support write multiple items into cgroup v2 file

[smileusd]

This pr is a supplement for cgroup v2 file value write. 
I met that after calculating in kubelet, i got a configs.Resources contains the cup,memory,pid and also cgroup v2 field like io.max and io.weight. the map of Unified in the config might be like: 
{"io.max": "252:1 riops=1000 wiops=1000 \n 252:2 riops=2000 wiops=2000", "io.weight": "252:1 300 \n 252:2 500" }
Currently the unified set can not support write multiple line in one cgroup file.

[rata]

Just curious, why is this writing two lines instead of overwriting them?

If you had a test case for this, I could have answered myself by looking at that ;)

[smileusd]

the cgroup file is also like a map. if the key is the same, the new one will override old one, but if it is different key, like different device number, the will add a new line

[smileusd]

Not sure how to add test case at here, there is the test by my environment

root@tess-node-6bg2h:/sys/fs/cgroup/kubepods.slice/kubepods-besteffort.slice/kubepods-besteffort-pod02d6c4d8_0888_464a_9037_7238679d4174.slice# cat io.max 
252:0 rbps=1048576 wbps=1048576 riops=100 wiops=100
253:0 rbps=1048576 wbps=1048576 riops=100 wiops=100
root@tess-node-6bg2h:/sys/fs/cgroup/kubepods.slice/kubepods-besteffort.slice/kubepods-besteffort-pod02d6c4d8_0888_464a_9037_7238679d4174.slice# echo  "252:16 rbps=1048576 wbps=1048576 riops=100 wiops=100" > io.max 
root@tess-node-6bg2h:/sys/fs/cgroup/kubepods.slice/kubepods-besteffort.slice/kubepods-besteffort-pod02d6c4d8_0888_464a_9037_7238679d4174.slice# cat io.max 
252:16 rbps=1048576 wbps=1048576 riops=100 wiops=100
252:0 rbps=1048576 wbps=1048576 riops=100 wiops=100
253:0 rbps=1048576 wbps=1048576 riops=100 wiops=100

[smileusd]

I added test case in pr @rata

[kolyshkin]

@smileusd perhaps you can set io.max not via unified but via linux.resources.blockIO.weightDevice

[smileusd]

@smileusd perhaps you can set io.max not via unified but via linux.resources.blockIO.weightDevice

@kolyshkin

// WeightDevice struct holds a `major:minor weight`|`major:minor leaf_weight` pair
type WeightDevice struct {
	blockIODevice
	// Weight is the bandwidth rate for the device, range is from 10 to 1000
	Weight uint16 `json:"weight"`
	// LeafWeight is the bandwidth rate for the device while competing with the cgroup's child cgroups, range is from 10 to 1000, cfq scheduler only
	LeafWeight uint16 `json:"leafWeight"`
}

Looks like it is only for io.weight not io.max. And actually cgroup v2 io.weight value range should be [1, 10000] not [10, 1000] So i think that should only for cgroup v1 might.

[kolyshkin]

@smileusd does #4186 fixes your issue?

[kolyshkin]

@smileusd it's a bit weird that you took my code (from #4186) and put it into your patch without even mentioning it.

[kolyshkin]

Looks like it is only for io.weight not io.max. And actually cgroup v2 io.weight value range should be [1, 10000] not [10, 1000] So i think that should only for cgroup v1 might.

I'm sorry for pointing your into the wrong setting.

What I meant to say is you should use BlkioThrottleReadBpsDevice, BlkioThrottleWriteBpsDevice, BlkioThrottleReadIOPSDevice, and BlkioThrottleWriteIOPSDevice. Those settings are translated by runc to writes into io.max (as you can see in the source code if you grep for io.max).

[kolyshkin]

Closing in favor of #4186.