Merge branch 'main' into feat/drainGPUPods

Author: lalitadithya

docs/designs/040-external-remediation-request.md

@@ -49,7 +49,6 @@ RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
     apt-get install -y --no-install-recommends \
         apt-transport-https \
         ca-certificates \
-        gnupg \
         python3-pip && \
     rm -rf /var/lib/apt/lists/*
 

health-monitors/gpu-health-monitor/Dockerfile

@@ -20,7 +20,7 @@
 FROM public.ecr.aws/docker/library/ubuntu:24.04
 
 ARG DEBIAN_FRONTEND=noninteractive
-ARG KUBECTL_VERSION=v1.34.1
+ARG KUBECTL_VERSION=v1.34.8
 
 RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
     --mount=type=cache,target=/var/lib/apt,sharing=locked \

log-collector/Dockerfile

@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-FROM public.ecr.aws/docker/library/python:3.13-alpine AS build
+FROM public.ecr.aws/docker/library/python:3.14-alpine AS build
 
 RUN --mount=type=cache,target=/root/.cache/pip \
     pip install poetry==2.3.3 poetry-plugin-export==1.10.0
@@ -31,7 +31,7 @@ RUN --mount=type=cache,target=/tmp/poetry_cache \
     poetry build --format wheel 
 RUN poetry export --format requirements.txt --output constraints.txt --without-hashes
 
-FROM public.ecr.aws/docker/library/python:3.13-alpine AS runtime
+FROM public.ecr.aws/docker/library/python:3.14-alpine AS runtime
 
 ENV PYTHONUNBUFFERED=1
 

log-collector/Dockerfile.cleanup

@@ -47,7 +47,6 @@ RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
     apt-get install -y --no-install-recommends \
         apt-transport-https \
         ca-certificates \
-        gnupg \
         python3-pip && \
     rm -rf /var/lib/apt/lists/*
 

preflight-checks/dcgm-diag/Dockerfile

@@ -22,8 +22,8 @@
 
 # PyTorch version determines CUDA runtime version - must match cluster's GPU driver
 # See: https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html#id4
-#   26.03-py3 = CUDA 13.2 (Ubuntu 24.04, Python 3.12)
-ARG PYTORCH_VERSION=26.03-py3
+#   26.04-py3 = CUDA 13.2.1 (Ubuntu 24.04, Python 3.12)
+ARG PYTORCH_VERSION=26.04-py3
 ARG VERSION="0.1.0"
 
 # =============================================================================
@@ -67,6 +67,8 @@ FROM nvcr.io/nvidia/pytorch:${PYTORCH_VERSION} AS runtime
 
 ARG VERSION
 
+RUN rm -f /usr/local/bin/uv /usr/local/bin/uvx
+
 # Remove the NGC image's bundled EFA/OFI-NCCL stack and stale RDMA libs.
 # On AWS, the host's /opt/amazon is mounted at runtime and provides the
 # EFA libs (libfabric, ofi-nccl plugin) that match the host's kernel

preflight-checks/nccl-allreduce/Dockerfile

@@ -22,7 +22,7 @@ include ../../make/common.mk
 include ../../make/python.mk
 include ../../make/docker.mk
 
-PYTORCH_VERSION ?= 25.01-py3
+PYTORCH_VERSION ?= 26.04-py3
 IMAGE_NAME := preflight-nccl-allreduce
 
 .PHONY: all

preflight-checks/nccl-allreduce/Makefile

@@ -14,8 +14,8 @@
 
 # PyTorch version determines CUDA runtime version - must match cluster's GPU driver
 # See: https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html#id4
-#   26.03-py3 = CUDA 13.2 (Ubuntu 24.04, Python 3.12)
-ARG PYTORCH_VERSION=26.03-py3
+#   26.04-py3 = CUDA 13.2.1 (Ubuntu 24.04, Python 3.12)
+ARG PYTORCH_VERSION=26.04-py3
 ARG NCCL_TESTS_VERSION=v2.13.11
 
 # =============================================================================
@@ -67,6 +67,8 @@ FROM nvcr.io/nvidia/pytorch:${PYTORCH_VERSION} AS runtime
 
 ARG NCCL_TESTS_VERSION
 
+RUN rm -f /usr/local/bin/uv /usr/local/bin/uvx
+
 # Build nccl-tests from source
 # Uses NCCL bundled with PyTorch image for version compatibility
 # NVCC_GENCODE targets modern data center GPUs: