Files
cozystack/packages
Mattia Eleuteri 392bbb3c96 fix(kubernetes): keep maxUnavailable 1 on GPU worker pools
A replacement VM in a pool with passed-through GPUs may need the very
GPU the old node holds. With maxUnavailable 0 it would stay Pending
and the rollout would never finish, so those pools keep removing a
node first.

Assisted-by: LLM
Signed-off-by: Mattia Eleuteri <mattia@hidora.io>
2026-10-29 10:45:21 +01:00
..