X-Git-Url: https://git.kernel.dk/?a=blobdiff_plain;f=backend.c;h=30f78b72f9d15acdca6c04930a8e26db53fa2d5e;hb=908119302fbb900a449988d03dacc95426c09e08;hp=ee75566af84cd6f07a8a4e824ba79e9788667d1b;hpb=c89318761586dbd77b8f31ce0ed68ff4fc086c89;p=fio.git

diff --git a/backend.c b/backend.c
index ee75566a..30f78b72 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;
 			}
 		}
@@ -921,7 +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;
@@ -1443,6 +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);
@@ -1524,6 +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");
 
@@ -1596,6 +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.
  */
@@ -1673,6 +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
 		 */
@@ -2020,9 +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);
@@ -2041,11 +2070,14 @@ int fio_backend(void)
 	if (!fio_abort) {
 		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);
+			}
 		}
 	}