From e497908b3f1ef2213ed1a0f1b70ce737e61b6768 Mon Sep 17 00:00:00 2001
From: Blake Devcich <blake.devcich@hpe.com>
Date: Tue, 11 Feb 2025 09:43:45 -0600
Subject: [PATCH] Add lustre support to MPIFileUtils

`dcp` has a `--grouplock` option and that requires the lustre client.
Lustre is now being built in the builder stage of the image and is used
to enable lustre build support in MPIFileUtils and also runtime support.

This image's OS is debian bookworm, which has deprecated `dpatch` - a
required build dependency for older versions of lustre. The latest cray
release of lustre 2.15 is being used, where `dpatch` is no longer
required.

Signed-off-by: Blake Devcich <blake.devcich@hpe.com>
---
 .github/workflows/main.yml |  2 ++
 Dockerfile                 | 47 +++++++++++++++++++++++++++++---------
 Makefile                   | 25 ++++++++++++++++----
 build_lustre.sh            | 26 +++++++++++++++++++++
 4 files changed, 84 insertions(+), 16 deletions(-)
 create mode 100755 build_lustre.sh

diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml
index 6e4009c..094e0fb 100644
--- a/.github/workflows/main.yml
+++ b/.github/workflows/main.yml
@@ -77,6 +77,7 @@ jobs:
         push: true
         target: production
         tags: ${{ steps.meta.outputs.tags }}
+        build-args: LUSTRE_VERSION=cray-2.15.B19
 
     - name: "Build the final Docker image in nektos/act"
       # The docker/build-push-action does actions/checkout, and that can't be
@@ -152,6 +153,7 @@ jobs:
         push: true
         target: debug
         tags: ${{ steps.meta.outputs.tags }}
+        build-args: LUSTRE_VERSION=cray-2.15.B19
 
   create_release:
     needs: production
diff --git a/Dockerfile b/Dockerfile
index bf60d94..e979063 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -18,8 +18,7 @@
 # These ARGs must be before the first FROM. This allows them to be valid for
 # use in FROM instructions.
 ARG MPI_OPERATOR_VERSION=0.6.0
-# See https://www.open-mpi.org/software/ompi/v4.1/ for releases and their
-# checkums.
+# See https://www.open-mpi.org/software/ompi/v4.1/ for releases and their checksums.
 ARG OPENMPI_VERSION=4.1.7
 ARG OPENMPI_MD5=787d2bc8b3db336db97c34236934b3df
 
@@ -27,17 +26,35 @@ FROM mpioperator/openmpi-builder:v$MPI_OPERATOR_VERSION AS builder
 
 ARG OPENMPI_VERSION
 ARG OPENMPI_MD5
+ARG LUSTRE_VERSION
 ENV OPENMPI_VERSION=$OPENMPI_VERSION
 ENV OPENMPI_MD5=$OPENMPI_MD5
+ENV LUSTRE_VERSION=$LUSTRE_VERSION
 
 RUN apt-get update && apt-get install -y \
     ca-certificates \
     wget tar make gcc cmake perl libbz2-dev pkg-config openssl libssl-dev libcap-dev \
     git libattr1-dev \
     openssh-server openssh-client  \
+    linux-headers-generic \
+    libtool libyaml-dev ed libreadline-dev libsnmp-dev mpi-default-dev libselinux-dev libncurses5-dev libncurses-dev \
+    bison flex gnupg libelf-dev gcc libssl-dev bc wget bzip2 build-essential udev kmod cpio module-assistant \
+    libmount-dev libnl-genl-3-dev \
     && rm -rf /var/lib/apt/lists/* \
     && update-ca-certificates
 
+# Build lustre to include in mpifileutils
+COPY build_lustre.sh /build_lustre.sh
+RUN /build_lustre.sh ${LUSTRE_VERSION}
+
+# Create MPI File Utils dependencies directory
+RUN mkdir -p /deps /mfu
+WORKDIR /deps
+
+# Stash the lustre libraries to make it easier to copy out in later stages
+RUN mkdir -p /deps/lustre/lib \
+    && cp -r /usr/lib/*lustre* /deps/lustre/lib/
+
 # Remove the OS binary of openmpi and build from source
 RUN apt-get remove -y openmpi-bin
 RUN wget --no-check-certificate https://download.open-mpi.org/release/open-mpi/v4.1/openmpi-$OPENMPI_VERSION.tar.gz
@@ -45,36 +62,34 @@ RUN [ $(md5sum openmpi-$OPENMPI_VERSION.tar.gz | awk '{print $1}') = "$OPENMPI_M
 RUN gunzip -c openmpi-$OPENMPI_VERSION.tar.gz | tar xf - \
     && cd openmpi-$OPENMPI_VERSION \
     && ./configure --prefix=/opt/openmpi-$OPENMPI_VERSION \
-    && make all install
+    && make -j $(nproc) all \
+    && make install
 RUN cp -r /opt/openmpi-$OPENMPI_VERSION/* /usr
 
 # Build and install MPI File Utils and all dependencies
-RUN mkdir -p /deps /mfu
-WORKDIR /deps
-
 RUN wget https://github.com/hpc/libcircle/releases/download/v0.3/libcircle-0.3.0.tar.gz \
     && tar xfz libcircle-0.3.0.tar.gz \
     && cd libcircle-0.3.0 \
     && ./configure --prefix=/deps/libcircle/lib \
-    && make install
+    && make -j $(nproc) install
 
 RUN wget https://github.com/libarchive/libarchive/releases/download/v3.5.3/libarchive-3.5.3.tar.gz \
     && tar xfz libarchive-3.5.3.tar.gz \
     && cd libarchive-3.5.3 \
     && ./configure --prefix=/deps/libarchive/lib \
-    && make install
+    && make -j $(nproc) install
 
 RUN wget https://github.com/llnl/lwgrp/releases/download/v1.0.3/lwgrp-1.0.3.tar.gz \
     && tar xfz lwgrp-1.0.3.tar.gz \
     && cd lwgrp-1.0.3 \
     && ./configure --prefix=/deps/lwgrp/lib \
-    && make install
+    && make -j $(nproc) install
 
 RUN wget https://github.com/llnl/dtcmp/releases/download/v1.1.1/dtcmp-1.1.1.tar.gz \
     && tar xfz dtcmp-1.1.1.tar.gz \
     && cd dtcmp-1.1.1 \
     && ./configure --prefix=/deps/dtcmp/lib --with-lwgrp=/deps/lwgrp/lib \
-    && make install
+    && make -j $(nproc) install
 
 # Until a new release of mpifileutils is cut, we need to use a tagged commit on
 # our fork to include our dcp changes (i.e. --uid, --gid)
@@ -86,6 +101,7 @@ RUN git clone --depth 1 https://github.com/nearnodeflash/mpifileutils.git --bran
     -DWITH_DTCMP_PREFIX=/deps/dtcmp/lib \
     -DWITH_LibArchive_PREFIX=/deps/libarchive/lib \
     -DCMAKE_INSTALL_PREFIX=/mfu \
+    -DENABLE_LUSTRE=ON \
     && make install
 
 
@@ -104,6 +120,7 @@ RUN cd build \
     -DWITH_DTCMP_PREFIX=/deps/dtcmp/lib \
     -DWITH_LibArchive_PREFIX=/deps/libarchive/lib \
     -DCMAKE_INSTALL_PREFIX=/mfu \
+    -DENABLE_LUSTRE=ON \
     -DCMAKE_BUILD_TYPE=Debug \
     && make install
 
@@ -113,7 +130,8 @@ RUN [ $(md5sum openmpi-$OPENMPI_VERSION.tar.gz | awk '{print $1}') = "$OPENMPI_M
 RUN gunzip -c openmpi-$OPENMPI_VERSION.tar.gz | tar xf - \
     && cd openmpi-$OPENMPI_VERSION \
     && ./configure --prefix=/opt/openmpi-$OPENMPI_VERSION-debug --enable-debug \
-    && make all install
+    && make -j $(nproc) all \
+    && make install
 
 ###############################################################################
 FROM mpioperator/openmpi:v$MPI_OPERATOR_VERSION AS production
@@ -129,12 +147,19 @@ COPY --from=builder /deps/libarchive/lib/ /usr
 COPY --from=builder /deps/lwgrp/lib/ /usr
 COPY --from=builder /deps/dtcmp/lib/ /usr
 
+COPY --from=builder /deps/lustre/lib/ /usr/lib
+
 COPY --from=builder /mfu/ /usr
 
 RUN apt-get remove -y openmpi-bin
 COPY --from=builder /opt/openmpi-$OPENMPI_VERSION /opt/openmpi-$OPENMPI_VERSION
 RUN cp -r /opt/openmpi-$OPENMPI_VERSION/* /usr && rm -rf /openmpi*
 
+# libreadline8 is necessary for dcp with lustre support
+RUN apt-get update && apt-get install -y \
+    libreadline8 \
+    && rm -rf /var/lib/apt/lists/*
+
 # Remove timezone configuration so we can inherit from host
 RUN rm -rf /etc/timezone && rm -rf /etc/localtime
 
diff --git a/Makefile b/Makefile
index e786eef..e8fc5d5 100644
--- a/Makefile
+++ b/Makefile
@@ -18,24 +18,40 @@
 # NOTE: git-version-gen will generate a value for VERSION, unless you override it.
 IMAGE_TAG_BASE ?= ghcr.io/nearnodeflash/nnf-mfu
 
+# CONTAINER_TOOL defines the container tool to be used for building images.
+# Be aware that the target commands are only tested with Docker which is
+# scaffolded by default. However, you might want to replace it to use other
+# tools. (i.e. podman)
+CONTAINER_TOOL ?= docker
+
+# Which tag to pull from the cray lustre source code.
+# **Note:** This must also be set in .github/workflows/main.yaml for both
+# production and debug docker builds.
+CRAY_LUSTRE_VERSION ?= cray-2.15.B19
+
+# Only support linux/amd64 builds. This reduces complexity with installing
+# lustre, which is an option enabled for mpifileutils.
+PLATFORM ?= linux/amd64
+
 docker-build: VERSION ?= $(shell cat .version)
 docker-build: TARGET ?= production
 docker-build: .version
-	docker build --target $(TARGET) -t $(IMAGE_TAG_BASE):$(VERSION) .
+	${CONTAINER_TOOL} build --platform=$(PLATFORM) --target $(TARGET) --build-arg LUSTRE_VERSION=${CRAY_LUSTRE_VERSION} -t $(IMAGE_TAG_BASE):$(VERSION) .
 
 docker-build-debug: VERSION ?= $(shell cat .version)
+docker-build-debug: TARGET ?= debug
 docker-build-debug: IMAGE_TAG_BASE := $(IMAGE_TAG_BASE)-debug
 docker-build-debug: .version
-	docker build --target debug -t $(IMAGE_TAG_BASE):$(VERSION) .
+	${CONTAINER_TOOL} build --platform=$(PLATFORM) --target $(TARGET) --build-arg LUSTRE_VERSION=${CRAY_LUSTRE_VERSION} -t $(IMAGE_TAG_BASE):$(VERSION) .
 
 docker-push: VERSION ?= $(shell cat .version)
 docker-push: .version
-	docker push $(IMAGE_TAG_BASE):$(VERSION)
+	${CONTAINER_TOOL} push $(IMAGE_TAG_BASE):$(VERSION)
 
 docker-push-debug: VERSION ?= $(shell cat .version)
 docker-push-debug: IMAGE_TAG_BASE := $(IMAGE_TAG_BASE)-debug
 docker-push-debug: .version
-	docker push $(IMAGE_TAG_BASE):$(VERSION)
+	${CONTAINER_TOOL} push $(IMAGE_TAG_BASE):$(VERSION)
 
 kind-push: VERSION ?= $(shell cat .version)
 kind-push: .version
@@ -54,4 +70,3 @@ kind-push-debug: .version
 
 clean:
 	rm -f .version
-  
diff --git a/build_lustre.sh b/build_lustre.sh
new file mode 100755
index 0000000..95b6768
--- /dev/null
+++ b/build_lustre.sh
@@ -0,0 +1,26 @@
+#!/bin/bash
+
+set -ex
+
+if [[ $1 == "" ]]; then
+    echo "Usage: $0 <Cray lustre tag e.g. cray-2.15.B19>"
+    exit 1
+fi
+CRAY_LUSTRE_VERSION=$1
+
+# Determine kernel version and set variables
+ls -l /usr/src/
+KERNEL_FLAVOR=$(ls /lib/modules | head -1 | tr '-' '\n' | tail -1)
+KERNEL_BASE_VER=$(ls /lib/modules | head -1 | grep -oP '\d+\.\d+\.\d+-\d+')
+# LINUX_DIR=$(ls -d -1 /usr/src/linux-headers-${KERNEL_BASE_VER})
+LINUX_DIR=$(ls -d -1 /usr/src/linux-headers-${KERNEL_BASE_VER}-${KERNEL_FLAVOR})
+
+git clone --depth=1 https://github.com/Cray/lustre.git
+cd lustre
+git checkout $CRAY_LUSTRE_VERSION
+sh autogen.sh
+./configure --disable-server --enable-client --disable-tests --enable-mpitests=no \
+    --disable-gss-keyring --enable-gss=no \
+    --with-linux=${LINUX_DIR}
+make -j "$(nproc || true)"
+make install