Overview
Request 592108 superseded
- Enhance IO scheduler tweaking rules to support blk-mq as well
* Set optimal blk-mq schedulers (bfq and mq-deadline for HDD and
SSD respectively)
* Explicitly set CFQ for rotational disks when no `elevator`
is specified
* Verbose comments explaining what is going on
* Rename the rules file since it is not ssd-only anymore
Tweaked the rules file per the @mwilks suggestions
- Created by develop7
- In state superseded
- Package maintainers: fbui, systemd-maintainers, and tsaupe
- Supersedes 591774
- Superseded by 592117
These rules apply to SCSI only. Why don't you use my more general approach from https://github.com/openSUSE/systemd/pull/7?
@mwilck multiple reasons, to be fair:
- introducing changes by small steps to decrease the possibility being clogged by the QA facing dramatically increased testing surface
- maintain the backward compatibility, with
elevator
cmdline parameter in particular - Not sure testing for the
/mq
directory is fool-proof enough - I have no devices not handled by scsi_mod, so I couldn't test your approach myself.
Thank you for the suggestion though: it looks much terser than one I'm proposing here.
Not sure testing for the /mq directory is fool-proof enough
I stand corrected, SUBSYSTEM!="block", GOTO="ssd_scheduler_end"
looks good enough.
I'll look into it. Thank you for the insight.
Request History
develop7 created request
- Enhance IO scheduler tweaking rules to support blk-mq as well
* Set optimal blk-mq schedulers (bfq and mq-deadline for HDD and
SSD respectively)
* Explicitly set CFQ for rotational disks when no `elevator`
is specified
* Verbose comments explaining what is going on
* Rename the rules file since it is not ssd-only anymore
Tweaked the rules file per the @mwilks suggestions
develop7 added mwilck as a reviewer
mwilck declined review
I'd move the test for elevator further up (before the TEST== statement). After all, if "elevator" is set, you won't don any further actions anyway.
Other than that, I'm fine with this.
mwilck declined request
I'd move the test for elevator further up (before the TEST== statement). After all, if "elevator" is set, you won't don any further actions anyway.
Other than that, I'm fine with this.
superseded by 592117
@mwilck the trick here is the
elevator
cmdline affects non-BLK-MQ config only, that's why it's put after MQ-related part@mwilck OTOH we shouldn't care whether elevator affects BLK-MQ configs or not. Fixing. Thank you again.