Final 1.0

Dockerfile

@@ -6,14 +6,32 @@ FROM ${BASE_REGISTRY}/${BASE_IMAGE}:${BASE_TAG}
 
 ARG VERSION=23.03
 
-#Non-root user
 RUN useradd -u 1818 triton
 
-#ubi 8 rpms
-RUN dnf upgrade && dnf clean all
+ENV CUDA_VERSION 11.2.1
+
+RUN dnf upgrade
+RUN dnf -y install pkg-config libgomp libpng python38
+RUN dnf clean all
+
+# nvidia-docker 1.0
+RUN echo "/usr/local/nvidia/lib" >> /etc/ld.so.conf.d/nvidia.conf && \
+    echo "/usr/local/nvidia/lib64" >> /etc/ld.so.conf.d/nvidia.conf
+
+#ENV
+ENV PATH /usr/local/nvidia/bin:/usr/local/cuda/bin:${PATH}
+ENV LD_LIBRARY_PATH /usr/local/nvidia/lib:/usr/local/nvidia/lib64:/opt/tritonserver/lib
+
+#Link
+RUN ln -s cuda-11.2 /usr/local/cuda
+
+# nvidia-container-runtime
+ENV NVIDIA_VISIBLE_DEVICES all
+ENV NVIDIA_DRIVER_CAPABILITIES compute,utility
+ENV NVIDIA_REQUIRE_CUDA "cuda>=11.2 brand=tesla,driver>=418,driver<419 brand=tesla,driver>=440,driver<441 driver>=450,driver<451"
 
 #Args
-#ARG TARBALL1=whls.tar.gz
+ARG TARBALL1=OCV.tar
 ARG TARBALL2=rpms.tar.gz
 ARG TARBALL6=libtensorflow_cc.so.1.tar
 ARG TARBALL7=libtensorflow_cc.so.tar
@@ -23,23 +41,29 @@ ARG TARBALL10=libtensorflow_triton.so.1.tar
 ARG TARBALL11=libtensorflow_triton.so.tar
 ARG TARBALL12=libtriton_tensorflow1.so.tar
 ARG TARBALL13=tritonserverwobe.tar.gz
-ARG TARBALLC0=cuda-parts0.tar
-ARG TARBALLC1=cuda-parts1.tar
-ARG TARBALLC2=cuda-parts2.tar
 
-#"${TARBALL1}",
-COPY ["${TARBALL2}", "/opt/"]
+ARG TARBALLN1=nvidia1.tar.gz
+ARG TARBALLN2=nvidia2.tar.gz
+ARG TARBALLN3=nvidia3.tar.gz
+ARG TARBALLN4=nvidia4.tar.gz
 
-COPY ["${TARBALL6}", "${TARBALL7}", "${TARBALL8}", "${TARBALL9}", \
-       "${TARBALL10}", "${TARBALL11}", "${TARBALL12}", "${TARBALL13}", \
-       "/opt/"]
+COPY ["${TARBALL1}", "/"]
 
-COPY ["${TARBALLC0}", "${TARBALLC1}", "${TARBALLC2}", "/"]
+COPY ["${TARBALLN1}", "${TARBALLN2}", "${TARBALLN3}", "${TARBALLN4}", \
+  "${TARBALL6}", "${TARBALL7}", "${TARBALL8}", "${TARBALL9}", \
+  "${TARBALL10}", "${TARBALL11}", "${TARBALL12}", "${TARBALL13}", \
+  "${TARBALL2}","/opt/"]
+
+WORKDIR /
+RUN tar -xvf ./${TARBALL1} && rm -fv ./${TARBALL1}
 
 WORKDIR /opt
 
 #Untar
-#RUN tar -zxf ./${TARBALL1} && rm -fv ./${TARBALL1}
+RUN tar -xvf ./${TARBALLN1} && rm -fv ./${TARBALLN1}
+RUN tar -xvf ./${TARBALLN2} && rm -fv ./${TARBALLN2}
+RUN tar -xvf ./${TARBALLN3} && rm -fv ./${TARBALLN3}
+RUN tar -xvf ./${TARBALLN4} && rm -fv ./${TARBALLN4}
 RUN tar -zxf ./${TARBALL2} && rm -fv ./${TARBALL2}
 RUN tar -xvf ./${TARBALL6} && rm -fv ./${TARBALL6}
 RUN tar -xvf ./${TARBALL7} && rm -fv ./${TARBALL7}
@@ -55,24 +79,37 @@ RUN mv libtensorflow_cc.so.1 libtensorflow_cc.so libtensorflow_framework.so.1 \
  libtensorflow_framework.so libtensorflow_triton.so.1 libtensorflow_triton.so \
  libtriton_tensorflow1.so LICENSE.tensorflow tritonserver/backends/tensorflow1/
 
-WORKDIR /
-
-#split cuda folder
-RUN tar -xvf ./${TARBALLC0} && rm -fv ./${TARBALLC0}
-RUN tar -xvf ./${TARBALLC1} && rm -fv ./${TARBALLC1}
-RUN tar -xvf ./${TARBALLC2} && rm -fv ./${TARBALLC2}
-
 WORKDIR /opt/baserpms
 
 #re2
 RUN rpm -i re2-20190801-1.el8.x86_64.rpm
 
+ENV NCCL_VERSION 2.8.4
+ENV LIBRARY_PATH /usr/local/cuda/lib64/stubs
+
+ENV CUDNN_VERSION 8.1.0.77
+
+WORKDIR /opt/nvidia1
+RUN rpm --nodeps -i *.rpm
+
+WORKDIR /opt/nvidia2
+RUN rpm --nodeps -i *.rpm
+
+WORKDIR /opt/nvidia3
+RUN rpm --nodeps -i *.rpm
+
+WORKDIR /opt/nvidia4
+RUN rpm --nodeps -i *.rpm
+
 WORKDIR /
 
-RUN rm -r usr/local/cuda-11.2/libnvvp
+#RUN rm -r usr/local/cuda-11.2/libnvvp
 
 RUN mkdir /models
 
+RUN ln -s /usr/local/cuda-11.2/targets/x86_64-linux/lib/stubs/libnvidia-ml.so \
+ /usr/local/cuda-11.2/targets/x86_64-linux/lib/stubs/libnvidia-ml.so.1
+
 USER triton
 
 EXPOSE 8000

hardening_manifest.yaml

@@ -32,7 +32,6 @@ labels:
   # Product the image belongs to for grouping multiple images
   mil.dso.ironbank.product.name: "triton-inference-server"
 
-
 resources:
   - url: "https://iron-bank-tars.s3-us-gov-west-1.amazonaws.com/baserpms.tar.gz"
     filename: "rpms.tar.gz"
@@ -79,21 +78,31 @@ resources:
     validation:
       type: "sha256"
       value: "a14c2b8ff2495299f99ccce9f58f53b75e06a54496e298dba7960d0646598851"
-  - url: "https://iron-bank-tars.s3-us-gov-west-1.amazonaws.com/cuda-parts0.tar"
-    filename: "cuda-parts0.tar"
+  - url: "https://iron-bank-tars.s3-us-gov-west-1.amazonaws.com/nvidia1.tar.gz"
+    filename: "nvidia1.tar.gz"
+    validation:
+      type: "sha256"
+      value: "81e9ed95f6aeb17b54f1d610fe644246c7888ded1dea396bc6c31b67998b2d3c"
+  - url: "https://iron-bank-tars.s3-us-gov-west-1.amazonaws.com/nvidia2.tar.gz"
+    filename: "nvidia2.tar.gz"
+    validation:
+      type: "sha256"
+      value: "5dab39a8348d2ff16999ec1fbbb43015f9ab7c00bbbf3ede544aee360bc01b1c"
+  - url: "https://iron-bank-tars.s3-us-gov-west-1.amazonaws.com/steve3.tar.gz"
+    filename: "nvidia3.tar.gz"
     validation:
       type: "sha256"
-      value: "6f46d963e08f7a151c8a420a2b75c49a1bbd67fc00e77dad6967b910f4c52ed6"
-  - url: "https://iron-bank-tars.s3-us-gov-west-1.amazonaws.com/cuda-parts1.tar"
-    filename: "cuda-parts1.tar"
+      value: "738de83c902922adf0137fe8e86b9757854a63dbe42fc7e17c84b065ad3a92ae"
+  - url: "https://iron-bank-tars.s3-us-gov-west-1.amazonaws.com/nvidia4.tar.gz"
+    filename: "nvidia4.tar.gz"
     validation:
       type: "sha256"
-      value: "d5a11547d71ad6c735ab3ead231f4d39f0e53dd20cd4b14f931301908e8e374c"
-  - url: "https://iron-bank-tars.s3-us-gov-west-1.amazonaws.com/cuda-parts2.tar"
-    filename: "cuda-parts2.tar"
+      value: "f06dcff4d63af8b522a04078991a8fabd020b613994c821b40f777d51aaf1fda"
+  - url: "https://iron-bank-tars.s3-us-gov-west-1.amazonaws.com/OCV.tar"
+    filename: "OCV.tar"
     validation:
       type: "sha256"
-      value: "51d4a76ac59a76e4bee1cab5632d1a452276520153d8098d179a55b729d278f7"
+      value: "02a342f9eb90f09338f0cffa19378befd09e9e0ed24da93b239c65d40ad87cf6"
 
 maintainers:
   - name: "Daniel Mumpton"