int main(int argc, char *argv[])
{
char *msg;
int seq;
int argi;
int alertcolors, alertinterval;
char *configfn = NULL;
char *checkfn = NULL;
int checkpointinterval = 900;
char acklogfn[PATH_MAX];
FILE *acklogfd = NULL;
char notiflogfn[PATH_MAX];
FILE *notiflogfd = NULL;
char *tracefn = NULL;
struct sigaction sa;
int configchanged;
time_t lastxmit = 0;
MEMDEFINE(acklogfn);
MEMDEFINE(notiflogfn);
libxymon_init(argv[0]);
/* Dont save the error buffer */
save_errbuf = 0;
/* Load alert config */
alertcolors = colorset(xgetenv("ALERTCOLORS"), ((1 << COL_GREEN) | (1 << COL_BLUE)));
alertinterval = 60*atoi(xgetenv("ALERTREPEAT"));
/* Create our loookup-trees */
hostnames = xtreeNew(strcasecmp);
testnames = xtreeNew(strcasecmp);
locations = xtreeNew(strcasecmp);
for (argi=1; (argi < argc); argi++) {
if (argnmatch(argv[argi], "--config=")) {
configfn = strdup(strchr(argv[argi], '=')+1);
}
else if (argnmatch(argv[argi], "--checkpoint-file=")) {
checkfn = strdup(strchr(argv[argi], '=')+1);
}
else if (argnmatch(argv[argi], "--checkpoint-interval=")) {
char *p = strchr(argv[argi], '=') + 1;
checkpointinterval = atoi(p);
}
else if (argnmatch(argv[argi], "--dump-config")) {
load_alertconfig(configfn, alertcolors, alertinterval);
dump_alertconfig(1);
return 0;
}
else if (argnmatch(argv[argi], "--cfid")) {
include_configid = 1;
}
else if (argnmatch(argv[argi], "--test")) {
char *testhost = NULL, *testservice = NULL, *testpage = NULL,
*testcolor = "red", *testgroups = NULL;
void *hinfo;
int testdur = 0;
FILE *logfd = NULL;
activealerts_t *awalk = NULL;
int paramno = 0;
argi++; if (argi < argc) testhost = argv[argi];
argi++; if (argi < argc) testservice = argv[argi];
argi++;
while (argi < argc) {
if (strncasecmp(argv[argi], "--duration=", 11) == 0) {
testdur = durationvalue(strchr(argv[argi], '=')+1);
}
else if (strncasecmp(argv[argi], "--color=", 8) == 0) {
testcolor = strchr(argv[argi], '=')+1;
}
else if (strncasecmp(argv[argi], "--group=", 8) == 0) {
testgroups = strchr(argv[argi], '=')+1;
}
else if (strncasecmp(argv[argi], "--time=", 7) == 0) {
fakestarttime = (time_t)atoi(strchr(argv[argi], '=')+1);
}
else {
paramno++;
if (paramno == 1) testdur = atoi(argv[argi]);
else if (paramno == 2) testcolor = argv[argi];
else if (paramno == 3) fakestarttime = (time_t) atoi(argv[argi]);
}
argi++;
}
if ((testhost == NULL) || (testservice == NULL)) {
printf("Usage: xymond_alert --test HOST SERVICE [options]\n");
printf("Possible options:\n\t[--duration=MINUTES]\n\t[--color=COLOR]\n\t[--group=GROUPNAME]\n\t[--time=TIMESPEC]\n");
return 1;
}
load_hostnames(xgetenv("HOSTSCFG"), NULL, get_fqdn());
hinfo = hostinfo(testhost);
if (hinfo) {
testpage = strdup(xmh_item(hinfo, XMH_ALLPAGEPATHS));
}
else {
errprintf("Host not found in hosts.cfg - assuming it is on the top page\n");
testpage = "";
}
awalk = (activealerts_t *)calloc(1, sizeof(activealerts_t));
awalk->hostname = find_name(hostnames, testhost);
awalk->testname = find_name(testnames, testservice);
awalk->location = find_name(locations, testpage);
awalk->ip = strdup("127.0.0.1");
awalk->color = awalk->maxcolor = parse_color(testcolor);
awalk->pagemessage = "Test of the alert configuration";
awalk->eventstart = getcurrenttime(NULL) - testdur*60;
awalk->groups = (testgroups ? strdup(testgroups) : NULL);
awalk->state = A_PAGING;
awalk->cookie = 12345;
awalk->next = NULL;
logfd = fopen("/dev/null", "w");
starttrace(NULL);
testonly = 1;
load_alertconfig(configfn, alertcolors, alertinterval);
load_holidays(0);
send_alert(awalk, logfd);
return 0;
}
else if (argnmatch(argv[argi], "--trace=")) {
tracefn = strdup(strchr(argv[argi], '=')+1);
starttrace(tracefn);
}
else if (net_worker_option(argv[argi])) {
/* Handled in the subroutine */
}
else if (standardoption(argv[argi])) {
if (showhelp) return 0;
}
else {
errprintf("Unknown option '%s'\n", argv[argi]);
}
}
/* Do the network stuff if needed */
net_worker_run(ST_ALERT, LOC_SINGLESERVER, NULL);
if (checkfn) {
load_checkpoint(checkfn);
nextcheckpoint = gettimer() + checkpointinterval;
dbgprintf("Next checkpoint at %d, interval %d\n", (int) nextcheckpoint, checkpointinterval);
}
setup_signalhandler("xymond_alert");
/* Need to handle these ourselves, so we can shutdown and save state-info */
memset(&sa, 0, sizeof(sa));
sa.sa_handler = sig_handler;
sigaction(SIGPIPE, &sa, NULL);
sigaction(SIGTERM, &sa, NULL);
sigaction(SIGINT, &sa, NULL);
sigaction(SIGCHLD, &sa, NULL);
sigaction(SIGUSR1, &sa, NULL);
sigaction(SIGHUP, &sa, NULL);
if (xgetenv("XYMONSERVERLOGS")) {
sprintf(acklogfn, "%s/acknowledge.log", xgetenv("XYMONSERVERLOGS"));
acklogfd = fopen(acklogfn, "a");
sprintf(notiflogfn, "%s/notifications.log", xgetenv("XYMONSERVERLOGS"));
notiflogfd = fopen(notiflogfn, "a");
}
/*
* The general idea here is that this loop handles receiving of alert-
* and ack-messages from the master daemon, and maintains a list of
* host+test combinations that may have alerts going out.
*
* This module does not deal with any specific alert-configuration,
* it just picks up the alert messages, maintains the list of
* known tests that are in some sort of critical condition, and
* periodically pushes alerts to the do_alert.c module for handling.
*
* The only modification of alerts that happen here is the handling
* of when the next alert is due. It calls into the next_alert()
* routine to learn when an alert should be repeated, and also
* deals with Acknowledgments that stop alerts from going out for
* a period of time.
*/
while (running) {
char *eoln, *restofmsg;
char *metadata[20];
char *p;
int metacount;
char *hostname = NULL, *testname = NULL;
struct timespec timeout;
time_t now, nowtimer;
int anytogo;
activealerts_t *awalk;
int childstat;
nowtimer = gettimer();
if (checkfn && (nowtimer > nextcheckpoint)) {
dbgprintf("Saving checkpoint\n");
nextcheckpoint = nowtimer + checkpointinterval;
save_checkpoint(checkfn);
if (acklogfd) acklogfd = freopen(acklogfn, "a", acklogfd);
if (notiflogfd) notiflogfd = freopen(notiflogfn, "a", notiflogfd);
}
timeout.tv_sec = 60; timeout.tv_nsec = 0;
msg = get_xymond_message(C_PAGE, "xymond_alert", &seq, &timeout);
if (msg == NULL) {
running = 0;
continue;
}
/* See what time it is - must happen AFTER the timeout */
now = getcurrenttime(NULL);
/* Split the message in the first line (with meta-data), and the rest */
eoln = strchr(msg, '\n');
if (eoln) {
*eoln = '\0';
restofmsg = eoln+1;
}
else {
restofmsg = "";
}
/*
* Now parse the meta-data into elements.
* We use our own "gettok()" routine which works
* like strtok(), but can handle empty elements.
*/
metacount = 0;
memset(&metadata, 0, sizeof(metadata));
p = gettok(msg, "|");
while (p && (metacount < 19)) {
metadata[metacount] = p;
metacount++;
p = gettok(NULL, "|");
}
metadata[metacount] = NULL;
if (metacount > 3) hostname = metadata[3];
if (metacount > 4) testname = metadata[4];
if ((metacount > 10) && (strncmp(metadata[0], "@@page", 6) == 0)) {
/* @@page|timestamp|sender|hostname|testname|hostip|expiretime|color|prevcolor|changetime|location|cookie|osname|classname|grouplist|modifiers */
int newcolor, newalertstatus, oldalertstatus;
dbgprintf("Got page message from %s:%s\n", hostname, testname);
traceprintf("@@page %s:%s:%s=%s\n", hostname, testname, metadata[10], metadata[7]);
awalk = find_active(hostname, testname);
if (awalk == NULL) {
char *hwalk = find_name(hostnames, hostname);
char *twalk = find_name(testnames, testname);
char *pwalk = find_name(locations, metadata[10]);
awalk = (activealerts_t *)calloc(1, sizeof(activealerts_t));
awalk->hostname = hwalk;
awalk->testname = twalk;
awalk->location = pwalk;
awalk->cookie = -1;
awalk->state = A_DEAD;
/*
* Use changetime here, if we restart the alert module then
* this gets the duration values more right than using "now".
* Also, define this only when a new alert arrives - we should
* NOT clear this when a status goes yellow->red, or if it
* flaps between yellow and red.
*/
awalk->eventstart = atoi(metadata[9]);
add_active(awalk->hostname, awalk);
traceprintf("New record\n");
}
newcolor = parse_color(metadata[7]);
oldalertstatus = ((alertcolors & (1 << awalk->color)) != 0);
newalertstatus = ((alertcolors & (1 << newcolor)) != 0);
traceprintf("state %d->%d\n", oldalertstatus, newalertstatus);
if (newalertstatus) {
/* It's in an alert state. */
awalk->color = newcolor;
awalk->state = A_PAGING;
if (newcolor > awalk->maxcolor) {
if (awalk->maxcolor != 0) {
/*
* Severity has increased (yellow -> red).
* Clear the repeat-interval, and set maxcolor to
* the new color. If it drops to yellow again,
* maxcolor stays at red, so a test that flaps
* between yellow and red will only alert on red
* the first time, and then follow the repeat
* interval.
*/
dbgprintf("Severity increased, cleared repeat interval: %s/%s %s->%s\n",
awalk->hostname, awalk->testname,
colorname(awalk->maxcolor), colorname(newcolor));
clear_interval(awalk);
}
awalk->maxcolor = newcolor;
}
}
else {
/*
* Send one "recovered" message out now, then go to A_DEAD.
* Dont update the color here - we want recoveries to go out
* only if the alert color triggered an alert
*/
awalk->state = (newcolor == COL_BLUE) ? A_DISABLED : A_RECOVERED;
}
if (oldalertstatus != newalertstatus) {
dbgprintf("Alert status changed from %d to %d\n", oldalertstatus, newalertstatus);
clear_interval(awalk);
}
if (awalk->ip) xfree(awalk->ip);
awalk->ip = strdup(metadata[5]);
awalk->cookie = atoi(metadata[11]);
if (awalk->osname) xfree(awalk->osname);
awalk->osname = (metadata[12] ? strdup(metadata[12]) : NULL);
if (awalk->classname) xfree(awalk->classname);
awalk->classname = (metadata[13] ? strdup(metadata[13]) : NULL);
if (awalk->groups) xfree(awalk->groups);
awalk->groups = (metadata[14] ? strdup(metadata[14]) : NULL);
if (awalk->pagemessage) xfree(awalk->pagemessage);
if (metadata[15]) {
/* Modifiers are more interesting than the message itself */
awalk->pagemessage = (char *)malloc(strlen(awalk->hostname) + strlen(awalk->testname) + strlen(colorname(awalk->color)) + strlen(metadata[15]) + strlen(restofmsg) + 10);
sprintf(awalk->pagemessage, "%s:%s %s\n%s\n%s",
awalk->hostname, awalk->testname, colorname(awalk->color), metadata[15], restofmsg);
}
else {
awalk->pagemessage = strdup(restofmsg);
}
}
else if ((metacount > 5) && (strncmp(metadata[0], "@@ack", 5) == 0)) {
/* @@ack|timestamp|sender|hostname|testname|hostip|expiretime */
/*
* An ack is handled simply by setting the next
* alert-time to when the ack expires.
*/
time_t nextalert = atoi(metadata[6]);
dbgprintf("Got ack message from %s:%s\n", hostname, testname);
traceprintf("@@ack: %s:%s now=%d, ackeduntil %d\n",
hostname, testname, (int)now, (int)nextalert);
awalk = find_active(hostname, testname);
if (acklogfd) {
int cookie = (awalk ? awalk->cookie : -1);
int color = (awalk ? awalk->color : 0);
fprintf(acklogfd, "%d\t%d\t%d\t%d\t%s\t%s.%s\t%s\t%s\n",
(int)now, cookie,
(int)((nextalert - now) / 60), cookie,
"np_filename_not_used",
hostname, testname,
colorname(color),
nlencode(restofmsg));
fflush(acklogfd);
}
if (awalk && (awalk->state == A_PAGING)) {
traceprintf("Record updated\n");
awalk->state = A_ACKED;
awalk->nextalerttime = nextalert;
if (awalk->ackmessage) xfree(awalk->ackmessage);
awalk->ackmessage = strdup(restofmsg);
}
else {
traceprintf("No record\n");
}
}
else if ((metacount > 4) && (strncmp(metadata[0], "@@notify", 5) == 0)) {
/* @@notify|timestamp|sender|hostname|testname|pagepath */
char *hwalk = find_name(hostnames, hostname);
char *twalk = find_name(testnames, testname);
char *pwalk = find_name(locations, (metadata[5] ? metadata[5] : ""));
awalk = (activealerts_t *)calloc(1, sizeof(activealerts_t));
awalk->hostname = hwalk;
awalk->testname = twalk;
awalk->location = pwalk;
awalk->cookie = -1;
awalk->pagemessage = strdup(restofmsg);
awalk->eventstart = getcurrenttime(NULL);
awalk->state = A_NOTIFY;
add_active(awalk->hostname, awalk);
}
else if ((metacount > 3) &&
((strncmp(metadata[0], "@@drophost", 10) == 0) || (strncmp(metadata[0], "@@dropstate", 11) == 0))) {
/* @@drophost|timestamp|sender|hostname */
/* @@dropstate|timestamp|sender|hostname */
xtreePos_t handle;
handle = xtreeFind(hostnames, hostname);
if (handle != xtreeEnd(hostnames)) {
alertanchor_t *anchor = (alertanchor_t *)xtreeData(hostnames, handle);
for (awalk = anchor->head; (awalk); awalk = awalk->next) awalk->state = A_DEAD;
}
}
else if ((metacount > 4) && (strncmp(metadata[0], "@@droptest", 10) == 0)) {
/* @@droptest|timestamp|sender|hostname|testname */
awalk = find_active(hostname, testname);
if (awalk) awalk->state = A_DEAD;
}
else if ((metacount > 4) && (strncmp(metadata[0], "@@renamehost", 12) == 0)) {
/* @@renamehost|timestamp|sender|hostname|newhostname */
/*
* We handle rename's simply by dropping the alert. If there is still an
* active alert for the host, it will have to be dealt with when the next
* status update arrives.
*/
xtreePos_t handle;
handle = xtreeFind(hostnames, hostname);
if (handle != xtreeEnd(hostnames)) {
alertanchor_t *anchor = (alertanchor_t *)xtreeData(hostnames, handle);
for (awalk = anchor->head; (awalk); awalk = awalk->next) awalk->state = A_DEAD;
}
}
else if ((metacount > 5) && (strncmp(metadata[0], "@@renametest", 12) == 0)) {
/* @@renametest|timestamp|sender|hostname|oldtestname|newtestname */
/*
* We handle rename's simply by dropping the alert. If there is still an
* active alert for the host, it will have to be dealt with when the next
* status update arrives.
*/
awalk = find_active(hostname, testname);
if (awalk) awalk->state = A_DEAD;
}
else if (strncmp(metadata[0], "@@shutdown", 10) == 0) {
running = 0;
errprintf("Got a shutdown message\n");
continue;
}
else if (strncmp(metadata[0], "@@logrotate", 11) == 0) {
char *fn = xgetenv("XYMONCHANNEL_LOGFILENAME");
if (fn && strlen(fn)) {
reopen_file(fn, "a", stdout);
reopen_file(fn, "a", stderr);
if (tracefn) {
stoptrace();
starttrace(tracefn);
}
}
continue;
}
else if (strncmp(metadata[0], "@@reload", 8) == 0) {
/* Nothing ... right now */
}
else if (strncmp(metadata[0], "@@idle", 6) == 0) {
/* Timeout */
}
/*
* When a burst of alerts happen, we get lots of alert messages
* coming in quickly. So lets handle them in bunches and only
* do the full alert handling once every 10 secs - that lets us
* combine a bunch of alerts into one transmission process.
*/
if (nowtimer < (lastxmit+10)) continue;
lastxmit = nowtimer;
/*
* Loop through the activealerts list and see if anything is pending.
* This is an optimization, we could just as well just fork off the
* notification child and let it handle all of it. But there is no
* reason to fork a child process unless it is going to do something.
*/
configchanged = load_alertconfig(configfn, alertcolors, alertinterval);
configchanged += load_holidays(0);
anytogo = 0;
for (awalk = alistBegin(); (awalk); awalk = alistNext()) {
int anymatch = 0;
switch (awalk->state) {
case A_NORECIP:
if (!configchanged) break;
/* The configuration has changed - switch NORECIP -> PAGING */
awalk->state = A_PAGING;
clear_interval(awalk);
/* Fall through */
case A_PAGING:
if (have_recipient(awalk, &anymatch)) {
if (awalk->nextalerttime <= now) anytogo++;
}
else {
if (!anymatch) {
awalk->state = A_NORECIP;
cleanup_alert(awalk);
}
}
break;
case A_ACKED:
if (awalk->nextalerttime <= now) {
/* An ack has expired, so drop the ack message and switch to A_PAGING */
anytogo++;
if (awalk->ackmessage) xfree(awalk->ackmessage);
awalk->state = A_PAGING;
}
break;
case A_RECOVERED:
case A_DISABLED:
case A_NOTIFY:
anytogo++;
break;
case A_DEAD:
break;
}
}
dbgprintf("%d alerts to go\n", anytogo);
if (anytogo) {
pid_t childpid;
childpid = fork();
if (childpid == 0) {
/* The child */
start_alerts();
for (awalk = alistBegin(); (awalk); awalk = alistNext()) {
switch (awalk->state) {
case A_PAGING:
if (awalk->nextalerttime <= now) {
send_alert(awalk, notiflogfd);
}
break;
case A_ACKED:
/* Cannot be A_ACKED unless the ack is still valid, so no alert. */
break;
case A_RECOVERED:
case A_DISABLED:
case A_NOTIFY:
send_alert(awalk, notiflogfd);
break;
case A_NORECIP:
case A_DEAD:
break;
}
}
finish_alerts();
/* Child does not continue */
exit(0);
}
else if (childpid < 0) {
errprintf("Fork failed, cannot send alerts: %s\n", strerror(errno));
}
}
/* Update the state flag and the next-alert timestamp */
for (awalk = alistBegin(); (awalk); awalk = alistNext()) {
switch (awalk->state) {
case A_PAGING:
if (awalk->nextalerttime <= now) awalk->nextalerttime = next_alert(awalk);
break;
case A_NORECIP:
break;
case A_ACKED:
/* Still cannot get here except if ack is still valid */
break;
case A_RECOVERED:
case A_DISABLED:
case A_NOTIFY:
awalk->state = A_DEAD;
/* Fall through */
case A_DEAD:
cleanup_alert(awalk);
break;
}
}
clean_all_active();
/* Pickup any finished child processes to avoid zombies */
while (wait3(&childstat, WNOHANG, NULL) > 0) ;
}
if (checkfn) save_checkpoint(checkfn);
if (acklogfd) fclose(acklogfd);
if (notiflogfd) fclose(notiflogfd);
stoptrace();
MEMUNDEFINE(notiflogfn);
MEMUNDEFINE(acklogfn);
if (termsig >= 0) {
errprintf("Terminated by signal %d\n", termsig);
}
return 0;
}
int main(int argc, char *argv[])
{
char *msg;
int argi;
struct sigaction sa;
char *exthandler = NULL;
char *extids = NULL;
char *processor = NULL;
struct sockaddr_un ctlsockaddr;
int ctlsocket;
/* Handle program options. */
for (argi = 1; (argi < argc); argi++) {
if (strcmp(argv[argi], "--debug") == 0) {
debug = 1;
}
else if (argnmatch(argv[argi], "--rrddir=")) {
char *p = strchr(argv[argi], '=');
rrddir = strdup(p+1);
}
else if (argnmatch(argv[argi], "--extra-script=")) {
char *p = strchr(argv[argi], '=');
exthandler = strdup(p+1);
}
else if (argnmatch(argv[argi], "--extra-tests=")) {
char *p = strchr(argv[argi], '=');
extids = strdup(p+1);
}
else if (argnmatch(argv[argi], "--processor=")) {
char *p = strchr(argv[argi], '=');
processor = strdup(p+1);
}
else if (strcmp(argv[argi], "--no-cache") == 0) {
use_rrd_cache = 0;
}
else if (net_worker_option(argv[argi])) {
/* Handled in the subroutine */
}
}
save_errbuf = 0;
if ((rrddir == NULL) && xgetenv("XYMONRRDS")) {
rrddir = strdup(xgetenv("XYMONRRDS"));
}
if (exthandler && extids) setup_exthandler(exthandler, extids);
/* Do the network stuff if needed */
net_worker_run(ST_RRD, LOC_STICKY, update_locator_hostdata);
setup_signalhandler("xymond_rrd");
memset(&sa, 0, sizeof(sa));
sa.sa_handler = sig_handler;
sigaction(SIGHUP, &sa, NULL);
sigaction(SIGCHLD, &sa, NULL);
sigaction(SIGTERM, &sa, NULL);
sigaction(SIGINT, &sa, NULL);
signal(SIGPIPE, SIG_DFL);
/* Setup the control socket that receives cache-flush commands */
memset(&ctlsockaddr, 0, sizeof(ctlsockaddr));
sprintf(ctlsockaddr.sun_path, "%s/rrdctl.%d", xgetenv("XYMONTMP"), getpid());
unlink(ctlsockaddr.sun_path); /* In case it was accidentally left behind */
ctlsockaddr.sun_family = AF_UNIX;
ctlsocket = socket(AF_UNIX, SOCK_DGRAM, 0);
if (ctlsocket == -1) {
errprintf("Cannot create cache-control socket (%s)\n", strerror(errno));
return 1;
}
fcntl(ctlsocket, F_SETFL, O_NONBLOCK);
if (bind(ctlsocket, (struct sockaddr *)&ctlsockaddr, sizeof(ctlsockaddr)) == -1) {
errprintf("Cannot bind to cache-control socket (%s)\n", strerror(errno));
return 1;
}
/* Linux obeys filesystem permissions on the socket file, so make it world-accessible */
if (chmod(ctlsockaddr.sun_path, S_IRUSR|S_IWUSR|S_IRGRP|S_IWGRP|S_IROTH|S_IWOTH) == -1) {
errprintf("Setting permissions on cache-control socket failed: %s\n", strerror(errno));
}
/* Load the RRD definitions */
load_rrddefs();
/* If we are passing data to an external processor, create the pipe to it */
setup_extprocessor(processor);
running = 1;
while (running) {
char *eoln, *restofmsg = NULL;
char *metadata[MAX_META+1];
int metacount;
char *p;
char *hostname = NULL, *testname = NULL, *sender = NULL, *classname = NULL, *pagepaths = NULL;
xymonrrd_t *ldef = NULL;
time_t tstamp;
int childstat;
ssize_t n;
char ctlbuf[PATH_MAX];
int gotcachectlmessage;
time_t now;
/* See if we have any cache-control messages pending */
do {
n = recv(ctlsocket, ctlbuf, sizeof(ctlbuf), 0);
gotcachectlmessage = (n > 0);
if (gotcachectlmessage) {
/* We have a control message */
char *bol, *eol;
ctlbuf[n] = '\0';
bol = ctlbuf;
do {
eol = strchr(bol, '\n'); if (eol) *eol = '\0';
rrdcacheflushhost(bol);
if (eol) { bol = eol+1; } else bol = NULL;
} while (bol && *bol);
}
} while (gotcachectlmessage);
/* Get next message */
msg = get_xymond_message(C_LAST, argv[0], &seq, NULL);
if (msg == NULL) {
running = 0;
continue;
}
now = gettimer();
if (reloadtime < now) {
/* Reload configuration files */
load_hostnames(xgetenv("HOSTSCFG"), NULL, get_fqdn());
load_client_config(NULL);
reloadtime = now + 600;
}
/* Split the message in the first line (with meta-data), and the rest */
eoln = strchr(msg, '\n');
if (eoln) {
*eoln = '\0';
restofmsg = eoln+1;
}
/* Parse the meta-data */
metacount = 0;
memset(&metadata, 0, sizeof(metadata));
p = gettok(msg, "|");
while (p && (metacount < MAX_META)) {
metadata[metacount++] = p;
p = gettok(NULL, "|");
}
metadata[metacount] = NULL;
if ((metacount >= 14) && (strncmp(metadata[0], "@@status", 8) == 0)) {
/*
* @@status|timestamp|sender|origin|hostname|testname|expiretime|color|testflags|\
* prevcolor|changetime|ackexpiretime|ackmessage|disableexpiretime|disablemessage|\
* clienttstamp|flapping|classname|pagepaths
*/
int color = parse_color(metadata[7]);
switch (color) {
case COL_GREEN:
case COL_YELLOW:
case COL_RED:
case COL_BLUE: /* Blue is OK, because it only arrives here when an update is sent */
tstamp = atoi(metadata[1]);
sender = metadata[2];
hostname = metadata[4];
testname = metadata[5];
classname = (metadata[17] ? metadata[17] : "");
pagepaths = (metadata[18] ? metadata[18] : "");
ldef = find_xymon_rrd(testname, metadata[8]);
update_rrd(hostname, testname, restofmsg, tstamp, sender, ldef, classname, pagepaths);
break;
default:
/* Ignore reports with purple, blue or clear - they have no data we want. */
break;
}
}
else if ((metacount > 5) && (strncmp(metadata[0], "@@data", 6) == 0)) {
/* @@data|timestamp|sender|origin|hostname|testname|classname|pagepaths */
tstamp = atoi(metadata[1]);
sender = metadata[2];
hostname = metadata[4];
testname = metadata[5];
classname = (metadata[6] ? metadata[6] : "");
pagepaths = (metadata[7] ? metadata[7] : "");
ldef = find_xymon_rrd(testname, "");
update_rrd(hostname, testname, restofmsg, tstamp, sender, ldef, classname, pagepaths);
}
else if (strncmp(metadata[0], "@@shutdown", 10) == 0) {
running = 0;
continue;
}
else if (strncmp(metadata[0], "@@idle", 6) == 0) {
/* Ignored */
continue;
}
else if (strncmp(metadata[0], "@@logrotate", 11) == 0) {
char *fn = xgetenv("XYMONCHANNEL_LOGFILENAME");
if (fn && strlen(fn)) {
freopen(fn, "a", stdout);
freopen(fn, "a", stderr);
}
continue;
}
else if (strncmp(metadata[0], "@@reload", 8) == 0) {
reloadtime = 0;
}
else if ((metacount > 3) && (strncmp(metadata[0], "@@drophost", 10) == 0)) {
char hostdir[PATH_MAX];
hostname = metadata[3];
MEMDEFINE(hostdir);
sprintf(hostdir, "%s/%s", rrddir, hostname);
dropdirectory(hostdir, 1);
MEMUNDEFINE(hostdir);
}
else if ((metacount > 4) && (strncmp(metadata[0], "@@droptest", 10) == 0)) {
/*
* Not implemented. Mappings of testnames -> rrd files is
* too complex, so on the rare occasion that a single test
* is deleted, they will have to delete the rrd files themselves.
*/
}
else if ((metacount > 4) && (strncmp(metadata[0], "@@renamehost", 12) == 0)) {
char oldhostdir[PATH_MAX];
char newhostdir[PATH_MAX];
char *newhostname;
MEMDEFINE(oldhostdir);
MEMDEFINE(newhostdir);
hostname = metadata[3];
newhostname = metadata[4];
sprintf(oldhostdir, "%s/%s", rrddir, hostname);
sprintf(newhostdir, "%s/%s", rrddir, newhostname);
rename(oldhostdir, newhostdir);
if (net_worker_locatorbased()) locator_rename_host(hostname, newhostname, ST_RRD);
MEMUNDEFINE(newhostdir);
MEMUNDEFINE(oldhostdir);
}
else if ((metacount > 5) && (strncmp(metadata[0], "@@renametest", 12) == 0)) {
/* Not implemented. See "droptest". */
}
/*
* We fork a subprocess when processing drophost requests.
* Pickup any finished child processes to avoid zombies
*/
while (wait3(&childstat, WNOHANG, NULL) > 0) ;
}
/* Flush all cached updates to disk */
errprintf("Shutting down, flushing cached updates to disk\n");
rrdcacheflushall();
errprintf("Cache flush completed\n");
/* Close the external processor */
shutdown_extprocessor();
/* Close the control socket */
close(ctlsocket);
unlink(ctlsockaddr.sun_path);
return 0;
}
int main(int argc, char *argv[])
{
char *msg;
int running;
int argi, seq;
int recentperiod = 3600;
int maxrecentcount = 5;
int logdirfull = 0;
int minlogspace = 5;
struct sigaction sa;
/* Handle program options. */
for (argi = 1; (argi < argc); argi++) {
if (argnmatch(argv[argi], "--logdir=")) {
clientlogdir = strchr(argv[argi], '=')+1;
}
else if (argnmatch(argv[argi], "--recent-period=")) {
char *p = strchr(argv[argi], '=');
recentperiod = 60*atoi(p+1);
}
else if (argnmatch(argv[argi], "--recent-count=")) {
char *p = strchr(argv[argi], '=');
maxrecentcount = atoi(p+1);
}
else if (argnmatch(argv[argi], "--minimum-free=")) {
minlogspace = atoi(strchr(argv[argi], '=')+1);
}
else if (strcmp(argv[argi], "--debug") == 0) {
/*
* A global "debug" variable is available. If
* it is set, then "dbgprintf()" outputs debug messages.
*/
debug = 1;
}
else if (net_worker_option(argv[argi])) {
/* Handled in the subroutine */
}
}
if (clientlogdir == NULL) clientlogdir = xgetenv("CLIENTLOGS");
if (clientlogdir == NULL) {
clientlogdir = (char *)malloc(strlen(xgetenv("XYMONVAR")) + 10);
sprintf(clientlogdir, "%s/hostdata", xgetenv("XYMONVAR"));
}
save_errbuf = 0;
/* Do the network stuff if needed */
net_worker_run(ST_HOSTDATA, LOC_STICKY, update_locator_hostdata);
setup_signalhandler("xymond_hostdata");
memset(&sa, 0, sizeof(sa));
sa.sa_handler = sig_handler;
signal(SIGCHLD, SIG_IGN);
sigaction(SIGHUP, &sa, NULL);
signal(SIGPIPE, SIG_DFL);
savetimes = xtreeNew(strcasecmp);
running = 1;
while (running) {
char *eoln, *restofmsg, *p;
char *metadata[MAX_META+1];
int metacount;
msg = get_xymond_message(C_CLICHG, "xymond_hostdata", &seq, NULL);
if (msg == NULL) {
/*
* get_xymond_message will return NULL if xymond_channel closes
* the input pipe. We should shutdown when that happens.
*/
running = 0;
continue;
}
if (nextfscheck < gettimer()) {
logdirfull = (chkfreespace(clientlogdir, minlogspace, minlogspace) != 0);
if (logdirfull) errprintf("Hostdata directory %s has less than %d%% free space - disabling save of data for 5 minutes\n", clientlogdir, minlogspace);
nextfscheck = gettimer() + 300;
}
/* Split the message in the first line (with meta-data), and the rest */
eoln = strchr(msg, '\n');
if (eoln) {
*eoln = '\0';
restofmsg = eoln+1;
}
else {
restofmsg = "";
}
metacount = 0;
memset(&metadata, 0, sizeof(metadata));
p = gettok(msg, "|");
while (p && (metacount < MAX_META)) {
metadata[metacount++] = p;
p = gettok(NULL, "|");
}
metadata[metacount] = NULL;
if (strncmp(metadata[0], "@@clichg", 8) == 0) {
xtreePos_t handle;
savetimes_t *itm;
int i, recentcount;
time_t now = gettimer();
char hostdir[PATH_MAX];
char fn[PATH_MAX];
FILE *fd;
/* metadata[3] is the hostname */
handle = xtreeFind(savetimes, metadata[3]);
if (handle != xtreeEnd(savetimes)) {
itm = (savetimes_t *)xtreeData(savetimes, handle);
}
else {
itm = (savetimes_t *)calloc(1, sizeof(savetimes_t));
itm->hostname = strdup(metadata[3]);
xtreeAdd(savetimes, itm->hostname, itm);
}
/* See how many times we've saved the hostdata recently (within the past 'recentperiod' seconds) */
for (i=0, recentcount=0; ((i < 12) && (itm->tstamp[i] > (now - recentperiod))); i++) recentcount++;
/* If it's been saved less than 'maxrecentcount' times, then save it. Otherwise just drop it */
if (!logdirfull && (recentcount < maxrecentcount)) {
int written, closestatus, ok = 1;
for (i = 10; (i > 0); i--) itm->tstamp[i+1] = itm->tstamp[i];
itm->tstamp[0] = now;
sprintf(hostdir, "%s/%s", clientlogdir, metadata[3]);
mkdir(hostdir, S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH);
sprintf(fn, "%s/%s", hostdir, metadata[4]);
fd = fopen(fn, "w");
if (fd == NULL) {
errprintf("Cannot create file %s: %s\n", fn, strerror(errno));
continue;
}
written = fwrite(restofmsg, 1, strlen(restofmsg), fd);
if (written != strlen(restofmsg)) {
errprintf("Cannot write hostdata file %s: %s\n", fn, strerror(errno));
closestatus = fclose(fd); /* Ignore any close errors */
ok = 0;
}
else {
closestatus = fclose(fd);
if (closestatus != 0) {
errprintf("Cannot write hostdata file %s: %s\n", fn, strerror(errno));
ok = 0;
}
}
if (!ok) remove(fn);
}
}
/*
* A "shutdown" message is sent when the master daemon
* terminates. The child workers should shutdown also.
*/
else if (strncmp(metadata[0], "@@shutdown", 10) == 0) {
running = 0;
continue;
}
else if (strncmp(metadata[0], "@@idle", 6) == 0) {
/* Ignored */
continue;
}
/*
* A "logrotate" message is sent when the Xymon logs are
* rotated. The child workers must re-open their logfiles,
* typically stdin and stderr - the filename is always
* provided in the XYMONCHANNEL_LOGFILENAME environment.
*/
else if (strncmp(metadata[0], "@@logrotate", 11) == 0) {
char *fn = xgetenv("XYMONCHANNEL_LOGFILENAME");
if (fn && strlen(fn)) {
reopen_file(fn, "a", stdout);
reopen_file(fn, "a", stderr);
}
continue;
}
else if ((metacount > 3) && (strncmp(metadata[0], "@@drophost", 10) == 0)) {
/* @@drophost|timestamp|sender|hostname */
char hostdir[PATH_MAX];
snprintf(hostdir, sizeof(hostdir), "%s/%s", clientlogdir, basename(metadata[3]));
dropdirectory(hostdir, 1);
}
else if ((metacount > 4) && (strncmp(metadata[0], "@@renamehost", 12) == 0)) {
/* @@renamehost|timestamp|sender|hostname|newhostname */
char oldhostdir[PATH_MAX], newhostdir[PATH_MAX];
snprintf(oldhostdir, sizeof(oldhostdir), "%s/%s", clientlogdir, basename(metadata[3]));
snprintf(newhostdir, sizeof(newhostdir), "%s/%s", clientlogdir, basename(metadata[4]));
rename(oldhostdir, newhostdir);
if (net_worker_locatorbased()) locator_rename_host(metadata[3], metadata[4], ST_HOSTDATA);
}
else if (strncmp(metadata[0], "@@reload", 8) == 0) {
/* Do nothing */
}
}
return 0;
}