Add Kata docs#370
Conversation
Documentation preview |
|
@manuelh-dev if you have time to review :) |
Signed-off-by: Abigail McCarthy <20771501+a-mccarthy@users.noreply.github.com>
Signed-off-by: Abigail McCarthy <20771501+a-mccarthy@users.noreply.github.com>
Signed-off-by: Abigail McCarthy <20771501+a-mccarthy@users.noreply.github.com>
a-mccarthy
force-pushed
the
add-kata-docs
branch
from
98104b7 to
cbcc63e
Compare
Signed-off-by: Abigail McCarthy <20771501+a-mccarthy@users.noreply.github.com>
| * ``NVIDIA GPU Feature Discovery`` -- to detect NVIDIA GPUs and label worker nodes. | ||
| - * ``NVIDIA Confidential Computing Manager for Kubernetes`` -- to set the confidential computing (CC) mode on the NVIDIA GPUs. | ||
| This component is deployed to all nodes configured for Kata Containers, even if you are not planning to run Confidential Containers. | ||
| Refer to the :doc:`Confidential Containers <cc:index>` documentation for more details. |
There was a problem hiding this comment.
maybe we should additionally link to the section in the deployment guide which talks about mode changes? If a node is CC capable, the GPU will be in CC mode won't it?
There was a problem hiding this comment.
I see this is aspect is being iterated on further below again. It might be good to make it clear at one place that a mode change may be required. Also, @jojimt is it possible to directly control this via GPU operator helm install flags?
| $ kubectl delete -f cuda-vectoradd-kata.yaml | ||
|
|
||
|
|
||
| Troubleshooting Workloads |
There was a problem hiding this comment.
Reminder for the future: Let's think about forking this out into a separate page. Please feel free to close.
| :alt: Logical diagram of software components between Kubelet and containers when using Kata Containers. | ||
|
|
||
| flowchart LR | ||
| a[Kubelet] --> b[CRI] --> c[Kata\nRuntime] --> d[Lightweight\nQEMU VM] --> e[Lightweight\nGuest OS] --> f[Pod] --> g[Container] |
There was a problem hiding this comment.
Potential to discuss: The Pod construct could be a box wrapping the VM, Guest OS, Container. Here, it looks like the runtime somewhat starts a VM, which somewhat starts a Guest OS which somewhat starts a Pod. Is this really correct? Maybe I'm over-thinking here, but would be good to run internally.
Signed-off-by: Abigail McCarthy <20771501+a-mccarthy@users.noreply.github.com>
Signed-off-by: Abigail McCarthy <20771501+a-mccarthy@users.noreply.github.com>
Signed-off-by: Abigail McCarthy <20771501+a-mccarthy@users.noreply.github.com>
2 participants
split from to #365 to just work on kata sandbox docs.