fio: allow to combine terse output with any selected output type
[fio.git] / stat.c
diff --git a/stat.c b/stat.c
index 8bcd566e0b93c6b162b3ed0020f3ecaead4e18ea..77275d94efe4de77f3fb3bc6a39fa6b70121c07a 100644 (file)
--- a/stat.c
+++ b/stat.c
@@ -14,6 +14,8 @@
 #include "lib/getrusage.h"
 #include "idletime.h"
 
+static struct fio_mutex *stat_mutex;
+
 void update_rusage_stat(struct thread_data *td)
 {
        struct thread_stat *ts = &td->ts;
@@ -65,12 +67,12 @@ static unsigned int plat_val_to_idx(unsigned int val)
 
        /*
         * Discard the error bits and apply the mask to find the
-         * index for the buckets in the group
+        * index for the buckets in the group
         */
        offset = (FIO_IO_U_PLAT_VAL - 1) & (val >> error_bits);
 
        /* Make sure the index does not exceed (array size - 1) */
-       idx = (base + offset) < (FIO_IO_U_PLAT_NR - 1)?
+       idx = (base + offset) < (FIO_IO_U_PLAT_NR - 1) ?
                (base + offset) : (FIO_IO_U_PLAT_NR - 1);
 
        return idx;
@@ -88,11 +90,11 @@ static unsigned int plat_idx_to_val(unsigned int idx)
 
        /* MSB <= (FIO_IO_U_PLAT_BITS-1), cannot be rounded off. Use
         * all bits of the sample as index */
-       if (idx < (FIO_IO_U_PLAT_VAL << 1) )
+       if (idx < (FIO_IO_U_PLAT_VAL << 1))
                return idx;
 
        /* Find the group and compute the minimum value of that group */
-       error_bits = (idx >> FIO_IO_U_PLAT_BITS) -1;
+       error_bits = (idx >> FIO_IO_U_PLAT_BITS) - 1;
        base = 1 << (error_bits + FIO_IO_U_PLAT_BITS);
 
        /* Find its bucket number of the group */
@@ -116,11 +118,9 @@ static int double_cmp(const void *a, const void *b)
        return cmp;
 }
 
-static unsigned int calc_clat_percentiles(unsigned int *io_u_plat,
-                                         unsigned long nr, fio_fp64_t *plist,
-                                         unsigned int **output,
-                                         unsigned int *maxv,
-                                         unsigned int *minv)
+unsigned int calc_clat_percentiles(unsigned int *io_u_plat, unsigned long nr,
+                                  fio_fp64_t *plist, unsigned int **output,
+                                  unsigned int *maxv, unsigned int *minv)
 {
        unsigned long sum = 0;
        unsigned int len, i, j = 0;
@@ -144,7 +144,7 @@ static unsigned int calc_clat_percentiles(unsigned int *io_u_plat,
         * isn't a worry. Also note that this does not work for NaN values.
         */
        if (len > 1)
-               qsort((void*)plist, len, sizeof(plist[0]), double_cmp);
+               qsort((void *)plist, len, sizeof(plist[0]), double_cmp);
 
        /*
         * Calculate bucket values, note down max and min values
@@ -182,12 +182,12 @@ static unsigned int calc_clat_percentiles(unsigned int *io_u_plat,
  * Find and display the p-th percentile of clat
  */
 static void show_clat_percentiles(unsigned int *io_u_plat, unsigned long nr,
-                                 fio_fp64_t *plist, uint64_t precision)
+                                 fio_fp64_t *plist, unsigned int precision)
 {
        unsigned int len, j = 0, minv, maxv;
        unsigned int *ovals;
-       int is_last, scale_down;
-       char buf[32];
+       int is_last, per_line, scale_down;
+       char fmt[32];
 
        len = calc_clat_percentiles(io_u_plat, nr, plist, &ovals, &maxv, &minv);
        if (!len)
@@ -205,13 +205,14 @@ static void show_clat_percentiles(unsigned int *io_u_plat, unsigned long nr,
                log_info("    clat percentiles (usec):\n     |");
        }
 
-       snprintf(buf, sizeof(buf), "%%1.%luf", precision);
+       snprintf(fmt, sizeof(fmt), "%%1.%uf", precision);
+       per_line = (80 - 7) / (precision + 14);
 
        for (j = 0; j < len; j++) {
                char fbuf[16], *ptr = fbuf;
 
                /* for formatting */
-               if (j != 0 && (j % 4) == 0)
+               if (j != 0 && (j % per_line) == 0)
                        log_info("     |");
 
                /* end of the list */
@@ -220,7 +221,7 @@ static void show_clat_percentiles(unsigned int *io_u_plat, unsigned long nr,
                if (plist[j].u.f < 10.0)
                        ptr += sprintf(fbuf, " ");
 
-               snprintf(ptr, sizeof(fbuf), buf, plist[j].u.f);
+               snprintf(ptr, sizeof(fbuf), fmt, plist[j].u.f);
 
                if (scale_down)
                        ovals[j] = (ovals[j] + 999) / 1000;
@@ -230,7 +231,7 @@ static void show_clat_percentiles(unsigned int *io_u_plat, unsigned long nr,
                if (is_last)
                        break;
 
-               if (j % 4 == 3) /* for formatting */
+               if ((j % per_line) == per_line - 1)     /* for formatting */
                        log_info("\n");
        }
 
@@ -239,18 +240,16 @@ out:
                free(ovals);
 }
 
-static int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max,
-                   double *mean, double *dev)
+int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max,
+            double *mean, double *dev)
 {
-       double n = is->samples;
+       double n = (double) is->samples;
 
-       if (is->samples == 0)
+       if (n == 0)
                return 0;
 
        *min = is->min_val;
        *max = is->max_val;
-
-       n = (double) is->samples;
        *mean = is->mean.u.f;
 
        if (n > 1.0)
@@ -275,15 +274,17 @@ void show_group_stats(struct group_run_stats *rs)
                if (!rs->max_run[i])
                        continue;
 
-               p1 = num2str(rs->io_kb[i], 6, rs->kb_base, i2p);
-               p2 = num2str(rs->agg[i], 6, rs->kb_base, i2p);
-               p3 = num2str(rs->min_bw[i], 6, rs->kb_base, i2p);
-               p4 = num2str(rs->max_bw[i], 6, rs->kb_base, i2p);
+               p1 = num2str(rs->io_kb[i], 6, rs->kb_base, i2p, 8);
+               p2 = num2str(rs->agg[i], 6, rs->kb_base, i2p, rs->unit_base);
+               p3 = num2str(rs->min_bw[i], 6, rs->kb_base, i2p, rs->unit_base);
+               p4 = num2str(rs->max_bw[i], 6, rs->kb_base, i2p, rs->unit_base);
 
-               log_info("%s: io=%sB, aggrb=%sB/s, minb=%sB/s, maxb=%sB/s,"
+               log_info("%s: io=%s, aggrb=%s/s, minb=%s/s, maxb=%s/s,"
                         " mint=%llumsec, maxt=%llumsec\n",
                                rs->unified_rw_rep ? "  MIXED" : ddir_str[i],
-                               p1, p2, p3, p4, rs->min_run[i], rs->max_run[i]);
+                               p1, p2, p3, p4,
+                               (unsigned long long) rs->min_run[i],
+                               (unsigned long long) rs->max_run[i]);
 
                free(p1);
                free(p2);
@@ -292,12 +293,7 @@ void show_group_stats(struct group_run_stats *rs)
        }
 }
 
-#define ts_total_io_u(ts)      \
-       ((ts)->total_io_u[DDIR_READ] + (ts)->total_io_u[DDIR_WRITE] +\
-               (ts)->total_io_u[DDIR_TRIM])
-
-static void stat_calc_dist(unsigned int *map, unsigned long total,
-                          double *io_u_dist)
+void stat_calc_dist(unsigned int *map, unsigned long total, double *io_u_dist)
 {
        int i;
 
@@ -318,7 +314,7 @@ static void stat_calc_dist(unsigned int *map, unsigned long total,
 static void stat_calc_lat(struct thread_stat *ts, double *dst,
                          unsigned int *src, int nr)
 {
-       unsigned long total = ts_total_io_u(ts);
+       unsigned long total = ddir_rw_sum(ts->total_io_u);
        int i;
 
        /*
@@ -335,28 +331,33 @@ static void stat_calc_lat(struct thread_stat *ts, double *dst,
        }
 }
 
-static void stat_calc_lat_u(struct thread_stat *ts, double *io_u_lat)
+void stat_calc_lat_u(struct thread_stat *ts, double *io_u_lat)
 {
        stat_calc_lat(ts, io_u_lat, ts->io_u_lat_u, FIO_IO_U_LAT_U_NR);
 }
 
-static void stat_calc_lat_m(struct thread_stat *ts, double *io_u_lat)
+void stat_calc_lat_m(struct thread_stat *ts, double *io_u_lat)
 {
        stat_calc_lat(ts, io_u_lat, ts->io_u_lat_m, FIO_IO_U_LAT_M_NR);
 }
 
-static int usec_to_msec(unsigned long *min, unsigned long *max, double *mean,
-                       double *dev)
+static void display_lat(const char *name, unsigned long min, unsigned long max,
+                       double mean, double dev)
 {
-       if (*min > 1000 && *max > 1000 && *mean > 1000.0 && *dev > 1000.0) {
-               *min /= 1000;
-               *max /= 1000;
-               *mean /= 1000.0;
-               *dev /= 1000.0;
-               return 0;
-       }
+       const char *base = "(usec)";
+       char *minp, *maxp;
 
-       return 1;
+       if (!usec_to_msec(&min, &max, &mean, &dev))
+               base = "(msec)";
+
+       minp = num2str(min, 6, 1, 0, 0);
+       maxp = num2str(max, 6, 1, 0, 0);
+
+       log_info("    %s %s: min=%s, max=%s, avg=%5.02f,"
+                " stdev=%5.02f\n", name, base, minp, maxp, mean, dev);
+
+       free(minp);
+       free(maxp);
 }
 
 static void show_ddir_status(struct group_run_stats *rs, struct thread_stat *ts,
@@ -378,68 +379,28 @@ static void show_ddir_status(struct group_run_stats *rs, struct thread_stat *ts,
        runt = ts->runtime[ddir];
 
        bw = (1000 * ts->io_bytes[ddir]) / runt;
-       io_p = num2str(ts->io_bytes[ddir], 6, 1, i2p);
-       bw_p = num2str(bw, 6, 1, i2p);
+       io_p = num2str(ts->io_bytes[ddir], 6, 1, i2p, 8);
+       bw_p = num2str(bw, 6, 1, i2p, ts->unit_base);
 
        iops = (1000 * (uint64_t)ts->total_io_u[ddir]) / runt;
-       iops_p = num2str(iops, 6, 1, 0);
+       iops_p = num2str(iops, 6, 1, 0, 0);
 
-       log_info("  %s: io=%sB, bw=%sB/s, iops=%s, runt=%6llumsec\n",
+       log_info("  %s: io=%s, bw=%s/s, iops=%s, runt=%6llumsec\n",
                                rs->unified_rw_rep ? "mixed" : ddir_str[ddir],
-                               io_p, bw_p, iops_p, ts->runtime[ddir]);
+                               io_p, bw_p, iops_p,
+                               (unsigned long long) ts->runtime[ddir]);
 
        free(io_p);
        free(bw_p);
        free(iops_p);
 
-       if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev)) {
-               const char *base = "(usec)";
-               char *minp, *maxp;
-
-               if (!usec_to_msec(&min, &max, &mean, &dev))
-                       base = "(msec)";
-
-               minp = num2str(min, 6, 1, 0);
-               maxp = num2str(max, 6, 1, 0);
-
-               log_info("    slat %s: min=%s, max=%s, avg=%5.02f,"
-                        " stdev=%5.02f\n", base, minp, maxp, mean, dev);
-
-               free(minp);
-               free(maxp);
-       }
-       if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev)) {
-               const char *base = "(usec)";
-               char *minp, *maxp;
-
-               if (!usec_to_msec(&min, &max, &mean, &dev))
-                       base = "(msec)";
-
-               minp = num2str(min, 6, 1, 0);
-               maxp = num2str(max, 6, 1, 0);
-
-               log_info("    clat %s: min=%s, max=%s, avg=%5.02f,"
-                        " stdev=%5.02f\n", base, minp, maxp, mean, dev);
-
-               free(minp);
-               free(maxp);
-       }
-       if (calc_lat(&ts->lat_stat[ddir], &min, &max, &mean, &dev)) {
-               const char *base = "(usec)";
-               char *minp, *maxp;
-
-               if (!usec_to_msec(&min, &max, &mean, &dev))
-                       base = "(msec)";
-
-               minp = num2str(min, 6, 1, 0);
-               maxp = num2str(max, 6, 1, 0);
-
-               log_info("     lat %s: min=%s, max=%s, avg=%5.02f,"
-                        " stdev=%5.02f\n", base, minp, maxp, mean, dev);
+       if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev))
+               display_lat("slat", min, max, mean, dev);
+       if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev))
+               display_lat("clat", min, max, mean, dev);
+       if (calc_lat(&ts->lat_stat[ddir], &min, &max, &mean, &dev))
+               display_lat(" lat", min, max, mean, dev);
 
-               free(minp);
-               free(maxp);
-       }
        if (ts->clat_percentiles) {
                show_clat_percentiles(ts->io_u_plat[ddir],
                                        ts->clat_stat[ddir].samples,
@@ -447,8 +408,15 @@ static void show_ddir_status(struct group_run_stats *rs, struct thread_stat *ts,
                                        ts->percentile_precision);
        }
        if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
-               double p_of_agg = 100.0;
-               const char *bw_str = "KB";
+               double p_of_agg = 100.0, fkb_base = (double)rs->kb_base;
+               const char *bw_str = (rs->unit_base == 1 ? "Kbit" : "KB");
+
+               if (rs->unit_base == 1) {
+                       min *= 8.0;
+                       max *= 8.0;
+                       mean *= 8.0;
+                       dev *= 8.0;
+               }
 
                if (rs->agg[ddir]) {
                        p_of_agg = mean * 100 / (double) rs->agg[ddir];
@@ -456,15 +424,15 @@ static void show_ddir_status(struct group_run_stats *rs, struct thread_stat *ts,
                                p_of_agg = 100.0;
                }
 
-               if (mean > 999999.9) {
-                       min /= 1000.0;
-                       max /= 1000.0;
-                       mean /= 1000.0;
-                       dev /= 1000.0;
-                       bw_str = "MB";
+               if (mean > fkb_base * fkb_base) {
+                       min /= fkb_base;
+                       max /= fkb_base;
+                       mean /= fkb_base;
+                       dev /= fkb_base;
+                       bw_str = (rs->unit_base == 1 ? "Mbit" : "MB");
                }
 
-               log_info("    bw (%s/s)  : min=%5lu, max=%5lu, per=%3.2f%%,"
+               log_info("    bw (%-4s/s): min=%5lu, max=%5lu, per=%3.2f%%,"
                         " avg=%5.02f, stdev=%5.02f\n", bw_str, min, max,
                                                        p_of_agg, mean, dev);
        }
@@ -517,19 +485,23 @@ static void show_lat_m(double *io_u_lat_m)
        show_lat(io_u_lat_m, FIO_IO_U_LAT_M_NR, ranges, "msec");
 }
 
-static void show_latencies(double *io_u_lat_u, double *io_u_lat_m)
+static void show_latencies(struct thread_stat *ts)
 {
+       double io_u_lat_u[FIO_IO_U_LAT_U_NR];
+       double io_u_lat_m[FIO_IO_U_LAT_M_NR];
+
+       stat_calc_lat_u(ts, io_u_lat_u);
+       stat_calc_lat_m(ts, io_u_lat_m);
+
        show_lat_u(io_u_lat_u);
        show_lat_m(io_u_lat_m);
 }
 
-void show_thread_status(struct thread_stat *ts, struct group_run_stats *rs)
+void show_thread_status_normal(struct thread_stat *ts, struct group_run_stats *rs)
 {
        double usr_cpu, sys_cpu;
        unsigned long runtime;
        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];
        time_t time_p;
        char time_buf[64];
 
@@ -562,9 +534,7 @@ void show_thread_status(struct thread_stat *ts, struct group_run_stats *rs)
        if (ts->io_bytes[DDIR_TRIM])
                show_ddir_status(rs, ts, DDIR_TRIM);
 
-       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);
+       show_latencies(ts);
 
        runtime = ts->total_run_time;
        if (runtime) {
@@ -577,10 +547,13 @@ void show_thread_status(struct thread_stat *ts, struct group_run_stats *rs)
                sys_cpu = 0;
        }
 
-       log_info("  cpu          : usr=%3.2f%%, sys=%3.2f%%, ctx=%lu, majf=%lu,"
-                " minf=%lu\n", usr_cpu, sys_cpu, ts->ctx, ts->majf, ts->minf);
+       log_info("  cpu          : usr=%3.2f%%, sys=%3.2f%%, ctx=%llu,"
+                " majf=%llu, minf=%llu\n", usr_cpu, sys_cpu,
+                       (unsigned long long) ts->ctx,
+                       (unsigned long long) ts->majf,
+                       (unsigned long long) ts->minf);
 
-       stat_calc_dist(ts->io_u_map, ts_total_io_u(ts), io_u_dist);
+       stat_calc_dist(ts->io_u_map, ddir_rw_sum(ts->total_io_u), io_u_dist);
        log_info("  IO depths    : 1=%3.1f%%, 2=%3.1f%%, 4=%3.1f%%, 8=%3.1f%%,"
                 " 16=%3.1f%%, 32=%3.1f%%, >=64=%3.1f%%\n", io_u_dist[0],
                                        io_u_dist[1], io_u_dist[2],
@@ -599,18 +572,27 @@ 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    : 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]);
+       log_info("     issued    : total=r=%llu/w=%llu/d=%llu,"
+                                " short=r=%llu/w=%llu/d=%llu\n",
+                                       (unsigned long long) ts->total_io_u[0],
+                                       (unsigned long long) ts->total_io_u[1],
+                                       (unsigned long long) ts->total_io_u[2],
+                                       (unsigned long long) ts->short_io_u[0],
+                                       (unsigned long long) ts->short_io_u[1],
+                                       (unsigned long long) ts->short_io_u[2]);
        if (ts->continue_on_error) {
-               log_info("     errors    : total=%lu, first_error=%d/<%s>\n",
-                                       ts->total_err_count,
+               log_info("     errors    : total=%llu, first_error=%d/<%s>\n",
+                                       (unsigned long long)ts->total_err_count,
                                        ts->first_error,
                                        strerror(ts->first_error));
        }
+       if (ts->latency_depth) {
+               log_info("     latency   : target=%llu, window=%llu, percentile=%.2f%%, depth=%u\n",
+                                       (unsigned long long)ts->latency_target,
+                                       (unsigned long long)ts->latency_window,
+                                       ts->latency_percentile.u.f,
+                                       ts->latency_depth);
+       }
 }
 
 static void show_ddir_status_terse(struct thread_stat *ts,
@@ -633,8 +615,9 @@ static void show_ddir_status_terse(struct thread_stat *ts,
                iops = (1000 * (uint64_t) ts->total_io_u[ddir]) / runt;
        }
 
-       log_info(";%llu;%llu;%llu;%llu", ts->io_bytes[ddir] >> 10, bw, iops,
-                                                       ts->runtime[ddir]);
+       log_info(";%llu;%llu;%llu;%llu",
+               (unsigned long long) ts->io_bytes[ddir] >> 10, bw, iops,
+                                       (unsigned long long) ts->runtime[ddir]);
 
        if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev))
                log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
@@ -774,7 +757,7 @@ static void add_ddir_status_json(struct thread_stat *ts,
        if (ovals)
                free(ovals);
 
-       if (!calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
+       if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
                if (rs->agg[ddir]) {
                        p_of_agg = mean * 100 / (double) rs->agg[ddir];
                        if (p_of_agg > 100.0)
@@ -792,7 +775,7 @@ static void add_ddir_status_json(struct thread_stat *ts,
 }
 
 static void show_thread_status_terse_v2(struct thread_stat *ts,
-                                       struct group_run_stats *rs)
+                                       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];
@@ -820,11 +803,13 @@ static void show_thread_status_terse_v2(struct thread_stat *ts,
                sys_cpu = 0;
        }
 
-       log_info(";%f%%;%f%%;%lu;%lu;%lu", usr_cpu, sys_cpu, ts->ctx, ts->majf,
-                                                               ts->minf);
+       log_info(";%f%%;%f%%;%llu;%llu;%llu", usr_cpu, sys_cpu,
+                                               (unsigned long long) ts->ctx,
+                                               (unsigned long long) ts->majf,
+                                               (unsigned long long) 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_dist(ts->io_u_map, ddir_rw_sum(ts->total_io_u), io_u_dist);
        stat_calc_lat_u(ts, io_u_lat_u);
        stat_calc_lat_m(ts, io_u_lat_m);
 
@@ -841,7 +826,7 @@ static void show_thread_status_terse_v2(struct thread_stat *ts,
                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(";%llu;%d", (unsigned long long) ts->total_err_count, ts->first_error);
        log_info("\n");
 
        /* Additional output if description is set */
@@ -882,11 +867,13 @@ static void show_thread_status_terse_v3_v4(struct thread_stat *ts,
                sys_cpu = 0;
        }
 
-       log_info(";%f%%;%f%%;%lu;%lu;%lu", usr_cpu, sys_cpu, ts->ctx, ts->majf,
-                                                               ts->minf);
+       log_info(";%f%%;%f%%;%llu;%llu;%llu", usr_cpu, sys_cpu,
+                                               (unsigned long long) ts->ctx,
+                                               (unsigned long long) ts->majf,
+                                               (unsigned long long) 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_dist(ts->io_u_map, ddir_rw_sum(ts->total_io_u), io_u_dist);
        stat_calc_lat_u(ts, io_u_lat_u);
        stat_calc_lat_m(ts, io_u_lat_m);
 
@@ -903,11 +890,12 @@ static void show_thread_status_terse_v3_v4(struct thread_stat *ts,
                log_info(";%3.2f%%", io_u_lat_m[i]);
 
        /* disk util stats, if any */
-       show_disk_util(1, NULL);
+       if (is_backend)
+               show_disk_util(1, NULL);
 
        /* 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(";%llu;%d", (unsigned long long) ts->total_err_count, ts->first_error);
 
        /* Additional output if description is set */
        if (strlen(ts->description))
@@ -953,7 +941,7 @@ static struct json_object *show_thread_status_json(struct thread_stat *ts,
 
 
        /* 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_dist(ts->io_u_map, ddir_rw_sum(ts->total_io_u), io_u_dist);
        stat_calc_lat_u(ts, io_u_lat_u);
        stat_calc_lat_m(ts, io_u_lat_m);
 
@@ -990,7 +978,14 @@ static struct json_object *show_thread_status_json(struct thread_stat *ts,
        /* Additional output if continue_on_error set - default off*/
        if (ts->continue_on_error) {
                json_object_add_value_int(root, "total_err", ts->total_err_count);
-               json_object_add_value_int(root, "total_err", ts->first_error);
+               json_object_add_value_int(root, "first_error", ts->first_error);
+       }
+
+       if (ts->latency_depth) {
+               json_object_add_value_int(root, "latency_depth", ts->latency_depth);
+               json_object_add_value_int(root, "latency_target", ts->latency_target);
+               json_object_add_value_float(root, "latency_percentile", ts->latency_percentile.u.f);
+               json_object_add_value_int(root, "latency_window", ts->latency_window);
        }
 
        /* Additional output if description is set */
@@ -1011,6 +1006,18 @@ static void show_thread_status_terse(struct thread_stat *ts,
                log_err("fio: bad terse version!? %d\n", terse_version);
 }
 
+struct json_object *show_thread_status(struct thread_stat *ts,
+                                      struct group_run_stats *rs)
+{
+       if (output_format == FIO_OUTPUT_TERSE)
+               show_thread_status_terse(ts, rs);
+       else if (output_format == FIO_OUTPUT_JSON)
+               return(show_thread_status_json(ts, rs));
+       else
+               show_thread_status_normal(ts, rs);
+       return NULL;
+}
+
 static void sum_stat(struct io_stat *dst, struct io_stat *src, int nr)
 {
        double mean, S;
@@ -1161,13 +1168,14 @@ void init_thread_stat(struct thread_stat *ts)
        ts->groupid = -1;
 }
 
-void show_run_stats(void)
+static void __show_run_stats(void)
 {
        struct group_run_stats *runstats, *rs;
        struct thread_data *td;
        struct thread_stat *threadstats, *ts;
        int i, j, nr_ts, last_ts, idx;
        int kb_base_warned = 0;
+       int unit_base_warned = 0;
        struct json_object *root = NULL;
        struct json_array *array = NULL;
 
@@ -1231,6 +1239,11 @@ void show_run_stats(void)
                        else
                                memset(ts->description, 0, FIO_JOBNAME_SIZE);
 
+                       /*
+                        * If multiple entries in this group, this is
+                        * the first member.
+                        */
+                       ts->thread_number = td->thread_number;
                        ts->groupid = td->groupid;
 
                        /*
@@ -1239,11 +1252,16 @@ void show_run_stats(void)
                        ts->pid = td->pid;
 
                        ts->kb_base = td->o.kb_base;
+                       ts->unit_base = td->o.unit_base;
                        ts->unified_rw_rep = td->o.unified_rw_rep;
                } else if (ts->kb_base != td->o.kb_base && !kb_base_warned) {
                        log_info("fio: kb_base differs for jobs in group, using"
                                 " %u as the base\n", ts->kb_base);
                        kb_base_warned = 1;
+               } else if (ts->unit_base != td->o.unit_base && !unit_base_warned) {
+                       log_info("fio: unit_base differs for jobs in group, using"
+                                " %u as the base\n", ts->unit_base);
+                       unit_base_warned = 1;
                }
 
                ts->continue_on_error = td->o.continue_on_error;
@@ -1260,6 +1278,11 @@ void show_run_stats(void)
                        }
                }
 
+               ts->latency_depth = td->latency_qd;
+               ts->latency_target = td->o.latency_target;
+               ts->latency_percentile = td->o.latency_percentile;
+               ts->latency_window = td->o.latency_window;
+
                sum_thread_stats(ts, &td->ts, idx);
        }
 
@@ -1269,6 +1292,7 @@ void show_run_stats(void)
                ts = &threadstats[i];
                rs = &runstats[ts->groupid];
                rs->kb_base = ts->kb_base;
+               rs->unit_base = ts->unit_base;
                rs->unified_rw_rep += ts->unified_rw_rep;
 
                for (j = 0; j < DDIR_RWDIR_CNT; j++) {
@@ -1332,7 +1356,7 @@ void show_run_stats(void)
                        struct json_object *tmp = show_thread_status_json(ts, rs);
                        json_array_add_value_object(array, tmp);
                } else
-                       show_thread_status(ts, rs);
+                       show_thread_status_normal(ts, rs);
        }
        if (output_format == FIO_OUTPUT_JSON) {
                /* disk util stats, if any */
@@ -1357,16 +1381,34 @@ void show_run_stats(void)
 
        if (is_backend)
                fio_server_send_du();
-       else if (output_format == FIO_OUTPUT_NORMAL)
+       else if (output_format == FIO_OUTPUT_NORMAL) {
                show_disk_util(0, NULL);
+               show_idle_prof_stats(FIO_OUTPUT_NORMAL, NULL);
+       }
 
-       show_idle_prof_stats(FIO_OUTPUT_NORMAL, NULL);
+       if ( !(output_format == FIO_OUTPUT_TERSE) && append_terse_output) {
+               log_info("\nAdditional Terse Output:\n");
 
+               for (i = 0; i < nr_ts; i++) {
+                       ts = &threadstats[i];
+                       rs = &runstats[ts->groupid];
+                       show_thread_status_terse(ts, rs);
+               }
+       }
+
+       log_info_flush();
        free(runstats);
        free(threadstats);
 }
 
-static void *__show_running_run_stats(void *arg)
+void show_run_stats(void)
+{
+       fio_mutex_down(stat_mutex);
+       __show_run_stats();
+       fio_mutex_up(stat_mutex);
+}
+
+static void *__show_running_run_stats(void fio_unused *arg)
 {
        struct thread_data *td;
        unsigned long long *rt;
@@ -1385,14 +1427,22 @@ static void *__show_running_run_stats(void *arg)
                if (td_trim(td) && td->io_bytes[DDIR_TRIM])
                        td->ts.runtime[DDIR_TRIM] += rt[i];
 
-               update_rusage_stat(td);
+               td->update_rusage = 1;
                td->ts.io_bytes[DDIR_READ] = td->io_bytes[DDIR_READ];
                td->ts.io_bytes[DDIR_WRITE] = td->io_bytes[DDIR_WRITE];
                td->ts.io_bytes[DDIR_TRIM] = td->io_bytes[DDIR_TRIM];
                td->ts.total_run_time = mtime_since(&td->epoch, &tv);
        }
 
-       show_run_stats();
+       for_each_td(td, i) {
+               if (td->rusage_sem) {
+                       td->update_rusage = 1;
+                       fio_mutex_down(td->rusage_sem);
+               }
+               td->update_rusage = 0;
+       }
+
+       __show_run_stats();
 
        for_each_td(td, i) {
                if (td_read(td) && td->io_bytes[DDIR_READ])
@@ -1404,6 +1454,7 @@ static void *__show_running_run_stats(void *arg)
        }
 
        free(rt);
+       fio_mutex_up(stat_mutex);
        return NULL;
 }
 
@@ -1416,8 +1467,58 @@ void show_running_run_stats(void)
 {
        pthread_t thread;
 
-       pthread_create(&thread, NULL, __show_running_run_stats, NULL);
-       pthread_detach(thread);
+       fio_mutex_down(stat_mutex);
+
+       if (!pthread_create(&thread, NULL, __show_running_run_stats, NULL)) {
+               pthread_detach(thread);
+               return;
+       }
+
+       fio_mutex_up(stat_mutex);
+}
+
+static int status_interval_init;
+static struct timeval status_time;
+
+#define FIO_STATUS_FILE                "/tmp/fio-dump-status"
+
+static int check_status_file(void)
+{
+       struct stat sb;
+       const char *temp_dir;
+       char fio_status_file_path[PATH_MAX];
+
+       temp_dir = getenv("TMPDIR");
+       if (temp_dir == NULL)
+               temp_dir = getenv("TEMP");
+       if (temp_dir == NULL)
+               temp_dir = "/tmp";
+
+       snprintf(fio_status_file_path, sizeof(fio_status_file_path), "%s/%s", temp_dir, FIO_STATUS_FILE);
+
+       if (stat(fio_status_file_path, &sb))
+               return 0;
+
+       unlink(fio_status_file_path);
+       return 1;
+}
+
+void check_for_running_stats(void)
+{
+       if (status_interval) {
+               if (!status_interval_init) {
+                       fio_gettime(&status_time, NULL);
+                       status_interval_init = 1;
+               } else if (mtime_since_now(&status_time) >= status_interval) {
+                       show_running_run_stats();
+                       fio_gettime(&status_time, NULL);
+                       return;
+               }
+       }
+       if (check_status_file()) {
+               show_running_run_stats();
+               return;
+       }
 }
 
 static inline void add_stat_sample(struct io_stat *is, unsigned long data)
@@ -1445,13 +1546,23 @@ static void __add_log_sample(struct io_log *iolog, unsigned long val,
 {
        const int nr_samples = iolog->nr_samples;
 
+       if (iolog->disabled)
+               return;
+
        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;
+               void *new_log;
 
-               iolog->log = realloc(iolog->log, new_size);
+               new_log = realloc(iolog->log, new_size);
+               if (!new_log) {
+                       log_err("fio: failed extending iolog! Will stop logging.\n");
+                       iolog->disabled = 1;
+                       return;
+               }
+               iolog->log = new_log;
                iolog->max_samples <<= 1;
        }
 
@@ -1632,7 +1743,7 @@ void add_bw_sample(struct thread_data *td, enum fio_ddir ddir, unsigned int bs,
        fio_gettime(&td->bw_sample_time, NULL);
 }
 
-void add_iops_sample(struct thread_data *td, enum fio_ddir ddir,
+void add_iops_sample(struct thread_data *td, enum fio_ddir ddir, unsigned int bs,
                     struct timeval *t)
 {
        struct thread_stat *ts = &td->ts;
@@ -1659,10 +1770,25 @@ void add_iops_sample(struct thread_data *td, enum fio_ddir ddir,
                add_stat_sample(&ts->iops_stat[ddir], iops);
 
                if (td->iops_log)
-                       add_log_sample(td, td->iops_log, iops, ddir, 0);
+                       add_log_sample(td, td->iops_log, iops, ddir, bs);
 
                td->stat_io_blocks[ddir] = td->this_io_blocks[ddir];
        }
 
        fio_gettime(&td->iops_sample_time, NULL);
 }
+
+void stat_init(void)
+{
+       stat_mutex = fio_mutex_init(FIO_MUTEX_UNLOCKED);
+}
+
+void stat_exit(void)
+{
+       /*
+        * When we have the mutex, we know out-of-band access to it
+        * have ended.
+        */
+       fio_mutex_down(stat_mutex);
+       fio_mutex_remove(stat_mutex);
+}