Ensure that stat outputs are serialized
[fio.git] / stat.c
diff --git a/stat.c b/stat.c
index b16c55c6bfd0c42db41e99f2ff4283b1187a1216..b3861887e2a204b404acb9c043f1c7302a7c32be 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;
@@ -180,11 +182,12 @@ unsigned int calc_clat_percentiles(unsigned int *io_u_plat, unsigned long nr,
  * 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)
+                                 fio_fp64_t *plist, unsigned int precision)
 {
        unsigned int len, j = 0, minv, maxv;
        unsigned int *ovals;
-       int is_last, scale_down;
+       int is_last, per_line, scale_down;
+       char fmt[32];
 
        len = calc_clat_percentiles(io_u_plat, nr, plist, &ovals, &maxv, &minv);
        if (!len)
@@ -202,20 +205,23 @@ static void show_clat_percentiles(unsigned int *io_u_plat, unsigned long nr,
                log_info("    clat percentiles (usec):\n     |");
        }
 
+       snprintf(fmt, sizeof(fmt), "%%1.%uf", precision);
+       per_line = (80 - 7) / (precision + 14);
+
        for (j = 0; j < len; j++) {
-               char fbuf[8];
+               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 */
                is_last = (j == len - 1);
 
                if (plist[j].u.f < 10.0)
-                       sprintf(fbuf, " %2.2f", plist[j].u.f);
-               else
-                       sprintf(fbuf, "%2.2f", plist[j].u.f);
+                       ptr += sprintf(fbuf, " ");
+
+               snprintf(ptr, sizeof(fbuf), fmt, plist[j].u.f);
 
                if (scale_down)
                        ovals[j] = (ovals[j] + 999) / 1000;
@@ -225,7 +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");
        }
 
@@ -270,15 +276,17 @@ 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);
@@ -344,8 +352,8 @@ static void display_lat(const char *name, unsigned long min, unsigned long max,
        if (!usec_to_msec(&min, &max, &mean, &dev))
                base = "(msec)";
 
-       minp = num2str(min, 6, 1, 0);
-       maxp = num2str(max, 6, 1, 0);
+       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);
@@ -373,15 +381,16 @@ 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);
@@ -397,11 +406,19 @@ static void show_ddir_status(struct group_run_stats *rs, struct thread_stat *ts,
        if (ts->clat_percentiles) {
                show_clat_percentiles(ts->io_u_plat[ddir],
                                        ts->clat_stat[ddir].samples,
-                                       ts->percentile_list);
+                                       ts->percentile_list,
+                                       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];
@@ -409,15 +426,15 @@ 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);
        }
@@ -532,8 +549,11 @@ 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, 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%%,"
@@ -554,15 +574,17 @@ 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));
        }
@@ -588,8 +610,9 @@ 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);
@@ -614,7 +637,7 @@ static void show_ddir_status_terse(struct thread_stat *ts,
                        log_info(";0%%=0");
                        continue;
                }
-               log_info(";%2.2f%%=%u", ts->percentile_list[i].u.f, ovals[i]);
+               log_info(";%f%%=%u", ts->percentile_list[i].u.f, ovals[i]);
        }
 
        if (calc_lat(&ts->lat_stat[ddir], &min, &max, &mean, &dev))
@@ -712,7 +735,7 @@ static void add_ddir_status_json(struct thread_stat *ts,
                        json_object_add_value_int(percentile_object, "0.00", 0);
                        continue;
                }
-               snprintf(buf, sizeof(buf), "%2.2f", ts->percentile_list[i].u.f);
+               snprintf(buf, sizeof(buf), "%f", ts->percentile_list[i].u.f);
                json_object_add_value_int(percentile_object, (const char *)buf, ovals[i]);
        }
 
@@ -729,7 +752,7 @@ 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)
@@ -775,8 +798,10 @@ 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, ddir_rw_sum(ts->total_io_u), io_u_dist);
@@ -796,7 +821,7 @@ 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 */
@@ -837,8 +862,10 @@ 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, ddir_rw_sum(ts->total_io_u), io_u_dist);
@@ -862,7 +889,7 @@ static void show_thread_status_terse_v3_v4(struct thread_stat *ts,
 
        /* 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))
@@ -1116,13 +1143,14 @@ 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;
 
@@ -1169,6 +1197,7 @@ void show_run_stats(void)
                ts = &threadstats[j];
 
                ts->clat_percentiles = td->o.clat_percentiles;
+               ts->percentile_precision = td->o.percentile_precision;
                memcpy(ts->percentile_list, td->o.percentile_list, sizeof(td->o.percentile_list));
 
                idx++;
@@ -1198,11 +1227,16 @@ 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;
@@ -1228,6 +1262,7 @@ 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++) {
@@ -1316,16 +1351,24 @@ 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);
-
+       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;
@@ -1344,14 +1387,22 @@ 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])
@@ -1363,6 +1414,7 @@ static void *__show_running_run_stats(void *arg)
        }
 
        free(rt);
+       fio_mutex_up(stat_mutex);
        return NULL;
 }
 
@@ -1375,8 +1427,48 @@ 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;
+
+       if (stat(FIO_STATUS_FILE, &sb))
+               return 0;
+
+       unlink(FIO_STATUS_FILE);
+       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)
@@ -1625,3 +1717,18 @@ void add_iops_sample(struct thread_data *td, enum fio_ddir 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);
+}