[RFC] Object ID about agent

[sameo]

cc @tallclair

[sameo]

I want to point out that processes, storage, devices and NICs have unique host and guest identifiers. The kernel manages and uniquely identifies them, so I wonder why we need another abstraction layer on top of it.

[bergwolf]

@sameo I proposed the abstraction because it's easier to think from a higher level. And then we do not need to look into details of each object to find out which is a possible unique identifier and thus end the ever-running discussions like whether pid/MAC is validly unique in a guest.

BTW what do you think can uniquely identify a storage? Device path cannot work because it may not have one at all. And mountpoints can easily duplicate.

[WeiZhang555]

I was also thinking about this recently, I think this is a good idea, which will end all unique key related talks. With something like UUID system, we can make the design more consolidated. One good example is in database world, every record has a unique number index when we design the table.

/cc @jodh-intel @sboeuf for more discussions.

[bergwolf]

@sameo now that we've agreed that a runtime generated process ID works better than guest pid in #72, is it OK to push the similar object ID rule globally?

[sboeuf]

I am fine with that, since it is not going to change anything, it's more about a concept we all agree on.

[c3d]

This was not done since it was suggested. I believe we should close it, and re-open the issue if we actually start working on it. @bergwolf please reopen if you think it is wrong to close it.