/**
* Initializes the high availability (ha) structure
*
* returns
* 0 on success
* -1 on error
*/
inline int init_pingtable(struct ha *table,int timeout,int maxpings)
{
if(maxpings<=0)
maxpings=1;
table->begin=0;
table->end=0;
table->timed_out_pings=0;
table->size=maxpings;
table->timeout=timeout;
if (!(table->mutex=lock_alloc())){
LM_ERR("Unable to allocate a lock for the ping table\n");
goto error;
}else
lock_init(table->mutex);
LM_ERR("alloc'ing %d bytes for %d pings\n",(int)(maxpings*sizeof(struct ping)),maxpings);
if (0==(table->pings=shm_malloc(maxpings*sizeof(struct ping)))){
LM_ERR("Unable to shm_malloc %d bytes for %d pings\n",(int)(maxpings*sizeof(struct ping)),maxpings);
goto error;
}else{
memset(table->pings,0,(maxpings*sizeof(struct ping)));
}
return 0;
error:
destroy_pingtable(table);
return -1;
}
/** Main loop for the Event Dispatcher process.
*
*/
int dispatcher_main_loop(void)
{
struct pollfd poll_fds[3+MAX_AS_NR],*poll_tmp;
int clean_index,i,j,k,fd,poll_events=0,socks[2],chld_status;
int as_nr,unc_as_nr;
pid_t chld;
struct timeval last_ping,now;
struct as_entry *as;
sig_flag=0;
is_dispatcher=1;
as_nr=0;
timerclear(&last_ping);
timerclear(&now);
signal(SIGCHLD,seas_sighandler);
signal(SIGTERM,seas_sighandler);
signal(SIGUSR1,seas_sighandler);
signal(SIGINT, seas_sighandler);
signal(SIGKILL,seas_sighandler);
strcpy(whoami,"Seas Event Dispatcher process");
/*I set process_no to -1 because otherwise, the logging process confuses this process with another from SER
* (see LM_*() and dprint() and my_pid())*/
process_no = -1;
if(open_server_sockets(seas_listen_ip,seas_listen_port,socks)==-1){
LM_ERR("unable to open server sockets on dispatcher\n");
return -1;
}
for(i=0;i<2;i++){
poll_fds[i].fd=socks[i];
poll_fds[i].revents=0;
poll_fds[i].events=POLLIN;
}
poll_fds[2].fd=read_pipe;
poll_fds[2].revents=0;
poll_fds[2].events=POLLIN;/*pollhup ?*/
poll_events=0;
unc_as_nr=0;
if(use_ha)
spawn_pinger();
while(1){
if(sig_flag==SIGCHLD){
while ((chld=waitpid( -1, &chld_status, WNOHANG ))>0) {
if (WIFEXITED(chld_status)){
LM_INFO("child process %d exited normally, status=%d\n",
chld,WEXITSTATUS(chld_status));
}else if (WIFSIGNALED(chld_status)) {
LM_INFO("child process %d exited by a signal %d\n",
chld,WTERMSIG(chld_status));
}else if (WIFSTOPPED(chld_status))
LM_INFO("child process %d stopped by a signal %d\n",
chld,WSTOPSIG(chld_status));
for (as=as_list;as;as=as->next) {
if(as->type!=AS_TYPE)
continue;
if(as->u.as.action_pid==chld){
for(i=0;i<as_nr && ((poll_fds[3+i].fd)!=(as->u.as.event_fd));i++)
;
if(i==as_nr){
LM_ERR("Either the pinger has died or BUG found..\n");
continue;
}
/*overwrite the obsolete 'i' position with the next position*/
for(j=3+i;j<(as_nr+unc_as_nr+3-1);i++){
poll_fds[j].fd=poll_fds[j+1].fd;
poll_fds[j].events=poll_fds[j+1].events;
poll_fds[j].revents=poll_fds[j+1].revents;
}
close(as->u.as.event_fd);/*close the socket fd*/
if (as->u.as.ev_buffer.s) {
pkg_free(as->u.as.ev_buffer.s);
as->u.as.ev_buffer.s=(char *)0;
as->u.as.ev_buffer.len=0;
}
as->u.as.event_fd=as->u.as.action_fd=-1;
as->connected=0;
destroy_pingtable(&as->u.as.jain_pings);
destroy_pingtable(&as->u.as.servlet_pings);
as_nr--;
LM_WARN("client [%.*s] leaving (Action Dispatcher Process died !)\n",
as->name.len,as->name.s);
break;
}/*if(action_pid==chld)*/
}/*for(as=as_list;as;as=as->next)*/
}/*while(waitpid(-1)>0)*/
}else if (sig_flag) {
LM_WARN("received signal != sigchld(%d)\n",sig_flag);
}
sig_flag=0;
clean_index=0;
LM_INFO("polling [2 ServSock] [1 pipe] [%d App Servers]"
" [%d Uncomplete AS]\n",as_nr,unc_as_nr);
poll_events = poll(poll_fds,3+unc_as_nr+as_nr,-1);
if (poll_events == -1) {
if(errno==EINTR){
/*handle the case a child has died.
* It will be done in the next iteration in if(seas_sigchld_received)*/
continue;
}
if(errno==EBADF){
LM_ERR("invalid file descriptor pased to poll (%s)\n",
strerror(errno));
return -1;/*??*/
}
/* errors */
LM_ERR("poll'ing:%s\n",strerror(errno));
poll_events=0;
continue;
} else if (poll_events == 0) {/*timeout*/
continue;
} else {/*there are events !*/
/*handle connections from server sockets*/
for(i=0;i<2;i++){
if(poll_fds[i].revents)
poll_events--;
if(poll_fds[i].revents & POLLIN){
poll_fds[i].revents &= (~POLLIN);
if((fd=new_as_connect(socks[i],i==0?'e':'a'))>=0){
poll_tmp=&poll_fds[3+as_nr+unc_as_nr];
poll_tmp->fd=fd;
poll_tmp->events=POLLIN|POLLHUP;
unc_as_nr++;
LM_DBG("Have new %s client\n",i==0?"event":"action");
}else{
LM_ERR("accepting connection from AS\n");
}
}
}
/*handle data from pipe*/
if(poll_fds[2].revents & POLLIN){
poll_fds[2].revents &= (~POLLIN);
poll_events--;
if(dispatch_relay()<0){
LM_ERR("dispatch_relay returned -1"
"should clean-up table\n");
}
}
/*now handle receive data from completed AS*/
clean_index=0;
LM_DBG("Scanning data from %d AS\n",as_nr);
for(i=0;(i<as_nr) && poll_events;i++){
clean_index=0;
poll_tmp=&poll_fds[3+i];
if(poll_tmp->revents)
poll_events--;
if(poll_tmp->revents & POLLIN){
LM_DBG("POLLIN found in AS #%i\n",i);
poll_tmp->revents &= (~POLLIN);
switch(handle_as_data(poll_tmp->fd)){
case -2:/*read returned 0 bytes, an AS client is leaving*/
clean_index=1;
break;
case -1:/*shouldnt happen*/
LM_ERR("reading from AS socket\n");
break;
case 0:/* event_response received and processed*/
break;
default:
LM_WARN("unknown return type from handle_as_data\n");
}
}
if(clean_index || (poll_tmp->revents & POLLHUP)){
LM_DBG("POLHUP or read==0 found in %i AS \n",i);
clean_index=0;
poll_tmp->revents = 0;
for(as=as_list;as;as=as->next){
if(as->type==CLUSTER_TYPE)
continue;
if(as->connected && (as->u.as.event_fd == poll_tmp->fd)){
close(poll_tmp->fd);/*close the socket fd*/
/*TODO we should send a signal to the Action Dispatcher !!!*/
as->connected=0;
as_nr--;
/*overwrite the obsolete 'i' position with the next position*/
for(k=i;k<(as_nr+unc_as_nr);k++){
j=3+k;
poll_fds[j].fd=poll_fds[j+1].fd;
poll_fds[j].events=poll_fds[j+1].events;
poll_fds[j].revents=poll_fds[j+1].revents;
}
--i;
LM_WARN("client %.*s leaving !!!\n",as->name.len,as->name.s);
break;
}
}
if (!as) {
LM_ERR("the leaving client was not found in the as_list\n");
}
}
}
/*now handle data sent from uncompleted AS*/
LM_DBG("Scanning data from %d uncomplete AS \n",unc_as_nr);
clean_index=0;
for(i=0;i<unc_as_nr && poll_events;i++){
poll_tmp=&poll_fds[3+as_nr+i];
if(poll_tmp->revents)
poll_events--;
if(poll_tmp->revents & POLLIN){
LM_DBG("POLLIN found in %d uncomplete AS \n",i);
poll_tmp->revents &= (~POLLIN);
fd=handle_unc_as_data(poll_tmp->fd);
if(fd>0){
/* there's a new AS, push the uncomplete poll_fds up and set the AS */
for(k=i;k>0;k--){
j=3+as_nr+k;
poll_fds[j].fd=poll_fds[j-1].fd;
poll_fds[j].events=poll_fds[j-1].events;
poll_fds[j].revents=poll_fds[j-1].revents;
}
poll_fds[3+as_nr].fd=fd;
poll_fds[3+as_nr].events=POLLIN|POLLHUP;
poll_fds[3+as_nr].revents=0;
as_nr++;/*not very sure if this is thread-safe*/
unc_as_nr--;
}else if(fd<=0){/* pull the upper set of uncomplete AS down and take this one out*/
poll_tmp->revents=0;
for(k=i;k<(unc_as_nr-1);k++){
j=3+as_nr+k;
poll_fds[j].fd=poll_fds[j+1].fd;
poll_fds[j].events=poll_fds[j+1].events;
poll_fds[j].revents=poll_fds[j+1].revents;
}
unc_as_nr--;
/** we decrement i so that pulling down the upper part of the unc_as array so that
* it doesn't affect our for loop */
i--;
}
}
if(poll_tmp->revents & POLLHUP){
LM_DBG("POLLHUP found in %d uncomplete AS \n",i);
close(poll_tmp->fd);
for(k=i;k<(unc_as_nr-1);k++){
j=3+as_nr+k;
poll_fds[j].fd=poll_fds[j+1].fd;
poll_fds[j].events=poll_fds[j+1].events;
poll_fds[j].revents=poll_fds[j+1].revents;
}
unc_as_nr--;
i--;
poll_tmp->revents = 0;
}
}/*for*/
}/*else ...(poll_events>0)*/
}/*while(1)*/
}
/**
* receives 2 indexes in unc_as_t which correspond one to
* the events socket and the other to the actions socket
*
* returns
* 0 on success
* -1 on error
*/
static inline int add_new_as(int event_idx,int action_idx,struct as_entry *as)
{
struct unc_as *ev,*ac;
int j;
as_p the_as=0;
struct as_entry *tmp;
ev=&unc_as_t[event_idx];
ac=&unc_as_t[action_idx];
the_as=&(as->u.as);
the_as->action_fd=ac->fd;
the_as->event_fd=ev->fd;
the_as->name.len = strlen(ev->name);
if(use_ha){
if(jain_ping_timeout){
if (0>init_pingtable(&the_as->jain_pings,jain_ping_timeout,(jain_ping_timeout/jain_ping_period+1)*PING_OVER_FACTOR)){
LM_ERR("Unable to init jain pinging table...\n");
goto error;
}
}
if(servlet_ping_timeout){
if (0>init_pingtable(&the_as->servlet_pings,servlet_ping_timeout,(servlet_ping_timeout/servlet_ping_period+1)*PING_OVER_FACTOR)){
LM_ERR("Unable to init servlet pinging table...\n");
goto error;
}
}
}
/*TODO attention, this is pkg_malloc because only the Event_Dispatcher process
* has to use it !!*/
if(!(the_as->ev_buffer.s = pkg_malloc(AS_BUF_SIZE))){
LM_ERR("unable to alloc pkg mem for the event buffer\n");
goto error;
}
the_as->ev_buffer.len=0;
as->connected=1;
the_as->action_pid=0;
for(tmp=as_list;tmp;tmp=tmp->next){
if(tmp->type==AS_TYPE)
continue;
for (j=0;j<tmp->u.cs.num;j++) {
if (tmp->u.cs.as_names[j].len == the_as->name.len &&
!memcmp(tmp->u.cs.as_names[j].s,the_as->name.s,the_as->name.len)) {
if(tmp->u.cs.num==tmp->u.cs.registered){
LM_ERR("AS %.*s belongs to cluster %.*s which is already completed\n",
the_as->name.len,the_as->name.s,tmp->name.len,tmp->name.s);
break;
}
tmp->u.cs.registered++;
break;
}
}
}
if(0>spawn_action_dispatcher(as)){
LM_ERR("Unable to spawn Action Dispatcher for as %s\n",ev->name);
goto error;
}
if(send_sockinfo(the_as->event_fd)==-1){
LM_ERR("Unable to send socket info to as %s\n",ev->name);
goto error;
}
return 0;
error:
if(the_as->ev_buffer.s){
pkg_free(the_as->ev_buffer.s);
the_as->ev_buffer.s=(char*)0;
}
if(the_as->action_pid)
kill(the_as->action_pid,SIGTERM);
if(jain_ping_timeout)
destroy_pingtable(&the_as->jain_pings);
if(servlet_ping_timeout)
destroy_pingtable(&the_as->servlet_pings);
return -1;
}