stat: avoid hang with race in stat thread exit and rusage update
[fio.git] / stat.c
diff --git a/stat.c b/stat.c
index 3adb46eab8087cc53ae0fd2b7c15ad048596c040..3aae76df431e7d9acdb56ec31a2b8a172da195a1 100644 (file)
--- a/stat.c
+++ b/stat.c
@@ -506,9 +506,7 @@ static void show_thread_status_normal(struct thread_stat *ts,
        time_t time_p;
        char time_buf[64];
 
-       if (!(ts->io_bytes[DDIR_READ] + ts->io_bytes[DDIR_WRITE] +
-           ts->io_bytes[DDIR_TRIM]) && !(ts->total_io_u[DDIR_READ] +
-           ts->total_io_u[DDIR_WRITE] + ts->total_io_u[DDIR_TRIM]))
+       if (!ddir_rw_sum(ts->io_bytes) && !ddir_rw_sum(ts->total_io_u))
                return;
 
        time(&time_p);
@@ -574,13 +572,17 @@ static void show_thread_status_normal(struct thread_stat *ts,
                                        io_u_dist[3], io_u_dist[4],
                                        io_u_dist[5], io_u_dist[6]);
        log_info("     issued    : total=r=%llu/w=%llu/d=%llu,"
-                                " short=r=%llu/w=%llu/d=%llu\n",
+                                " short=r=%llu/w=%llu/d=%llu,"
+                                " drop=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]);
+                                       (unsigned long long) ts->short_io_u[2],
+                                       (unsigned long long) ts->drop_io_u[0],
+                                       (unsigned long long) ts->drop_io_u[1],
+                                       (unsigned long long) ts->drop_io_u[2]);
        if (ts->continue_on_error) {
                log_info("     errors    : total=%llu, first_error=%d/<%s>\n",
                                        (unsigned long long)ts->total_err_count,
@@ -703,6 +705,9 @@ static void add_ddir_status_json(struct thread_stat *ts,
        json_object_add_value_int(dir_object, "bw", bw);
        json_object_add_value_int(dir_object, "iops", iops);
        json_object_add_value_int(dir_object, "runtime", ts->runtime[ddir]);
+       json_object_add_value_int(dir_object, "total_ios", ts->total_io_u[ddir]);
+       json_object_add_value_int(dir_object, "short_ios", ts->short_io_u[ddir]);
+       json_object_add_value_int(dir_object, "drop_ios", ts->drop_io_u[ddir]);
 
        if (!calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev)) {
                min = max = 0;
@@ -1072,6 +1077,10 @@ void sum_group_stats(struct group_run_stats *dst, struct group_run_stats *src)
                dst->agg[i] += src->agg[i];
        }
 
+       if (!dst->kb_base)
+               dst->kb_base = src->kb_base;
+       if (!dst->unit_base)
+               dst->unit_base = src->unit_base;
 }
 
 void sum_thread_stats(struct thread_stat *dst, struct thread_stat *src, int nr)
@@ -1123,9 +1132,11 @@ void sum_thread_stats(struct thread_stat *dst, struct thread_stat *src, int nr)
                if (!dst->unified_rw_rep) {
                        dst->total_io_u[k] += src->total_io_u[k];
                        dst->short_io_u[k] += src->short_io_u[k];
+                       dst->drop_io_u[k] += src->drop_io_u[k];
                } else {
                        dst->total_io_u[0] += src->total_io_u[k];
                        dst->short_io_u[0] += src->short_io_u[k];
+                       dst->drop_io_u[0] += src->drop_io_u[k];
                }
        }
 
@@ -1169,7 +1180,7 @@ void init_thread_stat(struct thread_stat *ts)
        ts->groupid = -1;
 }
 
-static void __show_run_stats(void)
+void __show_run_stats(void)
 {
        struct group_run_stats *runstats, *rs;
        struct thread_data *td;
@@ -1411,13 +1422,15 @@ void show_run_stats(void)
        fio_mutex_up(stat_mutex);
 }
 
-static void *__show_running_run_stats(void fio_unused *arg)
+static void __show_running_run_stats(void)
 {
        struct thread_data *td;
        unsigned long long *rt;
        struct timeval tv;
        int i;
 
+       fio_mutex_down(stat_mutex);
+
        rt = malloc(thread_number * sizeof(unsigned long long));
        fio_gettime(&tv, NULL);
 
@@ -1438,6 +1451,8 @@ static void *__show_running_run_stats(void fio_unused *arg)
        }
 
        for_each_td(td, i) {
+               if (td->runstate >= TD_EXITED)
+                       continue;
                if (td->rusage_sem) {
                        td->update_rusage = 1;
                        fio_mutex_down(td->rusage_sem);
@@ -1458,31 +1473,6 @@ static void *__show_running_run_stats(void fio_unused *arg)
 
        free(rt);
        fio_mutex_up(stat_mutex);
-       return NULL;
-}
-
-/*
- * Called from signal handler. It _should_ be safe to just run this inline
- * in the sig handler, but we should be disturbing the system less by just
- * creating a thread to do it.
- */
-void show_running_run_stats(void)
-{
-       pthread_t thread;
-
-       fio_mutex_down(stat_mutex);
-
-       if (!pthread_create(&thread, NULL, __show_running_run_stats, NULL)) {
-               int err;
-
-               err = pthread_detach(thread);
-               if (err)
-                       log_err("fio: DU thread detach failed: %s\n", strerror(err));
-
-               return;
-       }
-
-       fio_mutex_up(stat_mutex);
 }
 
 static int status_interval_init;
@@ -1563,9 +1553,10 @@ static inline void add_stat_sample(struct io_stat *is, unsigned long data)
 
 static void __add_log_sample(struct io_log *iolog, unsigned long val,
                             enum fio_ddir ddir, unsigned int bs,
-                            unsigned long t)
+                            unsigned long t, uint64_t offset)
 {
-       const int nr_samples = iolog->nr_samples;
+       uint64_t nr_samples = iolog->nr_samples;
+       struct io_sample *s;
 
        if (iolog->disabled)
                return;
@@ -1574,23 +1565,43 @@ static void __add_log_sample(struct io_log *iolog, unsigned long val,
                iolog->avg_last = t;
 
        if (iolog->nr_samples == iolog->max_samples) {
-               int new_size = sizeof(struct io_sample) * iolog->max_samples*2;
+               size_t new_size;
                void *new_log;
 
-               new_log = realloc(iolog->log, new_size);
-               if (!new_log) {
-                       log_err("fio: failed extending iolog! Will stop logging.\n");
-                       iolog->disabled = 1;
-                       return;
+               new_size = 2 * iolog->max_samples * log_entry_sz(iolog);
+
+               if (iolog->log_gz && (new_size > iolog->log_gz)) {
+                       if (iolog_flush(iolog, 0)) {
+                               log_err("fio: failed flushing iolog! Will stop logging.\n");
+                               iolog->disabled = 1;
+                               return;
+                       }
+                       nr_samples = iolog->nr_samples;
+               } else {
+                       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;
                }
-               iolog->log = new_log;
-               iolog->max_samples <<= 1;
        }
 
-       iolog->log[nr_samples].val = val;
-       iolog->log[nr_samples].time = t;
-       iolog->log[nr_samples].ddir = ddir;
-       iolog->log[nr_samples].bs = bs;
+       s = get_sample(iolog, nr_samples);
+
+       s->val = val;
+       s->time = t;
+       io_sample_set_ddir(iolog, s, ddir);
+       s->bs = bs;
+
+       if (iolog->log_offset) {
+               struct io_sample_offset *so = (void *) s;
+
+               so->offset = offset;
+       }
+
        iolog->nr_samples++;
 }
 
@@ -1632,6 +1643,7 @@ void reset_io_stats(struct thread_data *td)
        for (i = 0; i < 3; i++) {
                ts->total_io_u[i] = 0;
                ts->short_io_u[i] = 0;
+               ts->drop_io_u[i] = 0;
        }
 }
 
@@ -1646,19 +1658,19 @@ static void _add_stat_to_log(struct io_log *iolog, unsigned long elapsed)
                unsigned long mr;
 
                mr = iolog->avg_window[DDIR_READ].mean.u.f + 0.50;
-               __add_log_sample(iolog, mr, DDIR_READ, 0, elapsed);
+               __add_log_sample(iolog, mr, DDIR_READ, 0, elapsed, 0);
        }
        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);
+               __add_log_sample(iolog, mw, DDIR_WRITE, 0, elapsed, 0);
        }
        if (iolog->avg_window[DDIR_TRIM].samples) {
                unsigned long mw;
 
                mw = iolog->avg_window[DDIR_TRIM].mean.u.f + 0.50;
-               __add_log_sample(iolog, mw, DDIR_TRIM, 0, elapsed);
+               __add_log_sample(iolog, mw, DDIR_TRIM, 0, elapsed, 0);
        }
 
        reset_io_stat(&iolog->avg_window[DDIR_READ]);
@@ -1668,7 +1680,7 @@ static void _add_stat_to_log(struct io_log *iolog, unsigned long elapsed)
 
 static void add_log_sample(struct thread_data *td, struct io_log *iolog,
                           unsigned long val, enum fio_ddir ddir,
-                          unsigned int bs)
+                          unsigned int bs, uint64_t offset)
 {
        unsigned long elapsed, this_window;
 
@@ -1681,7 +1693,7 @@ static void add_log_sample(struct thread_data *td, struct io_log *iolog,
         * If no time averaging, just add the log sample.
         */
        if (!iolog->avg_msec) {
-               __add_log_sample(iolog, val, ddir, bs, elapsed);
+               __add_log_sample(iolog, val, ddir, bs, elapsed, offset);
                return;
        }
 
@@ -1730,7 +1742,7 @@ void add_agg_sample(unsigned long val, enum fio_ddir ddir, unsigned int bs)
                return;
 
        iolog = agg_io_log[ddir];
-       __add_log_sample(iolog, val, ddir, bs, mtime_since_genesis());
+       __add_log_sample(iolog, val, ddir, bs, mtime_since_genesis(), 0);
 }
 
 static void add_clat_percentile_sample(struct thread_stat *ts,
@@ -1743,7 +1755,7 @@ static void add_clat_percentile_sample(struct thread_stat *ts,
 }
 
 void add_clat_sample(struct thread_data *td, enum fio_ddir ddir,
-                    unsigned long usec, unsigned int bs)
+                    unsigned long usec, unsigned int bs, uint64_t offset)
 {
        struct thread_stat *ts = &td->ts;
 
@@ -1753,14 +1765,14 @@ void add_clat_sample(struct thread_data *td, enum fio_ddir ddir,
        add_stat_sample(&ts->clat_stat[ddir], usec);
 
        if (td->clat_log)
-               add_log_sample(td, td->clat_log, usec, ddir, bs);
+               add_log_sample(td, td->clat_log, usec, ddir, bs, offset);
 
        if (ts->clat_percentiles)
                add_clat_percentile_sample(ts, usec, ddir);
 }
 
 void add_slat_sample(struct thread_data *td, enum fio_ddir ddir,
-                    unsigned long usec, unsigned int bs)
+                    unsigned long usec, unsigned int bs, uint64_t offset)
 {
        struct thread_stat *ts = &td->ts;
 
@@ -1770,11 +1782,11 @@ void add_slat_sample(struct thread_data *td, enum fio_ddir ddir,
        add_stat_sample(&ts->slat_stat[ddir], usec);
 
        if (td->slat_log)
-               add_log_sample(td, td->slat_log, usec, ddir, bs);
+               add_log_sample(td, td->slat_log, usec, ddir, bs, offset);
 }
 
 void add_lat_sample(struct thread_data *td, enum fio_ddir ddir,
-                   unsigned long usec, unsigned int bs)
+                   unsigned long usec, unsigned int bs, uint64_t offset)
 {
        struct thread_stat *ts = &td->ts;
 
@@ -1784,7 +1796,7 @@ void add_lat_sample(struct thread_data *td, enum fio_ddir ddir,
        add_stat_sample(&ts->lat_stat[ddir], usec);
 
        if (td->lat_log)
-               add_log_sample(td, td->lat_log, usec, ddir, bs);
+               add_log_sample(td, td->lat_log, usec, ddir, bs, offset);
 }
 
 void add_bw_sample(struct thread_data *td, enum fio_ddir ddir, unsigned int bs,
@@ -1818,7 +1830,7 @@ void add_bw_sample(struct thread_data *td, enum fio_ddir ddir, unsigned int bs,
                add_stat_sample(&ts->bw_stat[ddir], rate);
 
                if (td->bw_log)
-                       add_log_sample(td, td->bw_log, rate, ddir, bs);
+                       add_log_sample(td, td->bw_log, rate, ddir, bs, 0);
 
                td->stat_io_bytes[ddir] = td->this_io_bytes[ddir];
        }
@@ -1857,7 +1869,7 @@ void add_iops_sample(struct thread_data *td, enum fio_ddir ddir, unsigned int bs
                add_stat_sample(&ts->iops_stat[ddir], iops);
 
                if (td->iops_log)
-                       add_log_sample(td, td->iops_log, iops, ddir, bs);
+                       add_log_sample(td, td->iops_log, iops, ddir, bs, 0);
 
                td->stat_io_blocks[ddir] = td->this_io_blocks[ddir];
        }
@@ -1879,3 +1891,53 @@ void stat_exit(void)
        fio_mutex_down(stat_mutex);
        fio_mutex_remove(stat_mutex);
 }
+
+static pthread_t si_thread;
+static pthread_cond_t si_cond;
+static pthread_mutex_t si_lock;
+static int si_thread_exit;
+
+/*
+ * Called from signal handler. Wake up status thread.
+ */
+void show_running_run_stats(void)
+{
+       pthread_cond_signal(&si_cond);
+}
+
+static void *si_thread_main(void *unused)
+{
+       while (!si_thread_exit) {
+               pthread_cond_wait(&si_cond, &si_lock);
+               if (si_thread_exit)
+                       break;
+
+               __show_running_run_stats();
+       }
+
+       return NULL;
+}
+
+void create_status_interval_thread(void)
+{
+       int ret;
+
+       pthread_cond_init(&si_cond, NULL);
+       pthread_mutex_init(&si_lock, NULL);
+
+       ret = pthread_create(&si_thread, NULL, si_thread_main, NULL);
+       if (ret) {
+               log_err("Can't create status thread: %s\n", strerror(ret));
+               return;
+       }
+}
+
+void wait_for_status_interval_thread_exit(void)
+{
+       void *ret;
+
+       si_thread_exit = 1;
+       pthread_cond_signal(&si_cond);
+       pthread_join(si_thread, &ret);
+       pthread_cond_destroy(&si_cond);
+}