Get rid of __ prefix for internal frand state
[fio.git] / backend.c
index ac6ed3e45fc2c8cc319a6a614165214ae266464a..cae4e38de4b9fb3adc871863558dd1230bba32b2 100644 (file)
--- a/backend.c
+++ b/backend.c
 #include "lib/getrusage.h"
 #include "idletime.h"
 #include "err.h"
+#include "lib/tp.h"
+
+static pthread_t helper_thread;
+static pthread_mutex_t helper_lock;
+pthread_cond_t helper_cond;
+int helper_do_stat = 0;
 
-static pthread_t disk_util_thread;
-static struct fio_mutex *disk_thread_mutex;
 static struct fio_mutex *startup_mutex;
 static struct flist_head *cgroup_list;
 static char *cgroup_mnt;
@@ -72,7 +76,7 @@ unsigned int stat_number = 0;
 int shm_id = 0;
 int temp_stall_ts;
 unsigned long done_secs = 0;
-volatile int disk_util_exit = 0;
+volatile int helper_exit = 0;
 
 #define PAGE_ALIGN(buf)        \
        (char *) (((uintptr_t) (buf) + page_mask) & ~page_mask)
@@ -86,7 +90,7 @@ static void sig_int(int sig)
                        fio_server_got_signal(sig);
                else {
                        log_info("\nfio: terminating on signal %d\n", sig);
-                       fflush(stdout);
+                       log_info_flush();
                        exit_value = 128;
                }
 
@@ -390,7 +394,7 @@ static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir,
                         * fill_device option is set.
                         */
                        td_clear_error(td);
-                       td->terminate = 1;
+                       fio_mark_td_terminate(td);
                        return 1;
                } else {
                        /*
@@ -459,7 +463,7 @@ static void do_verify(struct thread_data *td, uint64_t verify_bytes)
                if (runtime_exceeded(td, &td->tv_cache)) {
                        __update_tv_cache(td);
                        if (runtime_exceeded(td, &td->tv_cache)) {
-                               td->terminate = 1;
+                               fio_mark_td_terminate(td);
                                break;
                        }
                }
@@ -713,7 +717,7 @@ static uint64_t do_io(struct thread_data *td)
                if (runtime_exceeded(td, &td->tv_cache)) {
                        __update_tv_cache(td);
                        if (runtime_exceeded(td, &td->tv_cache)) {
-                               td->terminate = 1;
+                               fio_mark_td_terminate(td);
                                break;
                        }
                }
@@ -749,9 +753,9 @@ static uint64_t do_io(struct thread_data *td)
                    ((io_u->flags & IO_U_F_VER_LIST) || !td_rw(td))) {
 
                        if (!td->o.verify_pattern_bytes) {
-                               io_u->rand_seed = __rand(&td->__verify_state);
+                               io_u->rand_seed = __rand(&td->verify_state);
                                if (sizeof(int) != sizeof(long *))
-                                       io_u->rand_seed *= __rand(&td->__verify_state);
+                                       io_u->rand_seed *= __rand(&td->verify_state);
                        }
 
                        if (td->o.verify_async)
@@ -921,7 +925,7 @@ reap:
 
        if (td->o.fill_device && td->error == ENOSPC) {
                td->error = 0;
-               td->terminate = 1;
+               fio_mark_td_terminate(td);
        }
        if (!td->error) {
                struct fio_file *f;
@@ -1443,6 +1447,9 @@ static void *thread_main(void *data)
                        goto err;
        }
 
+       if (td->flags & TD_F_COMPRESS_LOG)
+               tp_init(&td->tp_data);
+
        fio_verify_init(td);
 
        fio_gettime(&td->epoch, NULL);
@@ -1524,6 +1531,9 @@ static void *thread_main(void *data)
 
        fio_writeout_logs(td);
 
+       if (td->flags & TD_F_COMPRESS_LOG)
+               tp_exit(&td->tp_data);
+
        if (o->exec_postrun)
                exec_string(o, o->exec_postrun, (const char *)"postrun");
 
@@ -1555,13 +1565,17 @@ err:
        if (o->write_iolog_file)
                write_iolog_close(td);
 
-       fio_mutex_remove(td->rusage_sem);
-       td->rusage_sem = NULL;
-
        fio_mutex_remove(td->mutex);
        td->mutex = NULL;
 
        td_set_runstate(td, TD_EXITED);
+
+       /*
+        * Do this last after setting our runstate to exited, so we
+        * know that the stat thread is signaled.
+        */
+       check_update_rusage(td);
+
        return (void *) (uintptr_t) td->error;
 }
 
@@ -1596,6 +1610,13 @@ static int fork_main(int shmid, int offset)
        return (int) (uintptr_t) ret;
 }
 
+static void dump_td_info(struct thread_data *td)
+{
+       log_err("fio: job '%s' hasn't exited in %lu seconds, it appears to "
+               "be stuck. Doing forceful exit of this job.\n", td->o.name,
+                       (unsigned long) time_since_now(&td->terminate_time));
+}
+
 /*
  * Run over the job map and reap the threads that have exited, if any.
  */
@@ -1673,6 +1694,17 @@ static void reap_threads(unsigned int *nr_running, unsigned int *t_rate,
                        }
                }
 
+               /*
+                * If the job is stuck, do a forceful timeout of it and
+                * move on.
+                */
+               if (td->terminate &&
+                   time_since_now(&td->terminate_time) >= FIO_REAP_TIMEOUT) {
+                       dump_td_info(td);
+                       td_set_runstate(td, TD_REAPED);
+                       goto reaped;
+               }
+
                /*
                 * thread is not dead, continue
                 */
@@ -1738,7 +1770,7 @@ static void run_threads(void)
                                                nr_process > 1 ? "es" : "");
                }
                log_info("\n");
-               fflush(stdout);
+               log_info_flush();
        }
 
        todo = thread_number;
@@ -1945,57 +1977,70 @@ static void run_threads(void)
        update_io_ticks();
 }
 
-void wait_for_disk_thread_exit(void)
+static void wait_for_helper_thread_exit(void)
 {
-       fio_mutex_down(disk_thread_mutex);
+       void *ret;
+
+       helper_exit = 1;
+       pthread_cond_signal(&helper_cond);
+       pthread_join(helper_thread, &ret);
 }
 
 static void free_disk_util(void)
 {
-       disk_util_start_exit();
-       wait_for_disk_thread_exit();
        disk_util_prune_entries();
+
+       pthread_cond_destroy(&helper_cond);
 }
 
-static void *disk_thread_main(void *data)
+static void *helper_thread_main(void *data)
 {
        int ret = 0;
 
        fio_mutex_up(startup_mutex);
 
-       while (threads && !ret) {
-               usleep(DISK_UTIL_MSEC * 1000);
-               if (!threads)
-                       break;
+       while (!ret) {
+               uint64_t sec = DISK_UTIL_MSEC / 1000;
+               uint64_t nsec = (DISK_UTIL_MSEC % 1000) * 1000000;
+               struct timespec ts;
+               struct timeval tv;
+
+               gettimeofday(&tv, NULL);
+               ts.tv_sec = tv.tv_sec + sec;
+               ts.tv_nsec = (tv.tv_usec * 1000) + nsec;
+               if (ts.tv_nsec > 1000000000ULL) {
+                       ts.tv_nsec -= 1000000000ULL;
+                       ts.tv_sec++;
+               }
+
+               pthread_cond_timedwait(&helper_cond, &helper_lock, &ts);
+
                ret = update_io_ticks();
 
+               if (helper_do_stat) {
+                       helper_do_stat = 0;
+                       __show_running_run_stats();
+               }
+
                if (!is_backend)
                        print_thread_status();
        }
 
-       fio_mutex_up(disk_thread_mutex);
        return NULL;
 }
 
-static int create_disk_util_thread(void)
+static int create_helper_thread(void)
 {
        int ret;
 
        setup_disk_util();
 
-       disk_thread_mutex = fio_mutex_init(FIO_MUTEX_LOCKED);
-
-       ret = pthread_create(&disk_util_thread, NULL, disk_thread_main, NULL);
-       if (ret) {
-               fio_mutex_remove(disk_thread_mutex);
-               log_err("Can't create disk util thread: %s\n", strerror(ret));
-               return 1;
-       }
+       pthread_cond_init(&helper_cond, NULL);
+       pthread_mutex_init(&helper_lock, NULL);
 
-       ret = pthread_detach(disk_util_thread);
+       ret = pthread_create(&helper_thread, NULL, helper_thread_main, NULL);
        if (ret) {
-               fio_mutex_remove(disk_thread_mutex);
-               log_err("Can't detatch disk util thread: %s\n", strerror(ret));
+               log_err("Can't create helper thread: %s\n", strerror(ret));
                return 1;
        }
 
@@ -2020,9 +2065,13 @@ int fio_backend(void)
                return 0;
 
        if (write_bw_log) {
-               setup_log(&agg_io_log[DDIR_READ], 0, IO_LOG_TYPE_BW, 0);
-               setup_log(&agg_io_log[DDIR_WRITE], 0, IO_LOG_TYPE_BW, 0);
-               setup_log(&agg_io_log[DDIR_TRIM], 0, IO_LOG_TYPE_BW, 0);
+               struct log_params p = {
+                       .log_type = IO_LOG_TYPE_BW,
+               };
+
+               setup_log(&agg_io_log[DDIR_READ], &p, "agg-read_bw.log");
+               setup_log(&agg_io_log[DDIR_WRITE], &p, "agg-write_bw.log");
+               setup_log(&agg_io_log[DDIR_TRIM], &p, "agg-trim_bw.log");
        }
 
        startup_mutex = fio_mutex_init(FIO_MUTEX_LOCKED);
@@ -2031,26 +2080,34 @@ int fio_backend(void)
 
        set_genesis_time();
        stat_init();
-       create_disk_util_thread();
+       create_helper_thread();
 
        cgroup_list = smalloc(sizeof(*cgroup_list));
        INIT_FLIST_HEAD(cgroup_list);
 
        run_threads();
 
+       wait_for_helper_thread_exit();
+
        if (!fio_abort) {
-               show_run_stats();
+               __show_run_stats();
                if (write_bw_log) {
-                       __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log");
-                       __finish_log(agg_io_log[DDIR_WRITE],
-                                       "agg-write_bw.log");
-                       __finish_log(agg_io_log[DDIR_TRIM],
-                                       "agg-write_bw.log");
+                       int i;
+
+                       for (i = 0; i < DDIR_RWDIR_CNT; i++) {
+                               struct io_log *log = agg_io_log[i];
+
+                               flush_log(log);
+                               free_log(log);
+                       }
                }
        }
 
-       for_each_td(td, i)
+       for_each_td(td, i) {
                fio_options_free(td);
+               fio_mutex_remove(td->rusage_sem);
+               td->rusage_sem = NULL;
+       }
 
        free_disk_util();
        cgroup_kill(cgroup_list);
@@ -2058,7 +2115,6 @@ int fio_backend(void)
        sfree(cgroup_mnt);
 
        fio_mutex_remove(startup_mutex);
-       fio_mutex_remove(disk_thread_mutex);
        stat_exit();
        return exit_value;
 }