9ec47cf482
Due to the way /etc/udev/rules.d/60-io-scheduler.rules is
written, it doesn't check if the I/O scheduler is already
set correctly or not, before resetting it. Thus, a condition
was added to each "action", in such a way that only the
correct "action" is executed.
Using gparted/sgdisk in sysinv causes disks to be "reprobed"
by the kernel every minute, so the disk I/O scheduler is reset
multiple times. This can be seen by running the command below:
sysadmin@controller-0:~$ ( while true; do sched="$(cat /sys/block
/sda/queue/scheduler)"; if test "${sched}" != "${last_sched}";
then echo "$(date -Is): change: ${sched}"; last_sched="${sched}";
fi; done )
Test Plan:
PASS: It was checked disk I/O scheduler (sys/block/sda/queue
/scheduler) is not resetting every minute, multiple times on
AIO-SX fresh install.
Closes-Bug: 1996822
Signed-off-by: Erickson Silva de Oliveira <Erickson.SilvadeOliveira@windriver.com>
Change-Id: Ic4a9b963d00393b591fd23f2c1224ad6b8740e5e
131 lines
4.9 KiB
Plaintext
131 lines
4.9 KiB
Plaintext
# This file contains the rules to customize io scheduler.
|
|
|
|
# Need nodetype environment variable
|
|
IMPORT{file}="/etc/platform/platform.conf"
|
|
|
|
# 3.10 kernel
|
|
# Heuristics:
|
|
# 'deadline' io-scheduler tuned settings
|
|
# - deadline generally recommended for databases, servers, and SSDs,
|
|
# and for more deterministic latency
|
|
# - note that read_expire is a key tuning parameter here
|
|
# - the following is recommended by DRBD user guide
|
|
# front_merges: 0 (from 1)
|
|
# read_expire: 150 (from 500)
|
|
# write_expire: 1500 (from 5000)
|
|
#
|
|
# 'cfq' io scheduler tuned settings
|
|
# - recommended for HDD, and fairness (ionice)
|
|
# - the following tuned values greatly improve parallelism,
|
|
# and performance in large storage environments
|
|
# slice_idle: 0 (from 8)
|
|
# quantum: 32 (from 8)
|
|
#
|
|
# 'noop' io scheduler for variants of HW-RAID.
|
|
# - RAID controller will do its own separate scheduling
|
|
#
|
|
# Overall:
|
|
# - Historically we choose 'deadline' for low-latency, SSDs, and local storage;
|
|
# 'cfq' for worker non-sda HDDs, 'noop' for all variants of HW-RAID.
|
|
# - We prefer to guarantee latency more than fairness for all platform services,
|
|
# especially under extreme read and write load, e.g, when creating/deleting
|
|
# multiple heat stacks, or running disk intensive operations.
|
|
# - Kubernetes etcd server has better latency with cfq as compared
|
|
# to deadline even on SSDs.
|
|
# - This setting requires re-evaluation with io-schedulers
|
|
# (i.e., mq-deadline, bfq, kyber) available in new kernels.
|
|
|
|
# 5.10 kernel
|
|
# Heuristics:
|
|
# 'none'
|
|
# - Implements a first-in first-out (FIFO) scheduling algorithm. It
|
|
# merges requests at the generic block layer through a simple last-hit cache.
|
|
# 'mq-deadline'
|
|
# - Attempts to provide a guaranteed latency for requests from the point at
|
|
# which requests reach the scheduler.
|
|
# - This scheduler is suitable for most use cases, but particularly those in
|
|
# which the write operations are mostly asynchronous.
|
|
# 'bfq'
|
|
# - Targets desktop systems and interactive tasks.
|
|
# - The bfq scheduler ensures that a single application is never using all of
|
|
# the bandwidth. In effect, the storage device is always as responsive as if
|
|
# it was idle. In its default configuration, bfq focuses on delivering the
|
|
# lowest latency rather than achieving the maximum throughput.
|
|
# - bfq is based on cfq code. It does not grant the disk to each process for
|
|
# a fixed time slice but assigns a budget measured in number of sectors to the process.
|
|
# - This scheduler is suitable while copying large files and the system does not become
|
|
# unresponsive in this case.
|
|
# 'kyber'
|
|
# - The scheduler tunes itself to achieve a latency goal by calculating the latencies
|
|
# of every I/O request submitted to the block I/O layer. You can configure the target
|
|
# latencies for read, in the case of cache-misses, and synchronous write requests.
|
|
# - This scheduler is suitable for fast devices, for example NVMe, SSD, or other low
|
|
# latency devices.
|
|
#
|
|
# Overall:
|
|
# - Traditional HDD with a SCSI interface: Use mq-deadline or bfq
|
|
# - High-performance SSD or a CPU-bound system with fast storage: Use none,
|
|
# especially when running enterprise applications. Alternatively, use kyber.
|
|
# - Desktop or interactive tasks: Use bfq.
|
|
# - Virtual guest: Use mq-deadline. With a host bus adapter (HBA) driver that
|
|
# is multi-queue capable, use none.
|
|
|
|
ACTION!="add|change", GOTO="iosched_end"
|
|
SUBSYSTEM!="block", GOTO="iosched_end"
|
|
KERNEL!="sd[a-z]", GOTO="iosched_end"
|
|
|
|
PROGRAM="/usr/local/bin/is-rootdisk-device.sh %E{DEVNAME}", RESULT=="?*", ENV{rootdisk}="1"
|
|
|
|
# === === === ===
|
|
# Check the I/O scheduler settings are correct.
|
|
# === === === ===
|
|
|
|
ATTRS{raid_level}=="*RAID*", \
|
|
ATTR{queue/scheduler}=="*\[none\]*", \
|
|
GOTO="iosched_end"
|
|
|
|
ATTRS{label}=="*RAID*", \
|
|
ATTR{queue/scheduler}=="*\[none\]*", \
|
|
GOTO="iosched_end"
|
|
|
|
ENV{nodetype}=="controller", \
|
|
ENV{rootdisk}=="1", \
|
|
ATTR{queue/scheduler}=="*\[bfq\]*", \
|
|
ATTR{queue/iosched/strict_guarantees}=="1", \
|
|
GOTO="iosched_end"
|
|
|
|
ATTR{queue/scheduler}=="*\[mq-deadline\]*", \
|
|
ATTR{queue/iosched/front_merges}=="0", \
|
|
ATTR{queue/iosched/read_expire}=="150", \
|
|
ATTR{queue/iosched/write_expire}=="1500", \
|
|
GOTO="iosched_end"
|
|
|
|
# === === === ===
|
|
# I/O scheduler settings are not as expected, need to be defined.
|
|
# === === === ===
|
|
|
|
# Set none io scheduler for variants of HW-RAID.
|
|
# HP ProLiant DL360p Gen8; HP ProLiant DL380p Gen8
|
|
ATTRS{raid_level}=="*RAID*", \
|
|
ATTR{queue/scheduler}="none", \
|
|
GOTO="iosched_end"
|
|
|
|
# Dell Inc. PowerEdge R720
|
|
ATTRS{label}=="*RAID*", \
|
|
ATTR{queue/scheduler}="none", \
|
|
GOTO="iosched_end"
|
|
|
|
ENV{nodetype}=="controller", \
|
|
ENV{rootdisk}=="1", \
|
|
ATTR{queue/scheduler}="bfq", \
|
|
ATTR{queue/iosched/strict_guarantees}="1", \
|
|
GOTO="iosched_end"
|
|
|
|
ATTR{queue/scheduler}="mq-deadline", \
|
|
ATTR{queue/iosched/front_merges}="0", \
|
|
ATTR{queue/iosched/read_expire}="150", \
|
|
ATTR{queue/iosched/write_expire}="1500", \
|
|
GOTO="iosched_end"
|
|
|
|
LABEL="iosched_end"
|