fio: when not using a job file, timeouts are not applied

fio: when not using a job file, timeouts are not applied

[Daniel Farina]

Hello list,

Recently I found a bug whereby when when the --timeout option is
utilized yet one does not use a job file (instead packing all the jobs
into the arguments to fio) that timeouts to not work in fio 1.55.
Having poked around via gdb it does appear the step whereby the
timeout is propagated throughout the program does not occur when there
are precisely 0 job files.

Let me know if I can help with additional details, or it is thought
that I am in error.

Thanks,
fdr

Re: fio: when not using a job file, timeouts are not applied

[Chris Worley]

Can we see your fio command?  I never use job files, and always use
timeouts successfully.

Chris
On Mon, Jul 18, 2011 at 1:01 PM, Daniel Farina <daniel@heroku.com> wrote:
>
> Hello list,
>
> Recently I found a bug whereby when when the --timeout option is
> utilized yet one does not use a job file (instead packing all the jobs
> into the arguments to fio) that timeouts to not work in fio 1.55.
> Having poked around via gdb it does appear the step whereby the
> timeout is propagated throughout the program does not occur when there
> are precisely 0 job files.
>
> Let me know if I can help with additional details, or it is thought
> that I am in error.
>
> Thanks,
> fdr
> --
> To unsubscribe from this list: send the line "unsubscribe fio" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at �http://vger.kernel.org/majordomo-info.html

Re: fio: when not using a job file, timeouts are not applied

[Daniel Farina]

On Mon, Jul 18, 2011 at 12:35 PM, Chris Worley <worleys@gmail.com> wrote:
> Can we see your fio command? �I never use job files, and always use
> timeouts successfully.

Absolutely:

# Doesn't terminate for "quite a while" (I wasn't willing to wait)
time sudo fio --readonly --minimal --norandommap --timeout=5
-name=/dev/sda1 --rw=randread --size=1GB --direct=1
--filename=/dev/sda1

With a jobfile called jobs.fio:

[/dev/sda1]
rw=randread
size=1GB
filename=/dev/sda1
direct=1

Everything is great:

time sudo fio --readonly --minimal --norandommap --timeout=5 jobs.fio
fio 1.55
2;/dev/sda1;0;0;41820;8563;5001;0;0;0.000000;0.000000;322;90574;460.528551;1378.863841;322;90575;461.305308;1378.867201;6880;9232;101.283584%;8469.333333;915.606903;0;0;0;0;0;0.000000;0.000000;0;0;0.000000;0.000000;0;0;0.000000;0.000000;0;0;0.000000%;0.000000;0.000000;4.000000%;0.000000%;10467;0;35;100.0%;0.0%;0.0%;0.0%;0.0%;0.0%;0.0%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;95.88%;3.25%;0.60%;0.17%;0.05%;0.02%;0.00%;0.00%;0.03%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%


real	0m5.288s
user	0m0.200s
sys	0m0.150s


-- 
fdr

Re: fio: when not using a job file, timeouts are not applied

[Chris Worley]

On Mon, Jul 18, 2011 at 3:46 PM, Daniel Farina <daniel@heroku.com> wrote:
> On Mon, Jul 18, 2011 at 12:35 PM, Chris Worley <worleys@gmail.com> wrote:
>> Can we see your fio command? �I never use job files, and always use
>> timeouts successfully.
>
> Absolutely:
>
> # Doesn't terminate for "quite a while" (I wasn't willing to wait)
> time sudo fio --readonly --minimal --norandommap --timeout=5
> -name=/dev/sda1 --rw=randread --size=1GB --direct=1

Is that a single dash prior to "name"?

Chris
> --filename=/dev/sda1
>
> With a jobfile called jobs.fio:
>
> [/dev/sda1]
> rw=randread
> size=1GB
> filename=/dev/sda1
> direct=1
>
> Everything is great:
>
> time sudo fio --readonly --minimal --norandommap --timeout=5 jobs.fio
> fio 1.55
> 2;/dev/sda1;0;0;41820;8563;5001;0;0;0.000000;0.000000;322;90574;460.528551;1378.863841;322;90575;461.305308;1378.867201;6880;9232;101.283584%;8469.333333;915.606903;0;0;0;0;0;0.000000;0.000000;0;0;0.000000;0.000000;0;0;0.000000;0.000000;0;0;0.000000%;0.000000;0.000000;4.000000%;0.000000%;10467;0;35;100.0%;0.0%;0.0%;0.0%;0.0%;0.0%;0.0%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;95.88%;3.25%;0.60%;0.17%;0.05%;0.02%;0.00%;0.00%;0.03%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%
>
>
> real � �0m5.288s
> user � �0m0.200s
> sys � � 0m0.150s
>
>
> --
> fdr
>

Re: fio: when not using a job file, timeouts are not applied

[Daniel Farina]

On Mon, Jul 18, 2011 at 4:11 PM, Chris Worley <worleys@gmail.com> wrote:
> On Mon, Jul 18, 2011 at 3:46 PM, Daniel Farina <daniel@heroku.com> wrote:
>> On Mon, Jul 18, 2011 at 12:35 PM, Chris Worley <worleys@gmail.com> wrote:
>>> Can we see your fio command? �I never use job files, and always use
>>> timeouts successfully.
>>
>> Absolutely:
>>
>> # Doesn't terminate for "quite a while" (I wasn't willing to wait)
>> time sudo fio --readonly --minimal --norandommap --timeout=5
>> -name=/dev/sda1 --rw=randread --size=1GB --direct=1
>
> Is that a single dash prior to "name"?

Yes, although that could be due to some sort of oopsie on my part.  It
doesn't seem to matter (seen here with a small amount of data, hence
timeout does not come into play):

$ time sudo fio --readonly --minimal --norandommap --timeout=5
-name=/dev/sda1 --rw=randread --size=256k --direct=1
--filename=/dev/sda1
fio 1.55
2;/dev/sda1;0;0;256;529;495;0;0;0.000000;0.000000;366;14021;799.359375;1701.445859;367;14023;799.984375;1701.600223;0;0;0.000000%;0.000000;0.000000;0;0;0;0;0;0.000000;0.000000;0;0;0.000000;0.000000;0;0;0.000000;0.000000;0;0;0.000000%;0.000000;0.000000;2.024291%;0.000000%;76;0;30;100.0%;0.0%;0.0%;0.0%;0.0%;0.0%;0.0%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;35.94%;57.81%;1.56%;1.56%;1.56%;0.00%;1.56%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%


real	0m0.749s
user	0m0.030s
sys	0m0.100s

$ time sudo fio --readonly --minimal --norandommap --timeout=5
--name=/dev/sda1 --rw=randread --size=256k --direct=1
--filename=/dev/sda1
fio 1.55
2;/dev/sda1;0;0;256;655;400;0;0;0.000000;0.000000;407;568;435.375000;21.859218;408;570;435.984375;21.941837;0;0;0.000000%;0.000000;0.000000;0;0;0;0;0;0.000000;0.000000;0;0;0.000000;0.000000;0;0;0.000000;0.000000;0;0;0.000000%;0.000000;0.000000;2.506266%;0.000000%;78;0;30;100.0%;0.0%;0.0%;0.0%;0.0%;0.0%;0.0%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;98.44%;1.56%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%
real	0m0.679s
user	0m0.010s
sys	0m0.150s


-- 
fdr

Re: fio: when not using a job file, timeouts are not applied

[Chris Worley]

Could you add-in "--loops=10000"?

On Mon, Jul 18, 2011 at 5:28 PM, Daniel Farina <daniel@heroku.com> wrote:
> On Mon, Jul 18, 2011 at 4:11 PM, Chris Worley <worleys@gmail.com> wrote:
>> On Mon, Jul 18, 2011 at 3:46 PM, Daniel Farina <daniel@heroku.com> wrote:
>>> On Mon, Jul 18, 2011 at 12:35 PM, Chris Worley <worleys@gmail.com> wrote:
>>>> Can we see your fio command? �I never use job files, and always use
>>>> timeouts successfully.
>>>
>>> Absolutely:
>>>
>>> # Doesn't terminate for "quite a while" (I wasn't willing to wait)
>>> time sudo fio --readonly --minimal --norandommap --timeout=5
>>> -name=/dev/sda1 --rw=randread --size=1GB --direct=1
>>
>> Is that a single dash prior to "name"?
>
> Yes, although that could be due to some sort of oopsie on my part. �It
> doesn't seem to matter (seen here with a small amount of data, hence
> timeout does not come into play):
>
> $ time sudo fio --readonly --minimal --norandommap --timeout=5
> -name=/dev/sda1 --rw=randread --size=256k --direct=1
> --filename=/dev/sda1
> fio 1.55
> 2;/dev/sda1;0;0;256;529;495;0;0;0.000000;0.000000;366;14021;799.359375;1701.445859;367;14023;799.984375;1701.600223;0;0;0.000000%;0.000000;0.000000;0;0;0;0;0;0.000000;0.000000;0;0;0.000000;0.000000;0;0;0.000000;0.000000;0;0;0.000000%;0.000000;0.000000;2.024291%;0.000000%;76;0;30;100.0%;0.0%;0.0%;0.0%;0.0%;0.0%;0.0%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;35.94%;57.81%;1.56%;1.56%;1.56%;0.00%;1.56%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%
>
>
> real � �0m0.749s
> user � �0m0.030s
> sys � � 0m0.100s
>
> $ time sudo fio --readonly --minimal --norandommap --timeout=5
> --name=/dev/sda1 --rw=randread --size=256k --direct=1
> --filename=/dev/sda1
> fio 1.55
> 2;/dev/sda1;0;0;256;655;400;0;0;0.000000;0.000000;407;568;435.375000;21.859218;408;570;435.984375;21.941837;0;0;0.000000%;0.000000;0.000000;0;0;0;0;0;0.000000;0.000000;0;0;0.000000;0.000000;0;0;0.000000;0.000000;0;0;0.000000%;0.000000;0.000000;2.506266%;0.000000%;78;0;30;100.0%;0.0%;0.0%;0.0%;0.0%;0.0%;0.0%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;98.44%;1.56%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%
> real � �0m0.679s
> user � �0m0.010s
> sys � � 0m0.150s
>
>
> --
> fdr
>

Re: fio: when not using a job file, timeouts are not applied

[Daniel Farina]

On Mon, Jul 18, 2011 at 4:49 PM, Chris Worley <worleys@gmail.com> wrote:
> Could you add-in "--loops=10000"?

No timeout occurs.

FYI, I have poked around at init.c, and feel that somehow
fill_def_thread isn't doing its job when job files are not being used:

int parse_options(int argc, char *argv[])
{
	int job_files, i;

	f_out = stdout;
	f_err = stderr;

	fio_options_fill_optstring();
	fio_options_dup_and_init(l_opts);

	if (setup_thread_area())
		return 1;
	if (fill_def_thread())
		return 1;

	job_files = parse_cmd_line(argc, argv);

	for (i = 0; i < job_files; i++) {
		if (fill_def_thread())
			return 1;
		if (parse_jobs_ini(ini_file[i], i))
			return 1;
		free(ini_file[i]);
	}

I think something is fishy here with the order of events (see
parse_cmd_line, which interprets the timeout, but fill_def_thread is
not called after that point unless there are job files). gdb seemed to
confirm that the value wasn't getting copied to the thread-local
memory.  (I also did try the non-forked based backend to make sure it
wasn't some mechanism problem there, if memory serves).

-- 
fdr

Re: fio: when not using a job file, timeouts are not applied

[Chris Worley]

Do you mean to be using --timeout or --runtime?

On Mon, Jul 18, 2011 at 6:02 PM, Daniel Farina <daniel@heroku.com> wrote:
> On Mon, Jul 18, 2011 at 4:49 PM, Chris Worley <worleys@gmail.com> wrote:
>> Could you add-in "--loops=10000"?
>
> No timeout occurs.
>
> FYI, I have poked around at init.c, and feel that somehow
> fill_def_thread isn't doing its job when job files are not being used:
>
> int parse_options(int argc, char *argv[])
> {
> � � � �int job_files, i;
>
> � � � �f_out = stdout;
> � � � �f_err = stderr;
>
> � � � �fio_options_fill_optstring();
> � � � �fio_options_dup_and_init(l_opts);
>
> � � � �if (setup_thread_area())
> � � � � � � � �return 1;
> � � � �if (fill_def_thread())
> � � � � � � � �return 1;
>
> � � � �job_files = parse_cmd_line(argc, argv);
>
> � � � �for (i = 0; i < job_files; i++) {
> � � � � � � � �if (fill_def_thread())
> � � � � � � � � � � � �return 1;
> � � � � � � � �if (parse_jobs_ini(ini_file[i], i))
> � � � � � � � � � � � �return 1;
> � � � � � � � �free(ini_file[i]);
> � � � �}
>
> I think something is fishy here with the order of events (see
> parse_cmd_line, which interprets the timeout, but fill_def_thread is
> not called after that point unless there are job files). gdb seemed to
> confirm that the value wasn't getting copied to the thread-local
> memory. �(I also did try the non-forked based backend to make sure it
> wasn't some mechanism problem there, if memory serves).
>
> --
> fdr
>

Re: fio: when not using a job file, timeouts are not applied

[Daniel Farina]

On Mon, Jul 18, 2011 at 5:19 PM, Chris Worley <worleys@gmail.com> wrote:
> Do you mean to be using --timeout or --runtime?

--timeout.  It's pretty clear the semantics using a job file and those
without a job file are not the same, or are you not able to reproduce
this?

-- 
fdr

Re: fio: when not using a job file, timeouts are not applied

[Chris Worley]

On Mon, Jul 18, 2011 at 6:19 PM, Chris Worley <worleys@gmail.com> wrote:
> Do you mean to be using --timeout or --runtime?

I should word that differently: it works fine using "runtime=5",
"timeout=5" should work, but doesn't.

Chris
>
> On Mon, Jul 18, 2011 at 6:02 PM, Daniel Farina <daniel@heroku.com> wrote:
>> On Mon, Jul 18, 2011 at 4:49 PM, Chris Worley <worleys@gmail.com> wrote:
>>> Could you add-in "--loops=10000"?
>>
>> No timeout occurs.
>>
>> FYI, I have poked around at init.c, and feel that somehow
>> fill_def_thread isn't doing its job when job files are not being used:
>>
>> int parse_options(int argc, char *argv[])
>> {
>> � � � �int job_files, i;
>>
>> � � � �f_out = stdout;
>> � � � �f_err = stderr;
>>
>> � � � �fio_options_fill_optstring();
>> � � � �fio_options_dup_and_init(l_opts);
>>
>> � � � �if (setup_thread_area())
>> � � � � � � � �return 1;
>> � � � �if (fill_def_thread())
>> � � � � � � � �return 1;
>>
>> � � � �job_files = parse_cmd_line(argc, argv);
>>
>> � � � �for (i = 0; i < job_files; i++) {
>> � � � � � � � �if (fill_def_thread())
>> � � � � � � � � � � � �return 1;
>> � � � � � � � �if (parse_jobs_ini(ini_file[i], i))
>> � � � � � � � � � � � �return 1;
>> � � � � � � � �free(ini_file[i]);
>> � � � �}
>>
>> I think something is fishy here with the order of events (see
>> parse_cmd_line, which interprets the timeout, but fill_def_thread is
>> not called after that point unless there are job files). gdb seemed to
>> confirm that the value wasn't getting copied to the thread-local
>> memory. �(I also did try the non-forked based backend to make sure it
>> wasn't some mechanism problem there, if memory serves).
>>
>> --
>> fdr
>>
>

Re: fio: when not using a job file, timeouts are not applied

[Jens Axboe]

On 2011-07-19 02:21, Daniel Farina wrote:
> On Mon, Jul 18, 2011 at 5:19 PM, Chris Worley <worleys@gmail.com> wrote:
>> Do you mean to be using --timeout or --runtime?
> 
> --timeout.  It's pretty clear the semantics using a job file and those
> without a job file are not the same, or are you not able to reproduce
> this?

Sorry for the late reply, I'm on vacation. This should fix it.

diff --git a/init.c b/init.c
index 85bd0f7..98e10f7 100644
--- a/init.c
+++ b/init.c
@@ -47,7 +47,6 @@ int warnings_fatal = 0;
 int write_bw_log = 0;
 int read_only = 0;
 
-static int def_timeout;
 static int write_lat_log;
 
 static int prev_group_jobs;
@@ -951,8 +950,6 @@ static int fill_def_thread(void)
 	 * fill default options
 	 */
 	fio_fill_default_options(&def_thread);
-
-	def_thread.o.timeout = def_timeout;
 	return 0;
 }
 
@@ -1160,7 +1157,7 @@ static int parse_cmd_line(int argc, char *argv[])
 			smalloc_pool_size = atoi(optarg);
 			break;
 		case 't':
-			def_timeout = atoi(optarg);
+			def_thread.o.timeout = atoi(optarg);
 			break;
 		case 'l':
 			write_lat_log = 1;

-- 
Jens Axboe