fio.git
3 years agoFio 2.6 fio-2.6
Jens Axboe [Wed, 27 Jan 2016 02:02:07 +0000 (19:02 -0700)]
Fio 2.6

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoserver: split vectored commands into fragments
Jens Axboe [Tue, 26 Jan 2016 23:44:24 +0000 (16:44 -0700)]
server: split vectored commands into fragments

This fixes the fio_send_cmd_ext_pdu() path to split commands
into properly sized fragments. This also reverts the
previous commit, that bypassed the check since this path
was broken.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoserver: disable pdu length check for now
Jens Axboe [Tue, 26 Jan 2016 18:19:16 +0000 (11:19 -0700)]
server: disable pdu length check for now

Vectored entries don't always honor it. Allow longer payloads, until
we fix the breakdown again.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoclient: remove debug printf() for compressed iolog
Jens Axboe [Tue, 26 Jan 2016 18:02:16 +0000 (11:02 -0700)]
client: remove debug printf() for compressed iolog

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agopowerpc: fix cpus_online() to get correct max CPU number for powerpc64
Dan Qing Zhang [Mon, 25 Jan 2016 13:04:18 +0000 (13:04 +0000)]
powerpc: fix cpus_online() to get correct max CPU number for powerpc64

This patch tries to fix cpus_online() for powerpc64 when running fio
by using parameter "cpus_allowed", in order to obtain correct max CPU
number on powerpc64 architecture.

When changing SMT settings on powerpc architecture, the way of cpu
numbering is changing as well. For example, when SMT=off, CPUs are
numbered like 0,8,16,..., while when SMT=2, CPUs are numbered like
0,1,8,9,.... Therefore, sysconf(_SC_NPROCESSORS_ONLN) is not applied
to get exact max CPU number on powerpc. And the max_cpu value can be
correctly ontained by using sysconf(_SC_NPROCESSORS_CONF).

Signed-off-by: Dan Qing Zhang <dqzhang@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoFix Windows build: define rand_r()
Bruce Cran [Fri, 22 Jan 2016 17:54:47 +0000 (10:54 -0700)]
Fix Windows build: define rand_r()

3 years agoMakefile: fix duplicate strcasestr for windows
Jens Axboe [Fri, 22 Jan 2016 22:10:31 +0000 (15:10 -0700)]
Makefile: fix duplicate strcasestr for windows

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoserver: use the wrapped MSG_DONTWAIT
Jens Axboe [Fri, 22 Jan 2016 18:49:57 +0000 (11:49 -0700)]
server: use the wrapped MSG_DONTWAIT

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoioengine: remove duplicate queue handler check
David Disseldorp [Thu, 21 Jan 2016 16:45:44 +0000 (17:45 +0100)]
ioengine: remove duplicate queue handler check

Signed-off-by: David Disseldorp <ddiss@suse.de>
3 years agoFio 2.5 fio-2.5
Jens Axboe [Thu, 21 Jan 2016 05:00:17 +0000 (22:00 -0700)]
Fio 2.5

2.4 was short lived, 2.5 fixes an important regression that
happened between fio-2.2.12 and fio-2.2.13.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agomutex: fix double unlock in fio_mutex_down_timeout()
Jens Axboe [Thu, 21 Jan 2016 04:52:47 +0000 (21:52 -0700)]
mutex: fix double unlock in fio_mutex_down_timeout()

In case of error, we unlock twice.

Fixes: 09400a60042e ("mutex: fixup fio_mutex_down_timeout()")
Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoinit: fix compile for FIO_INC_DEBUG not being set
Jens Axboe [Thu, 21 Jan 2016 04:17:55 +0000 (21:17 -0700)]
init: fix compile for FIO_INC_DEBUG not being set

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoFio 2.4 fio-2.4
Jens Axboe [Thu, 21 Jan 2016 04:13:08 +0000 (21:13 -0700)]
Fio 2.4

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoinit: increment stat count if parent == default thread
Jens Axboe [Thu, 21 Jan 2016 00:45:13 +0000 (17:45 -0700)]
init: increment stat count if parent == default thread

We skip doing the stat increment if group_reporting is set,
but that means we can end up with stat_number == 0. That
isn't correct, increment for group_reporting as well, but
only if the parent is the default thread. That means we
still won't do it for sub-jobs.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoclient: fix non-text command leak
Jens Axboe [Wed, 20 Jan 2016 19:17:54 +0000 (12:17 -0700)]
client: fix non-text command leak

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoclient: drain text output when exiting
Jens Axboe [Wed, 20 Jan 2016 18:22:22 +0000 (11:22 -0700)]
client: drain text output when exiting

If we don't, we can lose error messages from the server side. One
example would be for protocol version mismatch - the server will
notify the client, but the client doesn't check for these messages
before exiting.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoFix issue with termination before io_size has been reached
Jens Axboe [Thu, 14 Jan 2016 21:44:23 +0000 (14:44 -0700)]
Fix issue with termination before io_size has been reached

If the size of the device isn't a multiple of the block size,
then we can exit prematurely for cases where an io_limit has
been set higher than the device size. This is because the file(s)
get marked completely done if bs is a multiple of the device size,
but doesn't for the opposite case.

We should not factor this in for deciding to keep running or not.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoio_u: ensure that we align new start offset properly for time_based
Jens Axboe [Thu, 14 Jan 2016 17:31:38 +0000 (10:31 -0700)]
io_u: ensure that we align new start offset properly for time_based

If io_size isn't a multiple of the block size, weird things can
happen, which usually means that fio will exit with an unaligned
IO error:

fio: io_u error on file /dev/nvme0n1: Invalid argument: write offset=125830, buflen=1048576

where offset=125830 really should have been aligned to the block
size. Ensure that we do that.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agot/verify-state: one more printf type fix
Jens Axboe [Wed, 13 Jan 2016 20:17:39 +0000 (13:17 -0700)]
t/verify-state: one more printf type fix

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agot/verify-state: fix type for printf
Jens Axboe [Wed, 13 Jan 2016 20:17:00 +0000 (13:17 -0700)]
t/verify-state: fix type for printf

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoDestroy sk_out server key when we are done
Jens Axboe [Wed, 13 Jan 2016 16:40:51 +0000 (09:40 -0700)]
Destroy sk_out server key when we are done

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoAlways create server sk_out key
Jens Axboe [Wed, 13 Jan 2016 16:34:46 +0000 (09:34 -0700)]
Always create server sk_out key

We use it from helper threads, so let's just unconditionally
set it up. This fixes a problem on fio exit with libraries or
engines that also create pthread keys.

Reported-by: Jens Rosenboom <j.rosenboom@x-ion.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agooptions: fix bs_cmp()
Jens Axboe [Tue, 12 Jan 2016 17:18:50 +0000 (10:18 -0700)]
options: fix bs_cmp()

Return the int difference between the two, not a bool.

Reported-by: Yury Gribov <>
Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agouse option smalloc_pool_size
Andreas Piesk [Fri, 8 Jan 2016 13:29:12 +0000 (14:29 +0100)]
use option smalloc_pool_size

3 years agot/verify-state: add helper to inspect verify dump state files
Jens Axboe [Thu, 7 Jan 2016 22:06:22 +0000 (15:06 -0700)]
t/verify-state: add helper to inspect verify dump state files

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoMerge branch 'master' of git://github.com/ezrapedersen/fio
Jens Axboe [Thu, 7 Jan 2016 22:01:09 +0000 (15:01 -0700)]
Merge branch 'master' of git://github.com/ezrapedersen/fio

3 years agoverify-state: forward declare io_u
Jens Axboe [Thu, 7 Jan 2016 21:26:57 +0000 (14:26 -0700)]
verify-state: forward declare io_u

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoDisable libaio for ESXi build - bug#80
root [Thu, 7 Jan 2016 21:25:10 +0000 (21:25 +0000)]
Disable libaio for ESXi build - bug#80

3 years agoverify: split out state header code
Jens Axboe [Thu, 7 Jan 2016 21:22:09 +0000 (14:22 -0700)]
verify: split out state header code

We'll need to use it in a separate helper utility, and we don't
need all the verify code for this.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoserver: sk_out exit error handling
Jens Axboe [Thu, 7 Jan 2016 18:19:28 +0000 (11:19 -0700)]
server: sk_out exit error handling

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoconfigure: lex force off
Jens Axboe [Thu, 7 Jan 2016 18:18:22 +0000 (11:18 -0700)]
configure: lex force off

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoserver: include name of verify file requested for error message
Jens Axboe [Thu, 7 Jan 2016 18:07:51 +0000 (11:07 -0700)]
server: include name of verify file requested for error message

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agooptions: move pattern_fmt_desc where we need it
Jens Axboe [Wed, 30 Dec 2015 16:27:05 +0000 (09:27 -0700)]
options: move pattern_fmt_desc where we need it

No need for that to be global.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agooptions: we don't need the crc32c verify include
Jens Axboe [Wed, 30 Dec 2015 16:23:13 +0000 (09:23 -0700)]
options: we don't need the crc32c verify include

We used to test for the availability of the hw accellerated code,
but that happens elsewhere now.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agooptions: split out option grouping code
Jens Axboe [Wed, 30 Dec 2015 16:03:55 +0000 (09:03 -0700)]
options: split out option grouping code

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agooptions: make the groups/categories constant
Jens Axboe [Wed, 30 Dec 2015 15:55:19 +0000 (08:55 -0700)]
options: make the groups/categories constant

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoconfigure: add --enable-lex
Jens Axboe [Wed, 30 Dec 2015 03:58:13 +0000 (20:58 -0700)]
configure: add --enable-lex

For AIX, default to lex being off. It can be enabled with
--enable-lex. For other platforms, rely on the regular
configure check to check enable/disable.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoconfigure: check if lex requires/fails with -o for output
Jens Axboe [Tue, 29 Dec 2015 19:23:17 +0000 (12:23 -0700)]
configure: check if lex requires/fails with -o for output

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agot/dedupe: silence bogus warning on 'bytes' being used uninitialized
Jens Axboe [Tue, 29 Dec 2015 15:30:33 +0000 (08:30 -0700)]
t/dedupe: silence bogus warning on 'bytes' being used uninitialized

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agot/genzipf: fix for strcasestr missing
Jens Axboe [Tue, 29 Dec 2015 00:39:22 +0000 (17:39 -0700)]
t/genzipf: fix for strcasestr missing

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoFix included getopt locations
Jens Axboe [Tue, 29 Dec 2015 00:14:46 +0000 (17:14 -0700)]
Fix included getopt locations

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agolib/pattern: include oslib/strcasestr.h
Jens Axboe [Mon, 28 Dec 2015 23:45:49 +0000 (16:45 -0700)]
lib/pattern: include oslib/strcasestr.h

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoclient: rename send_file() to fio_send_file()
Jens Axboe [Mon, 28 Dec 2015 23:16:06 +0000 (16:16 -0700)]
client: rename send_file() to fio_send_file()

AIX apparently has a symbol clash.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoFio 2.3 fio-2.3
Jens Axboe [Wed, 23 Dec 2015 16:39:04 +0000 (09:39 -0700)]
Fio 2.3

Let's switch to a simpler numbering scheme. 2.3 follows 2.2.13, so
it should have been 2.2.14. But from now on, let's just use a simpler
MAJOR.MINOR release numbering scheme.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agogfio: use ffz64 for 64-bit types
Jens Axboe [Wed, 23 Dec 2015 16:10:05 +0000 (09:10 -0700)]
gfio: use ffz64 for 64-bit types

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoarch: add ffs64 and ffz64
Jens Axboe [Wed, 23 Dec 2015 16:09:23 +0000 (09:09 -0700)]
arch: add ffs64 and ffz64

These just use the normal C software version, extended to 64-bit
even on 32-bit archs. The fast variant is still machine word native.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoFix runtime of terse output
shoichi.sawada@toshiba.co.jp [Tue, 22 Dec 2015 08:16:24 +0000 (08:16 +0000)]
Fix runtime of terse output

Hi,

When I used fio with option of '--status-interval=1 --minimal' and 'random_distribution=zipf:1.2',
'runtime' of 2nd line was 18446744073709551345.
This value is -271 in int_64.

This is because 'runtime' was subtracted in function '__show_running_run_stats'.
If 'td->io_bytes' is 0 when this function was called,
and 'td->io_bytes' is not 0 when '__show_run_stats' returned,
'runtime' is subtracted by 'rt' even though 'runtime' have not added.

I fixed the problem by applying this patch to fio-2.2.13.

Regards
Shoichi

----

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agobackend: continue on failed wait_for() check, not break
Jens Axboe [Tue, 22 Dec 2015 20:40:14 +0000 (13:40 -0700)]
backend: continue on failed wait_for() check, not break

If we have this job file:

[a]

[b]
wait_for=a

[c]
wait_for=a

[d]

then we expect a/d to run together, then b/c after a is done. However,
right now we break out of the job loop, so it acts like a stonewall.
That means that a will run first, then b/c/d. That isn't correct.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agogfio: update for 64-bit option category/group
Jens Axboe [Tue, 22 Dec 2015 18:11:14 +0000 (11:11 -0700)]
gfio: update for 64-bit option category/group

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoAdd support for waiting for other jobs by name
Andrey Kuzmin [Tue, 22 Dec 2015 16:21:25 +0000 (09:21 -0700)]
Add support for waiting for other jobs by name

We have stonewall to barrier against previous jobs before a new
job runs. This adds support for 'wait_for', which allows the
user to explicitly wait for a given job before starting.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoparse: ensure that option group/category hold 64-bit
Jens Axboe [Tue, 22 Dec 2015 16:04:27 +0000 (09:04 -0700)]
parse: ensure that option group/category hold 64-bit

options.c:3459:12: warning: large integer implicitly truncated to unsigned type [-Woverflow]
   .group = FIO_OPT_G_INVALID,

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agorewrote libhdfs engine
Fabrice Bacchella [Tue, 22 Dec 2015 11:26:51 +0000 (12:26 +0100)]
rewrote libhdfs engine

3 years agoMake options mask a 64-bit type
Jens Axboe [Mon, 21 Dec 2015 23:03:01 +0000 (16:03 -0700)]
Make options mask a 64-bit type

We're running out of bit numbers, double it up.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agobackend: only terminate multiple loops if we did no IO
Jens Axboe [Mon, 21 Dec 2015 17:01:40 +0000 (10:01 -0700)]
backend: only terminate multiple loops if we did no IO

Commit 095196b1e76b introduced a stop point if do_io() returned
nothing, but that only counts writes and trims. Change that to
count all types of IO.

Fixes: 095196b1e76b ("backend: terminate loop if we didn't do IO")
Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agobackend: count iterative bytes for progress
Jens Axboe [Sat, 19 Dec 2015 05:33:35 +0000 (22:33 -0700)]
backend: count iterative bytes for progress

Commit ac28d9053679 changed it so we don't clear ->bytes_done[]
necessarily for each loop. However, this introduced a busy loop
where we this we did IO and continue to loop, but we really
didn't.

Fixes: ac28d9053679 ("Fixup bw/iops logging for short runs")
Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoserver: command tag cleanup
Jens Axboe [Fri, 18 Dec 2015 17:36:57 +0000 (10:36 -0700)]
server: command tag cleanup

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoserver: don't store command tag on the stack
Jens Axboe [Fri, 18 Dec 2015 16:17:41 +0000 (09:17 -0700)]
server: don't store command tag on the stack

It'll go out-of-scope, making the server unhappy.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoclient: iolog cleanups
Jens Axboe [Fri, 18 Dec 2015 15:57:24 +0000 (08:57 -0700)]
client: iolog cleanups

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoserver: code cleanups
Jens Axboe [Thu, 17 Dec 2015 22:23:48 +0000 (15:23 -0700)]
server: code cleanups

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoserver: remove leftover debug statement
Jens Axboe [Thu, 17 Dec 2015 21:59:49 +0000 (14:59 -0700)]
server: remove leftover debug statement

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoconfigure: fix zlib typo
Jens Axboe [Thu, 17 Dec 2015 21:55:16 +0000 (14:55 -0700)]
configure: fix zlib typo

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoclient/server: transparent handling of storing compressed logs
Jens Axboe [Thu, 17 Dec 2015 21:54:15 +0000 (14:54 -0700)]
client/server: transparent handling of storing compressed logs

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoFix compile warning for !zlib
Jens Axboe [Thu, 17 Dec 2015 21:53:45 +0000 (14:53 -0700)]
Fix compile warning for !zlib

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoclient/server: various bug fixes
Jens Axboe [Thu, 17 Dec 2015 17:24:03 +0000 (10:24 -0700)]
client/server: various bug fixes

- Better handling of vectored commands
- Improve sk_out mutex handling (don't alloc separately)
- Add support for sync sending of network data
- Prep for network xmit of logs

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoEnsure that command line options also end up in json output
Jens Axboe [Thu, 17 Dec 2015 15:32:15 +0000 (08:32 -0700)]
Ensure that command line options also end up in json output

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agogclient: don't free pdu on iolog return
Jens Axboe [Wed, 16 Dec 2015 22:05:54 +0000 (15:05 -0700)]
gclient: don't free pdu on iolog return

The command is freed separately, don't free it twice.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoclient: fix double free of command on timeout
Jens Axboe [Wed, 16 Dec 2015 22:05:22 +0000 (15:05 -0700)]
client: fix double free of command on timeout

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoverify: dump failed header, if verification fails
Jens Axboe [Wed, 16 Dec 2015 15:59:38 +0000 (08:59 -0700)]
verify: dump failed header, if verification fails

And verify_dump=1 is enabled.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoparse: fix segfault if posval option doesn't have a value
Jens Axboe [Wed, 16 Dec 2015 15:32:30 +0000 (08:32 -0700)]
parse: fix segfault if posval option doesn't have a value

If the user did

continue_on_error

in the job file without setting one of the possible values
for that option, we'd crash in string compare. Now fio will
exit in error, showing the valid options instead.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoclient/server: ensure we don't overrun memory for long option values
Jens Axboe [Wed, 16 Dec 2015 04:02:51 +0000 (21:02 -0700)]
client/server: ensure we don't overrun memory for long option values

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoclient: fix segfault for !json output
Jens Axboe [Wed, 16 Dec 2015 03:47:14 +0000 (20:47 -0700)]
client: fix segfault for !json output

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoclient/server: pass back job options to client
Jens Axboe [Wed, 16 Dec 2015 03:33:29 +0000 (20:33 -0700)]
client/server: pass back job options to client

Also fill in a few more missing pieces from the server based
json output, like time stamp.

Now the json output should match the local, on all accounts.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoclient: cleanup json output
Jens Axboe [Tue, 15 Dec 2015 20:48:32 +0000 (13:48 -0700)]
client: cleanup json output

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agostat: work around 'numjobs' oddity in dumping job output
Jens Axboe [Tue, 15 Dec 2015 17:38:05 +0000 (10:38 -0700)]
stat: work around 'numjobs' oddity in dumping job output

Inherit options between duped jobs, and don't include 'numjobs' in the
global option output.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoFirst stab at adding job options to json output
Jens Axboe [Tue, 15 Dec 2015 17:30:57 +0000 (10:30 -0700)]
First stab at adding job options to json output

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoRework 'dump_cmdline'
Jens Axboe [Tue, 15 Dec 2015 16:53:22 +0000 (09:53 -0700)]
Rework 'dump_cmdline'

Store the options in &td->opt_list. Should be no change for
dump_cmdline, but it will prepare us for being able to jump
the job definition as part of the json output.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoMerge branch 'server'
Jens Axboe [Tue, 15 Dec 2015 15:42:13 +0000 (08:42 -0700)]
Merge branch 'server'

3 years agoserver: bump listen() backlog
Jens Axboe [Tue, 15 Dec 2015 02:26:18 +0000 (19:26 -0700)]
server: bump listen() backlog

With the default if 0, we trigger a lot of these:

[29055.817884] TCP: request_sock_TCP: Possible SYN flooding on port 8765. Sending cookies.  Check SNMP counters.

Set it to 4, even 2 seems to be enough to get rid of it.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoclient: make SEND_ETA timeout non-fatal
Jens Axboe [Tue, 15 Dec 2015 01:51:22 +0000 (18:51 -0700)]
client: make SEND_ETA timeout non-fatal

If we fail a single ETA, then just soldier on. If we fail 5 in a row,
then give up permanently.

Also bring the command timeout back to 5s per command.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoFio 2.2.13 fio-2.2.13
Jens Axboe [Mon, 14 Dec 2015 21:53:35 +0000 (14:53 -0700)]
Fio 2.2.13

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoserver: create sk_out_key earlier
Jens Axboe [Mon, 14 Dec 2015 21:46:32 +0000 (14:46 -0700)]
server: create sk_out_key earlier

The previous code worked on Linux, since we seem to get '0' as
the key always. It segfaults on other platforms, since we
call pthread_getspecific() before the key has been created.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoserver: don't pass sk_out into accept loop
Jens Axboe [Mon, 14 Dec 2015 21:26:07 +0000 (14:26 -0700)]
server: don't pass sk_out into accept loop

Previous commit killed the use of it, just define it locally
in accept_loop().

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoserver: cleanup exported functions
Jens Axboe [Mon, 14 Dec 2015 20:31:34 +0000 (13:31 -0700)]
server: cleanup exported functions

Lots of cruft left from previously.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoserver: comments, and clear sk_out->sk when we close the 'sk'
Jens Axboe [Mon, 14 Dec 2015 17:55:24 +0000 (10:55 -0700)]
server: comments, and clear sk_out->sk when we close the 'sk'

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoserver: ensure each connection sets up its own sk_out
Jens Axboe [Mon, 14 Dec 2015 17:48:29 +0000 (10:48 -0700)]
server: ensure each connection sets up its own sk_out

Otherwise multiple connections will screw each other up. This
makes multiple connections to a single backend work again.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoserver: comment sk_entry struct and remove hole
Jens Axboe [Fri, 11 Dec 2015 22:13:41 +0000 (15:13 -0700)]
server: comment sk_entry struct and remove hole

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoFixup and improve per-thread data
Jens Axboe [Fri, 11 Dec 2015 20:25:24 +0000 (13:25 -0700)]
Fixup and improve per-thread data

- Get rid of backend_data, just use sk_out
- Add reference counting to sk_out and proper assign/drop helpers
- Ensure we pass it to other threads created
- Fix leaks of sk_out

This should get us closer to the new model actually working as
intended.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoserver: cleanup and proper error returns
Jens Axboe [Fri, 11 Dec 2015 00:46:05 +0000 (17:46 -0700)]
server: cleanup and proper error returns

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoserver: make the io log transmit use the new infrastructure
Jens Axboe [Thu, 10 Dec 2015 23:07:32 +0000 (16:07 -0700)]
server: make the io log transmit use the new infrastructure

Add support for vectored commands, and pass them on the backend
pipe as well.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoserver: rewrite message handling
Jens Axboe [Thu, 10 Dec 2015 22:32:15 +0000 (15:32 -0700)]
server: rewrite message handling

Currently we don't do any synchronization when we send out
network messages, this can cause a variety of issues. Instead
of adding serialization for transmits, let the messages queue
up in the network backend, and let the backend send them out.

Work in progress, not everything has been tested with this.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agolog: remove log_local_buf()
Jens Axboe [Thu, 10 Dec 2015 22:31:02 +0000 (15:31 -0700)]
log: remove log_local_buf()

Have the server return -1 if it can't log, then we can handle
it internally.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoAdd backend related data
Jens Axboe [Thu, 10 Dec 2015 21:26:44 +0000 (14:26 -0700)]
Add backend related data

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agomutex: fixup fio_mutex_down_timeout()
Jens Axboe [Thu, 10 Dec 2015 21:22:43 +0000 (14:22 -0700)]
mutex: fixup fio_mutex_down_timeout()

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoAdd exitall_on_error option
Jens Axboe [Thu, 10 Dec 2015 18:15:52 +0000 (11:15 -0700)]
Add exitall_on_error option

Like exitall, but only if a job exits due to erroring out.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoeta: use bool for forcing jobs_eta return
Jens Axboe [Wed, 9 Dec 2015 19:44:09 +0000 (12:44 -0700)]
eta: use bool for forcing jobs_eta return

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoserver: ensure we send the right sized jobs_eta when faking it
Jens Axboe [Wed, 9 Dec 2015 19:40:52 +0000 (12:40 -0700)]
server: ensure we send the right sized jobs_eta when faking it

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoserver: always end back ETA reply
Jens Axboe [Wed, 9 Dec 2015 19:37:05 +0000 (12:37 -0700)]
server: always end back ETA reply

Otherwise the client will time out waiting for a response. If we don't
have a local jobs_eta, then just send a zero filled one.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agothread_options: kill converted cpu masks
Jens Axboe [Wed, 9 Dec 2015 18:23:15 +0000 (11:23 -0700)]
thread_options: kill converted cpu masks

We can't currently convert them in cconv.c, so let's just disable
them as they eat a lot of space in the thread options.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoiolog: don't copy log samples for compression
Jens Axboe [Wed, 9 Dec 2015 16:16:13 +0000 (09:16 -0700)]
iolog: don't copy log samples for compression

For larger settings of log_compression, the allocation and copying of
the log will adversely affect performance and logged latencies. Instead
just reference the log in the compression code, while resetting the
current log to a smaller size. We pay a slightly higher cost in memory
until compression has completd, but it saves us from allocating and
copying tens (or hundreds) of megabytes inline.

Signed-off-by: Jens Axboe <axboe@fb.com>
3 years agoeta: fix comparison reversal for time based jobs
Jens Axboe [Wed, 9 Dec 2015 16:08:10 +0000 (09:08 -0700)]
eta: fix comparison reversal for time based jobs

If time_based=1 and runtime=x, then we know the job will take x
time to run. Fix a bug introduced in a previous commit.

Fixes: ("Fix ETA calculations")
Signed-off-by: Jens Axboe <axboe@fb.com>