mirror of
https://github.com/truenas/charts.git
synced 2026-02-12 23:06:13 +08:00
929e60d801
* Adapt charts CI and improve/fix common * add check on permissions contaienr * add postgres template * update comments * Update create_app.sh * add check * update script * auto gen item.yaml from Chart,yaml * rename readme on dest * duplicate readme from the same source * correct comment * reoder * remove extra space * keep both README and app-readme * update regex, to also allow 2 letter names, which is also valid * No need to check host network if there aren't any pod values * use same pattern as the pod.name label (not prepending release-name * update deps * add chart dirs to ci * Add a validation to check if there is any yaml errors after merging files * update charts path on ci * common/1.0.0/ -> common/ * update common-test dep path * temp update create_app script * make permissions container name configurable, incase we want to change order of execution * update naming convention * fix typo and a missed name change * do not allow `--` in names
1.6 KiB
1.6 KiB
Scale GPU
| Key | Type | Required | Helm Template | Default | Description |
|---|---|---|---|---|---|
| scaleGPU | list |
❌ | ❌ | [] |
Define the external interfaces as list |
| scaleGPU.targetSelector | dict |
❌ | ❌ | {} |
Where to assign the GPU |
| scaleGPU.targetSelector.[pod-name] | list |
❌ | ❌ | [] |
The workload to select |
| scaleGPU.targetSelector.[pod-name].[container-name] | string |
✅ | ❌ | "" |
The container to select |
| scaleGPU.gpu | dict |
✅ | ❌ | {} |
The GPU key value pair to define in resources |
When
targetSelectoris a dict, each entry is a list, containing the name(s) of the container(s) to assign the GPU WhentargetSelectoris a empty, it will assign the GPU to the primary pod/container Selected pod's will get appended the group44insupplementalGroups. This is to allow rootless containers to access the GPU
Appears in:
.Values.scaleGPU
Examples:
scaleGPU:
- gpu:
# Injected from SCALE UI/Middleware using $ref
nvidia.com/gpu: "1"
targetSelector:
workload-name:
- container-name