Krisztian Litkey

It should also do [these things](https://github.com/intel/cri-resource-manager/pull/602#pullrequestreview-577805992), too. ``` It would also need to be improved with additional tightest-fit allocation rules beyond the current topology socket/core/thread hierarchy to get to a...

> hey @klihub did you have any desire to work on this? I forget where we were at with it Sorry for the late reply... I had no time to...

@estesp @soulseen This [CDI PR](https://github.com/container-orchestrated-devices/container-device-interface/pull/82) should fix the read/write race bug.

> When making this configurable, wouldn't it be even more future-proof, if one could give sysfs mountpoint, defaulting to `/sys`, rather than giving a root directory and expecting the mountpoint...

> I think the current approach is actually better wrt. what @askervin suggested, as this is more general. It could be used for other system directories, too, if needed in...

> I tested your branch a discovered some issue (I understand this is RFC but just wanted to test some use cases - I'll describe it in another comment). >...

> > > But configs looks fine, when using those configs with --fallback-config or --force-config it works fine.. > > Hmm..., okay thanks That would suggest the [configMap ->] agent...

> We had discusses what we would expected generally from "switching policy on runtime" and I have additional questions: > > * I don't remember how cri-rm behaved previously -...

> One small question - I want to make sure I deploy proper version of agent but I don't know how to check after it is build (I need build...

@guozijing Could you provide a detailed description of the steps leading to the diff between expected and observed behavior, including (enough details about) the pods/containers (so we can verify whether...