured routers
421 # ----------------------------------------------------------------
422 sub register_routers {
425 my $conf = OpenSRF::Utils::Config->current;
426 my $routers = $conf->bootstrap->routers;
427 my $router_name = $conf->bootstrap->router_name;
430 for my $router (@$routers) {
433 if( !$router->{services} ||
434 !$router->{services}->{service} ||
436 ref($router->{services}->{service}) eq 'ARRAY' and
437 grep { $_ eq $self->{service} } @{$router->{services}->{service}}
438 ) || $router->{services}->{service} eq $self->{service}) {
440 my $name = $router->{name};
441 my $domain = $router->{domain};
442 push(@targets, "$name\@$domain/router");
446 push(@targets, "$router_name\@$router/router");
451 $logger->info("server: registering with router $_");
452 $self->{osrf_handle}->send(
454 body => 'registering',
455 router_command => 'register',
456 router_class => $self->{service}
460 $self->{routers} = \@targets;
463 # ----------------------------------------------------------------
464 # Sends the unregister command to any routers we have registered
466 # ----------------------------------------------------------------
467 sub unregister_routers {
469 return unless $self->{osrf_handle}->tcp_connected;
471 for my $router (@{$self->{routers}}) {
472 $logger->info("server: disconnecting from router $router");
473 $self->{osrf_handle}->send(
475 body => "unregistering",
476 router_command => "unregister",
477 router_class => $self->{service}
483 package OpenSRF::Server::Child;
486 use OpenSRF::Transport;
487 use OpenSRF::Application;
488 use OpenSRF::Transport::PeerHandle;
489 use OpenSRF::Transport::SlimJabber::XMPPMessage;
490 use OpenSRF::Utils::Logger qw($logger);
491 use OpenSRF::DomainObject::oilsResponse qw/:status/;
492 use Fcntl qw(F_GETFL F_SETFL O_NONBLOCK);
493 use Time::HiRes qw(time usleep);
494 use POSIX qw/:sys_wait_h :errno_h/;
496 use overload '""' => sub { return '[' . shift()->{pid} . ']'; };
499 my($class, $parent) = @_;
500 my $self = bless({}, $class);
501 $self->{pid} = 0; # my process ID
502 $self->{parent} = $parent; # Controller parent process
503 $self->{num_requests} = 0; # total serviced requests
504 $self->{sig_pipe} = 0; # true if last syswrite failed
510 my $flags = fcntl($fh, F_GETFL, 0);
511 fcntl($fh, F_SETFL, $flags | O_NONBLOCK);
516 my $flags = fcntl($fh, F_GETFL, 0);
517 $flags &= ~O_NONBLOCK;
518 fcntl($fh, F_SETFL, $flags);
521 # ----------------------------------------------------------------
522 # Connects to Jabber and runs the application child_init
523 # ----------------------------------------------------------------
526 my $service = $self->{parent}->{service};
527 $0 = "OpenSRF Drone [$service]";
528 OpenSRF::Transport::PeerHandle->construct($service);
529 OpenSRF::Application->application_implementation->child_init
530 if (OpenSRF::Application->application_implementation->can('child_init'));
533 # ----------------------------------------------------------------
534 # Waits for messages from the parent process, handles the message,
535 # then goes into the keepalive loop if this is a stateful session.
536 # When max_requests is hit, the process exits.
537 # ----------------------------------------------------------------
540 my $network = OpenSRF::Transport::PeerHandle->retrieve;
542 # main child run loop. Ends when this child hits max requests.
545 my $data = $self->wait_for_request or next;
547 # Update process name to show activity
551 # Discard extraneous data from the jabber socket
552 if(!$network->flush_socket()) {
553 $logger->error("server: network disconnected! child dropping request and exiting: $data");
557 my $session = OpenSRF::Transport->handler(
558 $self->{parent}->{service},
559 OpenSRF::Transport::SlimJabber::XMPPMessage->new(xml => $data)
562 $self->keepalive_loop($session);
564 last if ++$self->{num_requests} == $self->{parent}->{max_requests};
566 # Tell the parent process we are available to process requests
569 # Repair process name
573 $chatty and $logger->internal("server: child process shutting down after reaching max_requests");
575 OpenSRF::Application->application_implementation->child_exit
576 if (OpenSRF::Application->application_implementation->can('child_exit'));
579 # ----------------------------------------------------------------
580 # waits for a request data on the parent pipe and returns it.
581 # ----------------------------------------------------------------
582 sub wait_for_request {
586 my $read_size = 1024;
590 # Start out blocking, when data is available, read it all
593 local $SIG{'PIPE'} = sub { $sig_pipe = 1 };
596 my $n = sysread($self->{pipe_to_parent}, $buf, $read_size);
601 $logger->info("server: $self got SIGPIPE reading data from parent, retrying...");
602 usleep(50000); # 50 msec
606 $logger->error("server: error reading data pipe: $!") unless EAGAIN == $!;
610 last if $n <= 0; # no data left to read
614 last if $n < $read_size; # done reading all data
616 $self->set_nonblock($self->{pipe_to_parent}) unless $nonblock;
620 $self->set_block($self->{pipe_to_parent}) if $nonblock;
625 # ----------------------------------------------------------------
626 # If this is a stateful opensrf session, wait up to $keepalive
627 # seconds for subsequent requests from the client
628 # ----------------------------------------------------------------
630 my($self, $session) = @_;
631 my $keepalive = $self->{parent}->{keepalive};
633 while($session->state and $session->state == $session->CONNECTED) {
635 unless( $session->queue_wait($keepalive) ) {
637 # client failed to disconnect before timeout
638 $logger->info("server: no request was received in $keepalive seconds, exiting stateful session");
640 my $res = OpenSRF::DomainObject::oilsConnectStatus->new(
641 status => "Disconnected on timeout",
642 statusCode => STATUS_TIMEOUT
645 $session->status($res);
646 $session->state($session->DISCONNECTED);
651 $chatty and $logger->internal("server: child done with request(s)");
655 # ----------------------------------------------------------------
656 # Report our availability to our parent process
657 # ----------------------------------------------------------------
663 $self->{sig_pipe} = 0;
664 local $SIG{'PIPE'} = sub { $self->{sig_pipe} = 1; };
667 $self->{pipe_to_parent},
668 sprintf("%*s", OpenSRF::Server::STATUS_PIPE_DATA_SIZE, $self->{pid})
671 last unless $self->{sig_pipe};
672 $logger->error("server: $self got SIGPIPE writing status to parent, retrying...");
673 usleep(50000); # 50 msec
676 $logger->error("server: $self unable to send status to parent") if $self->{sig_pipe};