ping903q: improve output in Nagios mode; treat initial state as OK

New option -N is provided to treat the initial state as UNKNOWN as
in previous versions.

2 files changed, 82 insertions, 21 deletions

diff --git a/doc/ping903q.1 b/doc/ping903q.1
index 5da343d..84bd850 100644
--- a/doc/ping903q.1
+++ b/doc/ping903q.1
@@ -13,7 +13,7 @@
 .\"
 .\" You should have received a copy of the GNU General Public License
 .\" along with Ping903.  If not, see <http://www.gnu.org/licenses/>.
-.TH PING903Q 1 "February 27, 2020" "PING903Q" "User Commands"
+.TH PING903Q 1 "March 5, 2020" "PING903Q" "User Commands"
 .SH NAME
 ping903q \- ping903 query tool
 .SH SYNOPSIS
@@ -24,6 +24,7 @@ ping903q \- ping903 query tool
  [\fIIP\fR]
 .PP
 \fBping903q\fR\
+  [\fB\-N\fR]\
   [\fB\-f \fIFILE\fR]\
   [\fB\-p \fIPREFIX\fR]\
   [\fB\-R \fIREALM\fR]\
@@ -83,7 +84,7 @@ request, without sending unauthorized request first and consulting the
 reply.
 .SH EXIT CODE
 When called with one argument, the program exits with code 0 (success)
-if the IP is alive and 2 otherwise.
+if the IP is alive, 2 if it is not, and 3 if the host status is unknown.
 .PP
 When called without arguments, the program exits with code 0 if all
 monitored IP addresses are alive, 2 if none of them is reachable and 1
@@ -99,6 +100,16 @@ Warning condition.
 .TP
 .B 2
 Critical condition.
+.TP
+.B 3
+Unknown.  This code is returned in the following cases: (1) the host
+is not monitored by the
+.B ping903
+server, (2) the server replied with a HTTP code other than 200 or the
+reply was otherwise non-compliant, and (3) if the server has not
+yet collected enough data for that host and the command line option
+.B \-N
+was given.
 .PP
 In match mode, the program exits with code 0 if the requested host is
 monitored by the server, and with code 2 if it is not.
@@ -155,6 +166,11 @@ arguments.
 .PP
 Other options valid in this mode:
 .TP
+\fB\-N\fR
+By default hosts in initial state (i.e. for which no data has been
+collected) are treated as "OK" (exit status 0).  This option changes
+the default to treat them as "UNKNOWN" (exit status 3).
+.TP
 \fB\-p \fIPREFIX\fR
 Supplies the prefix to be displayed before Nagios status string.  The
 default is "PING".  The \fIPREFIX\fR string can contain the
diff --git a/src/ping903q.c b/src/ping903q.c
index 8323f44..64c17bb 100644
--- a/src/ping903q.c
+++ b/src/ping903q.c
@@ -35,6 +35,7 @@ static char *config_file = DEFAULT_CONFIG_FILE;
 FILE *http;
 int verbose;
 int resolve_ip;
+int nodata_ok = 1;
 char *nagios_prefix_format = "PING";
 char const http_version[] = "HTTP/1.1";
 char *cred_file_name;
@@ -754,6 +755,14 @@ print_host_status(struct json_value *obj, void *unused)
 	return alive ? EX_NAGIOS_OK : EX_NAGIOS_CRITICAL;
 }
 
+static int
+unknown_host_status(char const *host, char const *msg)
+{
+	abend("%s: %s", host, msg);
+	return EX_NAGIOS_UNKNOWN;
+}
+
+
 static char *
 resolve_host(char const *name)
 {
@@ -777,7 +786,9 @@ resolve_host(char const *name)
 }
 
 static void
-query_host(char const *name, int (*report)(struct json_value *, void *),
+query_host(char const *name,
+	   int (*report)(struct json_value *, void *),
+	   int (*unknown)(char const*, char const*),
 	   void *report_data)
 {
 	int rc;
@@ -800,21 +811,27 @@ query_host(char const *name, int (*report)(struct json_value *, void *),
 	}
 	http_resp_init(&resp);
 	http_query("GET", url, std_headers, &resp);
-	if (resp.code != 200)
-		abend("%s", resp.reason);
+	switch (resp.code) {
+	case 200:
+		break;
+	case 404:
+		exit(unknown(name, "host is not monitored"));
+	default:
+		exit(unknown(name, resp.reason));
+	}
 
 	hval = http_resp_get_header(&resp, "content-type");
 	if (!hval || strcmp(hval, "application/json")) {
-		abend("missing or unsupported content type");
+		exit(unknown(name, "missing or unsupported content type"));
 	}
 
 	rc = json_parse_string(resp.content, &obj, &p);
 	if (rc != JSON_E_NOERR)
 		abend("%s near %s", json_strerror(rc), p);
 	if (obj->type == json_null)
-		abend("%s (%s): host is not being monitored", name, ipstr);
+		exit(unknown(name, "host is not monitored"));
 	if (obj->type != json_object)
-		abend("returned entity has wrong type");
+		exit(unknown(name, "returned entity has wrong type"));
 	exit(report(obj, report_data));
 }
 
@@ -959,13 +976,13 @@ static inline int newstatus(int old, int new) {
 	return old > new ? old : new;
 }
 
+/* Format initial prefix */
 static void
-nagios_format_output(int status, double loss, double rta,
-		     struct nagios_check_data *cd, char const *name)
+nagios_format_prefix(char const *name, int status)
 {
 	char *p;
 	size_t len;
-        /* Format initial prefix */
+
 	p = nagios_prefix_format;
 	while (*p) {
 		len = strcspn(p, "%");
@@ -983,10 +1000,13 @@ nagios_format_output(int status, double loss, double rta,
 		}
 	}
 	printf(" %s ", status_str[status]);
-	if (status == EX_NAGIOS_UNKNOWN) {
-		printf("%s\n", "- waiting for data to arrive");
-		return;
-	}
+}
+
+static void
+nagios_format_output(int status, double loss, double rta,
+		     struct nagios_check_data *cd, char const *name)
+{
+	nagios_format_prefix(name, status);
 	if (loss >= 100) {
 		printf("%s", "- Packet loss = 100%");
 		rta = cd->cth.round_trip;
@@ -1010,6 +1030,12 @@ nagios_check(struct json_value *obj, void *data)
 
 	jv = ejson_get(obj, "status", json_string);
 	if (strcmp(jv->v.s, "init") == 0) {
+		if (nodata_ok) {
+			status = EX_NAGIOS_OK;
+			nagios_format_prefix(name, status);
+			printf("- %s\n", "waiting for data to arrive");
+			return status;
+		}
 		status = EX_NAGIOS_UNKNOWN;
 	} else if (strcmp(jv->v.s, "invalid") == 0
 		   || json_object_get(obj, "stddev", &jv)) {
@@ -1031,14 +1057,24 @@ nagios_check(struct json_value *obj, void *data)
 			status = newstatus(status, EX_NAGIOS_WARNING);
 	}
 	nagios_format_output(status, loss, rta, cd, name);
-	exit(status);
+	return status;
 }
+
+static int
+nagios_unknown(char const *name, char const *msg)
+{
+	nagios_format_prefix(name, EX_NAGIOS_UNKNOWN);
+	printf("- %s\n", msg);
+	return EX_NAGIOS_UNKNOWN;
+}
+
 
 void
 usage(void)
 {
 	printf("Usage: %s [-hrVv] [-f FILE] [HOST]\n", progname);
-	printf("   or: %s -H HOST -c RTA,PCT%% -w RTA,PCT%%\n", progname);
+	printf("   or: %s [-N] [-p PREFIX] -H HOST -c RTA,PCT%% -w RTA,PCT%%\n",
+	       progname);
 	printf("   or: %s -m HOST\n", progname);
 	printf("Query ping903 daemon.\n");
 	printf("\n");
@@ -1053,7 +1089,11 @@ usage(void)
 	printf("  -H HOST        query statistics for this host\n");
 	printf("  -c RTA,PCT%%    set critical threshold\n");
 	printf("  -w RTA,PCT%%    set warning threshold\n");
-	printf("\n(all three must be given in this mode)\n");
+	printf("\n  (all three must be given in this mode)\n\n");
+	printf("  -N             treat initial state as UNKNOWN, "
+	       "instead of OK\n");
+	printf("  -p PREFIX      prefix output with this string, "
+	       "instead of the default \"PING\"\n");
 	printf("\nMatch mode:\n\n");
 	printf("  -m             check whether HOST is monitored\n");
 	printf("\n");
@@ -1109,7 +1149,7 @@ main(int argc, char **argv)
 			exit(0);
 		}
 	}
-	while ((c = getopt(argc, argv, "c:f:H:hmp:R:rVvw:")) != EOF) {
+	while ((c = getopt(argc, argv, "c:f:H:hNmp:R:rVvw:")) != EOF) {
 		switch (c) {
 		case 'c':
 			c_opt = optarg;
@@ -1125,6 +1165,9 @@ main(int argc, char **argv)
 		case 'h':
 			usage();
 			exit(0);
+		case 'N':
+			nodata_ok = 0;
+			break;
 		case 'm':
 			setmode(MODE_MATCH, c);
 			break;
@@ -1201,14 +1244,16 @@ main(int argc, char **argv)
 		case 0:
 			query_all();
 		case 1:
-			query_host(argv[0], print_host_status, NULL);
+			query_host(argv[0],
+				   print_host_status,
+				   unknown_host_status, NULL);
 		default:
 			abend("bad number of arguments");
 		}
 		break;
 
 	case MODE_NAGIOS_CHECK:
-		query_host(host, nagios_check, &chkdata);
+		query_host(host, nagios_check, nagios_unknown, &chkdata);
 		break;
 
 	case MODE_MATCH: