File tree Expand file tree Collapse file tree 6 files changed +12
-7
lines changed
health-monitors/gpu-health-monitor Expand file tree Collapse file tree 6 files changed +12
-7
lines changed Original file line number Diff line number Diff line change 5959 SAFE_REF_NAME : ${{ inputs.safe_ref_name }}
6060 NVCR_CONTAINER_REPO : ${{ inputs.nvcr_container_repo }}
6161 NGC_ORG : ${{ inputs.container_org }}
62- PLATFORMS : ' linux/amd64' # Only amd64 for faster PR validation
6362 DOCKER_BUILDKIT : 1
6463 BUILDX_CACHE_FROM : type=local,src=/tmp/.buildx-cache
6564 BUILDX_CACHE_TO : type=local,dest=/tmp/.buildx-cache-new,mode=max
Original file line number Diff line number Diff line change @@ -136,6 +136,8 @@ jobs:
136136 env :
137137 # Disable registry cache for pull requests to avoid permission issues
138138 DISABLE_REGISTRY_CACHE : ${{ (github.event_name == 'pull_request' || startsWith(github.ref, 'refs/heads/pull-request/')) && 'true' || 'false' }}
139+ # Disable --load flag in CI builds (causes issues with multi-platform builds)
140+ DOCKER_LOAD_ARG : ' '
139141 with :
140142 safe_ref_name : ${{ needs.prepare-environment.outputs.safe_ref_name }}
141143 nvcr_container_repo : ${{ needs.prepare-environment.outputs.nvcr_container_repo }}
Original file line number Diff line number Diff line change @@ -64,6 +64,7 @@ BINARY_SOURCE ?= .
6464
6565# Docker configuration (can be overridden per module)
6666DOCKER_EXTRA_ARGS ?=
67+ DOCKER_LOAD_ARG ?= --load
6768HAS_DOCKER ?= 1
6869
6970# Module type configuration (Go=1, Python=0 - Python modules override targets)
@@ -182,7 +183,7 @@ docker-build: setup-buildx
182183 $(CACHE_FROM_ARG ) \
183184 $(CACHE_TO_ARG ) \
184185 $(DOCKER_EXTRA_ARGS ) \
185- --load \
186+ $( DOCKER_LOAD_ARG ) \
186187 -t $(NVCR_CONTAINER_REPO ) /$(NGC_ORG ) /nvsentinel-$(MODULE_NAME ) :$(SAFE_REF_NAME ) \
187188 -f $(DOCKER_MODULE_PATH ) /Dockerfile \
188189 .
@@ -191,10 +192,10 @@ docker-build: setup-buildx
191192docker-build-local : setup-buildx
192193 @echo " Building Docker image for $( MODULE_NAME) (local, no remote cache)..."
193194 cd $(REPO_ROOT ) && docker buildx build \
194- --platform linux/amd64 \
195+ --platform $( PLATFORMS ) \
195196 --network=host \
196197 $(DOCKER_EXTRA_ARGS ) \
197- --load \
198+ $( DOCKER_LOAD_ARG ) \
198199 -t $(MODULE_NAME ) :local \
199200 -f $(DOCKER_MODULE_PATH ) /Dockerfile \
200201 .
Original file line number Diff line number Diff line change @@ -117,7 +117,7 @@ docker-build-dcgm3: setup-buildx
117117 $(CACHE_FROM_ARG_DCGM3 ) \
118118 $(CACHE_TO_ARG_DCGM3 ) \
119119 $(DOCKER_EXTRA_ARGS ) \
120- --load \
120+ $( DOCKER_LOAD_ARG ) \
121121 -t $(NVCR_CONTAINER_REPO ) /$(NGC_ORG ) /nvsentinel-$(MODULE_NAME ) :$(SAFE_REF_NAME ) -dcgm-3.x \
122122 --build-arg DCGM_VERSION=3.3.7-1-ubuntu22.04 \
123123 --build-arg PYTHONPATH=/usr/local/dcgm/bindings/python3 \
@@ -133,7 +133,7 @@ docker-build-dcgm4: setup-buildx
133133 $(CACHE_FROM_ARG_DCGM4 ) \
134134 $(CACHE_TO_ARG_DCGM4 ) \
135135 $(DOCKER_EXTRA_ARGS ) \
136- --load \
136+ $( DOCKER_LOAD_ARG ) \
137137 -t $(NVCR_CONTAINER_REPO ) /$(NGC_ORG ) /nvsentinel-$(MODULE_NAME ) :$(SAFE_REF_NAME ) -dcgm-4.x \
138138 -f $(DOCKER_MODULE_PATH ) /Dockerfile \
139139 .
Original file line number Diff line number Diff line change @@ -56,7 +56,7 @@ docker-build-with-cache: setup-buildx
5656 $(CACHE_FROM_ARG ) \
5757 $(CACHE_TO_ARG ) \
5858 $(DOCKER_EXTRA_ARGS ) \
59- --load \
59+ $( DOCKER_LOAD_ARG ) \
6060 -t $(NVCR_CONTAINER_REPO ) /$(NGC_ORG ) /nvsentinel-$(MODULE_NAME ) :$(SAFE_REF_NAME ) \
6161 -f $(MODULE_NAME ) /Dockerfile \
6262 .
Original file line number Diff line number Diff line change @@ -9,6 +9,7 @@ SAFE_REF_NAME ?= $(shell echo $(CI_COMMIT_REF_NAME) | sed 's/\//-/g')
99SAFE_REF_NAME := $(if $(SAFE_REF_NAME ) ,$(SAFE_REF_NAME ) ,local)
1010BUILDX_BUILDER ?= nvsentinel-builder
1111PLATFORMS ?= linux/arm64,linux/amd64
12+ DOCKER_LOAD_ARG ?= --load
1213
1314# Cache configuration (can be disabled via environment variables)
1415DISABLE_REGISTRY_CACHE ?= false
@@ -90,6 +91,7 @@ docker-build-log-collector: setup-buildx
9091 --network=host \
9192 $(CACHE_FROM_ARG_LOG ) \
9293 $(CACHE_TO_ARG_LOG ) \
94+ $(DOCKER_LOAD_ARG ) \
9395 -t $(NVCR_CONTAINER_REPO ) /$(NGC_ORG ) /nvsentinel-log-collector:$(SAFE_REF_NAME ) \
9496 .
9597
@@ -103,6 +105,7 @@ docker-build-file-server-cleanup: setup-buildx
103105 --network=host \
104106 $(CACHE_FROM_ARG_CLEANUP ) \
105107 $(CACHE_TO_ARG_CLEANUP ) \
108+ $(DOCKER_LOAD_ARG ) \
106109 -f Dockerfile.cleanup \
107110 -t $(NVCR_CONTAINER_REPO ) /$(NGC_ORG ) /nvsentinel-file-server-cleanup:$(SAFE_REF_NAME ) \
108111 .
You can’t perform that action at this time.
0 commit comments