stat: NaN fixes
[fio.git] / stat.c
diff --git a/stat.c b/stat.c
index 5e0fe43..40e3b1b 100644 (file)
--- a/stat.c
+++ b/stat.c
@@ -262,7 +262,7 @@ void show_group_stats(struct group_run_stats *rs)
        const char *ddir_str[] = { "   READ", "  WRITE" };
        int i;
 
-       log_info("Run status group %d (all jobs):\n", rs->groupid);
+       log_info("\nRun status group %d (all jobs):\n", rs->groupid);
 
        for (i = 0; i <= DDIR_WRITE; i++) {
                const int i2p = is_power_of_2(rs->kb_base);
@@ -440,27 +440,42 @@ static void show_ddir_status(struct group_run_stats *rs, struct thread_stat *ts,
                                        ts->percentile_list);
        }
        if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
-               double p_of_agg;
+               double p_of_agg = 100.0;
+               const char *bw_str = "KB";
 
-               p_of_agg = mean * 100 / (double) rs->agg[ddir];
-               log_info("     bw (KB/s) : min=%5lu, max=%5lu, per=%3.2f%%,"
-                        " avg=%5.02f, stdev=%5.02f\n", min, max, p_of_agg,
-                                                       mean, dev);
+               if (rs->agg[dir]) {
+                       p_of_agg = mean * 100 / (double) rs->agg[ddir];
+                       if (p_of_agg > 100.0)
+                               p_of_agg = 100.0;
+               }
+
+               if (mean > 999999.9) {
+                       min /= 1000.0;
+                       max /= 1000.0;
+                       mean /= 1000.0;
+                       dev /= 1000.0;
+                       bw_str = "MB";
+               }
+
+               log_info("    bw (%s/s)  : min=%5lu, max=%5lu, per=%3.2f%%,"
+                        " avg=%5.02f, stdev=%5.02f\n", bw_str, min, max,
+                                                       p_of_agg, mean, dev);
        }
 }
 
-static void show_lat(double *io_u_lat, int nr, const char **ranges,
-                    const char *msg)
+static int show_lat(double *io_u_lat, int nr, const char **ranges,
+                   const char *msg)
 {
-       int new_line = 1, i, line = 0;
+       int new_line = 1, i, line = 0, shown = 0;
 
        for (i = 0; i < nr; i++) {
                if (io_u_lat[i] <= 0.0)
                        continue;
+               shown = 1;
                if (new_line) {
                        if (line)
                                log_info("\n");
-                       log_info("     lat (%s): ", msg);
+                       log_info("    lat (%s) : ", msg);
                        new_line = 0;
                        line = 0;
                }
@@ -471,6 +486,11 @@ static void show_lat(double *io_u_lat, int nr, const char **ranges,
                if (line == 5)
                        new_line = 1;
        }
+
+       if (shown)
+               log_info("\n");
+
+       return shown;
 }
 
 static void show_lat_u(double *io_u_lat_u)
@@ -493,9 +513,7 @@ static void show_lat_m(double *io_u_lat_m)
 static void show_latencies(double *io_u_lat_u, double *io_u_lat_m)
 {
        show_lat_u(io_u_lat_u);
-       log_info("\n");
        show_lat_m(io_u_lat_m);
-       log_info("\n");
 }
 
 void show_thread_status(struct thread_stat *ts, struct group_run_stats *rs)
@@ -520,7 +538,7 @@ void show_thread_status(struct thread_stat *ts, struct group_run_stats *rs)
                                        ts->error, ts->verror, (int) ts->pid);
        }
 
-       if (ts->description)
+       if (strlen(ts->description))
                log_info("  Description  : [%s]\n", ts->description);
 
        if (ts->io_bytes[DDIR_READ])
@@ -528,6 +546,10 @@ void show_thread_status(struct thread_stat *ts, struct group_run_stats *rs)
        if (ts->io_bytes[DDIR_WRITE])
                show_ddir_status(rs, ts, DDIR_WRITE);
 
+       stat_calc_lat_u(ts, io_u_lat_u);
+       stat_calc_lat_m(ts, io_u_lat_m);
+       show_latencies(io_u_lat_u, io_u_lat_m);
+
        runtime = ts->total_run_time;
        if (runtime) {
                double runt = (double) runtime;
@@ -561,14 +583,12 @@ void show_thread_status(struct thread_stat *ts, struct group_run_stats *rs)
                                        io_u_dist[1], io_u_dist[2],
                                        io_u_dist[3], io_u_dist[4],
                                        io_u_dist[5], io_u_dist[6]);
-       log_info("     issued r/w/d: total=%lu/%lu/%lu, short=%lu/%lu/%lu\n",
+       log_info("     issued    : total=r=%lu/w=%lu/d=%lu,"
+                                " short=r=%lu/w=%lu/d=%lu\n",
                                        ts->total_io_u[0], ts->total_io_u[1],
                                        ts->total_io_u[2],
                                        ts->short_io_u[0], ts->short_io_u[1],
                                        ts->short_io_u[2]);
-       stat_calc_lat_u(ts, io_u_lat_u);
-       stat_calc_lat_m(ts, io_u_lat_m);
-       show_latencies(io_u_lat_u, io_u_lat_m);
        if (ts->continue_on_error) {
                log_info("     errors    : total=%lu, first_error=%d/<%s>\n",
                                        ts->total_err_count,
@@ -610,11 +630,6 @@ static void show_ddir_status_terse(struct thread_stat *ts,
        else
                log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);
 
-       if (calc_lat(&ts->lat_stat[ddir], &min, &max, &mean, &dev))
-               log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
-       else
-               log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);
-
        if (ts->clat_percentiles) {
                len = calc_clat_percentiles(ts->io_u_plat[ddir],
                                        ts->clat_stat[ddir].samples,
@@ -630,22 +645,91 @@ static void show_ddir_status_terse(struct thread_stat *ts,
                }
                log_info(";%2.2f%%=%u", ts->percentile_list[i].u.f, ovals[i]);
        }
+
+       if (calc_lat(&ts->lat_stat[ddir], &min, &max, &mean, &dev))
+               log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
+       else
+               log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);
+
        if (ovals)
                free(ovals);
 
        if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
-               double p_of_agg;
+               double p_of_agg = 100.0;
+
+               if (rs->agg[ddir]) {
+                       p_of_agg = mean * 100 / (double) rs->agg[ddir];
+                       if (p_of_agg > 100.0)
+                               p_of_agg = 100.0;
+               }
 
-               p_of_agg = mean * 100 / (double) rs->agg[ddir];
                log_info(";%lu;%lu;%f%%;%f;%f", min, max, p_of_agg, mean, dev);
        } else
                log_info(";%lu;%lu;%f%%;%f;%f", 0UL, 0UL, 0.0, 0.0, 0.0);
 }
 
+static void show_thread_status_terse_v2(struct thread_stat *ts,
+                                       struct group_run_stats *rs)
+{
+       double io_u_dist[FIO_IO_U_MAP_NR];
+       double io_u_lat_u[FIO_IO_U_LAT_U_NR];
+       double io_u_lat_m[FIO_IO_U_LAT_M_NR];
+       double usr_cpu, sys_cpu;
+       int i;
+
+       /* General Info */
+       log_info("2;%s;%d;%d", ts->name, ts->groupid, ts->error);
+       /* Log Read Status */
+       show_ddir_status_terse(ts, rs, 0);
+       /* Log Write Status */
+       show_ddir_status_terse(ts, rs, 1);
+
+       /* CPU Usage */
+       if (ts->total_run_time) {
+               double runt = (double) ts->total_run_time;
+
+               usr_cpu = (double) ts->usr_time * 100 / runt;
+               sys_cpu = (double) ts->sys_time * 100 / runt;
+       } else {
+               usr_cpu = 0;
+               sys_cpu = 0;
+       }
+
+       log_info(";%f%%;%f%%;%lu;%lu;%lu", usr_cpu, sys_cpu, ts->ctx, ts->majf,
+                                                               ts->minf);
+
+       /* Calc % distribution of IO depths, usecond, msecond latency */
+       stat_calc_dist(ts->io_u_map, ts_total_io_u(ts), io_u_dist);
+       stat_calc_lat_u(ts, io_u_lat_u);
+       stat_calc_lat_m(ts, io_u_lat_m);
+
+       /* Only show fixed 7 I/O depth levels*/
+       log_info(";%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%",
+                       io_u_dist[0], io_u_dist[1], io_u_dist[2], io_u_dist[3],
+                       io_u_dist[4], io_u_dist[5], io_u_dist[6]);
+
+       /* Microsecond latency */
+       for (i = 0; i < FIO_IO_U_LAT_U_NR; i++)
+               log_info(";%3.2f%%", io_u_lat_u[i]);
+       /* Millisecond latency */
+       for (i = 0; i < FIO_IO_U_LAT_M_NR; i++)
+               log_info(";%3.2f%%", io_u_lat_m[i]);
+       /* Additional output if continue_on_error set - default off*/
+       if (ts->continue_on_error)
+               log_info(";%lu;%d", ts->total_err_count, ts->first_error);
+       log_info("\n");
+
+       /* Additional output if description is set */
+       if (ts->description)
+               log_info(";%s", ts->description);
+
+       log_info("\n");
+}
+
 #define FIO_TERSE_VERSION      "3"
 
-static void show_thread_status_terse(struct thread_stat *ts,
-                                    struct group_run_stats *rs)
+static void show_thread_status_terse_v3(struct thread_stat *ts,
+                                       struct group_run_stats *rs)
 {
        double io_u_dist[FIO_IO_U_MAP_NR];
        double io_u_lat_u[FIO_IO_U_LAT_U_NR];
@@ -654,8 +738,8 @@ static void show_thread_status_terse(struct thread_stat *ts,
        int i;
 
        /* General Info */
-       log_info("%s;%s;%d;%d", FIO_TERSE_VERSION, ts->name, ts->groupid,
-                               ts->error);
+       log_info("%s;%s;%s;%d;%d", FIO_TERSE_VERSION, fio_version_string,
+                                       ts->name, ts->groupid, ts->error);
        /* Log Read Status */
        show_ddir_status_terse(ts, rs, 0);
        /* Log Write Status */
@@ -705,6 +789,17 @@ static void show_thread_status_terse(struct thread_stat *ts,
                log_info(";%s", ts->description);
 }
 
+static void show_thread_status_terse(struct thread_stat *ts,
+                                    struct group_run_stats *rs)
+{
+       if (terse_version == 2)
+               show_thread_status_terse_v2(ts, rs);
+       else if (terse_version == 3)
+               show_thread_status_terse_v3(ts, rs);
+       else
+               log_err("fio: bad terse version!? %d\n", terse_version);
+}
+
 static void sum_stat(struct io_stat *dst, struct io_stat *src, int nr)
 {
        double mean, S;
@@ -994,25 +1089,23 @@ void show_run_stats(void)
                        show_thread_status(ts, rs);
        }
 
-       if (!terse_output) {
-               for (i = 0; i < groupid + 1; i++) {
-                       rs = &runstats[i];
-
-                       rs->groupid = i;
-                       if (is_backend)
-                               fio_server_send_gs(rs);
-                       else
-                               show_group_stats(rs);
-               }
+       for (i = 0; i < groupid + 1; i++) {
+               rs = &runstats[i];
 
+               rs->groupid = i;
                if (is_backend)
-                       fio_server_send_du();
-               else
-                       show_disk_util(0);
-
-               free_disk_util();
+                       fio_server_send_gs(rs);
+               else if (!terse_output)
+                       show_group_stats(rs);
        }
 
+       if (is_backend)
+               fio_server_send_du();
+       else if (!terse_output)
+               show_disk_util(0);
+
+       free_disk_util();
+
        free(runstats);
        free(threadstats);
 }
@@ -1042,6 +1135,9 @@ static void __add_log_sample(struct io_log *iolog, unsigned long val,
 {
        const int nr_samples = iolog->nr_samples;
 
+       if (!iolog->nr_samples)
+               iolog->avg_last = t;
+
        if (iolog->nr_samples == iolog->max_samples) {
                int new_size = sizeof(struct io_sample) * iolog->max_samples*2;
 
@@ -1056,14 +1152,66 @@ static void __add_log_sample(struct io_log *iolog, unsigned long val,
        iolog->nr_samples++;
 }
 
+static inline void reset_io_stat(struct io_stat *ios)
+{
+       ios->max_val = ios->min_val = ios->samples = 0;
+       ios->mean.u.f = ios->S.u.f = 0;
+}
+
 static void add_log_sample(struct thread_data *td, struct io_log *iolog,
                           unsigned long val, enum fio_ddir ddir,
                           unsigned int bs)
 {
+       unsigned long elapsed, this_window;
+
        if (!ddir_rw(ddir))
                return;
 
-       __add_log_sample(iolog, val, ddir, bs, mtime_since_now(&td->epoch));
+       elapsed = mtime_since_now(&td->epoch);
+
+       /*
+        * If no time averaging, just add the log sample.
+        */
+       if (!iolog->avg_msec) {
+               __add_log_sample(iolog, val, ddir, bs, elapsed);
+               return;
+       }
+
+       /*
+        * Add the sample. If the time period has passed, then
+        * add that entry to the log and clear.
+        */
+       add_stat_sample(&iolog->avg_window[ddir], val);
+
+       /*
+        * If period hasn't passed, adding the above sample is all we
+        * need to do.
+        */
+       this_window = elapsed - iolog->avg_last;
+       if (this_window < iolog->avg_msec)
+               return;
+
+       /*
+        * Note an entry in the log. Use the mean from the logged samples,
+        * making sure to properly round up. Only write a log entry if we
+        * had actual samples done.
+        */
+       if (iolog->avg_window[DDIR_READ].samples) {
+               unsigned long mr;
+
+               mr = iolog->avg_window[DDIR_READ].mean.u.f + 0.50;
+               __add_log_sample(iolog, mr, DDIR_READ, 0, elapsed);
+       }
+       if (iolog->avg_window[DDIR_WRITE].samples) {
+               unsigned long mw;
+
+               mw = iolog->avg_window[DDIR_WRITE].mean.u.f + 0.50;
+               __add_log_sample(iolog, mw, DDIR_WRITE, 0, elapsed);
+       }
+
+       reset_io_stat(&iolog->avg_window[DDIR_READ]);
+       reset_io_stat(&iolog->avg_window[DDIR_WRITE]);
+       iolog->avg_last = elapsed;
 }
 
 void add_agg_sample(unsigned long val, enum fio_ddir ddir, unsigned int bs)
@@ -1172,10 +1320,8 @@ void add_iops_sample(struct thread_data *td, enum fio_ddir ddir,
 
        add_stat_sample(&ts->iops_stat[ddir], iops);
 
-       if (td->iops_log) {
-               assert(iops);
+       if (td->iops_log)
                add_log_sample(td, td->iops_log, iops, ddir, 0);
-       }
 
        fio_gettime(&td->iops_sample_time, NULL);
        td->stat_io_blocks[ddir] = td->this_io_blocks[ddir];