Qubes Admin API

[marmarek]

Reported by joanna on 10 May 2014 12:56 UTC
https://groups.google.com/d/msg/qubes-devel/f2gDpXE3NJ8/_kH7LUrzJ80J

Migrated-From: https://wiki.qubes-os.org/ticket/853

---

TODO list, aggregated from commits below:

• a method to get VM statistics (memory usage, cpu time etc), maybe also a variant returning this as events (to avoid polling in tools like qubes-manager)?
• missing handling of global properties: I propose mgmt.property. prefix with exactly the same set of actions as for mgmt.vm.property.. Or maybe mgmt.global.property. ?
• Very limited error reporting - for example the reason why VM startup failed (no such VM, not enough memory, missing files, etc)
• mgmt.vm.property.List and mgmt.vm.property.Help are bound to VM instance at API level, while in practice those are bound to VM class. Generally I'm ok with that, but I face a practical issue: don't know how to properly handle docstrings for client-side wrappers. Our API have vm.property_list, so this part is not a problem (IMO dir(vm) not working is not a problem), but no idea how to handle help(VMClass.property). Maybe shouldn't? And handle only docstring on the object(s) returned by vm.property_list?
• missing calls for cloning - either at VM level (mgmt.vm.Clone), or individual elements (mgmt.vm.property.Clone, mgmt.vm.volume.Clone etc). I'm in favor of the first option (full VM clone with just new name+qid).
• No way to distinguish empty return data from failed call (at local socket level). Maybe qubesd should always respond with OK/FAIL on the first line and only then with actual data - this could be translated by qubesd-query to exit code.
• A way to query for vm -> label colour for GUI domain.
• Some way of editing labels themselves? (see Add seven new colors #2523 where I got)
• A way to query state of a single VM (and maybe VM existence check at the same time). Proposal: allow calling mgmt.vm.List to a specific VM, not only to dom0.
• Serialization of VM type (like 'netvm' property). Should it be just string? Then stubs deserializing it will have no idea it's more than string (vm.netvm.ip will not work). Maybe we need explicit type declaration in result of *.Get methods - instead of default=yes|no value, it would be default=yes|no type=... value?
• qrexec call argument cannot have : - we need a better idea for passing volume identifier (pool:vid currently). Or allow : in qrexec.
• volume id (vid) may contains chars not allowed in qrexec argument - for example /; we need to encode it somehow, or think of some other way (keep only pool name in argument and move vid to data stream?)
• no way to create StandaloneVM out of TemplateVM (previously having separate add_new_vm and clone_disk_files was used for that),
• mgmt.vm.Clone operation may be used to bypass some policy - for example user is not allowed to change netvm property of some VM (like force using some application through Tor), but can clone VM under another name and then change it.
• admin.vm.Shutdown should shutdown netvm even if is has vms connected to it
• check if failed start sends any event about vm.kill() called from except: clause
• features: the tool, if choses to, should set empty string for common "false" values; (document it)
• draw nice FSM of events (possibly similar to https://qubes-os.readthedocs.io/projects/core-admin/en/latest/qubes-vm/qubesvm.html#qubes.vm.qubesvm.QubesVM.get_power_state)

[woju]

Just pushed a draft of the API, to the qubes-doc, since the table is messy and won't be readable in the issue.

https://www.qubes-os.org/doc/mgmt1/

Didn't link anywhere yet.

@andrewdavidwong could you look on the page? It breaks the layout horribly and I don't know how to make this neither pretty nor readable. I took a shortcut and embedded a <style> directly in the page, which is bad, too.

[andrewdavidwong]

@woju: No problem; I'll work on this soon.

[andrewdavidwong]

@woju: How does it look now? (Note: You might need to clear your cache to get the new CSS.)

[woju]

Thanks. It is prettier, though it's less readable, since it gets scrolled independent of page, and wraps the cells. I disabled it on purpose, because if there is <name> class=<class> state=<state>\n, I didn't want it to get on multiple lines.

[andrewdavidwong]

Thanks. It is prettier, though it's less readable, since it gets scrolled independent of page,

The only way to avoid that is to use a different layout for the page. It sounds like maybe you don't want it to have the doc layout with the right-hand side bar. If the table doesn't get clipped, it'll simply overlap that side bar and be even less readable.

and wraps the cells. I disabled it on purpose, because if there is class= state=\n, I didn't want it to get on multiple lines.

Ok.

[marmarek]

Scrolling horizontally IMO isn't a problem - vertically is.

[andrewdavidwong]

If the table doesn't get clipped vertically, it'll be larger than the height of most viewports. My impression is that this is bad practice in web design, but I don't have any source to back that up.

[andrewdavidwong]

What if we create a full layout version of the table that takes up an entire page and provide a link from the existing page underneath the clipped table?

Alternatively, we could convert the entire page into the full layout and lose the side bar.

[woju]

I don't mind the side bar disappearing at all. And, as Marek said, table shouldn't be vertically scrollable independent of the page, because it makes it very annoying on touchscreen.

You may be right in the sense that nowadays it is fashionable to design custom scrolling, those designers do it and people get used to that, but IMHO this is not usable.

It may be valuable to devise some CSS class for the tables in documentation, since I will have at least two more such tables, for VM "events" and "features" (new functionality in R4.0 core).

But we are getting sidetracked.

[andrewdavidwong]

@woju @marmarek: Let me know what you think now.

[woju]

Great! Thank you.

[kalkin]

@woju What should mgmt1.vm.volume.List list?

[woju]

@kalkin some identifier about the volume and possibly very basic info, but I'm not sure what exactly.

The idea is that there are two set of calls, mgmt1.vm.volume.* and mgmt1.pool.volume.*, to allow for expressing the policy decisions "allow management of pool" and "allow management of storage for VM". They overlap in scope, on purpose. Currently I don't know who will use which one and in what context.