Skip to content

Commit d4b716a

Browse files
ocaisaocaisa
authored
Merge pull request #922 from Darkless012/2023.06-software.eessi.io
Updated link_nvidia_host_libraries.sh for better edge case handling
2 parents abf439c + 27d940c commit d4b716a

File tree

4 files changed

+735
-199
lines changed

4 files changed

+735
-199
lines changed

‎.github/workflows/tests_link_nvidia_host_libraries.yml‎

Lines changed: 196 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,196 @@
1+
# documentation: https://help.github.com/en/articles/workflow-syntax-for-github-actions
2+
name: Test NVIDIA Host Libraries Linking
3+
on:
4+
push:
5+
branches:
6+
- '*-software.eessi.io' # Matches any branch ending with '-software.eessi.io'
7+
pull_request:
8+
paths:
9+
- 'scripts/gpu_support/nvidia/link_nvidia_host_libraries.sh' # PR changes only relevant for this specific file
10+
- '.github/workflows/tests_link_nvidia_host_libraries.yml' # Also test when changing the tests themselves
11+
permissions:
12+
contents: read # to fetch code (actions/checkout)
13+
jobs:
14+
build:
15+
runs-on: ubuntu-24.04
16+
steps:
17+
- name: checkout
18+
uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938 # v4.2.0
19+
20+
- name: Initialize EESSI
21+
uses: eessi/github-action-eessi@v3
22+
23+
- name: Setup mock NVIDIA libraries
24+
run: |
25+
# Run the script to create mock libraries
26+
chmod +x ./tests/nvidia-libs/mock-nvidia-libs.sh
27+
echo ">>> Running ./tests/nvidia-libs/mock-nvidia-libs.sh"
28+
./tests/nvidia-libs/mock-nvidia-libs.sh
29+
30+
# Create symlink to override host's ldconfig, since the script tries to use /sbin/ldconfig first.
31+
echo "Symlinking ldconfig to /sbin/ldconfig"
32+
sudo ln -sf /tmp/ldconfig/ldconfig /sbin/ldconfig
33+
34+
# Verify the symlink was created correctly
35+
ls -la /sbin/ldconfig
36+
37+
- name: Setup mock nvidia-smi
38+
run: |
39+
# Create directory for mock nvidia-smi
40+
mkdir -p /tmp/nvidia-bin
41+
42+
# Copy the mock script
43+
chmod +x ./tests/nvidia-libs/mock-nvidia-smi.sh
44+
echo ">>> Copying ./tests/nvidia-libs/mock-nvidia-smi.sh"
45+
cp ./tests/nvidia-libs/mock-nvidia-smi.sh /tmp/nvidia-bin/nvidia-smi
46+
47+
# Add to PATH
48+
echo "Updating PATH"
49+
echo "PATH=/tmp/nvidia-bin:$PATH" >> $GITHUB_ENV
50+
51+
- name: Test LD_PRELOAD mode
52+
run: |
53+
echo ">>> Testing LD_PRELOAD mode"
54+
55+
# Run the script with LD_PRELOAD option (shouldn't create symlinks)
56+
output=$(./scripts/gpu_support/nvidia/link_nvidia_host_libraries.sh --show-ld-preload || { echo "Script returned non-zero: $?"; echo $output; exit 1; })
57+
58+
echo "$output"
59+
60+
echo ">>> Running checks"
61+
62+
# Check for expected outputs
63+
echo "$output" | grep "export EESSI_GPU_COMPAT_LD_PRELOAD=" || { echo "EESSI_GPU_COMPAT_LD_PRELOAD not found in output"; exit 1; }
64+
echo "$output" | grep "export EESSI_GPU_LD_PRELOAD=" || { echo "EESSI_GPU_LD_PRELOAD not found in output"; exit 1; }
65+
echo "$output" | grep "export EESSI_OVERRIDE_GPU_CHECK=" || { echo "EESSI_OVERRIDE_GPU_CHECK not found in output"; exit 1; }
66+
67+
# Verify that no symlinks were created
68+
if [ -e "/opt/eessi/nvidia/x86_64/host/driver_version.txt" ]; then
69+
echo "Error: symlinks were created in LD_PRELOAD mode"
70+
exit 1
71+
fi
72+
73+
echo "LD_PRELOAD mode test passed."
74+
75+
- name: Test normal run (first time)
76+
run: |
77+
echo ">>> Testing normal run - first time"
78+
79+
# Run with verbose mode
80+
output=$(./scripts/gpu_support/nvidia/link_nvidia_host_libraries.sh --verbose || { echo "Script returned non-zero: $?"; echo $output; exit 1; })
81+
82+
echo "$output"
83+
84+
echo ">>> Running checks"
85+
86+
# Check if NVIDIA GPU was detected - Driver version and CUDA version are hardcoded in `tests/nvidia-libs/mock-nvidia-smi.sh`
87+
echo "$output" | grep "Found NVIDIA GPU driver version 535.129.03" || { echo "Failed to detect NVIDIA driver version"; exit 1; }
88+
echo "$output" | grep "Found host CUDA version 8.0" || { echo "Failed to detect CUDA version"; exit 1; }
89+
90+
# Check if libraries were found
91+
echo "$output" | grep "Matched.*CUDA Libraries" || { echo "Failed to match CUDA libraries"; exit 1; }
92+
93+
# Verify symlinks were created
94+
if [ ! -d "/opt/eessi/nvidia/x86_64/host" ]; then
95+
echo "Error: host directory wasn't created"
96+
exit 1
97+
fi
98+
99+
# Check if version files were created
100+
if [ ! -f "/opt/eessi/nvidia/x86_64/host/driver_version.txt" ]; then
101+
echo "Error: driver_version.txt wasn't created"
102+
exit 1
103+
fi
104+
105+
# Check driver version content
106+
grep "535.129.03" "/opt/eessi/nvidia/x86_64/host/driver_version.txt" || { echo "Incorrect driver version"; exit 1; }
107+
108+
# Check if latest symlink was created
109+
if [ ! -L "/opt/eessi/nvidia/x86_64/latest" ]; then
110+
echo "Error: 'latest' symlink wasn't created"
111+
exit 1
112+
fi
113+
114+
# Check if latest points to host
115+
readlink "/opt/eessi/nvidia/x86_64/latest" | grep "host" || { echo "latest doesn't point to host"; exit 1; }
116+
117+
# Check if symlinks to libraries were created and point to correct files
118+
echo ">>> Checking library symlinks"
119+
120+
# List dir with libraries
121+
echo "Showing content of /tmp/nvidia_libs"
122+
echo "$(ls -l /tmp/nvidia_libs)"
123+
echo "Showing content of /opt/eessi/nvidia/x86_64/host"
124+
echo "$(ls -l /opt/eessi/nvidia/x86_64/host)"
125+
126+
# List expected library names - list of libraries is hardcoded in `tests/nvidia-libs/mock-nvidia-libs.sh`
127+
libraries=(
128+
"libcuda.so"
129+
"libcuda.so.1"
130+
"libnvidia-ml.so"
131+
"libnvidia-ml.so.1"
132+
"libnvidia-ptxjitcompiler.so"
133+
"libnvidia-ptxjitcompiler.so.1"
134+
"libcudadebugger.so"
135+
"libcudadebugger.so.1"
136+
)
137+
138+
# Check each expected library symlink
139+
for lib in "${libraries[@]}"; do
140+
lib_path="/opt/eessi/nvidia/x86_64/host/$lib"
141+
142+
# Check if the symlink exists
143+
if [ ! -L "$lib_path" ]; then
144+
echo "Error: Symlink for $lib was not created"
145+
exit 1
146+
fi
147+
148+
# Check if symlink target exists
149+
target=$(readlink "$lib_path")
150+
if [ ! -e "$target" ]; then
151+
echo "Error: Symlink $lib_path points to non-existent file: $target"
152+
exit 1
153+
fi
154+
155+
# Verify it points to our mock library in /tmp/nvidia_libs
156+
if [[ "$target" != "/tmp/nvidia_libs/$lib"* && "$target" != *"/tmp/nvidia_libs/"* ]]; then
157+
echo "Error: Symlink $lib_path points to $target, which is not in our mock directory"
158+
exit 1
159+
fi
160+
161+
echo ">>> Verified symlink: $lib -> $target"
162+
done
163+
164+
echo "First normal run test passed"
165+
166+
- name: Test normal run (second time)
167+
run: |
168+
echo ">>> Testing normal run - second time - should be idempotent"
169+
# Remove all write permissions on /opt/eessi so any attempts to write files fail
170+
chmod -R a-w /opt/eessi
171+
172+
# Store file timestamps before second run (ignoring access time)
173+
stat_before=$(stat --format="%n %s %y %U %G %m %i" "/opt/eessi/nvidia/x86_64/host/driver_version.txt")
174+
175+
# Run script again
176+
output=$(./scripts/gpu_support/nvidia/link_nvidia_host_libraries.sh || { echo "Script returned non-zero: $?"; echo $output; exit 1; })
177+
178+
echo "$output"
179+
180+
echo ">>> Running checks"
181+
182+
# Store file timestamps after second run (ignoring access time)
183+
stat_after=$(stat --format="%n %s %y %U %G %m %i" "/opt/eessi/nvidia/x86_64/host/driver_version.txt")
184+
185+
# Compare timestamps - should be the same (files shouldn't be modified)
186+
if [[ "$stat_before" != "$stat_after" ]]; then
187+
echo "Error: files were modified on second run when they shouldn't have been"
188+
echo "Before: $stat_before"
189+
echo "After: $stat_after"
190+
exit 1
191+
fi
192+
193+
# Check for message indicating that libraries are already linked
194+
echo "$output" | grep "have already been linked" || { echo "Missing 'already linked' message"; exit 1; }
195+
196+
echo "Second normal run test passed"

0 commit comments

Comments
 (0)