[feature] group nodes by hardware features

[Jason Hobbs]

When using FCE to enlist and configure nodes, we generate buckets of nodes based on hardware characteristics of the nodes we derive from API calls made to maas - this is done by a step called 'generate_buckets.

We use the buckets for a couple of reasons:

1) Identifying outliers in hardware config - we generally expect machines to be have the same type of hardware, maybe 2 or 3 models used per deployment. If a machine has a different number of disks, number or speed of interfaces and network links, disks with different sizes, is attached to different vlans, different amount of RAM, different CPU characteristics, different firmware versions, different graphics cards, we want to know it about. Grouping by these characteristics allows this to stand out.

2) Applying a templated config. We generally want to apply the same config to all nodes with the same set of hardware configuration.

generate_buckets makes a couple of maas api calls per node (machine details and node results read)

We'd like for MAAS to speed this up. When there are 100's of nodes, it takes a long time to make all of these API calls. Also, we would like to be able to use this feature with the upcoming templating feature, to apply a template to a group of nodes. We'd also like to read the list of groups, and see which machines are in which groups, in an efficient manner, so we can see if it matches expectations for machines.

[Andres Rodriguez (andreserl)]

Hi Jason,

I don't understand what you actually mean with this request. Can you expand on what you would actually expect this request would do? The bug report talks about the fact that you do multiple API calls from a script, but it doesn't present any context whatsoever on what this request is about, what user stories, what is the expected result, what is the expected use for this.

Thanks

[Jason Hobbs (jason-hobbs)]

Andres, I've added some additional detail.

[Andres Rodriguez (andreserl)]

I'm setting this as a feature request, that said I also wanted to clarify something.

The storage template feature as it was initially designed is no more. MAAS is not expected to support storage templates that can be automatically applied to certain machines (in a bucket or not). This feature has changed to do something different, which is to provide a descriptive language that will change the way MAAS addresses storage.

This means that users will be able to request machine providing a descriptive language which would automatically provide them with a machine, and apply a potential storage layout for the given machine. This could mean that a storage layout could be applied differently to different machines bbut then again, that's not yet decided.

That said, this may mean that this feature request would become obsolete given of the chagnes of what storage templates are meant to do, or it may just be relevant.

[khb (khbkhb)]

From support case 00288146. Our support engineer suggested this as a solution to our use case. For clarity, let me be more specific about our needs.

Dell C6420s configured with 2 NVMe, 4 SSD and 1BOSS raid (so 2 onboard SSDs which are hw raided together with the special BOSS controller (mirrored). When discovered, Ubuntu is finding the NVMes, then the SSDs and finally the BOSS. Unfortunately this translates into MAAS formatting the first NVMe and assuming to boot from it. Sadly, the BIOS does not even permit booting from the NVMe, and even if it could, the expected result is that the BOSS should be boot drive. So each system is ENLISTED and COMMISSIONed and then has to be reconfigured. Assuming that Canonical doesn't want to detect this and fix (any Dell system should default to the BOSS, that's the reason it exists) (or even if you did detect it correctly, this would be an RFE, albeit a lower urgency item).

There should be a mechanism for detecting a system meeting some set of conditions, and when it "fires" to adjust the configuration as part of the COMMISSION rather than having to do it ex post facto.

[Jerzy Husakowski (jhusakowski)]

The request appears to have two parts: making MAAS faster, and certain specifics about applying templates to groups of nodes. The former is on the roadmap as a general activity, the latter is on the roadmap in the form of "storage templates" feature. Other feature requests are welcome via MAAS Discourse - https://discourse.maas.io/c/features/15