Keep deployment settings on hybrid mapped phase payloads

Hybrid mapped execution rebuilt its static, mapped, and cleanup phase payloads
from a fixed field list that dropped modal_gpu and modal_max_containers. Those
payloads resolved default settings and expected a different runtime fingerprint
for the same Modal app, so the sibling-validation guard refused them with
ModalRemoteAppOutOfDateError.

Define the fingerprint-shaping payload fields once in settings and copy them at
every derived-payload site (mapped phases, warmup requests, speculative
prewarm targets).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

docs/LEARNINGS.md

@@ -759,3 +759,10 @@ generation streams with an empty `modal.exception.RemoteError`. Fixes:
 - `_replace_outdated_modal_app` refuses to stop an app whose remote fingerprint
   this process already deployed or validated, so divergent callers can no longer
   stop each other's deployments.
+- Follow-up: hybrid mapped execution builds its static, mapped, and cleanup phase
+  payloads in `_shared_subgraph_payload_fields`, which also dropped `modal_gpu` and
+  `modal_max_containers`. The refusal above caught it, raising
+  `ModalRemoteAppOutOfDateError` instead of killing jobs. All derived payloads now
+  copy `settings.MODAL_DEPLOYMENT_PAYLOAD_FIELDS` through
+  `modal_deployment_payload_fields()`; add any new fingerprint-shaping payload
+  field there.

prompt_affinity_planning.py

@@ -24,6 +24,7 @@ if __package__:
         _node_output_refs,
     )
     from .remote_plan_types import InputTarget, LinkedOutputRef
+    from .settings import MODAL_DEPLOYMENT_PAYLOAD_FIELDS
 else:  # pragma: no cover - flat import inside the Modal container.
     from component_planning import _component_ancestors_of_local_source
     from execution_environments import ExecutionProvider
@@ -40,6 +41,7 @@ else:  # pragma: no cover - flat import inside the Modal container.
         _node_output_refs,
     )
     from remote_plan_types import InputTarget, LinkedOutputRef
+    from settings import MODAL_DEPLOYMENT_PAYLOAD_FIELDS
 
 logger = logging.getLogger(__name__)
 
@@ -48,8 +50,7 @@ _SPECULATIVE_PREWARM_PAYLOAD_FIELDS = frozenset(
     {
         "component_id",
         "prompt_id",
-        "modal_gpu",
-        "modal_max_containers",
+        *MODAL_DEPLOYMENT_PAYLOAD_FIELDS,
         "execution_provider",
         "execution_environment_id",
         "remote_worker_affinity_group",

remote/mapped_execution.py

@@ -43,7 +43,7 @@ from .modal_warmup import (
     _prompt_warmup_head_start_seconds,
     boost_mapped_component_warmup,
 )
-from ..settings import get_settings
+from ..settings import get_settings, modal_deployment_payload_fields
 
 logger = logging.getLogger(__name__)
 
@@ -144,6 +144,7 @@ def _shared_subgraph_payload_fields(payload: dict[str, Any]) -> dict[str, Any]:
             payload.get("terminate_container_on_error", True)
         ),
         "custom_nodes_bundle": payload.get("custom_nodes_bundle"),
+        **modal_deployment_payload_fields(payload),
     }
     snapshot_profile_key = payload.get("snapshot_profile_key")
     if isinstance(snapshot_profile_key, str) and snapshot_profile_key.strip():

remote/modal_warmup.py

@@ -16,7 +16,11 @@ import threading
 import time
 from typing import Any, Callable, Iterator, Mapping
 
-from ..settings import get_settings, modal_deployment_app_name
+from ..settings import (
+    get_settings,
+    modal_deployment_app_name,
+    modal_deployment_payload_fields,
+)
 from .local_execution import _is_link, _resolve_required_subgraph_nodes
 from .modal_deployment import (
     ModalRemoteInvocationError,
@@ -542,8 +546,8 @@ def _build_prompt_warmup_request(payload: dict[str, Any]) -> dict[str, Any] | No
                 payload.get("component_id", "modal-warmup"),
             )
         ),
+        **modal_deployment_payload_fields(payload),
         "modal_gpu": settings.modal_gpu,
-        "modal_max_containers": payload.get("modal_max_containers"),
         "remote_worker_affinity_group": payload.get(
             "remote_worker_affinity_group"
         ),

settings.py

@@ -10,6 +10,7 @@ import re
 from dataclasses import dataclass, replace
 from functools import lru_cache
 from pathlib import Path
+from typing import Any, Mapping
 
 if __package__:
     from .instance_identity import (
@@ -250,6 +251,23 @@ def modal_deployment_app_name(settings: ModalSyncSettings) -> str:
     return f"{prefix}-{digest}{suffix}"
 
 
+MODAL_DEPLOYMENT_PAYLOAD_FIELDS = ("modal_gpu", "modal_max_containers")
+"""Payload fields that select deployment settings and so shape the runtime fingerprint.
+
+Any payload derived from another (warmups, prewarm targets, mapped phases) must
+copy these, or it will expect a different fingerprint for the same Modal app.
+"""
+
+
+def modal_deployment_payload_fields(payload: Mapping[str, Any]) -> dict[str, Any]:
+    """Return the deployment-identity fields present on one payload."""
+    return {
+        field_name: payload[field_name]
+        for field_name in MODAL_DEPLOYMENT_PAYLOAD_FIELDS
+        if payload.get(field_name) is not None
+    }
+
+
 def _read_path_env(name: str) -> Path | None:
     """Resolve an environment variable into a path when present."""
     value = os.getenv(name)

tests/test_mapped_execution.py

@@ -2532,3 +2532,60 @@ def test_trim_subgraph_payload_to_required_nodes_drops_unrelated_mapped_branch(
     assert trimmed_payload["execute_node_ids"] == ["3"]
     assert trimmed_payload["mapped_execute_node_ids"] == []
     assert trimmed_payload["static_execute_node_ids"] == ["3"]
+
+
+def test_mapped_phase_payloads_preserve_parent_runtime_fingerprint(
+    mapped_execution_module: Any,
+    modal_deployment_module: Any,
+    monkeypatch: Any,
+) -> None:
+    """Static, mapped, and cleanup phases must expect the parent's deployed fingerprint."""
+    phase_definition = {
+        "component_node_ids": ["9"],
+        "subgraph_prompt": {},
+        "boundary_inputs": [],
+        "boundary_outputs": [],
+        "execute_node_ids": ["9"],
+    }
+    parent_payload = {
+        "payload_kind": "mapped_subgraph",
+        "component_id": "1",
+        "modal_gpu": "B300",
+        "modal_max_containers": 4,
+        "static_phase": phase_definition,
+        "mapped_phase": phase_definition,
+        "remote_session": {"session_id": "session-1"},
+        "clear_remote_session": True,
+    }
+    derived_payloads = [
+        mapped_execution_module._build_phase_subgraph_payload(
+            parent_payload, "static_phase", "1"
+        ),
+        mapped_execution_module._build_phase_subgraph_payload(
+            parent_payload, "mapped_phase", "1__mapped"
+        ),
+    ]
+    monkeypatch.setattr(
+        mapped_execution_module,
+        "is_remote_session_handle_payload",
+        lambda remote_session: True,
+    )
+    derived_payloads.append(
+        mapped_execution_module._build_remote_session_cleanup_payload(parent_payload)
+    )
+    parent_fingerprint = modal_deployment_module._expected_remote_runtime_fingerprint(
+        parent_payload
+    )
+
+    assert parent_fingerprint != (
+        modal_deployment_module._expected_remote_runtime_fingerprint(
+            {"modal_gpu": "B300"}
+        )
+    )
+    for derived_payload in derived_payloads:
+        assert derived_payload["modal_gpu"] == "B300"
+        assert derived_payload["modal_max_containers"] == 4
+        assert (
+            modal_deployment_module._expected_remote_runtime_fingerprint(derived_payload)
+            == parent_fingerprint
+        )