Let check_ntp_peer check the number of truechimers

Add support for checking the number of usable time sources (i.e., the
number of peers which are classified as so-called "truechimers" by NTP's
intersection algorithm).  The new "-m" and "-n" options allow for
specifying the according WARNING and CRITICAL thresholds (and thereby
activating the truechimers check), respectively.

diff --git a/NEWS b/NEWS
index 4a55502d45c361d3578edf3151f289f31a497f4d..c555dda00f98f6e93972c6f50a8774a82149aa84 100644 (file)
--- a/NEWS
+++ b/NEWS
@@ -2,6 +2,7 @@ This file documents the major additions and syntax changes between releases.
 
 1.4.15 ...
        ENHANCEMENTS
+       New check_ntp_peer -m and -n options to check the number of usable time sources ("truechimers")
        FIXES
        Fix check_ircd binding to wrong interface (#668778)
        Add proxy-authorization option to check_http (Marcel Kuiper - #1323230, Bryan Irvine - #2863925)

diff --git a/plugins/check_ntp_peer.c b/plugins/check_ntp_peer.c
index e8325bc82392c0d39e55ca4fe1b7102aff741512..d4689bc793b1a0826505788601b6a2add2acc44e 100644 (file)
--- a/plugins/check_ntp_peer.c
+++ b/plugins/check_ntp_peer.c
@@ -56,6 +56,9 @@ static char *scrit="-1:16";
 static short do_jitter=0;
 static char *jwarn="-1:5000";
 static char *jcrit="-1:10000";
+static short do_truechimers=0;
+static char *twarn="0:";
+static char *tcrit="0:";
 static int syncsource_found=0;
 static int li_alarm=0;
 
@@ -63,6 +66,7 @@ int process_arguments (int, char **);
 thresholds *offset_thresholds = NULL;
 thresholds *jitter_thresholds = NULL;
 thresholds *stratum_thresholds = NULL;
+thresholds *truechimer_thresholds = NULL;
 void print_help (void);
 void print_usage (void);
 
@@ -121,6 +125,7 @@ typedef struct {
 #define OP_READVAR  0x02
 /* In peer status bytes, bits 6,7,8 determine clock selection status */
 #define PEER_SEL(x) ((ntohs(x)>>8)&0x07)
+#define PEER_TRUECHIMER 0x02
 #define PEER_INCLUDED 0x04
 #define PEER_SYNCSOURCE 0x06
 
@@ -160,12 +165,12 @@ void print_ntp_control_message(const ntp_control_message *p){
                for(i=0;i<numpeers;i++){
                        printf("\tpeer id %.2x status %.2x",
                               ntohs(peer[i].assoc), ntohs(peer[i].status));
-                       if (PEER_SEL(peer[i].status) >= PEER_INCLUDED){
-                               if(PEER_SEL(peer[i].status) >= PEER_SYNCSOURCE){
-                                       printf(" <-- current sync source");
-                               } else {
-                                       printf(" <-- current sync candidate");
-                               }
+                       if(PEER_SEL(peer[i].status) >= PEER_SYNCSOURCE){
+                               printf(" <-- current sync source");
+                       } else if(PEER_SEL(peer[i].status) >= PEER_INCLUDED){
+                               printf(" <-- current sync candidate");
+                       } else if(PEER_SEL(peer[i].status) >= PEER_TRUECHIMER){
+                               printf(" <-- outlyer, but truechimer");
                        }
                        printf("\n");
                }
@@ -194,7 +199,7 @@ setup_control_request(ntp_control_message *p, uint8_t opcode, uint16_t seq){
  *  status is pretty much useless as syncsource_found is a global variable
  *  used later in main to check is the server was synchronized. It works
  *  so I left it alone */
-int ntp_request(const char *host, double *offset, int *offset_result, double *jitter, int *stratum){
+int ntp_request(const char *host, double *offset, int *offset_result, double *jitter, int *stratum, int *num_truechimers){
        int conn=-1, i, npeers=0, num_candidates=0;
        double tmp_offset = 0;
        int min_peer_sel=PEER_INCLUDED;
@@ -209,6 +214,7 @@ int ntp_request(const char *host, double *offset, int *offset_result, double *ji
        status = STATE_OK;
        *offset_result = STATE_UNKNOWN;
        *jitter = *stratum = -1;
+       *num_truechimers = 0;
 
        /* Long-winded explanation:
         * Getting the sync peer offset, jitter and stratum requires a number of
@@ -261,11 +267,14 @@ int ntp_request(const char *host, double *offset, int *offset_result, double *ji
         * at least some candidates. In the latter case we'll issue
         * a warning but go ahead with the check on them. */
        for (i = 0; i < npeers; i++){
-               if (PEER_SEL(peers[i].status) >= PEER_INCLUDED){
-                       num_candidates++;
-                       if(PEER_SEL(peers[i].status) >= PEER_SYNCSOURCE){
-                               syncsource_found=1;
-                               min_peer_sel=PEER_SYNCSOURCE;
+               if(PEER_SEL(peers[i].status) >= PEER_TRUECHIMER){
+                       (*num_truechimers)++;
+                       if(PEER_SEL(peers[i].status) >= PEER_INCLUDED){
+                               num_candidates++;
+                               if(PEER_SEL(peers[i].status) >= PEER_SYNCSOURCE){
+                                       syncsource_found=1;
+                                       min_peer_sel=PEER_SYNCSOURCE;
+                               }
                        }
                }
        }
@@ -413,6 +422,8 @@ int process_arguments(int argc, char **argv){
                {"scrit", required_argument, 0, 'C'},
                {"jwarn", required_argument, 0, 'j'},
                {"jcrit", required_argument, 0, 'k'},
+               {"twarn", required_argument, 0, 'm'},
+               {"tcrit", required_argument, 0, 'n'},
                {"timeout", required_argument, 0, 't'},
                {"hostname", required_argument, 0, 'H'},
                {"port", required_argument, 0, 'p'},
@@ -424,7 +435,7 @@ int process_arguments(int argc, char **argv){
                usage ("\n");
 
        while (1) {
-               c = getopt_long (argc, argv, "Vhv46qw:c:W:C:j:k:t:H:p:", longopts, &option);
+               c = getopt_long (argc, argv, "Vhv46qw:c:W:C:j:k:m:n:t:H:p:", longopts, &option);
                if (c == -1 || c == EOF || c == 1)
                        break;
 
@@ -467,6 +478,14 @@ int process_arguments(int argc, char **argv){
                        do_jitter=1;
                        jcrit = optarg;
                        break;
+               case 'm':
+                       do_truechimers=1;
+                       twarn = optarg;
+                       break;
+               case 'n':
+                       do_truechimers=1;
+                       tcrit = optarg;
+                       break;
                case 'H':
                        if(is_host(optarg) == FALSE)
                                usage2(_("Invalid hostname/address"), optarg);
@@ -526,8 +545,16 @@ char *perfd_stratum (int stratum)
                TRUE, 0, TRUE, 16);
 }
 
+char *perfd_truechimers (int num_truechimers)
+{
+       return perfdata ("truechimers", num_truechimers, "",
+               do_truechimers, (int)truechimer_thresholds->warning->end,
+               do_truechimers, (int)truechimer_thresholds->critical->end,
+               TRUE, 0, FALSE, 0);
+}
+
 int main(int argc, char *argv[]){
-       int result, offset_result, stratum;
+       int result, offset_result, stratum, num_truechimers;
        double offset=0, jitter=0;
        char *result_line, *perfdata_line;
 
@@ -544,6 +571,7 @@ int main(int argc, char *argv[]){
        set_thresholds(&offset_thresholds, owarn, ocrit);
        set_thresholds(&jitter_thresholds, jwarn, jcrit);
        set_thresholds(&stratum_thresholds, swarn, scrit);
+       set_thresholds(&truechimer_thresholds, twarn, tcrit);
 
        /* initialize alarm signal handling */
        signal (SIGALRM, socket_timeout_alarm_handler);
@@ -552,7 +580,7 @@ int main(int argc, char *argv[]){
        alarm (socket_timeout);
 
        /* This returns either OK or WARNING (See comment preceeding ntp_request) */
-       result = ntp_request(server_address, &offset, &offset_result, &jitter, &stratum);
+       result = ntp_request(server_address, &offset, &offset_result, &jitter, &stratum, &num_truechimers);
 
        if(offset_result == STATE_UNKNOWN) {
                /* if there's no sync peer (this overrides ntp_request output): */
@@ -564,6 +592,9 @@ int main(int argc, char *argv[]){
                result = max_state_alt(result, get_status(fabs(offset), offset_thresholds));
        }
 
+       if(do_truechimers)
+               result = max_state_alt(result, get_status(num_truechimers, truechimer_thresholds));
+
        if(do_stratum)
                result = max_state_alt(result, get_status(stratum, stratum_thresholds));
 
@@ -604,6 +635,10 @@ int main(int argc, char *argv[]){
                asprintf(&result_line, "%s, stratum=%i", result_line, stratum);
                asprintf(&perfdata_line, "%s %s", perfdata_line, perfd_stratum(stratum));
        }
+       if (do_truechimers) {
+               asprintf(&result_line, "%s, truechimers=%i", result_line, num_truechimers);
+               asprintf(&perfdata_line, "%s %s", perfdata_line, perfd_truechimers(num_truechimers));
+       }
        printf("%s|%s\n", result_line, perfdata_line);
 
        if(server_address!=NULL) free(server_address);
@@ -640,6 +675,10 @@ void print_help(void){
        printf ("    %s\n", _("Warning threshold for jitter"));
        printf (" %s\n", "-k, --jcrit=THRESHOLD");
        printf ("    %s\n", _("Critical threshold for jitter"));
+       printf (" %s\n", "-m, --twarn=THRESHOLD");
+       printf ("    %s\n", _("Warning threshold for number of usable time sources (\"truechimers\")"));
+       printf (" %s\n", "-n, --tcrit=THRESHOLD");
+       printf ("    %s\n", _("Critical threshold for number of usable time sources (\"truechimers\")"));
        printf (_(UT_TIMEOUT), DEFAULT_SOCKET_TIMEOUT);
        printf (_(UT_VERBOSE));
 
@@ -668,6 +707,9 @@ void print_help(void){
        printf(" %s\n", _("(See Notes above for more details on thresholds formats):"));
        printf("  %s\n", ("./check_ntp_peer -H ntpserv -w 0.5 -c 1 -j -1:100 -k -1:200"));
        printf("\n");
+       printf(" %s\n", _("Only check the number of usable time sources (\"truechimers\"):"));
+       printf("  %s\n", ("./check_ntp_peer -H ntpserv -m :5 -n :3"));
+       printf("\n");
        printf(" %s\n", _("Check only stratum:"));
        printf("  %s\n", ("./check_ntp_peer -H ntpserv -W 4 -C 6"));
 

diff --git a/plugins/t/check_ntp.t b/plugins/t/check_ntp.t
index b71c863190f74eb1e4cf470b9cb56e83574485cb..3eee6e17d1c47c828389f02e60b8377614f5a543 100644 (file)
--- a/plugins/t/check_ntp.t
+++ b/plugins/t/check_ntp.t
@@ -34,9 +34,9 @@ my $hostname_invalid = getTestParameter( "NP_HOSTNAME_INVALID",
 my $ntp_okmatch1 = '/^NTP\sOK:\sOffset\s-?[0-9]+(\.[0-9]+)?(e-[0-9]{2})?\ssecs/';
 my $ntp_warnmatch1 = '/^NTP\sWARNING:\sOffset\s-?[0-9]+(\.[0-9]+)?(e-[0-9]{2})?\ssecs/';
 my $ntp_critmatch1 = '/^NTP\sCRITICAL:\sOffset\s-?[0-9]+(\.[0-9]+)?(e-[0-9]{2})?\ssecs/';
-my $ntp_okmatch2 = '/^NTP\sOK:\sOffset\s-?[0-9]+(\.[0-9]+)?(e-[0-9]{2})?\ssecs,\sjitter=[0-9]+\.[0-9]+,\sstratum=[0-9]{1,2}/';
-my $ntp_warnmatch2 = '/^NTP\sWARNING:\sOffset\s-?[0-9]+(\.[0-9]+)?(e-[0-9]{2})?\ssecs,\sjitter=[0-9]+\.[0-9]+,\sstratum=[0-9]{1,2}/';
-my $ntp_critmatch2 = '/^NTP\sCRITICAL:\sOffset\s-?[0-9]+(\.[0-9]+)?(e-[0-9]{2})?\ssecs,\sjitter=[0-9]+\.[0-9]+,\sstratum=[0-9]{1,2}/';
+my $ntp_okmatch2 = '/^NTP\sOK:\sOffset\s-?[0-9]+(\.[0-9]+)?(e-[0-9]{2})?\ssecs,\sjitter=[0-9]+\.[0-9]+,\sstratum=[0-9]{1,2},\struechimers=[0-9]+/';
+my $ntp_warnmatch2 = '/^NTP\sWARNING:\sOffset\s-?[0-9]+(\.[0-9]+)?(e-[0-9]{2})?\ssecs,\sjitter=[0-9]+\.[0-9]+,\sstratum=[0-9]{1,2},\struechimers=[0-9]+/';
+my $ntp_critmatch2 = '/^NTP\sCRITICAL:\sOffset\s-?[0-9]+(\.[0-9]+)?(e-[0-9]{2})?\ssecs,\sjitter=[0-9]+\.[0-9]+,\sstratum=[0-9]{1,2},\struechimers=[0-9]+/';
 my $ntp_noresponse = '/^(CRITICAL - Socket timeout after 3 seconds)|(NTP CRITICAL: No response from NTP server)$/';
 my $ntp_nosuchhost = '/^check_ntp.*: Invalid hostname/address - ' . $hostname_invalid . '/';
 
@@ -90,21 +90,21 @@ foreach my $plugin (@PLUGINS2) {
        SKIP: {
                skip "No NTP server defined", 1 unless $ntp_service;
                $res = NPTest->testCmd(
-                       "./$plugin -H $ntp_service -w 1000 -c 2000 -W 20 -C 21 -j 100000 -k 200000"
+                       "./$plugin -H $ntp_service -w 1000 -c 2000 -W 20 -C 21 -j 100000 -k 200000 -m 1: -n 0:"
                        );
-               cmp_ok( $res->return_code, '==', 0, "$plugin: Good NTP result with jitter and stratum check" );
-               like( $res->output, $ntp_okmatch2, "$plugin: Output match OK with jitter and stratum" );
+               cmp_ok( $res->return_code, '==', 0, "$plugin: Good NTP result with jitter, stratum, and truechimers check" );
+               like( $res->output, $ntp_okmatch2, "$plugin: Output match OK with jitter, stratum, and truechimers" );
 
                $res = NPTest->testCmd(
-                       "./$plugin -H $ntp_service -w 1000 -c 2000 -W \\~:-1 -C 21 -j 100000 -k 200000"
+                       "./$plugin -H $ntp_service -w 1000 -c 2000 -W \\~:-1 -C 21 -j 100000 -k 200000 -m 1: -n 0:"
                        );
-               cmp_ok( $res->return_code, '==', 1, "$plugin: Warning NTP result with jitter and stratum check" );
-               like( $res->output, $ntp_warnmatch2, "$plugin: Output match WARNING with jitter and stratum" );
+               cmp_ok( $res->return_code, '==', 1, "$plugin: Warning NTP result with jitter, stratum, and truechimers check" );
+               like( $res->output, $ntp_warnmatch2, "$plugin: Output match WARNING with jitter, stratum, and truechimers" );
 
                $res = NPTest->testCmd(
-                       "./$plugin -H $ntp_service -w 1000 -c 2000 -W 20 -C 21 -j 100000 -k \\~:-1"
+                       "./$plugin -H $ntp_service -w 1000 -c 2000 -W 20 -C 21 -j 100000 -k \\~:-1 -m 1: -n 0:"
                        );
-               cmp_ok( $res->return_code, '==', 2, "$plugin: Critical NTP result with jitter and stratum check" );
-               like( $res->output, $ntp_critmatch2, "$plugin: Output match CRITICAL with jitter and stratum" );
+               cmp_ok( $res->return_code, '==', 2, "$plugin: Critical NTP result with jitter, stratum, and truechimers check" );
+               like( $res->output, $ntp_critmatch2, "$plugin: Output match CRITICAL with jitter, stratum, and truechimers" );
        }
 }