Record job start time to fix time pain points

Add a new key in the json per-job output, job_start, that records the job start time obtained via a call to clock_gettime using the clock_id specified by the new job_start_clock_id option. This allows times of fio jobs and log entries to be compared/ordered against each other and against other system events recorded against the same clock_id. Add a note to the documentation for group_reporting about how there are several per-job values for which only the first job's value is recorded in the json output format when group_reporting is enabled. Fixes #1544 Signed-off-by: Nick Neumann [email protected]
axboe · Sep 8, 2023 · 12d325c · 12d325c
1 parent 4a0c766
commit 12d325c
Show file tree

Hide file tree

Showing 15 changed files with 68 additions and 15 deletions.
diff --git a/HOWTO.rst b/HOWTO.rst
@@ -755,6 +755,10 @@ Time related parameters
 	calls will be excluded from other uses. Fio will manually clear it from the
 	CPU mask of other jobs.
 
+.. option:: job_start_clock_id=int
+   The clock_id passed to the call to `clock_gettime` used to record job_start
+   in the `json` output format. Default is 0, or CLOCK_REALTIME.
+
 
 Target file/device
 ~~~~~~~~~~~~~~~~~~
@@ -3966,6 +3970,13 @@ Measurements and reporting
 	same reporting group, unless if separated by a :option:`stonewall`, or by
 	using :option:`new_group`.
 
+    NOTE: When :option: `group_reporting` is used along with `json` output,
+    there are certain per-job properties which can be different between jobs
+    but do not have a natural group-level equivalent. Examples include
+    `kb_base`, `unit_base`, `sig_figs`, `thread_number`, `pid`, and
+    `job_start`. For these properties, the values for the first job are
+    recorded for the group.
+
 .. option:: new_group
 
 	Start a new reporting group. See: :option:`group_reporting`.  If not given,

diff --git a/backend.c b/backend.c
@@ -1858,7 +1858,7 @@ static void *thread_main(void *data)
 	if (rate_submit_init(td, sk_out))
 		goto err;
 
-	set_epoch_time(td, o->log_unix_epoch | o->log_alternate_epoch, o->log_alternate_epoch_clock_id);
+	set_epoch_time(td, o->log_alternate_epoch_clock_id, o->job_start_clock_id);
 	fio_getrusage(&td->ru_start);
 	memcpy(&td->bw_sample_time, &td->epoch, sizeof(td->epoch));
 	memcpy(&td->iops_sample_time, &td->epoch, sizeof(td->epoch));

diff --git a/cconv.c b/cconv.c
@@ -219,6 +219,7 @@ int convert_thread_options_to_cpu(struct thread_options *o,
 	o->log_unix_epoch = le32_to_cpu(top->log_unix_epoch);
 	o->log_alternate_epoch = le32_to_cpu(top->log_alternate_epoch);
 	o->log_alternate_epoch_clock_id = le32_to_cpu(top->log_alternate_epoch_clock_id);
+	o->job_start_clock_id = le32_to_cpu(top->job_start_clock_id);
 	o->norandommap = le32_to_cpu(top->norandommap);
 	o->softrandommap = le32_to_cpu(top->softrandommap);
 	o->bs_unaligned = le32_to_cpu(top->bs_unaligned);
@@ -458,6 +459,7 @@ void convert_thread_options_to_net(struct thread_options_pack *top,
 	top->log_unix_epoch = cpu_to_le32(o->log_unix_epoch);
 	top->log_alternate_epoch = cpu_to_le32(o->log_alternate_epoch);
 	top->log_alternate_epoch_clock_id = cpu_to_le32(o->log_alternate_epoch_clock_id);
+	top->job_start_clock_id = cpu_to_le32(o->job_start_clock_id);
 	top->norandommap = cpu_to_le32(o->norandommap);
 	top->softrandommap = cpu_to_le32(o->softrandommap);
 	top->bs_unaligned = cpu_to_le32(o->bs_unaligned);

diff --git a/client.c b/client.c
@@ -956,6 +956,7 @@ static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
 	dst->error		= le32_to_cpu(src->error);
 	dst->thread_number	= le32_to_cpu(src->thread_number);
 	dst->groupid		= le32_to_cpu(src->groupid);
+	dst->job_start		= le64_to_cpu(src->job_start);
 	dst->pid		= le32_to_cpu(src->pid);
 	dst->members		= le32_to_cpu(src->members);
 	dst->unified_rw_rep	= le32_to_cpu(src->unified_rw_rep);

diff --git a/fio.1 b/fio.1
@@ -537,6 +537,10 @@ copy that segment, instead of entering the kernel with a
 \fBgettimeofday\fR\|(2) call. The CPU set aside for doing these time
 calls will be excluded from other uses. Fio will manually clear it from the
 CPU mask of other jobs.
+.TP
+.BI job_start_clock_id \fR=\fPint
+The clock_id passed to the call to \fBclock_gettime\fR used to record job_start
+in the \fBjson\fR output format. Default is 0, or CLOCK_REALTIME.
 .SS "Target file/device"
 .TP
 .BI directory \fR=\fPstr
@@ -3664,6 +3668,15 @@ quickly becomes unwieldy. To see the final report per-group instead of
 per-job, use \fBgroup_reporting\fR. Jobs in a file will be part of the
 same reporting group, unless if separated by a \fBstonewall\fR, or by
 using \fBnew_group\fR.
+.RS
+.P
+NOTE: When \fBgroup_reporting\fR is used along with \fBjson\fR output, there
+are certain per-job properties which can be different between jobs but do not
+have a natural group-level equivalent. Examples include \fBkb_base\fR,
+\fBunit_base\fR, \fBsig_figs\fR, \fBthread_number\fR, \fBpid\fR, and
+\fBjob_start\fR. For these properties, the values for the first job are
+recorded for the group.
+.RE
 .TP
 .BI new_group
 Start a new reporting group. See: \fBgroup_reporting\fR. If not given,

diff --git a/fio.h b/fio.h
@@ -388,7 +388,8 @@ struct thread_data {
 
 	struct timespec start;	/* start of this loop */
 	struct timespec epoch;	/* time job was started */
-	unsigned long long alternate_epoch; /* Time job was started, clock_gettime's clock_id epoch based. */
+	unsigned long long alternate_epoch; /* Time job was started, as clock_gettime(log_alternate_epoch_clock_id) */
+	unsigned long long job_start; /* Time job was started, as clock_gettime(job_start_clock_id) */
 	struct timespec last_issue;
 	long time_offset;
 	struct timespec ts_cache;

diff --git a/fio_time.h b/fio_time.h
@@ -30,6 +30,6 @@ extern bool ramp_time_over(struct thread_data *);
 extern bool in_ramp_time(struct thread_data *);
 extern void fio_time_init(void);
 extern void timespec_add_msec(struct timespec *, unsigned int);
-extern void set_epoch_time(struct thread_data *, int, clockid_t);
+extern void set_epoch_time(struct thread_data *, clockid_t, clockid_t);
 
 #endif
diff --git a/libfio.c b/libfio.c
@@ -149,7 +149,7 @@ void reset_all_stats(struct thread_data *td)
 		td->ts.runtime[i] = 0;
 	}
 
-	set_epoch_time(td, td->o.log_unix_epoch | td->o.log_alternate_epoch, td->o.log_alternate_epoch_clock_id);
+	set_epoch_time(td, td->o.log_alternate_epoch_clock_id, td->o.job_start_clock_id);
 	memcpy(&td->start, &td->epoch, sizeof(td->epoch));
 	memcpy(&td->iops_sample_time, &td->epoch, sizeof(td->epoch));
 	memcpy(&td->bw_sample_time, &td->epoch, sizeof(td->epoch));

diff --git a/options.c b/options.c
@@ -4952,6 +4952,16 @@ struct fio_option fio_options[FIO_MAX_OPTS] = {
 		.category = FIO_OPT_C_GENERAL,
 		.group	= FIO_OPT_G_CLOCK,
 	},
+	{
+		.name	= "job_start_clock_id",
+		.lname	= "Job start clock_id",
+		.type	= FIO_OPT_INT,
+		.off1	= offsetof(struct thread_options, job_start_clock_id),
+		.help	= "The clock_id passed to the call to clock_gettime used to record job_start in the json output format. Default is 0, or CLOCK_REALTIME",
+		.verify	= gtod_cpu_verify,
+		.category = FIO_OPT_C_GENERAL,
+		.group	= FIO_OPT_G_CLOCK,
+	},
 	{
 		.name	= "unified_rw_reporting",
 		.lname	= "Unified RW Reporting",

diff --git a/rate-submit.c b/rate-submit.c
@@ -185,7 +185,7 @@ static int io_workqueue_init_worker_fn(struct submit_worker *sw)
 	if (td->io_ops->post_init && td->io_ops->post_init(td))
 		goto err_io_init;
 
-	set_epoch_time(td, td->o.log_unix_epoch | td->o.log_alternate_epoch, td->o.log_alternate_epoch_clock_id);
+	set_epoch_time(td, td->o.log_alternate_epoch_clock_id, td->o.job_start_clock_id);
 	fio_getrusage(&td->ru_start);
 	clear_io_state(td, 1);
 

diff --git a/server.c b/server.c
@@ -1706,6 +1706,7 @@ void fio_server_send_ts(struct thread_stat *ts, struct group_run_stats *rs)
 	p.ts.error		= cpu_to_le32(ts->error);
 	p.ts.thread_number	= cpu_to_le32(ts->thread_number);
 	p.ts.groupid		= cpu_to_le32(ts->groupid);
+	p.ts.job_start		= cpu_to_le64(ts->job_start);
 	p.ts.pid		= cpu_to_le32(ts->pid);
 	p.ts.members		= cpu_to_le32(ts->members);
 	p.ts.unified_rw_rep	= cpu_to_le32(ts->unified_rw_rep);

diff --git a/stat.c b/stat.c
@@ -1712,6 +1712,7 @@ static struct json_object *show_thread_status_json(struct thread_stat *ts,
 	root = json_create_object();
 	json_object_add_value_string(root, "jobname", ts->name);
 	json_object_add_value_int(root, "groupid", ts->groupid);
+	json_object_add_value_int(root, "job_start", ts->job_start);
 	json_object_add_value_int(root, "error", ts->error);
 
 	/* ETA Info */
@@ -2526,6 +2527,7 @@ void __show_run_stats(void)
 			 */
 			ts->thread_number = td->thread_number;
 			ts->groupid = td->groupid;
+			ts->job_start = td->job_start;
 
 			/*
 			 * first pid in group, not very useful...
@@ -3048,7 +3050,9 @@ static void __add_log_sample(struct io_log *iolog, union io_sample_data data,
 		s = get_sample(iolog, cur_log, cur_log->nr_samples);
 
 		s->data = data;
-		s->time = t + (iolog->td ? iolog->td->alternate_epoch : 0);
+		s->time = t;
+		if (iolog->td && (iolog->td->o.log_unix_epoch || iolog->td->o.log_alternate_epoch))
+			s->time += iolog->td->alternate_epoch;
 		io_sample_set_ddir(iolog, s, ddir);
 		s->bs = bs;
 		s->priority = priority;

diff --git a/stat.h b/stat.h
@@ -169,6 +169,7 @@ struct thread_stat {
 	uint32_t error;
 	uint32_t thread_number;
 	uint32_t groupid;
+	uint64_t job_start; /* Time job was started, as clock_gettime(job_start_clock_id) */
 	uint32_t pid;
 	char description[FIO_JOBDESC_SIZE];
 	uint32_t members;

diff --git a/thread_options.h b/thread_options.h
@@ -273,6 +273,7 @@ struct thread_options {
 	unsigned int unified_rw_rep;
 	unsigned int gtod_reduce;
 	unsigned int gtod_cpu;
+	unsigned int job_start_clock_id;
 	enum fio_cs clocksource;
 	unsigned int no_stall;
 	unsigned int trim_percentage;
@@ -422,7 +423,6 @@ struct thread_options_pack {
 	uint32_t iodepth_batch_complete_min;
 	uint32_t iodepth_batch_complete_max;
 	uint32_t serialize_overlap;
-	uint32_t pad;
 
 	uint64_t size;
 	uint64_t io_size;
@@ -433,13 +433,11 @@ struct thread_options_pack {
 	uint32_t fill_device;
 	uint32_t file_append;
 	uint32_t unique_filename;
-	uint32_t pad3;
 	uint64_t file_size_low;
 	uint64_t file_size_high;
 	uint64_t start_offset;
 	uint64_t start_offset_align;
 	uint32_t start_offset_nz;
-	uint32_t pad4;
 
 	uint64_t bs[DDIR_RWDIR_CNT];
 	uint64_t ba[DDIR_RWDIR_CNT];
@@ -511,6 +509,7 @@ struct thread_options_pack {
 	struct zone_split zone_split[DDIR_RWDIR_CNT][ZONESPLIT_MAX];
 	uint32_t zone_split_nr[DDIR_RWDIR_CNT];
 
+	uint32_t pad;
 	fio_fp64_t zipf_theta;
 	fio_fp64_t pareto_h;
 	fio_fp64_t gauss_dev;
@@ -593,6 +592,7 @@ struct thread_options_pack {
 	uint32_t unified_rw_rep;
 	uint32_t gtod_reduce;
 	uint32_t gtod_cpu;
+	uint32_t job_start_clock_id;
 	uint32_t clocksource;
 	uint32_t no_stall;
 	uint32_t trim_percentage;
@@ -603,6 +603,7 @@ struct thread_options_pack {
 	uint32_t lat_percentiles;
 	uint32_t slat_percentiles;
 	uint32_t percentile_precision;
+	uint32_t pad2;
 	fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
 
 	uint8_t read_iolog_file[FIO_TOP_STR_MAX];

diff --git a/time.c b/time.c
@@ -172,14 +172,22 @@ void set_genesis_time(void)
 	fio_gettime(&genesis, NULL);
 }
 
-void set_epoch_time(struct thread_data *td, int log_alternate_epoch, clockid_t clock_id)
+void set_epoch_time(struct thread_data *td, clockid_t log_alternate_epoch_clock_id, clockid_t job_start_clock_id)
 {
+	struct timespec ts;
 	fio_gettime(&td->epoch, NULL);
-	if (log_alternate_epoch) {
-		struct timespec ts;
-		clock_gettime(clock_id, &ts);
-		td->alternate_epoch = (unsigned long long)(ts.tv_sec) * 1000 +
-		                 (unsigned long long)(ts.tv_nsec) / 1000000;
+	clock_gettime(log_alternate_epoch_clock_id, &ts);
+	td->alternate_epoch = (unsigned long long)(ts.tv_sec) * 1000 +
+						  (unsigned long long)(ts.tv_nsec) / 1000000;
+	if (job_start_clock_id == log_alternate_epoch_clock_id)
+	{
+		td->job_start = td->alternate_epoch;
+	}
+	else
+	{
+		clock_gettime(job_start_clock_id, &ts);
+		td->job_start = (unsigned long long)(ts.tv_sec) * 1000 +
+						(unsigned long long)(ts.tv_nsec) / 1000000;
 	}
 }