Igor Fedotov [Fri, 9 Feb 2018 14:39:28 +0000 (17:39 +0300)]
Add support for Ceph Rados benchmarking.
Signed-off-by: Igor Fedotov <ifedotov@suse.com>
Jens Axboe [Mon, 12 Feb 2018 17:55:07 +0000 (10:55 -0700)]
debug: fix inverted logic in fio_did_warn()
Fixes:
5ec3bc5401f3 ("init: add global 'warned' state")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Mon, 12 Feb 2018 17:19:18 +0000 (10:19 -0700)]
iolog: convert drop warning to fio_did_warn()
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Mon, 12 Feb 2018 17:18:32 +0000 (10:18 -0700)]
io_u: convert zoned bug warning to fio_did_warn()
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Mon, 12 Feb 2018 17:14:35 +0000 (10:14 -0700)]
verify: convert verify buf too small warning to fio_did_warn()
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Mon, 12 Feb 2018 17:13:04 +0000 (10:13 -0700)]
filesetup: convert root flush warning to fio_did_warn()
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Mon, 12 Feb 2018 17:11:31 +0000 (10:11 -0700)]
init: add global 'warned' state
In various places we spew a warning if some static variable
isn't set, but this still means it can happen once per job.
Add some global state to allow to track this globally instead.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Mon, 12 Feb 2018 15:24:21 +0000 (08:24 -0700)]
Merge branch 'master' of https://github.com/brycepg/fio
* 'master' of https://github.com/brycepg/fio:
Make fiologparser_hist compatible with python3
Jens Axboe [Sat, 10 Feb 2018 21:44:49 +0000 (14:44 -0700)]
init: fixup some bad style in previous commit
Make it follow the fio coding style.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Damian Yurzola [Fri, 9 Feb 2018 01:29:38 +0000 (17:29 -0800)]
init: fix broken verify_interval
The operands seems to have been inverted which in turn
created the situation whereby the interval was always
changed to match the min_bs
Fixes: https://github.com/axboe/fio/issues/522
Signed-off-by: Damian Yurzola <damian@yurzola.net>
Jens Axboe [Thu, 8 Feb 2018 22:46:46 +0000 (15:46 -0700)]
Include 'numjobs' in global options output
Fixes: https://github.com/axboe/fio/issues/237
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 7 Feb 2018 18:30:59 +0000 (11:30 -0700)]
mmap: don't include MADV_FREE in fadvise_hint check
This really should be based on invalidate, leave it as it was
for now.
Fixes:
c712c97ab871 ("Let fadvise_hint also apply too mmap engine and madvise")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 7 Feb 2018 18:13:05 +0000 (11:13 -0700)]
Let fadvise_hint also apply too mmap engine and madvise
Fixes: https://github.com/axboe/fio/issues/528
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Bryce Guinta [Tue, 6 Feb 2018 00:36:33 +0000 (17:36 -0700)]
Make fiologparser_hist compatible with python3
In python3, map functions return generators, which caused numpy to break
in this script.
Use list comprehensions instead of map (more pythonic)
They also have the added benefit of behaving the same
between python2 and python3.
Jens Axboe [Thu, 25 Jan 2018 23:04:20 +0000 (16:04 -0700)]
stat: add total fsync ios to json output
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 25 Jan 2018 17:28:23 +0000 (10:28 -0700)]
io_ddir: revert separate ddir count change
This essentially reverts the previous commit. We need more
careful changes for looping and indexing, the easiest is just
to move the count values at the end as not to skew the values
of the actual ddir entries.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 25 Jan 2018 17:16:51 +0000 (10:16 -0700)]
io_ddir: move count values out of the enum fio_ddir
They are not directions, just counts.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 25 Jan 2018 16:51:07 +0000 (09:51 -0700)]
stat: ensure that we align ts->sync_stat appropriately
clang correctly complains that we don't know if the alignment is
correct when the structure is packed, so ensure that we align it
to an 8-byte boundary.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 25 Jan 2018 16:27:14 +0000 (09:27 -0700)]
Track fsync/fdatasync/sync_file_range issue count
Also fix an issue where we didn't initialize the sync minimum
latency correctly, so it was always reported as 0.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 25 Jan 2018 03:22:50 +0000 (20:22 -0700)]
Switch last_was_sync and terminate to bool and pack better
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 24 Jan 2018 22:32:07 +0000 (15:32 -0700)]
Add suppor for logging fsync (and friends) latencies
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Tomohiro Kusumi [Mon, 15 Jan 2018 19:52:12 +0000 (21:52 +0200)]
null: drop unneeded casts from void* to non-void*
Some functions in null ioengine are used by C/C++ (e.g. null_init()
which calls malloc(3)), but C specific ones (not __cplusplus) don't
need explicit cast from void* to non-void*, and one usually doesn't.
Signed-off-by: Tomohiro Kusumi <tkusumi@tuxera.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Tomohiro Kusumi [Mon, 15 Jan 2018 19:52:11 +0000 (21:52 +0200)]
null: make *impl_ private
This can/should be private, given struct is default public in C++.
The null ioengine only needs access to new'd NullData which
encapsulates malloc'd null_data *impl_.
Signed-off-by: Tomohiro Kusumi <tkusumi@tuxera.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Tomohiro Kusumi [Mon, 15 Jan 2018 19:52:10 +0000 (21:52 +0200)]
null: fix compile time warning on OpenBSD
Fix the following warning on OpenBSD 6.2 (gcc 4.2.1).
This can probably be silenced with void** argument and memcpy() too,
but returning a non-void pointer is clearer IMO when the code also
targets C++.
Compiles with regular make/gmake as well as
# g++ -O2 -g -shared -rdynamic -fPIC -o cpp_null null.c -DFIO_EXTERNAL_ENGINE
--
CC engines/null.o
engines/null.c: In function 'fio_null_init':
engines/null.c:142: warning: dereferencing type-punned pointer will break strict-aliasing rules
Signed-off-by: Tomohiro Kusumi <tkusumi@tuxera.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 12 Jan 2018 17:59:08 +0000 (10:59 -0700)]
Merge branch 'diskless_invalidate' of https://github.com/sitsofe/fio
Jens Axboe [Fri, 12 Jan 2018 15:22:41 +0000 (08:22 -0700)]
Merge branch 'fio-issue-450' of https://github.com/gvkovai/fio
Sitsofe Wheeler [Sat, 30 Dec 2017 20:58:59 +0000 (20:58 +0000)]
filesetup: skip fallback invalidation with diskless ioengines
Since commit
0bcf41cdc22dfee6b3f3b2ba9a533b4b103c70c2 ("io_u:
re-invalidate cache when looping around without file open/close") the
job:
./fio --ioengine=null --size=4k --bs=4k --rate=4k \
--time_based --runtime=2s --name=badinvalidate
repeatedly produces the following warning while running:
fio: cache invalidation of badinvalidate.0.0 failed: Bad file descriptor
Fix this by skipping cache invalidation and logging a debug message for
diskless ioengines that don't define an explicit invalidation op.
Signed-off-by: Sitsofe Wheeler <sitsofe@yahoo.com>
Robert Elliott [Wed, 10 Jan 2018 15:50:01 +0000 (09:50 -0600)]
ioengines: don't call munmap unless size boundary is exceeded
In ioengines using mmap() and munmap(), don't call munmap() when doing an
IO that accesses the last byte of a file. This affects the dev-dax,
libpmem, and mmap ioengines.
Example:
io 12987 DEBUG fio_libpmem_prep
io 12987 io_u->offset
7516061696 : fdd->libpmem_off 0 : io_u->buflen 131072 : fdd->libpmem_sz
7516192768
io 12987 munmap addr=0x7f5600000000 size=0x1c0000000
That accesses the last 128 KiB of a 7 GiB file, triggering an munmap
because the comparison to 7 GiB fails:
7516061696 + 131072 <
7516192768
Change < to <= so that doesn't occur unless an overflow really
happens.
In one test in linux, this increases performance from 5 GB/s to 20 GB/s
as fio stops spending time in mmap() page table mapping spinlocks and
instead spends its time in memmove(). The perf top results improve from:
36.81% [kernel] [k] queued_spin_lock_slowpath
24.26% libc-2.26.so [.] __memmove_avx_unaligned_erms
4.09% [kernel] [k] __radix_tree_lookup
3.85% [kernel] [k] find_next_iomem_res
3.70% [kernel] [k] _raw_spin_lock_irq
3.48% [kernel] [k] down_read
to:
79.76% libc-2.26.so [.] __memmove_avx_unaligned_erms
3.23% libc-2.26.so [.] vfprintf
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 10 Jan 2018 15:30:07 +0000 (08:30 -0700)]
io/examples/fio-seq-read: use direct=1
It's confusing to use direct=0 with libaio, since that isn't
queued.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Sat, 6 Jan 2018 21:47:01 +0000 (14:47 -0700)]
stat: don't add duplicate clat entries for json
For some reason the initial commit always filled all
entries with zeroes, when it should just stop printing.
I suspect this was carried over from the terse/csv output,
where we do the same. But there it's actually valid, as we
have to have a fixed number of fields. For json we both
don't need it, and it's causing us to create invalid json
when we have duplicate keys.
Fixes: https://github.com/axboe/fio/issues/511
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 5 Jan 2018 20:38:40 +0000 (13:38 -0700)]
verify: don't adjust verification length based on interval when unaligned
It's perfectly fine to have a verification length that exceeds the
default verify interval when we have unaligned block sizes and
using block size ranges, since the verify interval just defaults to
the smallest blocksize.
Fixes: https://github.com/axboe/fio/issues/509
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jeff Furlong [Wed, 3 Jan 2018 23:20:38 +0000 (23:20 +0000)]
Fix client/server "all clients" reporting
In commit
e883cb3501de0f99297282c18842d50aaa7fa531, the default significant figures was added. The default value was preserved for local jobs and when client/server was used if jobs=1, but not when client/server was used if jobs > 1. This patch restores some reporting outputs ("all clients" output for IOPS and BW) when client/server is used if > 1 jobs, for example:
OLD:
All clients: (groupid=0, jobs=2): err= 0: pid=0: Wed Jan 3 14:30:13 2018
read: IOPS=0, BW=0 (0)(0B/5001msec) <--ZERO VALUES
slat (nsec): min=762, max=108579, avg=1058.97, stdev=221.30
clat (nsec): min=254, max=4544.2k, avg=11297.64, stdev=8284.14
lat (usec): min=9, max=4545, avg=12.41, stdev= 8.30
bw ( KiB/s): min=290192, max=347160, per=49.35%, avg=303613.33, stdev=11582.87, samples=18
iops : min=72548, max=86790, avg=75903.33, stdev=2895.72, samples=18
NEW:
All clients: (groupid=0, jobs=2): err= 0: pid=0: Wed Jan 3 15:08:33 2018
read: IOPS=147k, BW=575Mi (603M)(2876MiB/5001msec) <--CORRECT VALUES
slat (nsec): min=841, max=61945, avg=1304.83, stdev=232.75
clat (nsec): min=544, max=4487.2k, avg=11736.68, stdev=9123.64
lat (usec): min=10, max=4489, avg=13.10, stdev= 9.13
bw ( KiB/s): min=260945, max=305976, per=49.99%, avg=294350.22, stdev=9192.53, samples=18
iops : min=65236, max=76494, avg=73587.56, stdev=2298.13, samples=18
Regards,
Jeff
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 5 Jan 2018 15:30:32 +0000 (08:30 -0700)]
Change bluestop link to be https
Doesn't answer on http:// anymore.
Also include link to AppVeyor builds for Windows, as they always build
the latest and greatest.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Tue, 2 Jan 2018 16:05:44 +0000 (09:05 -0700)]
Merge branch 'percentiles' of https://github.com/sitsofe/fio
Sitsofe Wheeler [Sun, 31 Dec 2017 12:42:59 +0000 (12:42 +0000)]
init: disable percentiles when latency gathering is disabled
When disable_clat/disable_lat is set also disable the corresponding
clat_percentiles/lat_percentiles option.
Signed-off-by: Sitsofe Wheeler <sitsofe@yahoo.com>
Sitsofe Wheeler [Sun, 31 Dec 2017 11:41:55 +0000 (11:41 +0000)]
stat: make lat_percentiles=1 use sample count from lat_stat
The following job
./fio --name=nolatencies --ioengine=posixaio --direct=1 --rw=randread \
--iodepth=1 --bs=4k --filename=/tmp/fio.tmp --runtime=2s --size=50m \
--disable_lat=0 --disable_clat=1 --disable_slat=1 --lat_percentiles=1
outputs the following:
fio-3.3
Starting 1 process
test: (groupid=0, jobs=1): err= 0: pid=7316: Sun Dec 31 11:55:46 2017
read: IOPS=41.0k, BW=164MiB/s (172MB/s)(50.0MiB/305msec)
lat (usec): min=8, max=132, avg=17.58, stdev= 4.36
lat percentiles (nsec):
| 1.00th=[ 0], 5.00th=[ 0], 10.00th=[ 0], 20.00th=[ 0],
| 30.00th=[ 0], 40.00th=[ 0], 50.00th=[ 0], 60.00th=[ 0],
| 70.00th=[ 0], 80.00th=[ 0], 90.00th=[ 0], 95.00th=[ 0],
| 99.00th=[ 0], 99.50th=[ 0], 99.90th=[ 0], 99.95th=[ 0],
| 99.99th=[ 0]
cpu : usr=56.39%, sys=29.18%, ctx=12043, majf=0, minf=85
IO depths : 1=100.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=64=0.0%
submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
issued rwt: total=12800,0,0, short=0,0,0, dropped=0,0,0
latency : target=0, window=0, percentile=100.00%, depth=1
Run status group 0 (all jobs):
READ: bw=164MiB/s (172MB/s), 164MiB/s-164MiB/s (172MB/s-172MB/s), io=50.0MiB (52.4MB), run=305-305msec
which isn't showing total latency percentiles (lat percentiles) when it
should. Additionally there is a segfault when running the following job:
./fio --name=segfault --ioengine=posixaio --direct=1 --rw=randread \
--iodepth=1 --bs=4k --filename=/tmp/fio.tmp --runtime=2s --size=50m \
--disable_lat=1 --disable_clat=0 --disable_slat=1 --lat_percentiles=1
Fix these by making latency printing use the number of samples from the
correct latency type (clat_stat for lat_percentiles=0/clat_percentiles=1
and lat_stat for lat_percentiles=1/clat_percentiles=0).
Fixes https://www.spinics.net/lists/fio/msg06628.html ("fio lat reporting
").
Reported-by: Jeff Furlong <jeff.furlong@wdc.com>
Signed-off-by: Sitsofe Wheeler <sitsofe@yahoo.com>
Jens Axboe [Fri, 29 Dec 2017 15:45:22 +0000 (08:45 -0700)]
log: fix bad < 0 check for unsigned
Fixes:
e5f9a813 ("debug: make debug=io readable with multiple threads")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Robert Elliott [Fri, 29 Dec 2017 01:13:06 +0000 (01:13 +0000)]
debug: make debug=io readable with multiple threads
When multiple threads are active, debug=io prints are unreadable as the output
is interleaved. Multiple log_info_buf() calls are used to construct each line,
but there is no mutual exclusion covering multiple calls.
Change the dprint call tree to construct the entire line before passing it to
log_info_buf(), rather than make several calls.
Other nits:
* print the thread ID rather than the process ID
* change offset and length from decimal to hex
* separate offset, length, ddir, and file with , rather than / since the
filename on the same line likely has / of its own
* change "fill_io_u" to "fill" to match the others
* change "io complete" to "complete" to match the others
* change "->prep()=%d" to "prep: io_u %p: ret=%d" to resemble the others
* change offset/buflen in an error print to better resemble the normal prints
* add "file=" prefix for the filename
* check the calloc() return values inside the valist_to_buf functions
Old:
fill_io_u: io_u 0x7feeac010b80: off=720896/len=65536/ddir=1
io 50692io 50692//dev/dax0.0io 50692fill_io_u: io_u 0x7fee98010b80: off=196608/len=65536/ddir=1io 50692io 50692
io 50692->prep(0x7fef10010b80)=0
//dev/dax0.0//dev/dax1.0io 50692io 50692io 50692io 50692->prep(0x7feeec010b80)=0
io 50692prep: io_u 0x7feec4010b80: off=
1966080/len=65536/ddir=1io 50692io complete: io_u 0x7feedc010b80: off=393216/len=65536/ddir=1io
50692io 50692prep: io_u 0x7feef4010b80: off=720896/len=65536/ddir=1io 50692io 50692
io 50692io 50692prep: io_u 0x7feef0010b80: off=851968/len=65536/ddir=1//dev/dax0.0io 50692//dev/dax0.0
//dev/dax0.0
io 50692io 50692//dev/dax0.0//dev/dax0.0io 50692
New:
io 71400 queue: io_u 0x7fd0f0010b80: off=0x2f0000,len=0x10000,ddir=1,file=/dev/dax0.0
io 71395 fill: io_u 0x7fd0fc010b80: off=0x80000,len=0x10000,ddir=1,file=/dev/dax0.0
io 71395 prep: io_u 0x7fd0fc010b80: off=0x80000,len=0x10000,ddir=1,file=/dev/dax0.0
io 71395 prep: io_u 0x7fd0fc010b80: ret=0
io 71395 queue: io_u 0x7fd0fc010b80: off=0x80000,len=0x10000,ddir=1,file=/dev/dax0.0
io 71430 complete: io_u 0x7fd05c010b80: off=0x180000,len=0x10000,ddir=1,file=/dev/dax1.0
io 71400 complete: io_u 0x7fd0f0010b80: off=0x2f0000,len=0x10000,ddir=1,file=/dev/dax0.0
io 71430 fill: io_u 0x7fd05c010b80: off=0x190000,len=0x10000,ddir=1,file=/dev/dax1.0
io 71400 fill: io_u 0x7fd0f0010b80: off=0x300000,len=0x10000,ddir=1,file=/dev/dax0.0
io 71400 prep: io_u 0x7fd0f0010b80: off=0x300000,len=0x10000,ddir=1,file=/dev/dax0.0
io 71400 prep: io_u 0x7fd0f0010b80: ret=0
io 71400 queue: io_u 0x7fd0f0010b80: off=0x300000,len=0x10000,ddir=1,file=/dev/dax0.0
io 71430 prep: io_u 0x7fd05c010b80: off=0x190000,len=0x10000,ddir=1,file=/dev/dax1.0
io 71395 complete: io_u 0x7fd0fc010b80: off=0x80000,len=0x10000,ddir=1,file=/dev/dax0.0
io 71430 prep: io_u 0x7fd05c010b80: ret=0
io 71430 queue: io_u 0x7fd05c010b80: off=0x190000,len=0x10000,ddir=1,file=/dev/dax1.0
io 71421 complete: io_u 0x7fd090010b80: off=0x320000,len=0x10000,ddir=1,file=/dev/dax0.0
io 71419 complete: io_u 0x7fd098010b80: off=0x320000,len=0x10000,ddir=1,file=/dev/dax0.0
io 71404 complete: io_u 0x7fd0d0010b80: off=0x0,len=0x10000,ddir=1,file=/dev/dax0.0
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Tomohiro Kusumi [Fri, 29 Dec 2017 13:38:23 +0000 (15:38 +0200)]
lib/memcpy: fix warning on FreeBSD
Include struct timeval via sys/time.h.
http://pubs.opengroup.org/onlinepubs/
009604499/basedefs/sys/time.h.html
--
In file included from lib/memcpy.c:7:
lib/../fio_time.h:14:45: warning: declaration of 'struct timeval' will not be visible outside of this function [-Wvisibility]
extern uint64_t mtime_since_tv(const struct timeval *, const struct timeval *);
Signed-off-by: Tomohiro Kusumi <tkusumi@tuxera.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 28 Dec 2017 15:35:11 +0000 (08:35 -0700)]
Merge branch 'master' of https://github.com/yashi/fio
Yasushi SHOJI [Thu, 28 Dec 2017 01:53:47 +0000 (10:53 +0900)]
mutex: down_timeout: check against the base time
When the commit
8b6a404cdd2c40715885e562416c3db039912773 changed
timeval to timespec, it accsidentally picked timeout time `t' instead
of the current `base' time for checking timewait() really timed out.
t is base + msecs as follows:
base = gettimeofday()
t = base + msecs
pthread_cond_timedwait()
while (!done)
if (timedout)
if (mutex_timed_out(t, msecs))
done = true
Thus, after a good timedout, it burns the cpu for the given msecs
seconds doing tight loop in the while shown above. This causes 30 to
40 % CPU load on my i5-3360M 2.80GHz while fio server is running.
Jens Axboe [Wed, 27 Dec 2017 21:05:46 +0000 (14:05 -0700)]
Merge branch 'eta_display' of https://github.com/sitsofe/fio
Jens Axboe [Wed, 27 Dec 2017 21:05:27 +0000 (14:05 -0700)]
Merge branch 'barak/mmap_verify_only' of https://github.com/barakp/fio
Sitsofe Wheeler [Sat, 23 Dec 2017 09:55:02 +0000 (09:55 +0000)]
eta: show complete status line with max job states
Make the ETA output buffer big enough to cope with the maximum possible
jobs when their states can't be compressed (i.e. when every job's state
is different to the state of the job that came before it).
Allows the following script to display the complete status line:
rw[0]='read'; rw[1]='write'; \
for i in {1..4096}; do \
echo -e "[job$i]\nrw=${rw[$((i % 2))]}\n" \
"runtime=15"; \
done | \
./fio --eta=always --group_reporting=1 --ioengine=null --size=1g \
--time_based --bs=1k --thread --rate_iops=1 -
Fixes: https://github.com/axboe/fio/issues/500 ("Large number of threads
result in Seg faults")
Signed-off-by: Sitsofe Wheeler <sitsofe@yahoo.com>
Sitsofe Wheeler [Sat, 23 Dec 2017 08:43:54 +0000 (08:43 +0000)]
eta: fix previous line length calculation
We work out how many stale characters we haven't yet overprinted from
the previous run by comparing the number of characters that snprintf
added to the buffer in both cases. Unfortunately this doesn't account
for earlier characters that may (or may not) have already been in the
output buffer before snprintf was called...
Change the code to just use the current and previous output buffer
position to calculate the remainder of the line to be cleared.
When fio is compiled under clang with
./configure --extra-cflags="-fsanitize=address -fno-builtin" \
--disable-optimizations
a buffer overrun is demonstrated by the following script
rw[0]='read'; rw[1]='write'; \
for i in {1..4096}; do \
echo -e "[job$i]\nrw=${rw[$((i % 2))]}\n" \
"ramp_time=$(((4096 - i) / 2049 * 10))\n" \
"runtime=15"; \
done | \
./fio --group_reporting --ioengine=null --size=1g --time_based \
--bs=512 --thread --rate_iops=2 -
Signed-off-by: Sitsofe Wheeler <sitsofe@yahoo.com>
Sitsofe Wheeler [Sat, 23 Dec 2017 08:39:11 +0000 (08:39 +0000)]
eta: skip clearing of remainder of line when starting a new line
Signed-off-by: Sitsofe Wheeler <sitsofe@yahoo.com>
Sitsofe Wheeler [Sat, 23 Dec 2017 08:01:01 +0000 (08:01 +0000)]
eta: fix run_str_condensed overflow with maximum jobs
When the maximum number of jobs all have a different state to their
previous job, run_str didn't have space for the terminating null. Fix
this by making run_str one character larger.
Signed-off-by: Sitsofe Wheeler <sitsofe@yahoo.com>
Sitsofe Wheeler [Fri, 22 Dec 2017 11:10:35 +0000 (11:10 +0000)]
eta: adjust truncation case
c076de85fa4901683f6ce23ecdc071c17801ffad ("eta: fix buffer overflow in
ETA output") made adjustments to cope with snprintf truncation. However
since we're replacing the null added by snprintf, left only needs to be
one smaller than the output buffer. Make this change and add a comment
about what we're doing.
Signed-off-by: Sitsofe Wheeler <sitsofe@yahoo.com>
Jens Axboe [Thu, 21 Dec 2017 15:22:39 +0000 (08:22 -0700)]
Merge branch 'eta_overflow' of https://github.com/sitsofe/fio
Sitsofe Wheeler [Thu, 21 Dec 2017 12:23:36 +0000 (12:23 +0000)]
eta: fix buffer overflow in ETA output
When 900 or more jobs all have different states it is possible for the
space required to display the ETA line to be larger than the ETA output
buffer. fio initially truncates what it puts in the output buffer but
incorrectly advances where it thinks the end of the buffer is - when
truncating snprintf returns the number of characters that _would_ have
been converted if there had been enough space...
This patch fixes the post truncation calculation and adjusts the "left"
variable so there is room for the carriage return and a null terminator.
The following script reproduces the problem when fio has been compiled
with -fstack-protector or -fsanitize=address:
rw[0]='read'; rw[1]='write'; \
for i in {1..1000}; do \
echo -e "[job$i]\nrw=${rw[$((i % 2))]}\n" \
"ramp_time=$(((1000 - i) / 350 * 5))\n" \
"runtime=$((60 - (i % 3) * 2))"; \
done | \
./fio --eta-newline=1s --group_reporting --ioengine=null --size=1g \
--time_based --bs=512 --thread --rate_iops=123456 -
Fixes: https://github.com/axboe/fio/issues/500 ("Large number of threads
result in Seg faults")
Signed-off-by: Sitsofe Wheeler <sitsofe@yahoo.com>
Jens Axboe [Tue, 19 Dec 2017 20:16:36 +0000 (13:16 -0700)]
Fio 3.3
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Tue, 19 Dec 2017 17:17:41 +0000 (10:17 -0700)]
backend: tweaks to missed rate thinktime
shimrot says:
"May I suggest a tweak to the backend.c::handle_thinktime calculation of
"over"? That is, adjust based on the rate per byte similar to what
you've done. That is, if thinktime is greater than one op worth of
bytes, then subtract one op worth of bytes. But, if think time is
greater than one ops worth, instead add the missing the number of bytes
a full op would have exceeded the think time."
See: https://github.com/axboe/fio/issues/497#issuecomment-
352816955
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 15 Dec 2017 20:35:56 +0000 (13:35 -0700)]
ioengines: clear out ->td_ops_dlhandle if we close it
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 15 Dec 2017 20:34:22 +0000 (13:34 -0700)]
init: fix missing dlhandle reference put
If we keep using the same engine, put the dlhandle reference that
we now hold twice.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 15 Dec 2017 16:13:28 +0000 (09:13 -0700)]
parse: don't check for < 0 on an unsigned type
We do strict errors on some builds, this breaks them. Remove the
<= 0 check, just check for a value that's too large.
Fixes:
c26438ad ("parse: dump option type when using --debug=parse")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 15 Dec 2017 16:08:26 +0000 (09:08 -0700)]
ioengines: improve "is this the same IO engine" check
We can't just compare the name, that assumes that name and filename
match for an IO engine. While that's generally the case for the
engines that fio ships with, it's not a requirement, and it's
definitely not true for external engines.
Fixup the check by re-loading the engine and checking the OPS
instead. That should be bullet proof.
Fixes:
800334d ("Correctly detect whether ioengine_load can exit early")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 15 Dec 2017 15:41:03 +0000 (08:41 -0700)]
parse: dump option type when using --debug=parse
Currently we do things like:
parse 8731 __handle_option=dummy, type=10, ptr=1
for the debug parsing, and then you have to look up what that
option type is. Add names to them so we get the below instead:
parse 9170 __handle_option=dummy, type=OPT_STR_SET, ptr=1
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 14 Dec 2017 18:51:41 +0000 (11:51 -0700)]
Allow configurable ETA intervals
By default, fio prints ETA output every second. For some client/server
setups, it's desirable to allow a much longer interval, to avoid
spending too much time getting and printing ETA time.
Takes a normal time input, allowing usec/msec/sec etc postfixes.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 14 Dec 2017 17:58:03 +0000 (10:58 -0700)]
client: respect --eta=never for networked connections
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 14 Dec 2017 16:47:35 +0000 (09:47 -0700)]
server: convert more memset to on-stack initialization
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 14 Dec 2017 16:25:10 +0000 (09:25 -0700)]
server: cleanup iolog pdu prep
We don't have to memset() the whole thing, we just need to ensure
that the members we don't assign are cleared. Remove the memset
and initialize the pdu on the stack.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jeff Furlong [Thu, 14 Dec 2017 01:19:09 +0000 (01:19 +0000)]
Fix Windows local time
The fio summary output shows when the job completed, but references
system time on Windows. On Linux, local time (such as with time zone
specifics) is referenced. This patch converts Windows system time to
local time.
Before:
fio-3.2-72-ge80d-dirty
Starting 1 thread
Jobs: 1 (f=0): [f(1)][100.0%][r=69.6MiB/s,w=0KiB/s][r=17.8k,w=0 IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=1360: Wed Dec 13 23:21:16 2017 <--System Time
After:
fio-3.2-72-ge80d-dirty
Starting 1 thread
Jobs: 1 (f=0): [f(1)][100.0%][r=73.2MiB/s,w=0KiB/s][r=18.7k,w=0 IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=8088: Wed Dec 13 15:22:07 2017 <--Local Time
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Robert Elliott [Mon, 4 Dec 2017 20:53:16 +0000 (14:53 -0600)]
.gitignore: ignore tags files and additional output binaries
Ignore tags files used by emacs and ctags/vim.
Ignore the additional output binaries alongside fio, such as gfio
and all the programs in t/.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 13 Dec 2017 15:44:34 +0000 (08:44 -0700)]
server: initialize first iolog header properly
We're sending potential stack garbage over the wire, init the
whole pdu to zeroes before filling it in. This shuts up valgrind
about using uninitialized memory.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 8 Dec 2017 19:50:28 +0000 (12:50 -0700)]
fio: kill td->nr_normal_files
It's unused.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 8 Dec 2017 19:38:27 +0000 (12:38 -0700)]
Remove old emails
Just use my private email, it's the one I prefer using anyway.
Remove the old fusionio/oracle emails, they just cause
confusion and people tend to just write to all of them at once
when reporting an issue.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 8 Dec 2017 18:59:38 +0000 (11:59 -0700)]
rate: fix bad math
To figure out how much data we missed when doing a thinktime sleep,
we're currently dividing by the time slept. This is wrong, it should
be multiplied by the time slept and divided by
1000000 to go from
usec to a second base.
Additionally, don't ever subtract more than a block of data, and
adjust down depending on sleep.
Fixes:
1aa39b0c ("rate: ensure IO issue restarts right after sleep")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 7 Dec 2017 16:06:04 +0000 (09:06 -0700)]
rate: ensure IO issue restarts right after sleep
If we have rate_ignore_thinktime set, we account the number of bytes
we would have done in the period of the sleep. However, we need
to subtract one block, otherwise we end up doing a rated sleep
right after the thinktime sleep.
Fixes:
1a9bf814 ("Add option to ignore thinktime for rated IO")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 6 Dec 2017 19:30:20 +0000 (12:30 -0700)]
io_u: rate cleanup and spelling error
Fixes:
50a8ce86 ("Implement new Rate Control")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 6 Dec 2017 19:27:07 +0000 (12:27 -0700)]
Add option to ignore thinktime for rated IO
By default, fio will ignore thinktime when calculating the next
time to issue and IO, if rated IO is specified. This leads to
fio entering a catch-up type of mode after doing the specified
sleep. For some workloads, that may not be useful. If someone
asks for a specific amount of IOPS and sets a thinktime, they
may want to exclude the sleep time.
Fixes: https://github.com/axboe/fio/issues/497
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Barak Pinhas [Tue, 5 Dec 2017 12:23:22 +0000 (14:23 +0200)]
fix verify_only when using ioengine=mmap
verify_only was opening the file with O_RDONLY and the mmap
engine was trying to mmap it with PROT_READ | PROTWRITE hence
getting -EACCESS and failing the job
Jens Axboe [Sun, 3 Dec 2017 17:11:53 +0000 (10:11 -0700)]
Merge branch 'wip-cleanup' of https://github.com/ZVampirEM77/fio
Enming Zhang [Sun, 3 Dec 2017 09:50:10 +0000 (17:50 +0800)]
configure: fix typos
check for rbd_invaidate_cache() --> check for rbd_invalidate_cache()
Signed-off-by: Enming Zhang <enming.zhang@umcloud.com>
Jens Axboe [Sat, 2 Dec 2017 23:29:44 +0000 (16:29 -0700)]
steadystate: style cleanup
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Sat, 2 Dec 2017 16:18:46 +0000 (09:18 -0700)]
steadystate: add free helper
Also fix a style issue.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Vincent Fu [Fri, 1 Dec 2017 21:52:08 +0000 (16:52 -0500)]
thread_options: drop fadvise_stream from thread_options
ae8e559e got rid of the fadvise_stream option but did
not remove the corresponding member from thread_options
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 1 Dec 2017 21:49:42 +0000 (14:49 -0700)]
memcpy: add hybrid
Use builtin memcpy for < 64 bytes, use simple variant for larger
ones. libc doesn't seem to like to use SSE, the hand rolled simple
variant uses that for larger copies.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 1 Dec 2017 20:56:09 +0000 (13:56 -0700)]
memcpy: free buffer in case of failure
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 1 Dec 2017 18:30:49 +0000 (11:30 -0700)]
memcpy: use malloc
We don't have posix_memalign() everywhere, let's just use malloc.
Also free memory when done.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 1 Dec 2017 18:21:58 +0000 (11:21 -0700)]
fio_time: should include time.h
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 1 Dec 2017 18:13:23 +0000 (11:13 -0700)]
Add basic memcpy test
Research is ongoing in how to improve (and make deterministic) the
memcpy speed, since it's especially applicable to the persistent
memory engines.
Not documented yet.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 1 Dec 2017 04:48:12 +0000 (21:48 -0700)]
io_u: don't account io issue blocks for verify backlog
We don't account the bytes, we should not account the blocks
either.
Fixes:
ae2fafc8 ("verify: verify bytes should not add to this_io_bytes")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 1 Dec 2017 04:29:32 +0000 (21:29 -0700)]
options: warn if we exceed the supported number of split entries
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 1 Dec 2017 04:28:15 +0000 (21:28 -0700)]
Documentation: add note about how many bssplit and zones fio supports
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 1 Dec 2017 03:24:54 +0000 (20:24 -0700)]
ioengine: don't account verify bytes
If we do that, we can terminate early for a readwrite+verify
backlog workload.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 1 Dec 2017 02:50:34 +0000 (19:50 -0700)]
client: fix use-after-free for client timeout
We free the reply, but if it times out, we dereference it afterwards.
Store the opcode before calling into handle_cmd_timeout().
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 1 Dec 2017 02:48:25 +0000 (19:48 -0700)]
engines/dev-dax: fix leak of 'sfile' in error case
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 1 Dec 2017 02:47:42 +0000 (19:47 -0700)]
t/verify-state: fix leak in error case
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 1 Dec 2017 02:45:05 +0000 (19:45 -0700)]
backend: make it clear that we passed 'fd' to the new thread
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 1 Dec 2017 02:41:22 +0000 (19:41 -0700)]
gettime-thread: fix failure to check setaffinity return value
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 1 Dec 2017 02:17:38 +0000 (19:17 -0700)]
Documentation cleanup
- Space at end-of-line for the significant_figures change
- > 80 char line section in HOWTO
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 30 Nov 2017 23:49:23 +0000 (16:49 -0700)]
options: don't overrun bssplit array
Since we now have a different number of splits we support for
zones and block size splits, make sure we pass in the right
max.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 30 Nov 2017 23:47:27 +0000 (16:47 -0700)]
Bump support of zones to 256 max
This moves it from 64 to 256.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 30 Nov 2017 23:43:18 +0000 (16:43 -0700)]
steadystate: make flags conform to usual fio standard
__VAL are usually bit shift values, with VAL being the proper
mask. For some reason, the steadystate code uses both as
masks, fix that up.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 30 Nov 2017 20:28:48 +0000 (13:28 -0700)]
server: process connection list before executing trigger
This shrinks the window a bit for allowing the backend to notice
that jobs have gone away, and notify the client.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 30 Nov 2017 20:27:40 +0000 (13:27 -0700)]
client: ignore a client timeout, if the last thing we saw as a trigger
If triggers are used and are fatal to the fio backend (killing power,
etc), then we don't have time to process the stop/quit commands. We
expect the client to go away, but the backend can't let us know.
If the last thing we saw was a trigger command, don't exit in error
if the client times out and goes away.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 30 Nov 2017 17:59:25 +0000 (10:59 -0700)]
fio: add check rate flag
One of the things we do a lot for each IO, is check if we should be
checking the rate. Add a specific flag for that case, so we can answer
that question very cheaply.
This is good for more than a 5% speedup for a null engine test case.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 30 Nov 2017 16:18:37 +0000 (09:18 -0700)]
io_u: speed up small_content_scramble()
This is a hot path for write workloads, since we don't want to send the
same buffers to the device again and again. The idea is to defeat basic
dedupe/compression, but slightly modifying the buffer for each write.
small_content_scramble() does this by filling in the io_u offset into a
random spot in each 512b chunk of an io buffer, and filling in the start
time (sec,nsec) at the end of each 512b chunk.
With this change, we still do those two things, but we generate a random
cacheline within each 512b chunk, and fill the offset at the beginning
of the cacheline, and the time at the end of it. This means that
instead of potentially dirtying 2 cachelines for each 512b chunk in an
IO buffer, we dirty just 1.
The results should still be random enough that small_content_scramble()
fullfils the promise to defeat basic dedupe and compression, but it is
lighter to run.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 30 Nov 2017 14:03:33 +0000 (07:03 -0700)]
io_u: cleanup check_get_trim()
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 30 Nov 2017 13:50:36 +0000 (06:50 -0700)]
io_u: tweak small content buffer scramble
We currently generate a 'random' offset in a 512b chunk to fill in
the offset. Since we don't want the later time scramble to
overwrite it, we check and adjust for that. Instead just ensure that
we generate a random offset in the first half of the 512b chunk,
then we know we never overlap.
Signed-off-by: Jens Axboe <axboe@kernel.dk>