ci: split runtime benchmark into 3 parallel size chunks

Author: Becheler

.github/workflows/benchmark.yml

@@ -69,7 +69,7 @@ jobs:
         retention-days: 1
 
   # ══════════════════════════════════════════════════════════════════════
-  # 2. Run each benchmark in parallel
+  # 2a. Non-runtime benchmarks in parallel
   # ══════════════════════════════════════════════════════════════════════
   benchmark:
     needs: build
@@ -78,7 +78,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        suite: [correctness, runtime, incremental, epsilon]
+        suite: [correctness, incremental, epsilon]
     steps:
     - uses: actions/checkout@v4
 
@@ -116,11 +116,67 @@ jobs:
         path: louvain/results/*.csv
         retention-days: 30
 
+  # ══════════════════════════════════════════════════════════════════════
+  # 2b. Runtime benchmark split by graph size
+  # ══════════════════════════════════════════════════════════════════════
+  runtime:
+    needs: build
+    runs-on: ubuntu-latest
+    timeout-minutes: 180
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - chunk: small
+            sizes: "1000,5000"
+          - chunk: medium
+            sizes: "10000,50000"
+          - chunk: large
+            sizes: "100000,250000"
+    steps:
+    - uses: actions/checkout@v4
+
+    - name: Set up Python
+      uses: actions/setup-python@v5
+      with:
+        python-version: '3.11'
+
+    - name: Install Python dependencies
+      working-directory: louvain
+      run: |
+        python -m pip install --upgrade pip
+        pip install -r ../requirements-common.txt -r requirements.txt
+
+    - name: Download build artifacts
+      uses: actions/download-artifact@v4
+      with:
+        name: louvain-build
+        path: louvain/build/
+
+    - name: Make binaries executable
+      run: chmod +x louvain/build/*
+
+    - name: Run runtime benchmark (${{ matrix.chunk }})
+      working-directory: louvain
+      run: |
+        mkdir -p results
+        python scripts/benchmark.py --only runtime \
+          --sizes ${{ matrix.sizes }} \
+          --output-suffix _${{ matrix.chunk }}
+
+    - name: Upload runtime ${{ matrix.chunk }} results
+      uses: actions/upload-artifact@v4
+      if: always()
+      with:
+        name: results-runtime-${{ matrix.chunk }}
+        path: louvain/results/*.csv
+        retention-days: 30
+
   # ══════════════════════════════════════════════════════════════════════
   # 3. Merge results, generate plots, and commit
   # ══════════════════════════════════════════════════════════════════════
   visualize:
-    needs: benchmark
+    needs: [benchmark, runtime]
     if: always()
     runs-on: ubuntu-latest
     timeout-minutes: 30
@@ -146,10 +202,24 @@ jobs:
         path: louvain/results/
       continue-on-error: true
 
-    - name: Download runtime results
+    - name: Download runtime (small) results
       uses: actions/download-artifact@v4
       with:
-        name: results-runtime
+        name: results-runtime-small
+        path: louvain/results/
+      continue-on-error: true
+
+    - name: Download runtime (medium) results
+      uses: actions/download-artifact@v4
+      with:
+        name: results-runtime-medium
+        path: louvain/results/
+      continue-on-error: true
+
+    - name: Download runtime (large) results
+      uses: actions/download-artifact@v4
+      with:
+        name: results-runtime-large
         path: louvain/results/
       continue-on-error: true
 
@@ -167,6 +237,24 @@ jobs:
         path: louvain/results/
       continue-on-error: true
 
+    - name: Merge runtime CSV chunks
+      working-directory: louvain
+      run: |
+        python -c "
+        import pandas as pd, glob, os
+        parts = sorted(glob.glob('results/runtime_*.csv'))
+        if parts:
+            df = pd.concat([pd.read_csv(p) for p in parts], ignore_index=True)
+            df.to_csv('results/runtime.csv', index=False)
+            print(f'Merged {len(parts)} runtime chunks into runtime.csv ({len(df)} rows)')
+            for p in parts:
+                os.remove(p)
+        elif os.path.exists('results/runtime.csv'):
+            print('runtime.csv already present, skipping merge')
+        else:
+            print('WARNING: no runtime CSVs found')
+        "
+
     - name: Generate visualizations
       working-directory: louvain
       run: python scripts/visualize.py

louvain/scripts/benchmark.py

@@ -290,7 +290,7 @@ def run_benchmark_correctness(n_trials=100):
 
 # ── runtime scalability ─────────────────────────────────────────────────
 
-def run_benchmark_runtime(n_trials=10, sizes=None):
+def run_benchmark_runtime(n_trials=10, sizes=None, output_suffix=''):
     """Run runtime scalability benchmark.
 
     Benchmark fairness notes:
@@ -458,8 +458,9 @@ def run_benchmark_runtime(n_trials=10, sizes=None):
             print(', '.join(summary_parts))
 
     df = pd.DataFrame(results)
-    df.to_csv('results/runtime.csv', index=False)
-    print(f"\nSaved to results/runtime.csv")
+    out_name = f'results/runtime{output_suffix}.csv'
+    df.to_csv(out_name, index=False)
+    print(f"\nSaved to {out_name}")
     return df
 
 
@@ -752,34 +753,48 @@ def run_benchmark_epsilon(n_trials=10, sizes=None):
 
 # ── main ─────────────────────────────────────────────────────────────────
 
-BENCHMARKS = {
-    'correctness':  lambda q: run_benchmark_correctness(n_trials=5 if q else 100),
-    'runtime':      lambda q: run_benchmark_runtime(
-                        n_trials=3 if q else 10,
-                        sizes=[1000, 5000, 25000] if q else None),
-    'incremental':  lambda q: run_benchmark_incremental(
-                        n_trials=3 if q else 10,
-                        sizes=[100, 500] if q else None),
-    'epsilon':      lambda q: run_benchmark_epsilon(
-                        n_trials=3 if q else 10,
-                        sizes=[1000, 5000, 10000] if q else None),
-}
+BENCHMARKS = ['correctness', 'runtime', 'incremental', 'epsilon']
 
 if __name__ == '__main__':
     parser = argparse.ArgumentParser(description='Louvain benchmark suite')
     parser.add_argument('--quick', action='store_true',
                         help='Run a fast smoke-test (fewer graphs, trials, sizes)')
-    parser.add_argument('--only', choices=list(BENCHMARKS.keys()),
+    parser.add_argument('--only', choices=BENCHMARKS,
                         help='Run only the specified benchmark')
+    parser.add_argument('--sizes', type=str, default=None,
+                        help='Comma-separated list of graph sizes (for runtime/incremental/epsilon)')
+    parser.add_argument('--output-suffix', type=str, default='',
+                        help='Suffix appended to output CSV filename (e.g. "_part0")')
     args = parser.parse_args()
 
     os.makedirs('results', exist_ok=True)
 
     if args.quick:
         print('*** QUICK MODE — reduced trials & sizes ***\n')
 
+    sizes_override = [int(s) for s in args.sizes.split(',')] if args.sizes else None
+    suffix = args.output_suffix
+
+    def _run(name):
+        q = args.quick
+        if name == 'correctness':
+            run_benchmark_correctness(n_trials=5 if q else 100)
+        elif name == 'runtime':
+            run_benchmark_runtime(
+                n_trials=3 if q else 10,
+                sizes=sizes_override or ([1000, 5000, 25000] if q else None),
+                output_suffix=suffix)
+        elif name == 'incremental':
+            run_benchmark_incremental(
+                n_trials=3 if q else 10,
+                sizes=sizes_override or ([100, 500] if q else None))
+        elif name == 'epsilon':
+            run_benchmark_epsilon(
+                n_trials=3 if q else 10,
+                sizes=sizes_override or ([1000, 5000, 10000] if q else None))
+
     if args.only:
-        BENCHMARKS[args.only](args.quick)
+        _run(args.only)
     else:
-        for name, fn in BENCHMARKS.items():
-            fn(args.quick)
+        for name in BENCHMARKS:
+            _run(name)