@tduncan / Nvidia-Optimization-Stack-VLLM
Code Commits Issues Pulls Cloud Agent CI/CD
main
main
sh 109 lines 3.85 KB
Raw
1 #!/usr/bin/env bash
2 # SPDX-License-Identifier: Apache-2.0
3 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
4
5 set -euxo pipefail
6
7 # Surface the isolated snapshot-runtime install cost in the CI build log. The
8 # trap captures the status first: the date substitution below would reset it.
9 _snapshot_install_started=$(date +%s)
10 code=0
11 trap 'code=$?; echo "[TIMING] snapshot runtime install: $(( $(date +%s) - _snapshot_install_started )) s (exit $code)"' EXIT
12
13 readonly CRIU_VERSION="4.2.1"
14 readonly CRIU_SOURCE_SHA256="feffdf4638125ebb12d2434754f80a1d7bbba85a3e6bee98c216f88fb99a5d96"
15 readonly DYNAMO_CHECKPOINT_HELPER_COMMIT="fdf25efad60f696a73393caf341c527a93d84190"
16 readonly DYNAMO_CHECKPOINT_HELPER_SHA256="7b8ea21baecb1da729202c32e0fecbbbafb789d65e5c1c0649ea253ad09b8f82"
17 readonly DYNAMO_LICENSE_SHA256="aeeca3d74a13b91d17ab9211c407b8d94a77bf8d9e5750c467818d3066b8adcf"
18
19 # Keep the compiler and NL headers required by runtime JIT and RDMA.
20 snapshot_shared_build_deps=(
21 build-essential
22 libnl-3-dev
23 libnl-route-3-dev
24 )
25 snapshot_build_deps=(
26 pkg-config
27 libbsd-dev
28 libcap-dev
29 libnet1-dev
30 libprotobuf-dev
31 libprotobuf-c-dev
32 protobuf-c-compiler
33 protobuf-compiler
34 python3-protobuf
35 libgnutls28-dev
36 libnftables-dev
37 uuid-dev
38 )
39 snapshot_runtime_deps=(
40 libbsd0
41 libcap2
42 libnet1
43 libnl-3-200
44 libnl-route-3-200
45 libprotobuf-c1
46 libgnutls30t64
47 libnftables1
48 iproute2
49 iptables
50 procps
51 uuid-runtime
52 util-linux
53 )
54
55 apt-get update -y
56 apt-get install -y --no-install-recommends \
57 "${snapshot_shared_build_deps[@]}" \
58 "${snapshot_build_deps[@]}" \
59 "${snapshot_runtime_deps[@]}"
60
61 snapshot_build_dir="$(mktemp -d)"
62 curl -fsSL \
63 "https://github.com/checkpoint-restore/criu/archive/refs/tags/v${CRIU_VERSION}.tar.gz" \
64 -o "${snapshot_build_dir}/criu.tar.gz"
65 echo "${CRIU_SOURCE_SHA256} ${snapshot_build_dir}/criu.tar.gz" | sha256sum -c -
66 mkdir "${snapshot_build_dir}/criu"
67 tar -xzf "${snapshot_build_dir}/criu.tar.gz" \
68 --strip-components=1 -C "${snapshot_build_dir}/criu"
69 make -C "${snapshot_build_dir}/criu" -j"${MAX_JOBS:-$(nproc)}" criu cuda_plugin
70 install -D -m 0755 "${snapshot_build_dir}/criu/criu/criu" \
71 /usr/local/sbin/criu
72 install -D -m 0755 \
73 "${snapshot_build_dir}/criu/plugins/cuda/cuda_plugin.so" \
74 /usr/local/lib/criu/cuda_plugin.so
75
76 helper_url="https://raw.githubusercontent.com/ai-dynamo/dynamo/${DYNAMO_CHECKPOINT_HELPER_COMMIT}/deploy/snapshot/cmd/cuda-checkpoint-helper/main.c"
77 curl -fsSL "${helper_url}" -o "${snapshot_build_dir}/cuda-checkpoint.c"
78 echo "${DYNAMO_CHECKPOINT_HELPER_SHA256} ${snapshot_build_dir}/cuda-checkpoint.c" \
79 | sha256sum -c -
80 gcc -O2 -Wall -Wextra \
81 -I/usr/local/cuda/include \
82 -L/usr/local/cuda/lib64/stubs \
83 "${snapshot_build_dir}/cuda-checkpoint.c" -lcuda \
84 -o /usr/local/sbin/cuda-checkpoint
85
86 license_url="https://raw.githubusercontent.com/ai-dynamo/dynamo/${DYNAMO_CHECKPOINT_HELPER_COMMIT}/LICENSE"
87 curl -fsSL "${license_url}" -o "${snapshot_build_dir}/DYNAMO-LICENSE"
88 echo "${DYNAMO_LICENSE_SHA256} ${snapshot_build_dir}/DYNAMO-LICENSE" \
89 | sha256sum -c -
90
91 install -D -m 0644 "${snapshot_build_dir}/criu.tar.gz" \
92 /usr/share/doc/criu/source/criu.tar.gz
93 install -D -m 0644 "${snapshot_build_dir}/criu/COPYING" \
94 /usr/share/doc/criu/COPYING
95 install -D -m 0644 "${snapshot_build_dir}/cuda-checkpoint.c" \
96 /usr/share/doc/cuda-checkpoint-helper/main.c
97 install -D -m 0644 "${snapshot_build_dir}/DYNAMO-LICENSE" \
98 /usr/share/doc/cuda-checkpoint-helper/LICENSE
99
100 criu --version
101 ln -s /usr/local/cuda/lib64/stubs/libcuda.so \
102 "${snapshot_build_dir}/libcuda.so.1"
103 LD_LIBRARY_PATH="${snapshot_build_dir}" \
104 cuda-checkpoint --help | grep -q -- "--action"
105 test -x /usr/local/lib/criu/cuda_plugin.so
106
107 apt-get purge -y --auto-remove "${snapshot_build_deps[@]}"
108 rm -rf "${snapshot_build_dir}"
109 rm -rf /var/lib/apt/lists/*