build: expose image build parallelism knobs to workflows

.github/workflows/build-swebench-images.yml

@@ -24,6 +24,11 @@ on:
         required: false
         default: '12'
         type: string
+      build-batch-size:
+        description: 'Number of images to submit per batch'
+        required: false
+        default: '15'
+        type: string
       max-retries:
         description: 'Retries per image build'
         required: false
@@ -119,6 +124,7 @@ jobs:
           if [ -n "${{ inputs.dataset }}" ]; then echo "DATASET=${{ inputs.dataset }}" >> "$GITHUB_ENV"; fi
           if [ -n "${{ inputs.split }}" ]; then echo "SPLIT=${{ inputs.split }}" >> "$GITHUB_ENV"; fi
           if [ -n "${{ inputs.max-workers }}" ]; then echo "MAX_WORKERS=${{ inputs.max-workers }}" >> "$GITHUB_ENV"; fi
+          if [ -n "${{ inputs.build-batch-size }}" ]; then echo "BUILD_BATCH_SIZE=${{ inputs.build-batch-size }}" >> "$GITHUB_ENV"; fi
           if [ -n "${{ inputs.max-retries }}" ]; then echo "MAX_RETRIES=${{ inputs.max-retries }}" >> "$GITHUB_ENV"; fi
           # Empty string means "no limit"
           if [ -n "${{ inputs.n-limit }}" ]; then echo "N_LIMIT=${{ inputs.n-limit }}" >> "$GITHUB_ENV"; else echo "N_LIMIT=" >> "$GITHUB_ENV"; fi
@@ -238,6 +244,7 @@ jobs:
             --image ghcr.io/openhands/eval-agent-server \
             --push \
             --max-workers '${MAX_WORKERS}' \
+            --build-batch-size '${BUILD_BATCH_SIZE}' \
             --max-retries '${MAX_RETRIES}'"
 
           # Only include --n-limit if provided (non-empty)

.github/workflows/build-swtbench-images.yml

@@ -24,6 +24,11 @@ on:
         required: false
         default: '4'
         type: string
+      build-batch-size:
+        description: 'Number of agent-server images to submit per batch'
+        required: false
+        default: '15'
+        type: string
       n-limit:
         description: 'Limit number of images to build (0 for all)'
         required: false
@@ -53,7 +58,7 @@ on:
         required: false
         default: '2'
         type: string
-      build-batch-size:
+      eval-env-build-batch-size:
         description: 'Env images per batch for eval env builds'
         required: false
         default: '10'
@@ -88,6 +93,7 @@ jobs:
       DATASET: eth-sri/SWT-bench_Verified_bm25_27k_zsp
       SPLIT: test
       MAX_WORKERS: '4'
+      BUILD_BATCH_SIZE: '15'
       N_LIMIT: '0'
       INSTANCE_IDS: ''
       SELECT_FILE: ''
@@ -147,6 +153,7 @@ jobs:
           DATASET="${{ inputs.dataset || 'eth-sri/SWT-bench_Verified_bm25_27k_zsp' }}"
           SPLIT="${{ inputs.split || 'test' }}"
           MAX_WORKERS="${{ inputs.max-workers || '4' }}"
+          BUILD_BATCH_SIZE="${{ inputs.build-batch-size || '15' }}"
           N_LIMIT="${{ inputs.n-limit || '0' }}"
           INSTANCE_IDS="${{ inputs.instance-ids }}"
           FORCE_BUILD="${{ inputs.force-build || 'false' }}"
@@ -172,6 +179,7 @@ jobs:
             --image ghcr.io/openhands/eval-agent-server \
             --target ${TARGET} \
             --max-workers ${MAX_WORKERS} \
+            --build-batch-size ${BUILD_BATCH_SIZE} \
             --push"
 
           # Add n-limit if specified
@@ -210,7 +218,7 @@ jobs:
           MAX_WORKERS="${{ inputs.max-workers || '4' }}"
           BUILD_MODE="${{ inputs.build-mode || 'cli' }}"
           MAX_RETRIES="${{ inputs.max-retries || '2' }}"
-          BUILD_BATCH_SIZE="${{ inputs.build-batch-size || '10' }}"
+          BUILD_BATCH_SIZE="${{ inputs.eval-env-build-batch-size || '10' }}"
           FORCE_BUILD="${{ inputs.force-build || 'false' }}"
 
           echo "N_LIMIT=${N_LIMIT}" >> "$GITHUB_ENV"

benchmarks/commit0/build_images.py

@@ -124,6 +124,7 @@ def main(argv: list[str]) -> int:
         image=args.image,
         push=args.push,
         max_workers=args.max_workers,
+        build_batch_size=args.build_batch_size,
         dry_run=args.dry_run,
         force_build=args.force_build,
         max_retries=args.max_retries,

benchmarks/gaia/build_images.py

@@ -100,6 +100,7 @@ def tag_fn(_base: str) -> str:
         image=args.image,
         push=args.push,
         max_workers=1,  # Only building one image
+        build_batch_size=args.build_batch_size,
         dry_run=args.dry_run,
         force_build=args.force_build,
         max_retries=args.max_retries,

benchmarks/multiswebench/build_images.py

@@ -118,6 +118,7 @@ def main():
         ),
         base_image_to_custom_tag_fn=extract_custom_tag,
         max_workers=args.num_workers,
+        build_batch_size=args.build_batch_size,
         dry_run=False,
         force_build=args.force_build,
     )

benchmarks/swebench/build_images.py

@@ -177,6 +177,7 @@ def main(argv: list[str]) -> int:
         image=args.image,
         push=args.push,
         max_workers=args.max_workers,
+        build_batch_size=args.build_batch_size,
         dry_run=args.dry_run,
         force_build=args.force_build,
         max_retries=args.max_retries,

benchmarks/swebenchmultilingual/build_images.py

@@ -177,7 +177,9 @@ def main(argv: list[str]) -> int:
         image=args.image,
         push=args.push,
         max_workers=args.max_workers,
+        build_batch_size=args.build_batch_size,
         dry_run=args.dry_run,
+        force_build=args.force_build,
         max_retries=args.max_retries,
         base_image_to_custom_tag_fn=extract_custom_tag,
         post_build_fn=_wrap_if_needed,

benchmarks/swebenchmultimodal/build_images.py

@@ -83,6 +83,7 @@ def main(argv: list[str]) -> int:
         image=args.image,
         push=args.push,
         max_workers=args.max_workers,
+        build_batch_size=args.build_batch_size,
         dry_run=args.dry_run,
         force_build=args.force_build,
         max_retries=args.max_retries,

benchmarks/swegym/build_images.py

@@ -87,6 +87,7 @@ def main(argv: list[str]) -> int:
         image=args.image,
         push=args.push,
         max_workers=args.max_workers,
+        build_batch_size=args.build_batch_size,
         dry_run=args.dry_run,
         force_build=args.force_build,
         max_retries=args.max_retries,

benchmarks/swesmith/build_images.py

@@ -84,6 +84,7 @@ def main(argv: list[str]) -> int:
         image=args.image,
         push=args.push,
         max_workers=args.max_workers,
+        build_batch_size=args.build_batch_size,
         dry_run=args.dry_run,
         force_build=args.force_build,
         max_retries=args.max_retries,

benchmarks/swtbench/build_images.py

@@ -44,6 +44,7 @@ def main(argv: list[str]) -> int:
         image=args.image,
         push=args.push,
         max_workers=args.max_workers,
+        build_batch_size=args.build_batch_size,
         dry_run=args.dry_run,
         force_build=args.force_build,
         max_retries=args.max_retries,

benchmarks/utils/build_utils.py

@@ -342,6 +342,15 @@ def get_build_parser() -> argparse.ArgumentParser:
     parser.add_argument(
         "--max-workers", type=int, default=1, help="Concurrent builds (be cautious)"
     )
+    parser.add_argument(
+        "--build-batch-size",
+        type=int,
+        default=None,
+        help=(
+            "Number of images to submit per batch. Defaults to BUILD_BATCH_SIZE "
+            "when unset."
+        ),
+    )
     parser.add_argument(
         "--dry-run", action="store_true", help="List base images only, don’t build"
     )
@@ -666,6 +675,7 @@ def build_all_images(
     push: bool = False,
     base_image_to_custom_tag_fn: Callable[[str], str] | None = None,
     max_workers: int = 1,
+    build_batch_size: int | None = None,
     dry_run: bool = False,
     force_build: bool = False,
     max_retries: int = 3,
@@ -684,6 +694,8 @@ def build_all_images(
         base_image_to_custom_tag_fn: Function to extract a custom tag from a base image.
             Evaluated before scheduling builds so it can safely be a closure.
         max_workers: Number of concurrent builds.
+        build_batch_size: Number of images to submit per batch. If None, use the
+            BUILD_BATCH_SIZE environment variable.
         dry_run: If True, only list base images without building.
         force_build: If True, rebuild even when matching remote images already exist.
         max_retries: Number of times to retry each failed build (default: 3).
@@ -712,7 +724,11 @@ def build_all_images(
 
     # Batch/prune settings (tunable via env to control disk usage on sticky runners)
     # Default to smaller batches and more aggressive pruning on shared runners.
-    batch_size = int(os.getenv("BUILD_BATCH_SIZE", "15"))
+    batch_size = (
+        build_batch_size
+        if build_batch_size is not None
+        else int(os.getenv("BUILD_BATCH_SIZE", "15"))
+    )
     prune_keep_storage_gb = int(os.getenv("BUILDKIT_PRUNE_KEEP_GB", "60"))
     prune_threshold_pct = float(os.getenv("BUILDKIT_PRUNE_THRESHOLD_PCT", "60"))
     # Prune aggressively by default; filters like "unused-for=12h" prevented GC from

tests/test_image_utils.py

@@ -391,6 +391,70 @@ def test_build_parser_accepts_force_build(self):
         assert args.force_build is True
 
 
+class TestBuildBatchSizeConfig:
+    def test_build_parser_accepts_build_batch_size(self):
+        from benchmarks.utils.build_utils import get_build_parser
+
+        args = get_build_parser().parse_args(["--build-batch-size", "50"])
+
+        assert args.build_batch_size == 50
+
+    @patch.dict(os.environ, {"BUILD_BATCH_SIZE": "99"})
+    def test_build_all_images_prefers_explicit_batch_size_over_env(
+        self,
+        tmp_path: Path,
+    ):
+        from benchmarks.utils import build_utils
+
+        seen_batches: list[list[str]] = []
+
+        class FakeFuture:
+            def __init__(self, result: BuildOutput):
+                self._result = result
+
+            def result(self) -> BuildOutput:
+                return self._result
+
+        class FakeExecutor:
+            def __init__(self, *args, **kwargs):
+                self._batch: list[str] = []
+
+            def __enter__(self):
+                seen_batches.append(self._batch)
+                return self
+
+            def __exit__(self, exc_type, exc, tb):
+                return False
+
+            def submit(self, fn, **kwargs):
+                self._batch.append(kwargs["base_image"])
+                return FakeFuture(
+                    BuildOutput(
+                        base_image=kwargs["base_image"],
+                        tags=[f"tag:{kwargs['base_image']}"],
+                        error=None,
+                    )
+                )
+
+        with (
+            patch.object(build_utils, "ProcessPoolExecutor", FakeExecutor),
+            patch.object(
+                build_utils, "as_completed", side_effect=lambda futures: futures
+            ),
+            patch.object(build_utils, "buildkit_disk_usage", return_value=(0, 0)),
+            patch.object(build_utils, "maybe_prune_buildkit_cache", return_value=False),
+        ):
+            exit_code = build_utils.build_all_images(
+                base_images=["base-1", "base-2", "base-3"],
+                target="source-minimal",
+                build_dir=tmp_path,
+                build_batch_size=2,
+            )
+
+        assert exit_code == 0
+        assert seen_batches == [["base-1", "base-2"], ["base-3"]]
+
+
 class TestCachedSdistReuse:
     def test_build_image_passes_cached_sdist_to_sdk_build_module(
         self,