Merge branch 'main' into ladithyav-fix-versions

lalitadithya · web-flow · commit 7b2b315cbb95 · 2025-10-18T18:19:44.000+05:30
diff --git a/.github/actions/build-container/action.yml b/.github/actions/build-container/action.yml
@@ -59,7 +59,6 @@ runs:
         SAFE_REF_NAME: ${{ inputs.safe_ref_name }}
         NVCR_CONTAINER_REPO: ${{ inputs.nvcr_container_repo }}
         NGC_ORG: ${{ inputs.container_org }}
-        PLATFORMS: 'linux/amd64'  # Only amd64 for faster PR validation
         DOCKER_BUILDKIT: 1
         BUILDX_CACHE_FROM: type=local,src=/tmp/.buildx-cache
         BUILDX_CACHE_TO: type=local,dest=/tmp/.buildx-cache-new,mode=max
diff --git a/.github/workflows/container-build-test.yml b/.github/workflows/container-build-test.yml
@@ -136,6 +136,8 @@ jobs:
         env:
           # Disable registry cache for pull requests to avoid permission issues
           DISABLE_REGISTRY_CACHE: ${{ (github.event_name == 'pull_request' || startsWith(github.ref, 'refs/heads/pull-request/')) && 'true' || 'false' }}
+          # Disable --load flag in CI builds (causes issues with multi-platform builds)
+          DOCKER_LOAD_ARG: ''
         with:
           safe_ref_name: ${{ needs.prepare-environment.outputs.safe_ref_name }}
           nvcr_container_repo: ${{ needs.prepare-environment.outputs.nvcr_container_repo }}
diff --git a/common.mk b/common.mk
@@ -64,6 +64,7 @@ BINARY_SOURCE ?= .
 
 # Docker configuration (can be overridden per module)
 DOCKER_EXTRA_ARGS ?=
+DOCKER_LOAD_ARG ?= --load
 HAS_DOCKER ?= 1
 
 # Module type configuration (Go=1, Python=0 - Python modules override targets)
@@ -182,7 +183,7 @@ docker-build: setup-buildx
 		$(CACHE_FROM_ARG) \
 		$(CACHE_TO_ARG) \
 		$(DOCKER_EXTRA_ARGS) \
-		--load \
+		$(DOCKER_LOAD_ARG) \
 		-t $(NVCR_CONTAINER_REPO)/$(NGC_ORG)/nvsentinel-$(MODULE_NAME):$(SAFE_REF_NAME) \
 		-f $(DOCKER_MODULE_PATH)/Dockerfile \
 		.
@@ -191,10 +192,10 @@ docker-build: setup-buildx
 docker-build-local: setup-buildx
 	@echo "Building Docker image for $(MODULE_NAME) (local, no remote cache)..."
 	cd $(REPO_ROOT) && docker buildx build \
-		--platform linux/amd64 \
+		--platform $(PLATFORMS) \
 		--network=host \
 		$(DOCKER_EXTRA_ARGS) \
-		--load \
+		$(DOCKER_LOAD_ARG) \
 		-t $(MODULE_NAME):local \
 		-f $(DOCKER_MODULE_PATH)/Dockerfile \
 		.
diff --git a/fault-quarantine-module/Dockerfile b/fault-quarantine-module/Dockerfile
@@ -14,7 +14,7 @@
 
 # syntax=docker/dockerfile:1.6
 
-ARG TAG=v3.1.12
+ARG TAG=v3.1.13
 ARG IMG=nvcr.io/nvidia/distroless/go
 
 FROM public.ecr.aws/docker/library/golang:1.24-bullseye AS builder
diff --git a/fault-remediation-module/Dockerfile b/fault-remediation-module/Dockerfile
@@ -14,7 +14,7 @@
 
 # syntax=docker/dockerfile:1.6
 
-ARG TAG=v3.1.12
+ARG TAG=v3.1.13
 ARG IMG=nvcr.io/nvidia/distroless/go
 
 FROM public.ecr.aws/docker/library/golang:1.24-bullseye AS builder
diff --git a/health-events-analyzer/Dockerfile b/health-events-analyzer/Dockerfile
@@ -14,7 +14,7 @@
 
 # syntax=docker/dockerfile:1.6
 
-ARG TAG=v3.1.12
+ARG TAG=v3.1.13
 ARG IMG=nvcr.io/nvidia/distroless/go
 
 FROM public.ecr.aws/docker/library/golang:1.24-bullseye AS builder
diff --git a/health-monitors/csp-health-monitor/Dockerfile b/health-monitors/csp-health-monitor/Dockerfile
@@ -14,7 +14,7 @@
 
 # syntax=docker/dockerfile:1.6
 
-ARG TAG=v3.1.12
+ARG TAG=v3.1.13
 ARG IMG=nvcr.io/nvidia/distroless/go
 
 FROM public.ecr.aws/docker/library/golang:1.24-bullseye AS builder
diff --git a/health-monitors/gpu-health-monitor/Makefile b/health-monitors/gpu-health-monitor/Makefile
@@ -117,7 +117,7 @@ docker-build-dcgm3: setup-buildx
 		$(CACHE_FROM_ARG_DCGM3) \
 		$(CACHE_TO_ARG_DCGM3) \
 		$(DOCKER_EXTRA_ARGS) \
-		--load \
+		$(DOCKER_LOAD_ARG) \
 		-t $(NVCR_CONTAINER_REPO)/$(NGC_ORG)/nvsentinel-$(MODULE_NAME):$(SAFE_REF_NAME)-dcgm-3.x \
 		--build-arg DCGM_VERSION=3.3.7-1-ubuntu22.04 \
 		--build-arg PYTHONPATH=/usr/local/dcgm/bindings/python3 \
@@ -133,7 +133,7 @@ docker-build-dcgm4: setup-buildx
 		$(CACHE_FROM_ARG_DCGM4) \
 		$(CACHE_TO_ARG_DCGM4) \
 		$(DOCKER_EXTRA_ARGS) \
-		--load \
+		$(DOCKER_LOAD_ARG) \
 		-t $(NVCR_CONTAINER_REPO)/$(NGC_ORG)/nvsentinel-$(MODULE_NAME):$(SAFE_REF_NAME)-dcgm-4.x \
 		-f $(DOCKER_MODULE_PATH)/Dockerfile \
 		.
diff --git a/labeler-module/Dockerfile b/labeler-module/Dockerfile
@@ -14,7 +14,7 @@
 
 # syntax=docker/dockerfile:1.6
 
-ARG TAG=v3.1.12
+ARG TAG=v3.1.13
 ARG IMG=nvcr.io/nvidia/distroless/go
 
 FROM public.ecr.aws/docker/library/golang:1.24-bullseye AS builder
diff --git a/node-drainer-module/Dockerfile b/node-drainer-module/Dockerfile
@@ -14,7 +14,7 @@
 
 # syntax=docker/dockerfile:1.6
 
-ARG TAG=v3.1.12
+ARG TAG=v3.1.13
 ARG IMG=nvcr.io/nvidia/distroless/go
 
 FROM public.ecr.aws/docker/library/golang:1.24-bullseye AS builder
diff --git a/node-drainer-module/Makefile b/node-drainer-module/Makefile
@@ -56,7 +56,7 @@ docker-build-with-cache: setup-buildx
 		$(CACHE_FROM_ARG) \
 		$(CACHE_TO_ARG) \
 		$(DOCKER_EXTRA_ARGS) \
-		--load \
+		$(DOCKER_LOAD_ARG) \
 		-t $(NVCR_CONTAINER_REPO)/$(NGC_ORG)/nvsentinel-$(MODULE_NAME):$(SAFE_REF_NAME) \
 		-f $(MODULE_NAME)/Dockerfile \
 		.
diff --git a/nvsentinel-log-collector/Makefile b/nvsentinel-log-collector/Makefile
@@ -9,6 +9,7 @@ SAFE_REF_NAME ?= $(shell echo $(CI_COMMIT_REF_NAME) | sed 's/\//-/g')
 SAFE_REF_NAME := $(if $(SAFE_REF_NAME),$(SAFE_REF_NAME),local)
 BUILDX_BUILDER ?= nvsentinel-builder
 PLATFORMS ?= linux/arm64,linux/amd64
+DOCKER_LOAD_ARG ?= --load
 
 # Cache configuration (can be disabled via environment variables)
 DISABLE_REGISTRY_CACHE ?= false
@@ -90,6 +91,7 @@ docker-build-log-collector: setup-buildx
 		--network=host \
 		$(CACHE_FROM_ARG_LOG) \
 		$(CACHE_TO_ARG_LOG) \
+		$(DOCKER_LOAD_ARG) \
 		-t $(NVCR_CONTAINER_REPO)/$(NGC_ORG)/nvsentinel-log-collector:$(SAFE_REF_NAME) \
 		.
 
@@ -103,6 +105,7 @@ docker-build-file-server-cleanup: setup-buildx
 		--network=host \
 		$(CACHE_FROM_ARG_CLEANUP) \
 		$(CACHE_TO_ARG_CLEANUP) \
+		$(DOCKER_LOAD_ARG) \
 		-f Dockerfile.cleanup \
 		-t $(NVCR_CONTAINER_REPO)/$(NGC_ORG)/nvsentinel-file-server-cleanup:$(SAFE_REF_NAME) \
 		.
diff --git a/platform-connectors/Dockerfile b/platform-connectors/Dockerfile
@@ -14,7 +14,7 @@
 
 # syntax=docker/dockerfile:1.6
 
-ARG TAG=v3.1.12
+ARG TAG=v3.1.13
 ARG IMG=nvcr.io/nvidia/distroless/go
 
 FROM public.ecr.aws/docker/library/golang:1.24-bullseye AS builder
diff --git a/tilt/simple-health-client/Dockerfile b/tilt/simple-health-client/Dockerfile
@@ -14,7 +14,7 @@
 
 # syntax=docker/dockerfile:1.6
 
-ARG TAG=v3.1.12
+ARG TAG=v3.1.13
 ARG IMG=nvcr.io/nvidia/distroless/go
 
 FROM public.ecr.aws/docker/library/golang:1.24-bullseye AS builder
