3 @brief Spawn and manage a collection of child process to service requests.
5 Spawn a collection of child processes, replacing them as needed. Forward requests to them
6 and let the children do the work.
8 Each child processes some maximum number of requests before it terminates itself. When a
9 child dies, either deliberately or otherwise, we can spawn another one to replace it,
10 keeping the number of children within a predefined range.
12 Use a doubly-linked circular list to keep track of the children to whom we have forwarded
13 a request, and who are still working on them. Use a separate linear linked list to keep
14 track of children that are currently idle. Move them back and forth as needed.
16 For each child, set up two pipes:
17 - One for the parent to send requests to the child.
18 - One for the child to notify the parent that it is available for another request.
20 The message sent to the child represents an XML stanza as received from Jabber.
22 When the child finishes processing the request, it writes the string "available" back
23 to the parent. Then the parent knows that it can send that child another request.
28 #include <sys/types.h>
34 #include <sys/select.h>
37 #include "opensrf/utils.h"
38 #include "opensrf/log.h"
39 #include "opensrf/transport_client.h"
40 #include "opensrf/osrf_stack.h"
41 #include "opensrf/osrf_settings.h"
42 #include "opensrf/osrf_application.h"
44 #define READ_BUFSIZE 1024
45 #define ABS_MAX_CHILDREN 256
48 int max_requests; /**< How many requests a child processes before terminating. */
49 int min_children; /**< Minimum number of children to maintain. */
50 int max_children; /**< Maximum number of children to maintain. */
51 int fd; /**< Unused. */
52 int data_to_child; /**< Unused. */
53 int data_to_parent; /**< Unused. */
54 int current_num_children; /**< How many children are currently on the list. */
55 int keepalive; /**< Keepalive time for stateful sessions. */
56 char* appname; /**< Name of the application. */
57 /** Points to a circular linked list of children. */
58 struct prefork_child_struct* first_child;
59 /** List of of child processes that aren't doing anything at the moment and are
60 therefore available to service a new request. */
61 struct prefork_child_struct* idle_list;
62 /** List of allocated but unused prefork_children, available for reuse. Each one is just
63 raw memory, apart from the "next" pointer used to stitch them together. In particular,
64 there is no child process for them, and the file descriptors are not open. */
65 struct prefork_child_struct* free_list;
66 transport_client* connection; /**< Connection to Jabber. */
69 struct prefork_child_struct {
70 pid_t pid; /**< Process ID of the child. */
71 int read_data_fd; /**< Child uses to read request. */
72 int write_data_fd; /**< Parent uses to write request. */
73 int read_status_fd; /**< Parent reads to see if child is available. */
74 int write_status_fd; /**< Child uses to notify parent when it's available again. */
75 int max_requests; /**< How many requests a child can process before terminating. */
76 const char* appname; /**< Name of the application. */
77 int keepalive; /**< Keepalive time for stateful sessions. */
78 struct prefork_child_struct* next; /**< Linkage pointer for linked list. */
79 struct prefork_child_struct* prev; /**< Linkage pointer for linked list. */
82 typedef struct prefork_child_struct prefork_child;
84 /** Boolean. Set to true by a signal handler when it traps SIGCHLD. */
85 static volatile sig_atomic_t child_dead;
87 static int prefork_simple_init( prefork_simple* prefork, transport_client* client,
88 int max_requests, int min_children, int max_children );
89 static prefork_child* launch_child( prefork_simple* forker );
90 static void prefork_launch_children( prefork_simple* forker );
91 static void prefork_run( prefork_simple* forker );
92 static void add_prefork_child( prefork_simple* forker, prefork_child* child );
94 static void del_prefork_child( prefork_simple* forker, pid_t pid );
95 static int check_children( prefork_simple* forker, int forever );
96 static int prefork_child_process_request( prefork_child*, char* data );
97 static int prefork_child_init_hook( prefork_child* );
98 static prefork_child* prefork_child_init( prefork_simple* forker,
99 int read_data_fd, int write_data_fd,
100 int read_status_fd, int write_status_fd );
102 /* listens on the 'data_to_child' fd and wait for incoming data */
103 static void prefork_child_wait( prefork_child* child );
104 static void prefork_clear( prefork_simple*, bool graceful);
105 static void prefork_child_free( prefork_simple* forker, prefork_child* );
106 static void osrf_prefork_register_routers( const char* appname, bool unregister );
107 static void osrf_prefork_child_exit( prefork_child* );
109 static void sigchld_handler( int sig );
110 static void sigusr1_handler( int sig );
111 static void sigusr2_handler( int sig );
112 static void sigterm_handler( int sig );
113 static void sigint_handler( int sig );
115 /** Maintain a global pointer to the prefork_simple object
116 * for the current process so we can refer to it later
117 * for signal handling. There will only ever be one
118 * forker per process.
120 static prefork_simple *global_forker = NULL;
123 @brief Spawn and manage a collection of drone processes for servicing requests.
124 @param appname Name of the application.
125 @return 0 if successful, or -1 if error.
127 int osrf_prefork_run( const char* appname ) {
130 osrfLogError( OSRF_LOG_MARK, "osrf_prefork_run requires an appname to run!");
134 set_proc_title( "OpenSRF Listener [%s]", appname );
141 // Get configuration settings
142 osrfLogInfo( OSRF_LOG_MARK, "Loading config in osrf_forker for app %s", appname );
144 char* max_req = osrf_settings_host_value( "/apps/%s/unix_config/max_requests", appname );
145 char* min_children = osrf_settings_host_value( "/apps/%s/unix_config/min_children", appname );
146 char* max_children = osrf_settings_host_value( "/apps/%s/unix_config/max_children", appname );
147 char* keepalive = osrf_settings_host_value( "/apps/%s/keepalive", appname );
150 osrfLogWarning( OSRF_LOG_MARK, "Keepalive is not defined, assuming %d", kalive );
152 kalive = atoi( keepalive );
155 osrfLogWarning( OSRF_LOG_MARK, "Max requests not defined, assuming %d", maxr );
157 maxr = atoi( max_req );
160 osrfLogWarning( OSRF_LOG_MARK, "Min children not defined, assuming %d", minc );
162 minc = atoi( min_children );
165 osrfLogWarning( OSRF_LOG_MARK, "Max children not defined, assuming %d", maxc );
167 maxc = atoi( max_children );
171 free( min_children );
172 free( max_children );
173 /* --------------------------------------------------- */
175 char* resc = va_list_to_string( "%s_listener", appname );
177 // Make sure that we haven't already booted
178 if( !osrfSystemBootstrapClientResc( NULL, NULL, resc )) {
179 osrfLogError( OSRF_LOG_MARK, "Unable to bootstrap client for osrf_prefork_run()" );
186 prefork_simple forker;
188 if( prefork_simple_init( &forker, osrfSystemGetTransportClient(), maxr, minc, maxc )) {
189 osrfLogError( OSRF_LOG_MARK,
190 "osrf_prefork_run() failed to create prefork_simple object" );
194 // Finish initializing the prefork_simple.
195 forker.appname = strdup( appname );
196 forker.keepalive = kalive;
197 global_forker = &forker;
199 // Spawn the children; put them in the idle list.
200 prefork_launch_children( &forker );
202 // Tell the router that you're open for business.
203 osrf_prefork_register_routers( appname, false );
205 signal( SIGUSR1, sigusr1_handler);
206 signal( SIGUSR2, sigusr2_handler);
207 signal( SIGTERM, sigterm_handler);
208 signal( SIGINT, sigint_handler );
209 signal( SIGQUIT, sigint_handler );
211 // Sit back and let the requests roll in
212 osrfLogInfo( OSRF_LOG_MARK, "Launching osrf_forker for app %s", appname );
213 prefork_run( &forker );
215 osrfLogWarning( OSRF_LOG_MARK, "prefork_run() returned - how??" );
216 prefork_clear( &forker, false );
221 @brief Register the application with a specified router.
222 @param appname Name of the application.
223 @param routerName Name of the router.
224 @param routerDomain Domain of the router.
226 Tell the router that you're open for business so that it can route requests to you.
228 Called only by the parent process.
230 static void osrf_prefork_send_router_registration(
231 const char* appname, const char* routerName,
232 const char* routerDomain, bool unregister ) {
234 // Get a pointer to the global transport_client
235 transport_client* client = osrfSystemGetTransportClient();
237 // Construct the Jabber address of the router
238 char* jid = va_list_to_string( "%s@%s/router", routerName, routerDomain );
240 // Create the registration message, and send it
241 transport_message* msg;
244 osrfLogInfo( OSRF_LOG_MARK, "%s un-registering with router %s", appname, jid );
245 msg = message_init( "unregistering", NULL, NULL, jid, NULL );
246 message_set_router_info( msg, NULL, NULL, appname, "unregister", 0 );
250 osrfLogInfo( OSRF_LOG_MARK, "%s registering with router %s", appname, jid );
251 msg = message_init( "registering", NULL, NULL, jid, NULL );
252 message_set_router_info( msg, NULL, NULL, appname, "register", 0 );
255 client_send_message( client, msg );
263 @brief Register with a router, or not, according to some config settings.
264 @param appname Name of the application
265 @param RouterChunk A representation of part of the config file.
267 Parse a "complex" router configuration chunk.
269 Examine the services listed for a given router (normally in opensrf_core.xml). If
270 there is an entry for this service, or if there are @em no services listed, then
271 register with this router. Otherwise don't.
273 Called only by the parent process.
275 static void osrf_prefork_parse_router_chunk(
276 const char* appname, const jsonObject* routerChunk, bool unregister ) {
278 const char* routerName = jsonObjectGetString( jsonObjectGetKeyConst( routerChunk, "name" ));
279 const char* domain = jsonObjectGetString( jsonObjectGetKeyConst( routerChunk, "domain" ));
280 const jsonObject* services = jsonObjectGetKeyConst( routerChunk, "services" );
281 osrfLogDebug( OSRF_LOG_MARK, "found router config with domain %s and name %s",
282 routerName, domain );
284 if( services && services->type == JSON_HASH ) {
285 osrfLogDebug( OSRF_LOG_MARK, "investigating router information..." );
286 const jsonObject* service_obj = jsonObjectGetKeyConst( services, "service" );
288 ; // do nothing (shouldn't happen)
289 else if( JSON_ARRAY == service_obj->type ) {
290 // There are multiple services listed. Register with this router
291 // if and only if this service is on the list.
293 for( j = 0; j < service_obj->size; j++ ) {
294 const char* service = jsonObjectGetString( jsonObjectGetIndex( service_obj, j ));
295 if( service && !strcmp( appname, service ))
296 osrf_prefork_send_router_registration( appname, routerName, domain, unregister );
299 else if( JSON_STRING == service_obj->type ) {
300 // There's only one service listed. Register with this router
301 // if and only if this service is the one listed.
302 if( !strcmp( appname, jsonObjectGetString( service_obj )) )
303 osrf_prefork_send_router_registration( appname, routerName, domain, unregister );
306 // This router is not restricted to any set of services,
307 // so go ahead and register with it.
308 osrf_prefork_send_router_registration( appname, routerName, domain, unregister );
313 @brief Register the application with one or more routers, according to the configuration.
314 @param appname Name of the application.
316 Called only by the parent process.
318 static void osrf_prefork_register_routers( const char* appname, bool unregister ) {
320 jsonObject* routerInfo = osrfConfigGetValueObject( NULL, "/routers/router" );
323 for( i = 0; i < routerInfo->size; i++ ) {
324 const jsonObject* routerChunk = jsonObjectGetIndex( routerInfo, i );
326 if( routerChunk->type == JSON_STRING ) {
327 /* this accomodates simple router configs */
328 char* routerName = osrfConfigGetValue( NULL, "/router_name" );
329 char* domain = osrfConfigGetValue( NULL, "/routers/router" );
330 osrfLogDebug( OSRF_LOG_MARK, "found simple router settings with router name %s",
332 osrf_prefork_send_router_registration( appname, routerName, domain, unregister );
337 osrf_prefork_parse_router_chunk( appname, routerChunk, unregister );
341 jsonObjectFree( routerInfo );
345 @brief Initialize a child process.
346 @param child Pointer to the prefork_child representing the new child process.
347 @return Zero if successful, or -1 if not.
349 Called only by child processes. Actions:
350 - Connect to one or more cache servers
351 - Reconfigure logger, if necessary
352 - Discard parent's Jabber connection and open a new one
353 - Dynamically call an application-specific initialization routine
354 - Change the command line as reported by ps
356 static int prefork_child_init_hook( prefork_child* child ) {
358 if( !child ) return -1;
359 osrfLogDebug( OSRF_LOG_MARK, "Child init hook for child %d", child->pid );
361 // Connect to cache server(s).
362 osrfSystemInitCache();
363 char* resc = va_list_to_string( "%s_drone", child->appname );
365 // If we're a source-client, tell the logger now that we're a new process.
366 char* isclient = osrfConfigGetValue( NULL, "/client" );
367 if( isclient && !strcasecmp( isclient,"true" ))
368 osrfLogSetIsClient( 1 );
371 // Remove traces of our parent's socket connection so we can have our own.
372 osrfSystemIgnoreTransportClient();
375 if( !osrfSystemBootstrapClientResc( NULL, NULL, resc )) {
376 osrfLogError( OSRF_LOG_MARK, "Unable to bootstrap client for osrf_prefork_run()" );
383 // Dynamically call the application-specific initialization function
384 // from a previously loaded shared library.
385 if( ! osrfAppRunChildInit( child->appname )) {
386 osrfLogDebug( OSRF_LOG_MARK, "Prefork child_init succeeded\n" );
388 osrfLogError( OSRF_LOG_MARK, "Prefork child_init failed\n" );
392 // Change the command line as reported by ps
393 set_proc_title( "OpenSRF Drone [%s]", child->appname );
398 @brief Respond to a client request forwarded by the parent.
399 @param child Pointer to the state of the child process.
400 @param data Pointer to the raw XMPP message received from the parent.
401 @return 0 on success; non-zero means that the child process should clean itself up
402 and terminate immediately, presumably due to a fatal error condition.
404 Called only by a child process.
406 static int prefork_child_process_request( prefork_child* child, char* data ) {
407 if( !child ) return 0;
409 transport_client* client = osrfSystemGetTransportClient();
411 // Make sure that we're still connected to Jabber; reconnect if necessary.
412 if( !client_connected( client )) {
413 osrfSystemIgnoreTransportClient();
414 osrfLogWarning( OSRF_LOG_MARK, "Reconnecting child to opensrf after disconnect..." );
415 if( !osrf_system_bootstrap_client( NULL, NULL )) {
416 osrfLogError( OSRF_LOG_MARK,
417 "Unable to bootstrap client in prefork_child_process_request()" );
419 osrf_prefork_child_exit( child );
423 // Construct the message from the xml.
424 transport_message* msg = new_message_from_xml( data );
426 // Respond to the transport message. This is where method calls are buried.
427 osrfAppSession* session = osrf_stack_transport_handler( msg, child->appname );
431 int rc = session->panic;
434 osrfLogWarning( OSRF_LOG_MARK,
435 "Drone for session %s terminating immediately", session->session_id );
436 osrfAppSessionFree( session );
440 if( session->stateless && session->state != OSRF_SESSION_CONNECTED ) {
441 // We're no longer connected to the client, which presumably means that
442 // we're done with this request. Bail out.
443 osrfAppSessionFree( session );
447 // If we get this far, then the client has opened an application connection so that it
448 // can send multiple requests directly to the same server drone, bypassing the router
449 // and the listener. For example, it may need to do a database transaction, requiring
450 // multiple method calls within the same database session.
452 // Hence we go into a loop, responding to successive requests from the same client, until
453 // either the client disconnects or an error occurs.
455 osrfLogDebug( OSRF_LOG_MARK, "Entering keepalive loop for session %s", session->session_id );
456 int keepalive = child->keepalive;
464 // Respond to any input messages. This is where the method calls are buried.
465 osrfLogDebug( OSRF_LOG_MARK,
466 "osrf_prefork calling queue_wait [%d] in keepalive loop", keepalive );
467 start = time( NULL );
468 retval = osrf_app_session_queue_wait( session, keepalive, &recvd );
471 osrfLogDebug( OSRF_LOG_MARK, "Data received == %d", recvd );
473 // Now we check a number of possible reasons to exit the loop.
475 // If the method call decided to terminate immediately,
476 // note that for future reference.
480 // If an error occurred when we tried to service the request, exit the loop.
482 osrfLogError( OSRF_LOG_MARK, "queue-wait returned non-success %d", retval );
486 // If the client disconnected, exit the loop.
487 if( session->state != OSRF_SESSION_CONNECTED )
490 // If we timed out while waiting for a request, exit the loop.
491 if( !recvd && (end - start) >= keepalive ) {
492 osrfLogInfo( OSRF_LOG_MARK,
493 "No request was received in %d seconds, exiting stateful session", keepalive );
494 osrfAppSessionStatus(
498 0, "Disconnected on timeout" );
503 // If the child process has decided to terminate immediately, exit the loop.
508 osrfLogDebug( OSRF_LOG_MARK, "Exiting keepalive loop for session %s", session->session_id );
509 osrfAppSessionFree( session );
514 @brief Partially initialize a prefork_simple provided by the caller.
515 @param prefork Pointer to a a raw prefork_simple to be initialized.
516 @param client Pointer to a transport_client (connection to Jabber).
517 @param max_requests The maximum number of requests that a child process may service
519 @param min_children Minimum number of child processes to maintain.
520 @param max_children Maximum number of child processes to maintain.
521 @return 0 if successful, or 1 if not (due to invalid parameters).
523 static int prefork_simple_init( prefork_simple* prefork, transport_client* client,
524 int max_requests, int min_children, int max_children ) {
526 if( min_children > max_children ) {
527 osrfLogError( OSRF_LOG_MARK, "min_children (%d) is greater "
528 "than max_children (%d)", min_children, max_children );
532 if( max_children > ABS_MAX_CHILDREN ) {
533 osrfLogError( OSRF_LOG_MARK, "max_children (%d) is greater than ABS_MAX_CHILDREN (%d)",
534 max_children, ABS_MAX_CHILDREN );
538 osrfLogInfo( OSRF_LOG_MARK, "Prefork launching child with max_request=%d,"
539 "min_children=%d, max_children=%d", max_requests, min_children, max_children );
541 /* flesh out the struct */
542 prefork->max_requests = max_requests;
543 prefork->min_children = min_children;
544 prefork->max_children = max_children;
546 prefork->data_to_child = 0;
547 prefork->data_to_parent = 0;
548 prefork->current_num_children = 0;
549 prefork->keepalive = 0;
550 prefork->appname = NULL;
551 prefork->first_child = NULL;
552 prefork->idle_list = NULL;
553 prefork->free_list = NULL;
554 prefork->connection = client;
560 @brief Spawn a new child process and put it in the idle list.
561 @param forker Pointer to the prefork_simple that will own the process.
562 @return Pointer to the new prefork_child, or not at all.
564 Spawn a new child process. Create a prefork_child for it and put it in the idle list.
566 After forking, the parent returns a pointer to the new prefork_child. The child
567 services its quota of requests and then terminates without returning.
569 static prefork_child* launch_child( prefork_simple* forker ) {
575 // Set up the data and status pipes
576 if( pipe( data_fd ) < 0 ) { /* build the data pipe*/
577 osrfLogError( OSRF_LOG_MARK, "Pipe making error" );
581 if( pipe( status_fd ) < 0 ) {/* build the status pipe */
582 osrfLogError( OSRF_LOG_MARK, "Pipe making error" );
588 osrfLogInternal( OSRF_LOG_MARK, "Pipes: %d %d %d %d",
589 data_fd[0], data_fd[1], status_fd[0], status_fd[1] );
591 // Create and initialize a prefork_child for the new process
592 prefork_child* child = prefork_child_init( forker, data_fd[0],
593 data_fd[1], status_fd[0], status_fd[1] );
595 if( (pid=fork()) < 0 ) {
596 osrfLogError( OSRF_LOG_MARK, "Forking Error" );
597 prefork_child_free( forker, child );
601 // Add the new child to the head of the idle list
602 child->next = forker->idle_list;
603 forker->idle_list = child;
605 if( pid > 0 ) { /* parent */
607 signal( SIGCHLD, sigchld_handler );
608 ( forker->current_num_children )++;
611 osrfLogDebug( OSRF_LOG_MARK, "Parent launched %d", pid );
612 /* *no* child pipe FD's can be closed or the parent will re-use fd's that
613 the children are currently using */
619 // we don't want to adopt our parent's handlers.
620 signal( SIGUSR1, SIG_DFL );
621 signal( SIGUSR2, SIG_DFL );
622 signal( SIGTERM, SIG_DFL );
623 signal( SIGINT, SIG_DFL );
624 signal( SIGQUIT, SIG_DFL );
625 signal( SIGCHLD, SIG_DFL );
627 osrfLogInternal( OSRF_LOG_MARK,
628 "I am new child with read_data_fd = %d and write_status_fd = %d",
629 child->read_data_fd, child->write_status_fd );
631 child->pid = getpid();
632 close( child->write_data_fd );
633 close( child->read_status_fd );
636 if( prefork_child_init_hook( child ) == -1 ) {
637 osrfLogError( OSRF_LOG_MARK,
638 "Forker child going away because we could not connect to OpenSRF..." );
639 osrf_prefork_child_exit( child );
642 prefork_child_wait( child ); // Should exit without returning
643 osrf_prefork_child_exit( child ); // Just to be sure
644 return NULL; // Unreachable, but it keeps the compiler happy
649 @brief Terminate a child process.
650 @param child Pointer to the prefork_child representing the child process (not used).
652 Called only by child processes. Dynamically call an application-specific shutdown
653 function from a previously loaded shared library; then exit.
655 static void osrf_prefork_child_exit( prefork_child* child ) {
656 osrfAppRunExitCode();
661 @brief Launch all the child processes, putting them in the idle list.
662 @param forker Pointer to the prefork_simple that will own the children.
664 Called only by the parent process (in order to become a parent).
666 static void prefork_launch_children( prefork_simple* forker ) {
667 if( !forker ) return;
669 while( c++ < forker->min_children )
670 launch_child( forker );
674 @brief Signal handler for SIGCHLD: note that a child process has terminated.
675 @param sig The value of the trapped signal; always SIGCHLD.
677 Set a boolean to be checked later.
679 static void sigchld_handler( int sig ) {
680 signal( SIGCHLD, sigchld_handler );
685 @brief Signal handler for SIGUSR1
686 @param sig The value of the trapped signal; always SIGUSR1.
688 Send unregister command to all registered routers.
690 static void sigusr1_handler( int sig ) {
691 if (!global_forker) return;
692 osrf_prefork_register_routers(global_forker->appname, true);
693 signal( SIGUSR1, sigusr1_handler );
697 @brief Signal handler for SIGUSR2
698 @param sig The value of the trapped signal; always SIGUSR2.
700 Send register command to all known routers
702 static void sigusr2_handler( int sig ) {
703 if (!global_forker) return;
704 osrf_prefork_register_routers(global_forker->appname, false);
705 signal( SIGUSR2, sigusr2_handler );
709 @brief Signal handler for SIGTERM
710 @param sig The value of the trapped signal; always SIGTERM
712 Perform a graceful prefork server shutdown.
714 static void sigterm_handler(int sig) {
715 if (!global_forker) return;
716 osrfLogInfo(OSRF_LOG_MARK, "server: received SIGTERM, shutting down");
717 prefork_clear(global_forker, true);
722 @brief Signal handler for SIGINT or SIGQUIT
723 @param sig The value of the trapped signal
725 Perform a non-graceful prefork server shutdown.
727 static void sigint_handler(int sig) {
728 if (!global_forker) return;
729 osrfLogInfo(OSRF_LOG_MARK, "server: received SIGINT/QUIT, shutting down");
730 prefork_clear(global_forker, false);
735 @brief Replenish the collection of child processes, after one has terminated.
736 @param forker Pointer to the prefork_simple that manages the child processes.
738 The parent calls this function when it notices (via a signal handler) that
739 a child process has died.
741 Wait on the dead children so that they won't be zombies. Spawn new ones as needed
742 to maintain at least a minimum number.
744 void reap_children( prefork_simple* forker ) {
748 // Reset our boolean so that we can detect any further terminations.
751 // Bury the children so that they won't be zombies. WNOHANG means that waitpid() returns
752 // immediately if there are no waitable children, instead of waiting for more to die.
753 // Ignore the return code of the child. We don't do an autopsy.
754 while( (child_pid = waitpid( -1, NULL, WNOHANG )) > 0 ) {
755 --forker->current_num_children;
756 del_prefork_child( forker, child_pid );
759 // Spawn more children as needed.
760 while( forker->current_num_children < forker->min_children )
761 launch_child( forker );
765 @brief Read transport_messages and dispatch them to child processes for servicing.
766 @param forker Pointer to the prefork_simple that manages the child processes.
768 This is the main loop of the parent process, and once entered, does not exit.
770 For each usable transport_message received: look for an idle child to service it. If
771 no idle children are available, either spawn a new one or, if we've already spawned the
772 maximum number of children, wait for one to become available. Once a child is available
773 by whatever means, write an XML version of the input message, to a pipe designated for
776 static void prefork_run( prefork_simple* forker ) {
778 if( NULL == forker->idle_list )
779 return; // No available children, and we haven't even started yet
781 transport_message* cur_msg = NULL;
785 if( forker->first_child == NULL && forker->idle_list == NULL ) {/* no more children */
786 osrfLogWarning( OSRF_LOG_MARK, "No more children..." );
790 // Wait indefinitely for an input message
791 osrfLogDebug( OSRF_LOG_MARK, "Forker going into wait for data..." );
792 cur_msg = client_recv( forker->connection, -1 );
794 if( cur_msg == NULL )
795 continue; // Error? Interrupted by a signal? Try again...
797 message_prepare_xml( cur_msg );
798 const char* msg_data = cur_msg->msg_xml;
799 if( ! msg_data || ! *msg_data ) {
800 osrfLogWarning( OSRF_LOG_MARK, "Received % message from %s, thread %",
801 (msg_data ? "empty" : "NULL"), cur_msg->sender, cur_msg->thread );
802 message_free( cur_msg );
803 continue; // Message not usable; go on to the next one.
806 int honored = 0; /* will be set to true when we service the request */
812 if(check_children( forker, 0 ) < 0) {
813 continue; // check failed, try again
818 osrfLogDebug( OSRF_LOG_MARK, "Server received inbound data" );
820 prefork_child* cur_child = NULL;
822 // Look for an available child in the idle list. Since the idle list operates
823 // as a stack, the child we get is the one that was most recently active, or
824 // most recently spawned. That means it's the one most likely still to be in
825 // physical memory, and the one least likely to have to be swapped in.
826 while( forker->idle_list ) {
828 osrfLogDebug( OSRF_LOG_MARK, "Looking for idle child" );
829 // Grab the prefork_child at the head of the idle list
830 cur_child = forker->idle_list;
831 forker->idle_list = cur_child->next;
832 cur_child->next = NULL;
834 osrfLogInternal( OSRF_LOG_MARK,
835 "Searching for available child. cur_child->pid = %d", cur_child->pid );
836 osrfLogInternal( OSRF_LOG_MARK, "Current num children %d",
837 forker->current_num_children );
839 osrfLogDebug( OSRF_LOG_MARK, "forker sending data to %d", cur_child->pid );
840 osrfLogInternal( OSRF_LOG_MARK, "Writing to child fd %d",
841 cur_child->write_data_fd );
843 int written = write( cur_child->write_data_fd, msg_data, strlen( msg_data ) + 1 );
845 // This child appears to be dead or unusable. Discard it.
846 osrfLogWarning( OSRF_LOG_MARK, "Write returned error %d: %s",
847 errno, strerror( errno ));
848 kill( cur_child->pid, SIGKILL );
849 del_prefork_child( forker, cur_child->pid );
853 add_prefork_child( forker, cur_child ); // Add it to active list
858 /* if none available, add a new child if we can */
860 osrfLogDebug( OSRF_LOG_MARK, "Not enough children, attempting to add..." );
862 if( forker->current_num_children < forker->max_children ) {
863 osrfLogDebug( OSRF_LOG_MARK, "Launching new child with current_num = %d",
864 forker->current_num_children );
866 launch_child( forker ); // Put a new child into the idle list
867 if( forker->idle_list ) {
869 // Take the new child from the idle list
870 prefork_child* new_child = forker->idle_list;
871 forker->idle_list = new_child->next;
872 new_child->next = NULL;
874 osrfLogDebug( OSRF_LOG_MARK, "Writing to new child fd %d : pid %d",
875 new_child->write_data_fd, new_child->pid );
878 new_child->write_data_fd, msg_data, strlen( msg_data ) + 1 );
880 // This child appears to be dead or unusable. Discard it.
881 osrfLogWarning( OSRF_LOG_MARK, "Write returned error %d: %s",
882 errno, strerror( errno ));
883 kill( cur_child->pid, SIGKILL );
884 del_prefork_child( forker, cur_child->pid );
886 add_prefork_child( forker, new_child );
891 osrfLogWarning( OSRF_LOG_MARK, "Could not launch a new child as %d children "
892 "were already running; consider increasing max_children for this "
893 "application higher than %d in the OpenSRF configuration if this "
894 "message occurs frequently",
895 forker->current_num_children, forker->max_children );
900 osrfLogWarning( OSRF_LOG_MARK, "No children available, waiting..." );
901 if( check_children( forker, 1 ) >= 0 ) {
902 // Tell the loop not to call check_children again, since we just successfully called it
908 reap_children( forker );
910 } // end while( ! honored )
912 message_free( cur_msg );
914 } /* end top level listen loop */
919 @brief See if any children have become available.
920 @param forker Pointer to the prefork_simple that owns the children.
921 @param forever Boolean: true if we should wait indefinitely.
922 @return 0 or greater if successful, -1 on select error/interrupt
924 Call select() for all the children in the active list. Read each active file
925 descriptor and move the corresponding child to the idle list.
927 If @a forever is true, wait indefinitely for input. Otherwise return immediately if
928 there are no active file descriptors.
930 static int check_children( prefork_simple* forker, int forever ) {
933 reap_children( forker );
935 if( NULL == forker->first_child ) {
936 // If forever is true, then we're here because we've run out of idle
937 // processes, so there should be some active ones around, except during
938 // graceful shutdown, as we wait for all active children to become idle.
939 // If forever is false, then the children may all be idle, and that's okay.
941 osrfLogDebug( OSRF_LOG_MARK, "No active child processes to check" );
947 FD_ZERO( &read_set );
951 // Prepare to select() on pipes from all the active children
952 prefork_child* cur_child = forker->first_child;
954 if( cur_child->read_status_fd > max_fd )
955 max_fd = cur_child->read_status_fd;
956 FD_SET( cur_child->read_status_fd, &read_set );
957 cur_child = cur_child->next;
958 } while( cur_child != forker->first_child );
960 FD_CLR( 0, &read_set ); /* just to be sure */
964 if( (select_ret=select( max_fd + 1, &read_set, NULL, NULL, NULL )) == -1 ) {
965 osrfLogWarning( OSRF_LOG_MARK, "Select returned error %d on check_children: %s",
966 errno, strerror( errno ));
968 osrfLogInfo( OSRF_LOG_MARK,
969 "select() completed after waiting on children to become available" );
977 if( (select_ret=select( max_fd + 1, &read_set, NULL, NULL, &tv )) == -1 ) {
978 osrfLogWarning( OSRF_LOG_MARK, "Select returned error %d on check_children: %s",
979 errno, strerror( errno ));
983 if( select_ret <= 0 ) // we're done here
986 // Check each child in the active list.
987 // If it has responded, move it to the idle list.
988 cur_child = forker->first_child;
989 prefork_child* next_child = NULL;
992 next_child = cur_child->next;
993 if( FD_ISSET( cur_child->read_status_fd, &read_set )) {
994 osrfLogDebug( OSRF_LOG_MARK,
995 "Server received status from a child %d", cur_child->pid );
999 /* now suck off the data */
1001 if( (n=read( cur_child->read_status_fd, buf, sizeof( buf ) - 1 )) < 0 ) {
1002 osrfLogWarning( OSRF_LOG_MARK,
1003 "Read error after select in child status read with errno %d: %s",
1004 errno, strerror( errno ));
1008 osrfLogDebug( OSRF_LOG_MARK, "Read %d bytes from status buffer: %s", n, buf );
1011 // Remove the child from the active list
1012 if( forker->first_child == cur_child ) {
1013 if( cur_child->next == cur_child )
1014 forker->first_child = NULL; // only child in the active list
1016 forker->first_child = cur_child->next;
1018 cur_child->next->prev = cur_child->prev;
1019 cur_child->prev->next = cur_child->next;
1021 // Add it to the idle list
1022 cur_child->prev = NULL;
1023 cur_child->next = forker->idle_list;
1024 forker->idle_list = cur_child;
1026 cur_child = next_child;
1027 } while( forker->first_child && forker->first_child != next_child );
1033 @brief Service up a set maximum number of requests; then shut down.
1034 @param child Pointer to the prefork_child representing the child process.
1036 Called only by child process.
1038 Enter a loop, for up to max_requests iterations. On each iteration:
1039 - Wait indefinitely for a request from the parent.
1040 - Service the request.
1041 - Increment a counter. If the limit hasn't been reached, notify the parent that you
1042 are available for another request.
1044 After exiting the loop, shut down and terminate the process.
1046 static void prefork_child_wait( prefork_child* child ) {
1049 growing_buffer* gbuf = buffer_init( READ_BUFSIZE );
1050 char buf[READ_BUFSIZE];
1052 for( i = 0; i < child->max_requests; i++ ) {
1055 int gotdata = 0; // boolean; set to true if we get data
1056 clr_fl( child->read_data_fd, O_NONBLOCK );
1058 // Read a request from the parent, via a pipe, into a growing_buffer.
1059 while( (n = read( child->read_data_fd, buf, READ_BUFSIZE-1 )) > 0 ) {
1061 osrfLogDebug( OSRF_LOG_MARK, "Prefork child read %d bytes of data", n );
1063 set_fl( child->read_data_fd, O_NONBLOCK );
1066 buffer_add_n( gbuf, buf, n );
1069 if( errno == EAGAIN )
1072 if( errno == EPIPE ) {
1073 osrfLogDebug( OSRF_LOG_MARK, "C child attempted read on broken pipe, exiting..." );
1077 int terminate_now = 0; // Boolean
1080 osrfLogWarning( OSRF_LOG_MARK,
1081 "Prefork child read returned error with errno %d", errno );
1084 } else if( gotdata ) {
1085 // Process the request
1086 osrfLogDebug( OSRF_LOG_MARK, "Prefork child got a request.. processing.." );
1087 terminate_now = prefork_child_process_request( child, gbuf->buf );
1088 buffer_reset( gbuf );
1091 if( terminate_now ) {
1092 // We're terminating prematurely -- presumably due to a fatal error condition.
1093 osrfLogWarning( OSRF_LOG_MARK, "Prefork child terminating abruptly" );
1097 if( i < child->max_requests - 1 ) {
1098 // Report back to the parent for another request.
1100 ssize_t len = write(
1101 child->write_status_fd, "available" /*less than 64 bytes*/, msg_len );
1102 if( len != msg_len ) {
1103 osrfLogError( OSRF_LOG_MARK,
1104 "Drone terminating: unable to notify listener of availability: %s",
1106 buffer_free( gbuf );
1107 osrf_prefork_child_exit( child );
1112 buffer_free( gbuf );
1114 osrfLogDebug( OSRF_LOG_MARK, "Child with max-requests=%d, num-served=%d exiting...[%ld]",
1115 child->max_requests, i, (long) getpid());
1117 osrf_prefork_child_exit( child );
1121 @brief Add a prefork_child to the end of the active list.
1122 @param forker Pointer to the prefork_simple that owns the list.
1123 @param child Pointer to the prefork_child to be added.
1125 static void add_prefork_child( prefork_simple* forker, prefork_child* child ) {
1127 if( forker->first_child == NULL ) {
1128 // Simplest case: list is initially empty.
1129 forker->first_child = child;
1130 child->next = child;
1131 child->prev = child;
1133 // Find the last node in the circular list.
1134 prefork_child* last_child = forker->first_child->prev;
1136 // Insert the new child between the last and first children.
1137 last_child->next = child;
1138 child->prev = last_child;
1139 child->next = forker->first_child;
1140 forker->first_child->prev = child;
1145 @brief Delete and destroy a dead child from our list.
1146 @param forker Pointer to the prefork_simple that owns the dead child.
1147 @param pid Process ID of the dead child.
1149 Look for the dead child first in the list of active children. If you don't find it
1150 there, look in the list of idle children. If you find it, remove it from whichever
1151 list it's on, and destroy it.
1153 static void del_prefork_child( prefork_simple* forker, pid_t pid ) {
1155 osrfLogDebug( OSRF_LOG_MARK, "Deleting Child: %d", pid );
1157 prefork_child* cur_child = NULL;
1159 // Look first in the active list
1160 if( forker->first_child ) {
1161 cur_child = forker->first_child; /* current pointer */
1162 while( cur_child->pid != pid && cur_child->next != forker->first_child )
1163 cur_child = cur_child->next;
1165 if( cur_child->pid == pid ) {
1166 // We found the right node. Remove it from the list.
1167 if( cur_child->next == cur_child )
1168 forker->first_child = NULL; // only child in the list
1170 if( forker->first_child == cur_child )
1171 forker->first_child = cur_child->next; // Reseat forker->first_child
1173 // Stitch the adjacent nodes together
1174 cur_child->prev->next = cur_child->next;
1175 cur_child->next->prev = cur_child->prev;
1178 cur_child = NULL; // Didn't find it in the active list
1182 // Maybe it's in the idle list. This can happen if, for example,
1183 // a child is killed by a signal while it's between requests.
1185 prefork_child* prev = NULL;
1186 cur_child = forker->idle_list;
1187 while( cur_child && cur_child->pid != pid ) {
1189 cur_child = cur_child->next;
1193 // Detach from the list
1195 prev->next = cur_child->next;
1197 forker->idle_list = cur_child->next;
1198 } // else we can't find it
1201 // If we found the node, destroy it.
1203 prefork_child_free( forker, cur_child );
1207 @brief Create and initialize a prefork_child.
1208 @param forker Pointer to the prefork_simple that will own the prefork_child.
1209 @param read_data_fd Used by child to read request from parent.
1210 @param write_data_fd Used by parent to write request to child.
1211 @param read_status_fd Used by parent to read status from child.
1212 @param write_status_fd Used by child to write status to parent.
1213 @return Pointer to the newly created prefork_child.
1215 The calling code is responsible for freeing the prefork_child by calling
1216 prefork_child_free().
1218 static prefork_child* prefork_child_init( prefork_simple* forker,
1219 int read_data_fd, int write_data_fd,
1220 int read_status_fd, int write_status_fd ) {
1222 // Allocate a prefork_child -- from the free list if possible, or from
1223 // the heap if necessary. The free list is a non-circular, singly-linked list.
1224 prefork_child* child;
1225 if( forker->free_list ) {
1226 child = forker->free_list;
1227 forker->free_list = child->next;
1229 child = safe_malloc( sizeof( prefork_child ));
1232 child->read_data_fd = read_data_fd;
1233 child->write_data_fd = write_data_fd;
1234 child->read_status_fd = read_status_fd;
1235 child->write_status_fd = write_status_fd;
1236 child->max_requests = forker->max_requests;
1237 child->appname = forker->appname; // We don't make a separate copy
1238 child->keepalive = forker->keepalive;
1246 @brief Terminate all child processes and clear out a prefork_simple.
1247 @param prefork Pointer to the prefork_simple to be cleared out.
1249 We do not deallocate the prefork_simple itself, just its contents.
1251 static void prefork_clear( prefork_simple* prefork, bool graceful ) {
1253 // always de-register routers before killing child processes (or waiting
1254 // for them to complete) so that new requests are directed elsewhere.
1255 osrf_prefork_register_routers(global_forker->appname, true);
1257 while( prefork->first_child ) {
1260 // wait for at least one active child to become idle, then repeat.
1261 // once complete, all children will be idle and cleaned up below.
1262 osrfLogInfo(OSRF_LOG_MARK, "graceful shutdown waiting...");
1263 check_children(prefork, 1);
1266 // Kill and delete all the active children
1267 kill( prefork->first_child->pid, SIGKILL );
1268 del_prefork_child( prefork, prefork->first_child->pid );
1273 osrfLogInfo(OSRF_LOG_MARK,
1274 "all active children are now idle in graceful shutdown");
1277 // Kill all the idle prefork children, close their file
1278 // descriptors, and move them to the free list.
1279 prefork_child* child = prefork->idle_list;
1280 prefork->idle_list = NULL;
1282 prefork_child* temp = child->next;
1283 kill( child->pid, SIGKILL );
1284 prefork_child_free( prefork, child );
1287 //prefork->current_num_children = 0;
1289 // Physically free the free list of prefork_children.
1290 child = prefork->free_list;
1291 prefork->free_list = NULL;
1293 prefork_child* temp = child->next;
1298 // Close the Jabber connection
1299 client_free( prefork->connection );
1300 prefork->connection = NULL;
1302 // After giving the child processes a second to terminate, wait on them so that they
1303 // don't become zombies. We don't wait indefinitely, so it's possible that some
1304 // children will survive a bit longer.
1306 while( (waitpid( -1, NULL, WNOHANG )) > 0 ) {
1307 --prefork->current_num_children;
1310 free( prefork->appname );
1311 prefork->appname = NULL;
1315 @brief Destroy and deallocate a prefork_child.
1316 @param forker Pointer to the prefork_simple that owns the prefork_child.
1317 @param child Pointer to the prefork_child to be destroyed.
1319 static void prefork_child_free( prefork_simple* forker, prefork_child* child ) {
1320 close( child->read_data_fd );
1321 close( child->write_data_fd );
1322 close( child->read_status_fd );
1323 close( child->write_status_fd );
1325 // Stick the prefork_child in a free list for potential reuse. This is a
1326 // non-circular, singly linked list.
1328 child->next = forker->free_list;
1329 forker->free_list = child;