[PATCHSET v2] block: Separating discards from writes in Linux IO statistics
From: Tejun Heo
Date: Tue Jun 05 2018 - 14:01:32 EST
This patchset was originally posted by Michael Callahan.
https://marc.info/?l=linux-block&m=146541910129172&w=2
The patchset is refreshed on top of the current git master
v4.17-1306-g716a685 and a patch was added to add discard stats for
cgroup io.stat. The original patchset description from Michael
follows.
This patch set separates block layer statistics for discards from
writes. Discards are currently bundled with writes in the various
/sys/block/*/stat files as well as in /proc/diskstats. However
discards are nearly always used to mark storage that is no longer in
use. There are many reasons having discard not counted with writes is
useful:
1) For many non volatile memory devices it is just nice to know
that discards are enabled and working properly.
2) Discards have different performance characteristics than
writes. They are generally much faster and larger and bundling
them makes performance statistics less meaningful.
3) Discards are not writes in terms of tracking device lifetime.
If a device supports six device writes per day it is nice to know
how many writes have actually been written to the device as
discards do not count against that total.
Separation of discard statistics is accomplished by expanding the
struct diskstat arrays to 3 entries for STAT_READ, STAT_WRITE,
and STAT_DISCARD. A new rw_stat_group function is then used to
convert from rw_flags (cmd_flags from requests, bi_rw from bios)
into the appropriate stat group which is then tracked as before.
Lastly the new statistics are appended to the current
/sys/bloc/*/stat and /proc/diskstats on output such that they are
the last four entries of each. These are analogous to the four
read and write statistics.
* Number of discard ios completed
* Number of discard ios merged
* Number of discard sectors completed
* Milliseconds spent on discard requests.
[before ~]# cat sys/block/nvme0/stat
296550701 0 2372405688 67317193 19672752 0 7972237312
9375167 0 2787238 79718726
[after ~]# cat sys/block/nvme0/stat
296550701 0 2372405688 67317193 18034352 0 4616794112
9125902 0 2787238 79718726 1638400 0 3355443200
249265
Note that the discards have moved out of the write fields to the
end and that the write fields are now smaller by the difference.
Adding the new statistics to the end of /sys/block/*/stat and
/proc/diskstats is backwards compatible with both iostat and
vmstat which pick up just the old fields:
[root@after ~]# iostat -x
Linux 4.5.0_68319_ge5065f4-dirty (##hostname###) 05/17/2016
_x86_64_ (48 CPU)
avg-cpu: %user %nice %system %iowait %steal %idle
0.41 0.00 0.23 0.01 0.00 99.35
Device: rrqm/s wrqm/s r/s w/s rsec/s wsec/s
avgrq-sz avgqu-sz await svctm %util
sda 0.01 7.50 0.20 3.61 16.65 708.41
190.03 0.09 22.53 1.16 0.44
nvme0n1 0.00 0.00 587.03 35.70 4696.25 9139.08
22.22 0.16 0.24 0.01 0.55
[root@after ~]# vmstat -d
disk- ------------reads------------ ------------writes----------- -----IO------
total merged sectors ms total merged sectors ms cur sec
ram0 0 0 0 0 0 0 0 0 0 0
ram1 0 0 0 0 0 0 0 0 0 0
ram2 0 0 0 0 0 0 0 0 0 0
ram3 0 0 0 0 0 0 0 0 0 0
ram4 0 0 0 0 0 0 0 0 0 0
ram5 0 0 0 0 0 0 0 0 0 0
ram6 0 0 0 0 0 0 0 0 0 0
ram7 0 0 0 0 0 0 0 0 0 0
ram8 0 0 0 0 0 0 0 0 0 0
ram9 0 0 0 0 0 0 0 0 0 0
ram10 0 0 0 0 0 0 0 0 0 0
ram11 0 0 0 0 0 0 0 0 0 0
ram12 0 0 0 0 0 0 0 0 0 0
ram13 0 0 0 0 0 0 0 0 0 0
ram14 0 0 0 0 0 0 0 0 0 0
ram15 0 0 0 0 0 0 0 0 0 0
sda 102903 5247 8408420 47424 1820306 3782041 357153613 43320121
0 2228
nvme0n1 145633279 0 1165066312 18981333 13107200 0
3355443200 6663655 0 1796
loop0 0 0 0 0 0 0 0 0 0 0
loop1 0 0 0 0 0 0 0 0 0 0
[chop rest of loop devices]
[root@after ~]# cat /sys/fs/cgroup/user.slice/io.stat
8:0 rbytes=3534848 wbytes=4096 rios=723 wios=1 dbytes=20592091136 dios=16189
This patchset contains the following six patches.
0001-block-make-bdev_ops-rw_page-take-a-REQ_OP-instead-of.patch
0002-block-Add-part_stat_read_accum-to-read-across-field-.patch
0003-block-Define-and-use-STAT_READ-and-STAT_WRITE.patch
0004-block-Add-and-use-op_stat_group-for-indexing-disk_st.patch
0005-block-Track-DISCARD-statistics-and-output-them-in-st.patch
0006-blkcg-Track-DISCARD-statistics-and-output-them-in-cg.patch
and also available in the following git branch.
git://git.kernel.org/pub/scm/linux/kernel/git/tj/misc.git block-discard-stat
diffstat follows. Thanks.
Documentation/ABI/testing/procfs-diskstats | 10 ++++++++++
Documentation/admin-guide/cgroup-v2.rst | 10 ++++++----
Documentation/block/stat.txt | 28 ++++++++++++++++------------
Documentation/iostats.txt | 15 +++++++++++++++
block/bio.c | 16 +++++++++-------
block/blk-cgroup.c | 14 ++++++++++----
block/blk-core.c | 12 ++++++------
block/genhd.c | 29 ++++++++++++++++++-----------
block/partition-generic.c | 25 +++++++++++++++----------
drivers/block/brd.c | 14 +++++++-------
drivers/block/drbd/drbd_receiver.c | 3 +--
drivers/block/drbd/drbd_req.c | 4 ++--
drivers/block/drbd/drbd_worker.c | 4 +---
drivers/block/rsxx/dev.c | 6 +++---
drivers/block/zram/zram_drv.c | 19 +++++++++----------
drivers/lightnvm/pblk-cache.c | 5 +++--
drivers/lightnvm/pblk-read.c | 5 +++--
drivers/md/bcache/request.c | 13 +++++--------
drivers/md/dm.c | 6 ++++--
drivers/md/md.c | 8 ++++----
drivers/nvdimm/btt.c | 12 ++++++------
drivers/nvdimm/nd.h | 7 +++----
drivers/nvdimm/pmem.c | 13 ++++++-------
fs/block_dev.c | 6 ++++--
fs/ext4/super.c | 5 +++--
fs/ext4/sysfs.c | 6 ++++--
fs/f2fs/f2fs.h | 2 +-
fs/f2fs/super.c | 3 ++-
fs/mpage.c | 4 ++--
include/linux/bio.h | 4 ++--
include/linux/blk-cgroup.h | 5 ++++-
include/linux/blk_types.h | 20 ++++++++++++++++++++
include/linux/blkdev.h | 2 +-
include/linux/genhd.h | 14 ++++++++++----
34 files changed, 215 insertions(+), 134 deletions(-)
--
tejun