Aleksei Sviridkin cff833ebb0 feat(api): refuse a vGPU class on a worker pool ...

Every KubernetesNodes pool runs Talos workers, which carry no vGPU
guest driver, so a pool that names a vGPU class would start workers
that never use their GPUs. The write is refused while it is made,
before any machine exists. A class must exist before
the pool is stored, so a later class cannot change the mode the pool
was checked against.

Assisted-by: LLM
Signed-off-by: Aleksei Sviridkin <f@lex.la>

2026-11-03 14:46:01 +03:00

..

apis

feat(api): offer GPU classes through a gpuclass option source

2026-11-03 14:46:01 +03:00

apiserver

feat(api): offer GPU classes through a gpuclass option source

2026-11-03 14:46:01 +03:00

cmd/server

docs(api): clarify kind-wide Helm timeout overrides

2026-10-10 22:36:54 +03:00

config

docs(api): clarify kind-wide Helm timeout overrides

2026-10-10 22:36:54 +03:00

generated

feat(api): offer GPU classes through a gpuclass option source

2026-11-03 14:46:01 +03:00

lineage

perf(lineage): cache owner GETs and short-circuit re-admission

2026-09-29 17:32:13 +02:00

ovnstatus

refactor: apply the Go 1.27 modernizers

2026-09-24 17:23:18 +03:00

registry

feat(api): refuse a vGPU class on a worker pool

2026-11-03 14:46:01 +03:00

version

test(version): clear ambient COZYSTACK_VERSION in the unset case

2026-07-06 10:03:56 +05:00