Please enable JavaScript.
Coggle requires JavaScript to display documents.
DM blkcg support - Coggle Diagram
DM blkcg support
investigation
observe io.bfq.weight on loop
work as expected
observe same io throttle on loop
work as expected on rhel9
not work on rhel8
on both loop dio or not
dm-thin
understand dm-thin implementation
bio allocation part
compare code path between normal lvm volumn and dm-thin
trace
trace bio->bi_blkg
issues reported
io.bfq.weight doesn't work on dm(unknown)
io throttle doesn't work on dm-thin-volumn
wiops isn't throttled on dm-thin-volume
how does blk-throttle works?
is it related with io submission from new context?
will it makes a difference on wiops throttle over dm-thin-volume?
dm-thin bio->bi_blkg should be correct
dm-thin is the disk which FS bios are sent to
it shouldn't since the throttle is done on thin-volume, not underlying thin-pool or other volume
so why doesn't it work?
why isn't bios to dm-thin throttled?
possible reasons
bio_split()
each bio is just throttled once
also need to understand dm bio account
block-throttle: avoid double charge
why no such issue for normal disk in case of bio splitting?
request based throttle
throttle after request is allocated
is it correct?
shouldn't be correct
one throttling blkcg will cause other processs
both rq_qos throttle and bio throttle