ntpd: promote log level 3 to production
authorDenys Vlasenko <vda.linux@googlemail.com>
Wed, 4 Dec 2013 15:32:09 +0000 (16:32 +0100)
committerDenys Vlasenko <vda.linux@googlemail.com>
Wed, 4 Dec 2013 15:32:09 +0000 (16:32 +0100)
Without it, peer selection logic is hard to debug

function                                             old     new   delta
update_local_clock                                   861     982    +121
select_and_cluster                                   962    1065    +103
common_ping_main                                    1788    1821     +33
huft_build                                          1287    1296      +9
chpst_main                                           708     704      -4
INET6_resolve                                        127     119      -8
------------------------------------------------------------------------------
(add/remove: 0/0 grow/shrink: 4/2 up/down: 266/-12)           Total: 254 bytes

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
networking/ntpd.c

index 13cca3e77f1d861c7b994c887f38ed24d7fb9e63..aeef0431ad72e10c8d733142a3af25b00d0d1767 100644 (file)
 
 
 /* Verbosity control (max level of -dddd options accepted).
- * max 5 is very talkative (and bloated). 2 is non-bloated,
+ * max 6 is very talkative (and bloated). 3 is non-bloated,
  * production level setting.
  */
-#define MAX_VERBOSE     2
+#define MAX_VERBOSE     3
 
 
 /* High-level description of the algorithm:
@@ -367,6 +367,7 @@ static const int const_IPTOS_LOWDELAY = IPTOS_LOWDELAY;
 #define VERB3 if (MAX_VERBOSE >= 3 && G.verbose >= 3)
 #define VERB4 if (MAX_VERBOSE >= 4 && G.verbose >= 4)
 #define VERB5 if (MAX_VERBOSE >= 5 && G.verbose >= 5)
+#define VERB6 if (MAX_VERBOSE >= 6 && G.verbose >= 6)
 
 
 static double LOG2D(int a)
@@ -568,7 +569,7 @@ filter_datapoints(peer_t *p)
        got_newest = 0;
        sum = 0;
        for (i = 0; i < NUM_DATAPOINTS; i++) {
-               VERB4 {
+               VERB5 {
                        bb_error_msg("datapoint[%d]: off:%f disp:%f(%f) age:%f%s",
                                i,
                                fdp[idx].d_offset,
@@ -665,7 +666,7 @@ filter_datapoints(peer_t *p)
        sum = SQRT(sum / NUM_DATAPOINTS);
        p->filter_jitter = sum > G_precision_sec ? sum : G_precision_sec;
 
-       VERB3 bb_error_msg("filter offset:%+f disp:%f jitter:%f",
+       VERB4 bb_error_msg("filter offset:%+f disp:%f jitter:%f",
                        p->filter_offset,
                        p->filter_dispersion,
                        p->filter_jitter);
@@ -700,7 +701,7 @@ reset_peer_stats(peer_t *p, double offset)
                p->lastpkt_recv_time = G.cur_time;
        }
        filter_datapoints(p); /* recalc p->filter_xxx */
-       VERB5 bb_error_msg("%s->lastpkt_recv_time=%f", p->p_dotted, p->lastpkt_recv_time);
+       VERB6 bb_error_msg("%s->lastpkt_recv_time=%f", p->p_dotted, p->lastpkt_recv_time);
 }
 
 static void
@@ -972,20 +973,20 @@ fit(peer_t *p, double rd)
 {
        if ((p->reachable_bits & (p->reachable_bits-1)) == 0) {
                /* One or zero bits in reachable_bits */
-               VERB3 bb_error_msg("peer %s unfit for selection: unreachable", p->p_dotted);
+               VERB4 bb_error_msg("peer %s unfit for selection: unreachable", p->p_dotted);
                return 0;
        }
 #if 0 /* we filter out such packets earlier */
        if ((p->lastpkt_status & LI_ALARM) == LI_ALARM
         || p->lastpkt_stratum >= MAXSTRAT
        ) {
-               VERB3 bb_error_msg("peer %s unfit for selection: bad status/stratum", p->p_dotted);
+               VERB4 bb_error_msg("peer %s unfit for selection: bad status/stratum", p->p_dotted);
                return 0;
        }
 #endif
        /* rd is root_distance(p) */
        if (rd > MAXDIST + FREQ_TOLERANCE * (1 << G.poll_exp)) {
-               VERB3 bb_error_msg("peer %s unfit for selection: root distance too high", p->p_dotted);
+               VERB4 bb_error_msg("peer %s unfit for selection: root distance too high", p->p_dotted);
                return 0;
        }
 //TODO
@@ -1025,7 +1026,7 @@ select_and_cluster(void)
                        continue;
                }
 
-               VERB4 bb_error_msg("interval: [%f %f %f] %s",
+               VERB5 bb_error_msg("interval: [%f %f %f] %s",
                                offset - rd,
                                offset,
                                offset + rd,
@@ -1050,7 +1051,7 @@ select_and_cluster(void)
        }
        num_candidates = num_points / 3;
        if (num_candidates == 0) {
-               VERB3 bb_error_msg("no valid datapoints, no peer selected");
+               VERB3 bb_error_msg("no valid datapoints%s", ", no peer selected");
                return NULL;
        }
 //TODO: sorting does not seem to be done in reference code
@@ -1108,12 +1109,13 @@ select_and_cluster(void)
                        break;
                num_falsetickers++;
                if (num_falsetickers * 2 >= num_candidates) {
-                       VERB3 bb_error_msg("too many falsetickers:%d (candidates:%d), no peer selected",
-                                       num_falsetickers, num_candidates);
+                       VERB3 bb_error_msg("falsetickers:%d, candidates:%d%s",
+                                       num_falsetickers, num_candidates,
+                                       ", no peer selected");
                        return NULL;
                }
        }
-       VERB3 bb_error_msg("selected interval: [%f, %f]; candidates:%d falsetickers:%d",
+       VERB4 bb_error_msg("selected interval: [%f, %f]; candidates:%d falsetickers:%d",
                        low, high, num_candidates, num_falsetickers);
 
        /* Clustering */
@@ -1131,7 +1133,7 @@ select_and_cluster(void)
                survivor[num_survivors].p = p;
                /* x.opt_rd == root_distance(p); */
                survivor[num_survivors].metric = MAXDIST * p->lastpkt_stratum + point[i].opt_rd;
-               VERB4 bb_error_msg("survivor[%d] metric:%f peer:%s",
+               VERB5 bb_error_msg("survivor[%d] metric:%f peer:%s",
                        num_survivors, survivor[num_survivors].metric, p->p_dotted);
                num_survivors++;
        }
@@ -1141,8 +1143,9 @@ select_and_cluster(void)
         * is acceptable.
         */
        if (num_survivors < MIN_SELECTED) {
-               VERB3 bb_error_msg("num_survivors %d < %d, no peer selected",
-                               num_survivors, MIN_SELECTED);
+               VERB3 bb_error_msg("survivors:%d%s",
+                               num_survivors,
+                               ", no peer selected");
                return NULL;
        }
 
@@ -1162,7 +1165,7 @@ select_and_cluster(void)
                double min_jitter = min_jitter;
 
                if (num_survivors <= MIN_CLUSTERED) {
-                       VERB3 bb_error_msg("num_survivors %d <= %d, not discarding more",
+                       VERB4 bb_error_msg("num_survivors %d <= %d, not discarding more",
                                        num_survivors, MIN_CLUSTERED);
                        break;
                }
@@ -1188,11 +1191,11 @@ select_and_cluster(void)
                                max_selection_jitter = selection_jitter_sq;
                                max_idx = i;
                        }
-                       VERB5 bb_error_msg("survivor %d selection_jitter^2:%f",
+                       VERB6 bb_error_msg("survivor %d selection_jitter^2:%f",
                                        i, selection_jitter_sq);
                }
                max_selection_jitter = SQRT(max_selection_jitter / num_survivors);
-               VERB4 bb_error_msg("max_selection_jitter (at %d):%f min_jitter:%f",
+               VERB5 bb_error_msg("max_selection_jitter (at %d):%f min_jitter:%f",
                                max_idx, max_selection_jitter, min_jitter);
 
                /* If the maximum selection jitter is less than the
@@ -1201,7 +1204,7 @@ select_and_cluster(void)
                 * as well stop.
                 */
                if (max_selection_jitter < min_jitter) {
-                       VERB3 bb_error_msg("max_selection_jitter:%f < min_jitter:%f, num_survivors:%d, not discarding more",
+                       VERB4 bb_error_msg("max_selection_jitter:%f < min_jitter:%f, num_survivors:%d, not discarding more",
                                        max_selection_jitter, min_jitter, num_survivors);
                        break;
                }
@@ -1209,7 +1212,7 @@ select_and_cluster(void)
                /* Delete survivor[max_idx] from the list
                 * and go around again.
                 */
-               VERB5 bb_error_msg("dropping survivor %d", max_idx);
+               VERB6 bb_error_msg("dropping survivor %d", max_idx);
                num_survivors--;
                while (max_idx < num_survivors) {
                        survivor[max_idx] = survivor[max_idx + 1];
@@ -1251,7 +1254,7 @@ select_and_cluster(void)
                /* Starting from 1 is ok here */
                for (i = 1; i < num_survivors; i++) {
                        if (G.last_update_peer == survivor[i].p) {
-                               VERB4 bb_error_msg("keeping old synced peer");
+                               VERB5 bb_error_msg("keeping old synced peer");
                                p = G.last_update_peer;
                                goto keep_old;
                        }
@@ -1259,7 +1262,7 @@ select_and_cluster(void)
        }
        G.last_update_peer = p;
  keep_old:
-       VERB3 bb_error_msg("selected peer %s filter_offset:%+f age:%f",
+       VERB4 bb_error_msg("selected peer %s filter_offset:%+f age:%f",
                        p->p_dotted,
                        p->filter_offset,
                        G.cur_time - p->lastpkt_recv_time
@@ -1278,7 +1281,7 @@ set_new_values(int disc_state, double offset, double recv_time)
         * of the last clock filter sample, which must be earlier than
         * the current time.
         */
-       VERB3 bb_error_msg("disc_state=%d last update offset=%f recv_time=%f",
+       VERB4 bb_error_msg("disc_state=%d last update offset=%f recv_time=%f",
                        disc_state, offset, recv_time);
        G.discipline_state = disc_state;
        G.last_update_offset = offset;
@@ -1316,8 +1319,8 @@ update_local_clock(peer_t *p)
         * an old sample or the same sample twice.
         */
        if (recv_time <= G.last_update_recv_time) {
-               VERB3 bb_error_msg("same or older datapoint: %f >= %f, not using it",
-                               G.last_update_recv_time, recv_time);
+               VERB3 bb_error_msg("update from %s: same or older datapoint, not using it",
+                       p->p_dotted);
                return 0; /* "leave poll interval as is" */
        }
 
@@ -1333,7 +1336,7 @@ update_local_clock(peer_t *p)
        if (G.discipline_state == STATE_FREQ) {
                /* Ignore updates until the stepout threshold */
                if (since_last_update < WATCH_THRESHOLD) {
-                       VERB3 bb_error_msg("measuring drift, datapoint ignored, %f sec remains",
+                       VERB4 bb_error_msg("measuring drift, datapoint ignored, %f sec remains",
                                        WATCH_THRESHOLD - since_last_update);
                        return 0; /* "leave poll interval as is" */
                }
@@ -1349,10 +1352,15 @@ update_local_clock(peer_t *p)
        if (abs_offset > STEP_THRESHOLD) {
                double remains;
 
+// TODO: this "spike state" seems to be useless, peer selection already drops
+// occassional "bad" datapoints. If we are here, there were _many_ large offsets -
+// looks like _our_ clock is off.
                switch (G.discipline_state) {
                case STATE_SYNC:
                        /* The first outlyer: ignore it, switch to SPIK state */
-                       VERB2 bb_error_msg("offset:%+f - spike", offset);
+                       VERB3 bb_error_msg("update from %s: offset:%+f, spike%s",
+                               p->p_dotted, offset,
+                               "");
                        G.discipline_state = STATE_SPIK;
                        return -1; /* "decrease poll interval" */
 
@@ -1362,8 +1370,9 @@ update_local_clock(peer_t *p)
                         */
                        remains = WATCH_THRESHOLD - since_last_update;
                        if (remains > 0) {
-                               VERB2 bb_error_msg("spike, datapoint ignored, %f sec remains",
-                                               remains);
+                               VERB3 bb_error_msg("update from %s: offset:%+f, spike%s",
+                                       p->p_dotted, offset,
+                                       ", datapoint ignored");
                                return -1; /* "decrease poll interval" */
                        }
                        /* fall through: we need to step */
@@ -1390,7 +1399,7 @@ update_local_clock(peer_t *p)
                 * is always suppressed, even at the longer poll
                 * intervals.
                 */
-               VERB3 bb_error_msg("stepping time by %+f; poll_exp=MINPOLL", offset);
+               VERB4 bb_error_msg("stepping time by %+f; poll_exp=MINPOLL", offset);
                step_time(offset);
                if (option_mask32 & OPT_q) {
                        /* We were only asked to set time once. Done. */
@@ -1415,7 +1424,7 @@ update_local_clock(peer_t *p)
        } else { /* abs_offset <= STEP_THRESHOLD */
 
                if (G.poll_exp < MINPOLL && G.initial_poll_complete) {
-                       VERB3 bb_error_msg("small offset:%+f, disabling burst mode", offset);
+                       VERB4 bb_error_msg("small offset:%+f, disabling burst mode", offset);
                        G.polladj_count = 0;
                        G.poll_exp = MINPOLL;
                }
@@ -1444,7 +1453,7 @@ update_local_clock(peer_t *p)
 #else
                        set_new_values(STATE_SYNC, offset, recv_time);
 #endif
-                       VERB3 bb_error_msg("transitioning to FREQ, datapoint ignored");
+                       VERB4 bb_error_msg("transitioning to FREQ, datapoint ignored");
                        return 0; /* "leave poll interval as is" */
 
 #if 0 /* this is dead code for now */
@@ -1513,7 +1522,7 @@ update_local_clock(peer_t *p)
        dtemp = p->filter_jitter; // SQRT(SQUARE(p->filter_jitter) + SQUARE(G.cluster_jitter));
        dtemp += MAXD(p->filter_dispersion + FREQ_TOLERANCE * (G.cur_time - p->lastpkt_recv_time) + abs_offset, MINDISP);
        G.rootdisp = p->lastpkt_rootdisp + dtemp;
-       VERB3 bb_error_msg("updating leap/refid/reftime/rootdisp from peer %s", p->p_dotted);
+       VERB4 bb_error_msg("updating leap/refid/reftime/rootdisp from peer %s", p->p_dotted);
 
        /* We are in STATE_SYNC now, but did not do adjtimex yet.
         * (Any other state does not reach this, they all return earlier)
@@ -1533,13 +1542,13 @@ update_local_clock(peer_t *p)
        dtemp = SQUARE(dtemp);
        G.discipline_wander = SQRT(etemp + (dtemp - etemp) / AVG);
 
-       VERB3 bb_error_msg("discipline freq_drift=%.9f(int:%ld corr:%e) wander=%f",
+       VERB4 bb_error_msg("discipline freq_drift=%.9f(int:%ld corr:%e) wander=%f",
                        G.discipline_freq_drift,
                        (long)(G.discipline_freq_drift * 65536e6),
                        freq_drift,
                        G.discipline_wander);
 #endif
-       VERB3 {
+       VERB4 {
                memset(&tmx, 0, sizeof(tmx));
                if (adjtimex(&tmx) < 0)
                        bb_perror_msg_and_die("adjtimex");
@@ -1587,7 +1596,7 @@ update_local_clock(peer_t *p)
        /* NB: here kernel returns constant == G.poll_exp, not == G.poll_exp - 4.
         * Not sure why. Perhaps it is normal.
         */
-       VERB3 bb_error_msg("adjtimex:%d freq:%ld offset:%+ld status:0x%x",
+       VERB4 bb_error_msg("adjtimex:%d freq:%ld offset:%+ld status:0x%x",
                                rc, tmx.freq, tmx.offset, tmx.status);
        G.kernel_freq_drift = tmx.freq / 65536;
        VERB2 bb_error_msg("update from:%s offset:%+f jitter:%f clock drift:%+.3fppm tc:%d",
@@ -1609,7 +1618,7 @@ retry_interval(void)
        interval = RETRY_INTERVAL;
        r = random();
        interval += r % (unsigned)(RETRY_INTERVAL / 4);
-       VERB3 bb_error_msg("chose retry interval:%u", interval);
+       VERB4 bb_error_msg("chose retry interval:%u", interval);
        return interval;
 }
 static unsigned
@@ -1622,7 +1631,7 @@ poll_interval(int exponent)
        interval = 1 << exponent;
        r = random();
        interval += ((r & (interval-1)) >> 4) + ((r >> 8) & 1); /* + 1/16 of interval, max */
-       VERB3 bb_error_msg("chose poll interval:%u (poll_exp:%d exp:%d)", interval, G.poll_exp, exponent);
+       VERB4 bb_error_msg("chose poll interval:%u (poll_exp:%d exp:%d)", interval, G.poll_exp, exponent);
        return interval;
 }
 static NOINLINE void
@@ -1716,7 +1725,7 @@ recv_and_process_peer_pkt(peer_t *p)
        T4 = G.cur_time;
 
        p->lastpkt_recv_time = T4;
-       VERB5 bb_error_msg("%s->lastpkt_recv_time=%f", p->p_dotted, p->lastpkt_recv_time);
+       VERB6 bb_error_msg("%s->lastpkt_recv_time=%f", p->p_dotted, p->lastpkt_recv_time);
 
        /* The delay calculation is a special case. In cases where the
         * server and client clocks are running at different rates and
@@ -1779,7 +1788,7 @@ recv_and_process_peer_pkt(peer_t *p)
                         * drop poll interval one step down.
                         */
                        if (fabs(q->filter_offset) >= POLLDOWN_OFFSET) {
-                               VERB3 bb_error_msg("offset:%+f > POLLDOWN_OFFSET", q->filter_offset);
+                               VERB4 bb_error_msg("offset:%+f > POLLDOWN_OFFSET", q->filter_offset);
                                goto poll_down;
                        }
                }
@@ -1801,11 +1810,11 @@ recv_and_process_peer_pkt(peer_t *p)
                                G.polladj_count = 0;
                                if (G.poll_exp < MAXPOLL) {
                                        G.poll_exp++;
-                                       VERB3 bb_error_msg("polladj: discipline_jitter:%f ++poll_exp=%d",
+                                       VERB4 bb_error_msg("polladj: discipline_jitter:%f ++poll_exp=%d",
                                                        G.discipline_jitter, G.poll_exp);
                                }
                        } else {
-                               VERB3 bb_error_msg("polladj: incr:%d", G.polladj_count);
+                               VERB4 bb_error_msg("polladj: incr:%d", G.polladj_count);
                        }
                } else {
                        G.polladj_count -= G.poll_exp * 2;
@@ -1827,11 +1836,11 @@ recv_and_process_peer_pkt(peer_t *p)
                                                if (pp->p_fd < 0)
                                                        pp->next_action_time -= (1 << G.poll_exp);
                                        }
-                                       VERB3 bb_error_msg("polladj: discipline_jitter:%f --poll_exp=%d",
+                                       VERB4 bb_error_msg("polladj: discipline_jitter:%f --poll_exp=%d",
                                                        G.discipline_jitter, G.poll_exp);
                                }
                        } else {
-                               VERB3 bb_error_msg("polladj: decr:%d", G.polladj_count);
+                               VERB4 bb_error_msg("polladj: decr:%d", G.polladj_count);
                        }
                }
        }