X-Git-Url: https://git.kernel.dk/?a=blobdiff_plain;f=backend.c;h=981625b61095dd42995d32291ec3fdc4b58cb597;hb=b470a02cade02049509d22442addfbc88d10116e;hp=e0f8aa76f584dc8f978c1f2649fae5292f563718;hpb=77731b29c35286a43ca6109ebf13389169ec20e9;p=fio.git diff --git a/backend.c b/backend.c index e0f8aa76..981625b6 100644 --- a/backend.c +++ b/backend.c @@ -53,6 +53,7 @@ #include "lib/getrusage.h" #include "idletime.h" #include "err.h" +#include "lib/tp.h" static pthread_t disk_util_thread; static struct fio_mutex *disk_thread_mutex; @@ -390,7 +391,7 @@ 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 +460,7 @@ 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 +714,7 @@ 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; } } @@ -780,16 +781,21 @@ static uint64_t do_io(struct thread_data *td) case FIO_Q_COMPLETED: if (io_u->error) { ret = -io_u->error; + unlog_io_piece(td, io_u); clear_io_u(td, io_u); } else if (io_u->resid) { int bytes = io_u->xfer_buflen - io_u->resid; struct fio_file *f = io_u->file; bytes_issued += bytes; + + trim_io_piece(td, io_u); + /* * zero read, fail */ if (!bytes) { + unlog_io_piece(td, io_u); td_verror(td, EIO, "full resid"); put_io_u(td, io_u); break; @@ -830,6 +836,7 @@ sync_done: bytes_issued += io_u->xfer_buflen; break; case FIO_Q_BUSY: + unlog_io_piece(td, io_u); requeue_io_u(td, &io_u); ret2 = td_io_commit(td); if (ret2 < 0) @@ -915,7 +922,7 @@ 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; @@ -1264,11 +1271,6 @@ static void *thread_main(void *data) } else td->pid = gettid(); - /* - * fio_time_init() may not have been called yet if running as a server - */ - fio_time_init(); - fio_local_clock_init(o->use_thread); dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); @@ -1339,6 +1341,7 @@ static void *thread_main(void *data) #ifdef CONFIG_LIBNUMA /* numa node setup */ if (o->numa_cpumask_set || o->numa_memmask_set) { + struct bitmask *mask; int ret; if (numa_available() < 0) { @@ -1347,7 +1350,9 @@ static void *thread_main(void *data) } if (o->numa_cpumask_set) { - ret = numa_run_on_node_mask(o->numa_cpunodesmask); + mask = numa_parse_nodestring(o->numa_cpunodes); + ret = numa_run_on_node_mask(mask); + numa_free_nodemask(mask); if (ret == -1) { td_verror(td, errno, \ "numa_run_on_node_mask failed\n"); @@ -1357,12 +1362,16 @@ static void *thread_main(void *data) if (o->numa_memmask_set) { + mask = NULL; + if (o->numa_memnodes) + mask = numa_parse_nodestring(o->numa_memnodes); + switch (o->numa_mem_mode) { case MPOL_INTERLEAVE: - numa_set_interleave_mask(o->numa_memnodesmask); + numa_set_interleave_mask(mask); break; case MPOL_BIND: - numa_set_membind(o->numa_memnodesmask); + numa_set_membind(mask); break; case MPOL_LOCAL: numa_set_localalloc(); @@ -1375,6 +1384,9 @@ static void *thread_main(void *data) break; } + if (mask) + numa_free_nodemask(mask); + } } #endif @@ -1432,6 +1444,9 @@ 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); @@ -1513,6 +1528,9 @@ 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"); @@ -1564,7 +1582,7 @@ static int fork_main(int shmid, int offset) struct thread_data *td; void *data, *ret; -#ifndef __hpux +#if !defined(__hpux) && !defined(CONFIG_NO_SHM) data = shmat(shmid, NULL, 0); if (data == (void *) -1) { int __err = errno; @@ -1585,6 +1603,13 @@ 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. */ @@ -1662,6 +1687,17 @@ 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 */ @@ -2009,9 +2045,13 @@ int fio_backend(void) return 0; if (write_bw_log) { - setup_log(&agg_io_log[DDIR_READ], 0, IO_LOG_TYPE_BW); - setup_log(&agg_io_log[DDIR_WRITE], 0, IO_LOG_TYPE_BW); - setup_log(&agg_io_log[DDIR_TRIM], 0, IO_LOG_TYPE_BW); + 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); @@ -2028,13 +2068,16 @@ int fio_backend(void) run_threads(); 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); + } } }