Optimize instance state queries by fetching only required fields

[gajeshbhat]

Now that canonical/lxd#17378 and canonical/lxd#17679 are merged to main on LXD we can make use of new query parameters to optimize instance state queries in lxd-ui. This was also added to canonical/pylxd#706 recently. Happy to help contribute this feature and send in a pull request. But we can discuss this first before making any code changes.

[gajeshbhat]

CC : @edlerd

[edlerd]

Hello @gajeshbhat ,

thanks for reaching out and suggesting support. I agree it should be a nice optimization to define exactly which fields to fetch for instances.

Let's avoid defining the fields needed per use case. That would complicate cache handling and reuse of the instance objects fetched for lists and other components present in the cache on the client side. Instead, I'd suggest a similar approach taken for the instanceEntitlements. A global list of all the fields ever used in the UI. We can attach that list then to both the fetchInstance and fetchInstances queries in the src/api/instances.tsx file.

Please let me know if you have further questions or suggestions.

[edlerd]

A good starting point to inventory the used fields could be the LxdInstance type and its usages.

[gajeshbhat]

Thank you @edlerd for those links. I will start there, while I work through this, does the CI in this repo has the mechanism to test our optimization against the latest build built off of main on canonical/lxd ?

[gajeshbhat]

In our case the latest release at this time which is https://github.com/canonical/lxd/releases/tag/lxd-6.7 where the latest changes are incorporated, I presume yes looking at the release notes as the LXD UI Changes were also included in that release.

[edlerd]

Yes, CI runs the changes from the PR against various versions. So if you changed the api/instances.tsx file, it would be tested against LXD edge with your changes and LXD 5.21 and 5.0. The latter two would just ignore the extra url parameters.

[gajeshbhat]

@edlerd I put up a draft PR and we can iterate on it and do test deployments as I have only tested it locally and documented my testing process with commands in the PR description. I also made a small UI-side change for fine-grained permissions. The current LXD backend rejects the combined selective-recursion + entitlements request and returns "Entitlements can only be requested when recursion is enabled" error on the UI in the instances list page. Fine-grained mode still needs access_entitlements for correct instance actions, so I kept the optimized selective-recursion fetch for instance data and fetch entitlements separately, then merge them client-side. There might be better way to do this and we can collaborate to clean this up.