From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from firstgate.proxmox.com (firstgate.proxmox.com [212.224.123.68]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature RSA-PSS (2048 bits)) (No client certificate requested) by lists.proxmox.com (Postfix) with ESMTPS id C27709F0CA for ; Sat, 4 Nov 2023 09:34:42 +0100 (CET) Received: from firstgate.proxmox.com (localhost [127.0.0.1]) by firstgate.proxmox.com (Proxmox) with ESMTP id AAD9F6085 for ; Sat, 4 Nov 2023 09:34:12 +0100 (CET) Received: from proxmox-new.maurer-it.com (proxmox-new.maurer-it.com [94.136.29.106]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature RSA-PSS (2048 bits) server-digest SHA256) (No client certificate requested) by firstgate.proxmox.com (Proxmox) with ESMTPS for ; Sat, 4 Nov 2023 09:34:11 +0100 (CET) Received: from proxmox-new.maurer-it.com (localhost.localdomain [127.0.0.1]) by proxmox-new.maurer-it.com (Proxmox) with ESMTP id A30E4444EA for ; Sat, 4 Nov 2023 09:34:10 +0100 (CET) Message-ID: <2cd139da-fe1b-4709-881a-603ea5de8427@proxmox.com> Date: Sat, 4 Nov 2023 09:34:08 +0100 MIME-Version: 1.0 User-Agent: Mozilla Thunderbird Beta Content-Language: en-GB To: Proxmox VE development discussion , Dominik Csapak References: <20231103115343.4133611-1-d.csapak@proxmox.com> From: Thomas Lamprecht In-Reply-To: <20231103115343.4133611-1-d.csapak@proxmox.com> Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit X-SPAM-LEVEL: Spam detection results: 0 AWL -0.219 Adjusted score from AWL reputation of From: address BAYES_00 -1.9 Bayes spam probability is 0 to 1% DMARC_MISSING 0.1 Missing DMARC policy KAM_DMARC_STATUS 0.01 Test Rule for DKIM or SPF Failure with Strict Alignment POISEN_SPAM_PILL 0.1 Meta: its spam POISEN_SPAM_PILL_1 0.1 random spam to be learned in bayes POISEN_SPAM_PILL_3 0.1 random spam to be learned in bayes SPF_HELO_NONE 0.001 SPF: HELO does not publish an SPF Record SPF_PASS -0.001 SPF: sender matches SPF record T_SCC_BODY_TEXT_LINE -0.01 - Subject: Re: [pve-devel] [RFC PATCH cluster/guest-common/qemu-server/container/manager] add backend profile support X-BeenThere: pve-devel@lists.proxmox.com X-Mailman-Version: 2.1.29 Precedence: list List-Id: Proxmox VE development discussion List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Sat, 04 Nov 2023 08:34:42 -0000 On 03/11/2023 12:53, Dominik Csapak wrote: > This series aims to provide profile support when creating guests (ct/vm) > so that users can reuse options without having to specify them every > time. > > Sending as RFC because I don't quite like some things with the current > implementation and I'm not quite sure in which direction I should take > this. Also the GUI part isn't done yet and i wanted to see if the > direction is OK. (Sorry for the wall of text) > > The major issues: > > Using a single section config for both VMs and CTs make handling the > properties a bit weird. For now i prefix the options with "$type_" > so vm options are e.g. 'vm_ostype', 'vm_name', and so on while container > options are 'ct_ostype', 'ct_hostname', etc. That properties are shared by different sections by default is IMO not really ideal in general, it's also making the storage API and its docs rather hard to understand & work with. One option could be to opt-into a newer behavior, e.g., via some property, or getter, that the section config implementation needs to set, or override and return true, which makes then all properties isolated if not explicitly marked as shared (via another new property), that way we could use it now, and move over existing ones where it makes sense, without risking wide breakage. > > Using the same config/plugin system also makes using it a bit weird. > We have to register/init them in the api where they're used, but for the > cli we have to register only the available type and then init. This > makes it necessary to always set 'allow_unknown' while parsing the > config so that 'qm' doesn't trip over the container profiles... > > A fix for both could be to separate the ct/vm configs into two files > (similar to how we did pci/usb mapping configs), that would fix the > prefixing, as well as the register/init issue (We have to have > separate APIs then ofc). > > Another fix would be to extend the section config to allow different > properties of the same name for different types. Should be possible, > although we have to be careful to not break existing ones, and also > the API interface has to be separate for each type then (cannot really > have confiflicting api parameter schemas for the same name?) > > We could also go in a completely different direction and create a config > per profile? (like we handle vm configs). Downside of that is, that the > current guest config handling part is partly in pmxcfs, so we'd have to > make that either more generic, or duplicate it for profiles. I don't see how this would have anything to do with pmxcfs and VMIDs, that map to an actual guest instance and thus needs special treatment compared to just some profile that can, e.g., live in a /etc/pve/guest-profiles/ as .profile (the extension just an example), and be handled only via perl – profiles are only used in the profile management API, where it's ok to fully parse one, and in guest creation POST calls, so even cfs_register* would be probably overkill. Having a file per profile makes some things relatively easy, but doesn't fits the commonly used section config, let's see if others have input (i.e., actively ask one/some of fabian/wolfgang/fiona. > (I don't quite like this one, since i think a single config for all > profiles should be enough? We could still do that later if we want > and the current way is impractical) We can always go from this to the other, or vice versa, if *really* needed, we should try hard to avoid either such move, though.. > > The minor issues: > > * Is the priv path ok? /mapping/profiles/ feels a bit weird What feels weird? s/profile/guest-profile/ might be a bit more telling though. > * We should probably introduce a 'meta' property for ct like we have for > vms? we could record the profile there and also e.g. the template used > to set the container up. I assume you mean in the CT config, IIRC there is even an older patch floating around for adding that for recording the ctime. But it shouldn't be needed, as meta stuff should be only informational, profile names can change (delete/re-add) or get modified, so saving the ID has no use, logging it to the task-log is enough for starters, I think. > * I did not restrict to any options of the config and i don't believe > this should make any issues (the critical ones are filtered out anyway) > but should we maybe have some kind of whitelist? if yes, which one > should be on there? I didn't do any evaluation of all properties, but things like force-machine and running-state properties make no sense, so at least disallowing those would be good. > * there is still an issue with the docs generation, i'll have to look > into it > * I'm not quite sure how the UI for creating/editing profiles should > look like. For creating i could imagine a 'create profile from guest' > type of thing (thanks @mira for the idea), but for editing or creating > from scratch I'm a bit conflicted. We could simply show the profiles > in the tree, and reuse the vm hw/options panels for that, but does > that seem overkill? We could also leave that API only for now, and > make the gui later? (Using it in the wizard would also not be trivial, > but there the constraints are rather straight forward) - view: simple grid with profile ID, profile type and the (unavoidable) profile comment as the tree columns (for starters). The admins can set a descriptive ID and comment to know what a profile is intended for. Showing all profile values is IMO bad UX, as it becomes crowded super fast for more than a few trivial profiles. - profile create & edit: - a per-type add, i.e., "Add VM Profile" and "Add CT Profile" - each provides the id and comment at the top and then has a panel with an add button to add a new property. - Most properties are simple single-fields, for storage we can add more complex widgets too, idealy recycling (parts of) the forms and/or panels we already use other where - I'd start out with defining good widgets for the most sensible properties first, that is naturally a bit subjective, but e.g., memory, max memory, cpu cores, cpu model, storage (mpX, scsiX, virtioX), network, including (host)name, DNS (for CTs). Anything else can then be a "raw" field for now, and we can add further "first-class" widgets for properties that are deemed relevant due to feedback from other devs and users. IMO a create wizard is relatively much work, but not necessarily better, as seeing all in one view has it's advantage.