Enable container run

docs/container_runs.md

@@ -0,0 +1,38 @@
+# Container Runs
+
+Container runs activate when `container_image` (or `container_runtime`) is set.
+Defaults:
+- Image: `registry.nersc.gov/amsc014/superfacility/pele:latest`
+- ALCF image: `docker://registry.nersc.gov/amsc014/superfacility/pele:latest`
+
+## Local (Docker)
+
+Example config override:
+
+```yaml
+environment: local
+container_image: registry.nersc.gov/amsc014/superfacility/pele:latest
+container_runtime: docker
+```
+
+## NERSC Perlmutter (podman-hpc)
+
+Example config override:
+
+```yaml
+environment: perlmutter
+container_image: registry.nersc.gov/amsc014/superfacility/pele:latest
+container_runtime: podman-hpc
+```
+
+Reference: https://docs.nersc.gov/development/containers/podman-hpc/overview/
+
+## ALCF (Apptainer)
+
+Example config override:
+
+```yaml
+environment: alcf
+container_image: docker://registry.nersc.gov/amsc014/superfacility/pele:latest
+container_runtime: apptainer
+```

docs/mcp.md

@@ -36,6 +36,10 @@ The MCP server exposes these tools:
 
 Use `tools/list` to discover schemas and required parameters.
 
+## Container runs
+
+Container usage examples live in `docs/container_runs.md`.
+
 ## Example: apply_plan
 
 Apply a previously generated plan to write inputs:

mcp_server.py

@@ -48,6 +48,7 @@
     from src.services.input_writer import InputWriterService
     from src.services.knowledge import PeleKnowledgeService
     from src.services.plan import SimulationPlan
+    from src.services.run_container import ContainerRunner
     from src.services.run_local import LocalRunner
     from src.services.run_superfacility import SuperfacilityRunner
     from src.services.validation import ValidationService
@@ -62,6 +63,7 @@
     from src.services.input_writer import InputWriterService
     from src.services.knowledge import PeleKnowledgeService
     from src.services.plan import SimulationPlan
+    from src.services.run_container import ContainerRunner
     from src.services.run_local import LocalRunner
     from src.services.run_superfacility import SuperfacilityRunner
     from src.services.validation import ValidationService
@@ -130,6 +132,12 @@ def _persist_session_context(session_id: str, context: dict[str, Any]) -> None:
 def _select_runner(active_config: AMReXAgentConfig):
     """Select execution runner based on config environment."""
     environment = (active_config.environment or "").lower()
+    container_image = getattr(active_config, "container_image", None)
+    container_runtime = getattr(active_config, "container_runtime", None)
+    has_container_image = isinstance(container_image, (str, Path)) and str(container_image).strip()
+    has_container_runtime = isinstance(container_runtime, str) and container_runtime.strip()
+    if has_container_image or has_container_runtime:
+        return ContainerRunner(active_config)
     if environment == "local":
         return LocalRunner(active_config)
     if environment in {"perlmutter", "mcp"}:

src/config.py

@@ -533,6 +533,47 @@ class AMReXAgentConfig(BaseModel):
                     "Auto-detected from PODMAN_HPC or SHIFTER env variables. "
                     "When True, visualization uses extraction (headless) + rendering (local) workflow."
     )
+    container_runtime: Optional[str] = Field(
+        default=None,
+        description="Container runtime override (docker, podman, podman-hpc, apptainer, singularity). "
+                    "When unset, inferred from environment."
+    )
+    container_image: Optional[str] = Field(
+        default=None,
+        description="Container image reference (e.g., docker image or apptainer .sif or docker:// ref)."
+    )
+    container_entrypoint: Optional[str] = Field(
+        default=None,
+        description="Container entrypoint or executable path to run (default: /usr/local/bin/run_pelelmex)."
+    )
+    container_workdir: str = Field(
+        default="/work",
+        description="Workdir inside the container for mounted run directory."
+    )
+    container_inputs_name: str = Field(
+        default="inputs",
+        description="Inputs filename passed to the container entrypoint."
+    )
+    container_extra_args: Optional[List[str]] = Field(
+        default_factory=list,
+        description="Extra runtime args passed to the container (e.g., --env, --network)."
+    )
+    container_account: Optional[str] = Field(
+        default=None,
+        description="Slurm account for container jobs (used for non-NERSC slurm submission)."
+    )
+    container_qos: Optional[str] = Field(
+        default=None,
+        description="Slurm QoS for container jobs (overrides default when set)."
+    )
+    container_constraint: Optional[str] = Field(
+        default=None,
+        description="Slurm constraint for container jobs (overrides default when set)."
+    )
+    container_walltime: Optional[str] = Field(
+        default=None,
+        description="Slurm walltime for container jobs (overrides default when set)."
+    )
 
     analysis_always_enabled: bool = Field(
         default=True,

src/nodes/runner_node.py

@@ -154,8 +154,17 @@ def runner_node(state: GraphState) -> dict[str, Any]:
         if compile_selection.get("value") == "compile_only":
             run_after_compile = False
 
-        # Select runner based on environment
-        if config.environment == "local":
+        # Select runner based on environment or container settings
+        container_image = getattr(config, "container_image", None)
+        container_runtime = getattr(config, "container_runtime", None)
+        has_container_image = isinstance(container_image, (str, Path)) and str(container_image).strip()
+        has_container_runtime = isinstance(container_runtime, str) and container_runtime.strip()
+
+        if has_container_image or has_container_runtime:
+            from src.services.run_container import ContainerRunner
+            runner = ContainerRunner(config)
+            logger.info("Using ContainerRunner for containerized execution")
+        elif config.environment == "local":
             from src.services.run_local import LocalRunner
             runner = LocalRunner(config)
             logger.info("Using LocalRunner for local execution")