Investigating faster workspace startup

[l0rd]

What problem do we want to address?

Loading a Che workspace is currently an operation that takes 45 secs or more.

We want to speed this up as much as possible. Ideally under 10 secs because fast software is the best software.

How are we going to address it?

We have the opportunity to experiment new ideas with the DevWorkspace that we could not test before.

In particular:

• Reduce the number of containers (i.e. running only a Theia container and tooling container) https://github.com/devfile/devworkspace-operator/issues/218
• Start an editor only workspace and asynchronously update the CR and "recreate" the workspace pod
• Cache vsx, images, DevWorkspace CRs result of devfile flattening
• Volume Snapshots
• Always have a Pod with minimal CPU and memory request up and running and ready to serve

In this gdoc @amisevsk has added some considerations and a proposal.

[amisevsk]

Related: https://github.com/devfile/devworkspace-operator/pull/208

[benoitf]

about reduce numbers: you could try merging 'user container' and 'tooling container' but keep 'theia container'

Details

- **currently we have**: one theia image, no default user image, and potentially many sidecar images : 2+ images - **issue is about**: one big image for each language (one che-theia-java, one che-theia-python, etc) but you still need to pick-up the ‘right che-theia’: one image - **suggestion is**: theia image, user image with all tools (java/python/go/etc) : two images (you’re doing composition, the user image is not tied to a specific theia version and if you run out of memory/cpu in the user/tooling container, you still have a responsive IDE)

[amoldhondse]

Are there any time lines for improving loading time of a Che workspace ?

[l0rd]

Hi @amoldhondse,

I am not sure how you found this issue but that's the DevWorkspace operator issue tracker, not Che. I am asking because although the DevWorkspace is planned to be the next workspace engine for eclipse Che it's not the case yet (here are the instructions to try it out).

That said, to answer your question:

• Starting a Che workspace today should take around 45 seconds if the images are pre-pulled on the nodes of your Kubernetes cluster. We provide a tool to pre-pull images: the image puller. If you are not using it and you want to improve the workspaces startup time that's the first thing you should look at.
• From my first tests using the DevWorkspace as the workspace engine of Che, workspaces startup is even faster and usually under 30 seconds. But be aware that such workspaces currently have have some limitations (no che-theia plugin support) and require a v2 devfile.
• Improving workspaces startup time is critical but the tasks mentioned above are currently on hold until the DevWorkspace engine to be able to fully support Che workspaces.

[amoldhondse]

@l0rd Thanks for the response.We are using image puller configuration as specified in the documentation but it is still taking more than 45 secs for workspace to load.Any other things that you can suggest ? We have this configurations in our image puller,do you suggest adding anything or augmenting - images: java11-maven=quay.io/eclipse/che-java11-maven:nightly;che-theia=quay.io/eclipse/che-theia:next;che-jwtproxy=quay.io/eclipse/che-jwtproxy@sha256:881d1c91e7f5840314f25104ef5c0acee59ed484a5f9ef39daf3008725ea1033;che-theia-endpoint-runtime-binary=quay.io/eclipse/che-theia-endpoint-runtime-binary:next;che-plugin-artifacts-broker=quay.io/eclipse/che-plugin-artifacts-broker@sha256:f8591171ab5ac51e2af37ee9969781097c41a2e9f952de283e084f258411e5ca;che-nodejs10-community=quay.io/eclipse/che-nodejs10-community:7.12.2;che-sidecar-node=quay.io/eclipse/che-sidecar-node@sha256:0370b2c3d52d31c9fd52bff3edb84bd36bbcc2aff6224957d6dd94935843ee59;che-sidecar-vale=quay.io/eclipse/che-sidecar-vale;

[l0rd]

You should verify that the images and their digests match the ones that are pulled when you start a workspace. If you don't know how to figure out the images that are pulled you can reference this article.