int
main ( int argc, char *argv[] )
{
int rc;
struct stat struct_stat;
pthread_t pid;
pthread_attr_t attr;
int i, num_sources;
uid_t gmetad_uid;
mode_t rrd_umask;
char * gmetad_username;
struct passwd *pw;
gmetad_config_t *c = &gmetad_config;
apr_interval_time_t sleep_time;
apr_time_t last_metadata;
double random_sleep_factor;
unsigned int rand_seed;
rc = apr_initialize();
if (rc != APR_SUCCESS) {
return -1;
}
/* create a memory pool. */
apr_pool_create(&global_context, NULL);
/* Ignore SIGPIPE */
signal( SIGPIPE, SIG_IGN );
initialize_scoreboard();
/* Mark the time this gmetad started */
started = apr_time_now();
if (cmdline_parser(argc, argv, &args_info) != 0)
err_quit("command-line parser error");
num_sources = number_of_datasources( args_info.conf_arg );
if(!num_sources)
{
err_quit("%s doesn't have any data sources specified", args_info.conf_arg);
}
memset(&root, 0, sizeof(root));
root.id = ROOT_NODE;
/* Get the real number of data sources later */
sources = hash_create( num_sources + 10 );
if (! sources )
{
err_quit("Unable to create sources hash\n");
}
root.authority = hash_create( num_sources + 10 );
if (!root.authority)
{
err_quit("Unable to create root authority (our grids and clusters) hash\n");
}
root.metric_summary = hash_create (DEFAULT_METRICSIZE);
if (!root.metric_summary)
{
err_quit("Unable to create root summary hash");
}
parse_config_file ( args_info.conf_arg );
/* If given, use command line directives over config file ones. */
if (args_info.debug_given)
{
c->debug_level = args_info.debug_arg;
}
debug_level = c->debug_level;
set_debug_msg_level(debug_level);
/* Setup our default authority pointer if the conf file hasnt yet.
* Done in the style of hash node strings. */
if (!root.stringslen)
{
gethostname(hostname, HOSTNAMESZ);
root.authority_ptr = 0;
sprintf(root.strings, "http://%s/ganglia/", hostname);
root.stringslen += strlen(root.strings) + 1;
}
rand_seed = apr_time_now() * (int)pthread_self();
for(i = 0; i < root.stringslen; rand_seed = rand_seed * root.strings[i++]);
/* Debug level 1 is error output only, and no daemonizing. */
if (!debug_level)
{
rrd_umask = c->umask;
daemon_init (argv[0], 0, rrd_umask);
}
if (args_info.pid_file_given)
{
update_pidfile (args_info.pid_file_arg);
}
/* The rrd_rootdir must be writable by the gmetad process */
if( c->should_setuid )
{
if(! (pw = getpwnam(c->setuid_username)))
{
err_sys("Getpwnam error");
}
gmetad_uid = pw->pw_uid;
gmetad_username = c->setuid_username;
}
else
{
gmetad_uid = getuid();
if(! (pw = getpwuid(gmetad_uid)))
{
err_sys("Getpwnam error");
}
gmetad_username = strdup(pw->pw_name);
}
debug_msg("Going to run as user %s", gmetad_username);
if( c->should_setuid )
{
become_a_nobody(c->setuid_username);
}
if( c->write_rrds )
{
if( stat( c->rrd_rootdir, &struct_stat ) )
{
err_sys("Please make sure that %s exists", c->rrd_rootdir);
}
if ( struct_stat.st_uid != gmetad_uid )
{
err_quit("Please make sure that %s is owned by %s", c->rrd_rootdir, gmetad_username);
}
if (! (struct_stat.st_mode & S_IWUSR) )
{
err_quit("Please make sure %s has WRITE permission for %s", gmetad_username, c->rrd_rootdir);
}
}
if(debug_level)
{
fprintf(stderr,"Sources are ...\n");
hash_foreach( sources, print_sources, NULL);
}
#ifdef WITH_MEMCACHED
if (c->memcached_parameters != NULL)
{
memcached_connection_pool = memcached_pool(c->memcached_parameters, strlen(c->memcached_parameters));
}
#endif /* WITH_MEMCACHED */
server_socket = g_tcp_socket_server_new( c->xml_port );
if (server_socket == NULL)
{
err_quit("tcp_listen() on xml_port failed");
}
debug_msg("xml listening on port %d", c->xml_port);
interactive_socket = g_tcp_socket_server_new( c->interactive_port );
if (interactive_socket == NULL)
{
err_quit("tcp_listen() on interactive_port failed");
}
debug_msg("interactive xml listening on port %d", c->interactive_port);
/* Forward metrics to Graphite using carbon protocol */
if (c->carbon_server != NULL)
{
if (!strcmp(c->carbon_protocol, "udp"))
{
carbon_udp_socket = init_carbon_udp_socket (c->carbon_server, c->carbon_port);
if (carbon_udp_socket == NULL)
err_quit("carbon %s socket failed for %s:%d", c->carbon_protocol, c->carbon_server, c->carbon_port);
}
debug_msg("carbon forwarding ready to send via %s to %s:%d", c->carbon_protocol, c->carbon_server, c->carbon_port);
}
#ifdef WITH_RIEMANN
if (c->riemann_server !=NULL)
{
if (!strcmp(c->riemann_protocol, "udp"))
{
riemann_udp_socket = init_riemann_udp_socket (c->riemann_server, c->riemann_port);
if (riemann_udp_socket == NULL)
err_quit("[riemann] %s socket failed for %s:%d", c->riemann_protocol, c->riemann_server, c->riemann_port);
} else {
err_quit("[riemann] TCP transport not supported yet.");
}
debug_msg("[riemann] ready to forward metrics via %s to %s:%d", c->riemann_protocol, c->riemann_server, c->riemann_port);
}
#endif /* WITH_RIEMANN */
/* initialize summary mutex */
root.sum_finished = (pthread_mutex_t *)
malloc(sizeof(pthread_mutex_t));
pthread_mutex_init(root.sum_finished, NULL);
pthread_attr_init( &attr );
pthread_attr_setdetachstate( &attr, PTHREAD_CREATE_DETACHED );
/* Spin off the non-interactive server threads. (Half as many as interactive). */
for (i=0; i < c->server_threads/2; i++)
pthread_create(&pid, &attr, server_thread, (void*) 0);
/* Spin off the interactive server threads. */
for (i=0; i < c->server_threads; i++)
pthread_create(&pid, &attr, server_thread, (void*) 1);
hash_foreach( sources, spin_off_the_data_threads, NULL );
/* A thread to cleanup old metrics and hosts */
pthread_create(&pid, &attr, cleanup_thread, (void *) NULL);
debug_msg("cleanup thread has been started");
/* Meta data */
last_metadata = 0;
for(;;)
{
/* Do at a random interval, between
(shortest_step/2) +/- METADATA_SLEEP_RANDOMIZE percent */
random_sleep_factor = (1 + (METADATA_SLEEP_RANDOMIZE / 50.0) * ((rand_r(&rand_seed) - RAND_MAX/2)/(float)RAND_MAX));
sleep_time = random_sleep_factor * apr_time_from_sec(c->shortest_step) / 2;
/* Make sure the sleep time is at least 1 second */
if(apr_time_sec(apr_time_now() + sleep_time) < (METADATA_MINIMUM_SLEEP + apr_time_sec(apr_time_now())))
sleep_time += apr_time_from_sec(METADATA_MINIMUM_SLEEP);
apr_sleep(sleep_time);
/* Need to be sure root is locked while doing summary */
pthread_mutex_lock(root.sum_finished);
/* Flush the old values */
hash_foreach(root.metric_summary, zero_out_summary, NULL);
root.hosts_up = 0;
root.hosts_down = 0;
/* Sum the new values */
hash_foreach(root.authority, do_root_summary, NULL );
/* summary completed */
pthread_mutex_unlock(root.sum_finished);
/* Save them to RRD */
hash_foreach(root.metric_summary, write_root_summary, NULL);
/* Remember our last run */
last_metadata = apr_time_now();
}
apr_pool_destroy(global_context);
apr_terminate();
return 0;
}
int main(int argc, char *argv[])
{
int rval;
gexec_cluster_t cluster;
gexec_host_t *host;
llist_entry *li;
debug_level = 1;
set_debug_msg_level(debug_level);
if (cmdline_parser (argc, argv, &args_info) != 0)
exit(1) ;
rval = gexec_cluster(&cluster, args_info.gmond_ip_arg, args_info.gmond_port_arg );
if ( rval != 0)
{
printf("Unable to get hostlist from %s %d!\n", args_info.gmond_ip_arg, args_info.gmond_port_arg);
exit(-1);
}
if( args_info.mpifile_flag )
{
if( args_info.all_flag )
li = cluster.hosts;
else
li = cluster.gexec_hosts;
for( ; li != NULL; li = li->next )
{
host = li->val;
if( host->name_resolved && ! args_info.numeric_flag )
{
if(!strcmp(host->domain, "unspecified"))
printf("%s:%d\n", host->name, host->cpu_num);
else
printf("%s.%s:%d\n", host->name, host->domain, host->cpu_num);
}
else
{
printf("%s:%d\n", host->ip, host->cpu_num);
}
}
exit(0);
}
if(! args_info.list_flag )
{
printf("CLUSTER INFORMATION\n");
printf(" Name: %s\n", cluster.name);
printf(" Hosts: %d\n", cluster.num_hosts);
printf("Gexec Hosts: %d\n", cluster.num_gexec_hosts);
printf(" Dead Hosts: %d\n", cluster.num_dead_hosts);
printf(" Localtime: %s\n", ctime(&(cluster.localtime)) );
}
if( args_info.dead_flag)
{
if(! args_info.list_flag )
printf("DEAD CLUSTER HOSTS\n");
if(cluster.num_dead_hosts)
{
if(! args_info.list_flag )
printf("%32.32s Last Reported\n", "Hostname");
for( li = cluster.dead_hosts; li != NULL; li = li->next)
{
host= li->val;
printf("%32.32s %s", host->name, ctime(&(host->last_reported)));
}
}
else
{
printf("There are no hosts down at this time\n");
}
gexec_cluster_free(&cluster);
exit(0);
}
if( args_info.all_flag )
{
li = cluster.hosts;
if(! cluster.num_hosts )
{
printf("There are no hosts up at this time\n");
gexec_cluster_free(&cluster);
exit(0);
}
}
else
{
li = cluster.gexec_hosts;
if(! cluster.num_gexec_hosts)
{
printf("There are no hosts running gexec at this time\n");
gexec_cluster_free(&cluster);
exit(0);
}
}
if(! args_info.list_flag )
{
printf("CLUSTER HOSTS\n");
printf("Hostname LOAD CPU Gexec\n");
printf(" CPUs (Procs/Total) [ 1, 5, 15min] [ User, Nice, System, Idle, Wio]\n\n");
}
for(; li != NULL; li = li->next)
{
host = li->val;
if( host->name_resolved && ! args_info.numeric_flag )
{
if(!strcmp(host->domain, "unspecified"))
printf("%s", host->name);
else
printf("%s.%s", host->name, host->domain);
}
else
{
printf("%s", host->ip);
}
if( args_info.single_line_flag )
printf(" ");
else
printf("\n");
printf(" %4d (%5d/%5d) [%6.2f,%6.2f,%6.2f] [%6.1f,%6.1f,%6.1f,%6.1f,%6.1f] ",
host->cpu_num, host->proc_run, host->proc_total,
host->load_one, host->load_five, host->load_fifteen,
host->cpu_user, host->cpu_nice, host->cpu_system, host->cpu_idle, host->cpu_wio);
if(host->gexec_on)
printf("%s\n", "ON");
else
printf("%s\n", "OFF");
}
gexec_cluster_free(&cluster);
return 0;
}
