int main(int argc, char **argv)
{
unsigned long k;
struct test_data *td;
t_set_colors(0);
t_start("fanout tests");
run_tests(10, 64);
run_tests(512, 64);
run_tests(64, 64);
run_tests(511, 17);
destroyed = 0;
fot = fanout_create(512);
ok_int(fanout_remove(fot, 12398) == NULL, 1,
"remove on empty table must yield NULL");
ok_int(fanout_get(fot, 123887987) == NULL, 1,
"get on empty table must yield NULL");
for (k = 0; k < 16385; k++) {
struct test_data *tdata = calloc(1, sizeof(*td));
tdata->key = k;
asprintf(&tdata->name, "%lu", k);
fanout_add(fot, k, tdata);
}
td = fanout_get(fot, k - 1);
ok_int(td != NULL, 1, "get must get what add inserts");
ok_int(fanout_remove(fot, k + 1) == NULL, 1,
"remove on non-inserted key must yield NULL");
ok_int(fanout_get(fot, k + 1) == NULL, 1,
"get on non-inserted must yield NULL");
fanout_destroy(fot, pdest);
ok_int((int)destroyed, (int)k, "destroy counter while free()'ing");
return t_end();
}
static void run_tests(int ntests, int fo_size)
{
struct tcase *tc;
unsigned long last_ptr, *ptr;
int i, added = 0, removed = 0;
fanout_table *fo;
last_ptr = ntests;
fo = fanout_create(fo_size);
tc = calloc(ntests, sizeof(*tc));
for (i = 0; i < ntests; i++) {
tc[i].value = i;
if (!fanout_add(fo, tc[i].key, &tc[i].value))
added++;
}
ok_int(added, ntests, "Adding stuff must work");
while ((ptr = (unsigned long *)fanout_remove(fo, 0))) {
ok_int((int)*ptr, (int)last_ptr - 1, "Removing a bunch of them");
removed++;
last_ptr = *ptr;
}
ok_int(added, removed, "Removing should work as well as adding");
fanout_destroy(fo, destructor);
ok_int(destroyed, 0, "Expected no entries in destructor");
fo = fanout_create(fo_size);
for (i = 0; i < ntests; i++) {
tc[i].value = i;
if (!fanout_add(fo, tc[i].key, &tc[i].value))
added++;
}
fanout_destroy(fo, destructor);
ok_int(destroyed, ntests, "Expected ntest entries in destructor");
destroyed = 0;
free(tc);
}
static int wproc_destroy(struct wproc_worker *wp, int flags)
{
int i = 0, force = 0, self;
if (!wp)
return 0;
force = !!(flags & WPROC_FORCE);
self = getpid();
/* master retains workers through restarts */
if (self == nagios_pid && !force)
return 0;
/* free all memory when either forcing or a worker called us */
iocache_destroy(wp->ioc);
wp->ioc = NULL;
my_free(wp->name);
fanout_destroy(wp->jobs, fo_destroy_job);
wp->jobs = NULL;
/* workers must never control other workers, so they return early */
if (self != nagios_pid)
return 0;
/* kill(0, SIGKILL) equals suicide, so we avoid it */
if (wp->pid) {
kill(wp->pid, SIGKILL);
}
iobroker_close(nagios_iobs, wp->sd);
/* reap our possibly lost children */
while (waitpid(-1, &i, WNOHANG) > 0)
; /* do nothing */
free(wp);
return 0;
}
/* frees memory allocated for the scheduled downtime data */
void free_downtime_data(void) {
scheduled_downtime *this_downtime = NULL;
scheduled_downtime *next_downtime = NULL;
fanout_destroy(dt_fanout, NULL);
/* free memory for the scheduled_downtime list */
for(this_downtime = scheduled_downtime_list; this_downtime != NULL; this_downtime = next_downtime) {
next_downtime = this_downtime->next;
my_free(this_downtime->host_name);
my_free(this_downtime->service_description);
my_free(this_downtime->author);
my_free(this_downtime->comment);
my_free(this_downtime);
}
/* reset list pointer */
scheduled_downtime_list = NULL;
return;
}
static int handle_worker_result(int sd, int events, void *arg)
{
wproc_object_job *oj = NULL;
char *buf, *error_reason = NULL;
unsigned long size;
int ret;
static struct kvvec kvv = KVVEC_INITIALIZER;
struct wproc_worker *wp = (struct wproc_worker *)arg;
if(iocache_capacity(wp->ioc) == 0) {
logit(NSLOG_RUNTIME_WARNING, TRUE, "wproc: iocache_capacity() is 0 for worker %s.\n", wp->name);
}
ret = iocache_read(wp->ioc, wp->sd);
if (ret < 0) {
logit(NSLOG_RUNTIME_WARNING, TRUE, "wproc: iocache_read() from %s returned %d: %s\n",
wp->name, ret, strerror(errno));
return 0;
} else if (ret == 0) {
logit(NSLOG_INFO_MESSAGE, TRUE, "wproc: Socket to worker %s broken, removing", wp->name);
wproc_num_workers_online--;
iobroker_unregister(nagios_iobs, sd);
if (workers.len <= 0) {
/* there aren't global workers left, we can't run any more checks
* we should try respawning a few of the standard ones
*/
logit(NSLOG_RUNTIME_ERROR, TRUE, "wproc: All our workers are dead, we can't do anything!");
}
remove_worker(wp);
fanout_destroy(wp->jobs, fo_reassign_wproc_job);
wp->jobs = NULL;
wproc_destroy(wp, 0);
return 0;
}
while ((buf = worker_ioc2msg(wp->ioc, &size, 0))) {
struct wproc_job *job;
wproc_result wpres;
/* log messages are handled first */
if (size > 5 && !memcmp(buf, "log=", 4)) {
logit(NSLOG_INFO_MESSAGE, TRUE, "wproc: %s: %s\n", wp->name, buf + 4);
continue;
}
/* for everything else we need to actually parse */
if (buf2kvvec_prealloc(&kvv, buf, size, '=', '\0', KVVEC_ASSIGN) <= 0) {
logit(NSLOG_RUNTIME_ERROR, TRUE,
"wproc: Failed to parse key/value vector from worker response with len %lu. First kv=%s",
size, buf ? buf : "(NULL)");
continue;
}
memset(&wpres, 0, sizeof(wpres));
wpres.job_id = -1;
wpres.type = -1;
wpres.response = &kvv;
parse_worker_result(&wpres, &kvv);
job = get_job(wp, wpres.job_id);
if (!job) {
logit(NSLOG_RUNTIME_WARNING, TRUE, "wproc: Job with id '%d' doesn't exist on %s.\n",
wpres.job_id, wp->name);
continue;
}
if (wpres.type != job->type) {
logit(NSLOG_RUNTIME_WARNING, TRUE, "wproc: %s claims job %d is type %d, but we think it's type %d\n",
wp->name, job->id, wpres.type, job->type);
break;
}
oj = (wproc_object_job *)job->arg;
/*
* ETIME ("Timer expired") doesn't really happen
* on any modern systems, so we reuse it to mean
* "program timed out"
*/
if (wpres.error_code == ETIME) {
wpres.early_timeout = TRUE;
}
if (wpres.early_timeout) {
asprintf(&error_reason, "timed out after %.2fs", tv_delta_f(&wpres.start, &wpres.stop));
}
else if (WIFSIGNALED(wpres.wait_status)) {
asprintf(&error_reason, "died by signal %d%s after %.2f seconds",
WTERMSIG(wpres.wait_status),
WCOREDUMP(wpres.wait_status) ? " (core dumped)" : "",
tv_delta_f(&wpres.start, &wpres.stop));
}
else if (job->type != WPJOB_CHECK && WEXITSTATUS(wpres.wait_status) != 0) {
asprintf(&error_reason, "is a non-check helper but exited with return code %d",
WEXITSTATUS(wpres.wait_status));
}
if (error_reason) {
logit(NSLOG_RUNTIME_ERROR, TRUE, "wproc: %s job %d from worker %s %s",
wpjob_type_name(job->type), job->id, wp->name, error_reason);
logit(NSLOG_RUNTIME_ERROR, TRUE, "wproc: command: %s\n", job->command);
if (job->type != WPJOB_CHECK && oj) {
logit(NSLOG_RUNTIME_ERROR, TRUE, "wproc: host=%s; service=%s; contact=%s\n",
oj->host_name ? oj->host_name : "(none)",
oj->service_description ? oj->service_description : "(none)",
oj->contact_name ? oj->contact_name : "(none)");
} else if (oj) {
struct check_result *cr = (struct check_result *)job->arg;
logit(NSLOG_RUNTIME_ERROR, TRUE, "wproc: host=%s; service=%s;\n",
cr->host_name, cr->service_description);
}
logit(NSLOG_RUNTIME_ERROR, TRUE, "wproc: early_timeout=%d; exited_ok=%d; wait_status=%d; error_code=%d;\n",
wpres.early_timeout, wpres.exited_ok, wpres.wait_status, wpres.error_code);
wproc_logdump_buffer(NSLOG_RUNTIME_ERROR, TRUE, "wproc: stderr", wpres.outerr);
wproc_logdump_buffer(NSLOG_RUNTIME_ERROR, TRUE, "wproc: stdout", wpres.outstd);
}
my_free(error_reason);
switch (job->type) {
case WPJOB_CHECK:
ret = handle_worker_check(&wpres, wp, job);
break;
case WPJOB_NOTIFY:
if (wpres.early_timeout) {
if (oj->service_description) {
logit(NSLOG_RUNTIME_WARNING, TRUE, "Warning: Notifying contact '%s' of service '%s' on host '%s' by command '%s' timed out after %.2f seconds\n",
oj->contact_name, oj->service_description,
oj->host_name, job->command,
tv2float(&wpres.runtime));
} else {
logit(NSLOG_RUNTIME_WARNING, TRUE, "Warning: Notifying contact '%s' of host '%s' by command '%s' timed out after %.2f seconds\n",
oj->contact_name, oj->host_name,
job->command, tv2float(&wpres.runtime));
}
}
break;
case WPJOB_OCSP:
if (wpres.early_timeout) {
logit(NSLOG_RUNTIME_WARNING, TRUE, "Warning: OCSP command '%s' for service '%s' on host '%s' timed out after %.2f seconds\n",
job->command, oj->service_description, oj->host_name,
tv2float(&wpres.runtime));
}
break;
case WPJOB_OCHP:
if (wpres.early_timeout) {
logit(NSLOG_RUNTIME_WARNING, TRUE, "Warning: OCHP command '%s' for host '%s' timed out after %.2f seconds\n",
job->command, oj->host_name, tv2float(&wpres.runtime));
}
break;
case WPJOB_GLOBAL_SVC_EVTHANDLER:
if (wpres.early_timeout) {
logit(NSLOG_EVENT_HANDLER | NSLOG_RUNTIME_WARNING, TRUE,
"Warning: Global service event handler command '%s' timed out after %.2f seconds\n",
job->command, tv2float(&wpres.runtime));
}
break;
case WPJOB_SVC_EVTHANDLER:
if (wpres.early_timeout) {
logit(NSLOG_EVENT_HANDLER | NSLOG_RUNTIME_WARNING, TRUE,
"Warning: Service event handler command '%s' timed out after %.2f seconds\n",
job->command, tv2float(&wpres.runtime));
}
break;
case WPJOB_GLOBAL_HOST_EVTHANDLER:
if (wpres.early_timeout) {
logit(NSLOG_EVENT_HANDLER | NSLOG_RUNTIME_WARNING, TRUE,
"Warning: Global host event handler command '%s' timed out after %.2f seconds\n",
job->command, tv2float(&wpres.runtime));
}
break;
case WPJOB_HOST_EVTHANDLER:
if (wpres.early_timeout) {
logit(NSLOG_EVENT_HANDLER | NSLOG_RUNTIME_WARNING, TRUE,
"Warning: Host event handler command '%s' timed out after %.2f seconds\n",
job->command, tv2float(&wpres.runtime));
}
break;
case WPJOB_CALLBACK:
run_job_callback(job, &wpres, 0);
break;
default:
logit(NSLOG_RUNTIME_WARNING, TRUE, "Worker %d: Unknown jobtype: %d\n", wp->pid, job->type);
break;
}
destroy_job(job);
}
return 0;
}