Skip to content

Commit a16f64d

Browse files
Copilottransientlunatic
Copilot
and
transientlunatic
committed
Add GitHub Actions workflow for Slurm testing
Co-authored-by: transientlunatic <4365778+transientlunatic@users.noreply.github.com>
1 parent eaf2c92 commit a16f64d

File tree

1 file changed

+230
-0
lines changed

1 file changed

+230
-0
lines changed

‎.github/workflows/slurm-tests.yml‎

Lines changed: 230 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,230 @@
1+
name: Tests with Slurm
2+
on: [push, pull_request]
3+
4+
jobs:
5+
tests:
6+
name: "Slurm Testing"
7+
runs-on: ubuntu-latest
8+
container:
9+
image: ghcr.io/natejenkins/slurm-docker-cluster:23.11.7
10+
options: --privileged --hostname slurmctld
11+
defaults:
12+
run:
13+
shell: bash -el {0}
14+
steps:
15+
16+
- uses: actions/checkout@v5
17+
with:
18+
fetch-depth: 0
19+
20+
- name: Install base tools
21+
run: |
22+
apt-get update
23+
apt-get install -y sudo git python3 python3-pip python3-venv curl
24+
25+
- name: Start Slurm services
26+
run: |
27+
# Start munge authentication service
28+
sudo -u munge /usr/sbin/munged
29+
30+
# Start Slurm controller
31+
/usr/sbin/slurmctld
32+
33+
# Start Slurm daemon
34+
/usr/sbin/slurmd
35+
36+
# Give services time to start
37+
sleep 5
38+
39+
# Verify Slurm is working
40+
sinfo
41+
squeue
42+
43+
- name: Create submit user
44+
run: |
45+
useradd -m -s /bin/bash submituser
46+
echo "submituser ALL=(ALL) NOPASSWD:ALL" >> /etc/sudoers
47+
48+
- name: Ensure submituser can access workspace
49+
run: |
50+
mkdir -p "$GITHUB_WORKSPACE/test_project"
51+
chmod -R a+rwX "$GITHUB_WORKSPACE"
52+
53+
- name: Install Python dependencies
54+
run: |
55+
python3 -m pip install --upgrade pip
56+
python3 -m pip install -e "$GITHUB_WORKSPACE"
57+
python3 -m pip install python-crontab
58+
59+
- name: Install additional packages
60+
run: |
61+
python3 -m pip install -U bilby bilby_pipe==1.4.0 pesummary
62+
python3 -m pip install git+https://git.ligo.org/asimov/pipelines/gwdata.git@update-htcondor || echo "gwdata install failed, continuing..."
63+
64+
- name: Set up git
65+
run: |
66+
su - submituser <<SUBMITUSER
67+
set -euo pipefail
68+
git config --global init.defaultBranch master
69+
git config --global user.email "you@example.com"
70+
git config --global user.name "Your Name"
71+
SUBMITUSER
72+
73+
- name: Start MockGWDataFindServer in background
74+
run: |
75+
python3 << EOF &
76+
from tests.mock_gwdatafind_server import MockGWDataFindServer
77+
import time
78+
79+
frame_configs = {
80+
('H', 'H1_LOSC_16_V1'): ['file://$GITHUB_WORKSPACE/tests/test_data/frames/H-H1_GWOSC_16KHZ_R1-1126259447-32.gwf'],
81+
('L', 'L1_LOSC_16_V1'): ['file://$GITHUB_WORKSPACE/tests/test_data/frames/L-L1_GWOSC_16KHZ_R1-1126259447-32.gwf']
82+
}
83+
84+
server = MockGWDataFindServer(port=8765, frame_configs=frame_configs)
85+
server.start()
86+
print("MockGWDataFindServer started on port 8765", flush=True)
87+
88+
try:
89+
while True:
90+
time.sleep(1)
91+
except KeyboardInterrupt:
92+
server.stop()
93+
EOF
94+
95+
# Give the server time to start
96+
sleep 3
97+
98+
- name: Verify server is running
99+
run: |
100+
curl -f http://localhost:8765/api/v1/gwf/H/H1_LOSC_16_V1/1126259460,1126259464/file.json || echo "Server check failed but continuing..."
101+
102+
- name: Create asimov project
103+
run: |
104+
su - submituser <<SUBMITUSER
105+
set -euo pipefail
106+
cd $GITHUB_WORKSPACE/test_project
107+
asimov init "Test Project"
108+
SUBMITUSER
109+
110+
- name: Verify Slurm scheduler was detected
111+
run: |
112+
su - submituser <<SUBMITUSER
113+
set -euo pipefail
114+
cd $GITHUB_WORKSPACE/test_project
115+
116+
# Check that scheduler type is set to slurm
117+
scheduler_type=\$(grep -A 1 "\\[scheduler\\]" .asimov/asimov.conf | grep "type" | awk -F'=' '{print \$2}' | tr -d ' ')
118+
echo "Detected scheduler type: \$scheduler_type"
119+
120+
if [ "\$scheduler_type" != "slurm" ]; then
121+
echo "ERROR: Expected scheduler type 'slurm' but got '\$scheduler_type'"
122+
exit 1
123+
fi
124+
125+
echo "✓ Slurm scheduler correctly detected"
126+
SUBMITUSER
127+
128+
- name: Configure asimov settings
129+
run: |
130+
su - submituser <<SUBMITUSER
131+
set -euo pipefail
132+
cd $GITHUB_WORKSPACE/test_project
133+
asimov configuration update slurm/user submituser
134+
asimov configuration update slurm/cron_minute '*/1'
135+
SUBMITUSER
136+
137+
- name: Apply GWOSC quick test blueprint
138+
run: |
139+
su - submituser <<SUBMITUSER
140+
set -euo pipefail
141+
cd $GITHUB_WORKSPACE/test_project
142+
asimov apply -f "$GITHUB_WORKSPACE/tests/test_blueprints/gwosc_quick_test.yaml"
143+
SUBMITUSER
144+
145+
- name: Add event from GWOSC
146+
run: |
147+
su - submituser <<SUBMITUSER
148+
set -euo pipefail
149+
cd $GITHUB_WORKSPACE/test_project
150+
asimov apply -f "$GITHUB_WORKSPACE/tests/test_blueprints/gwosc_event.yaml"
151+
SUBMITUSER
152+
153+
- name: Run basic scheduler tests
154+
run: |
155+
cd $GITHUB_WORKSPACE
156+
python3 -m unittest tests.test_scheduler -v
157+
158+
- name: Test Slurm job submission
159+
run: |
160+
su - submituser <<SUBMITUSER
161+
set -euo pipefail
162+
cd $GITHUB_WORKSPACE/test_project
163+
164+
# Test basic Slurm functionality
165+
echo "Testing basic Slurm job submission..."
166+
167+
# Create a simple test job
168+
cat > test_job.sh << 'TESTJOB'
169+
#!/bin/bash
170+
#SBATCH --job-name=test-asimov
171+
#SBATCH --output=test_output.log
172+
#SBATCH --error=test_error.log
173+
#SBATCH --time=00:01:00
174+
175+
echo "Slurm job running successfully"
176+
sleep 2
177+
echo "Job completed"
178+
TESTJOB
179+
180+
# Submit the job
181+
job_id=\$(sbatch --parsable test_job.sh)
182+
echo "Submitted test job: \$job_id"
183+
184+
# Wait for job to complete
185+
for i in {1..30}; do
186+
state=\$(squeue -j \$job_id -h -o "%t" 2>/dev/null || echo "DONE")
187+
echo "Job state: \$state"
188+
189+
if [ "\$state" == "DONE" ] || [ -z "\$state" ]; then
190+
echo "Job completed"
191+
break
192+
fi
193+
194+
sleep 2
195+
done
196+
197+
# Check output
198+
if [ -f test_output.log ]; then
199+
echo "Test job output:"
200+
cat test_output.log
201+
fi
202+
203+
echo "✓ Basic Slurm job submission works"
204+
SUBMITUSER
205+
206+
- name: Show Slurm configuration
207+
run: |
208+
echo "=== Slurm Configuration ==="
209+
sinfo -o "%P %a %l %D %N"
210+
echo ""
211+
echo "=== Slurm Queue ==="
212+
squeue
213+
echo ""
214+
echo "=== Slurm Version ==="
215+
sinfo --version
216+
217+
- name: Archive testing artifacts
218+
if: always()
219+
uses: actions/upload-artifact@v4
220+
with:
221+
name: slurm-test-artifacts
222+
path: |
223+
test_project/
224+
retention-days: 5
225+
226+
- name: Stop background server
227+
if: always()
228+
run: |
229+
# Kill the background Python process running the server
230+
pkill -f "MockGWDataFindServer" || true

0 commit comments

Comments
 (0)