tune: reduce default node cache size cap

README.md

@@ -348,7 +348,7 @@ If a remote-marked node depends on a model filename that cannot be resolved to a
 - `COMFY_MODAL_APP_NAME` and `COMFY_MODAL_VOLUME_NAME`: Override Modal app and volume naming.
 - `COMFY_MODAL_INTERRUPT_DICT_NAME`: Override the shared Modal `Dict` used for remote cancellation flags. Default: `<app_name>-interrupts`.
 - `COMFY_MODAL_NODE_CACHE_DICT_NAME`: Override the shared Modal `Dict` used for persisted transport-safe node outputs. Default: `<app_name>-node-cache`.
-- `COMFY_MODAL_NODE_CACHE_MAX_BYTES`: Maximum raw output size eligible for persisted node caching. Default: `52428800` (50 MiB). Set to `0` to disable the persisted node-output cache.
+- `COMFY_MODAL_NODE_CACHE_MAX_BYTES`: Maximum raw output size eligible for persisted node caching. Default: `5242880` (5 MiB). Set to `0` to disable the persisted node-output cache.
 - `COMFY_MODAL_TERMINATE_CONTAINER_ON_ERROR`: When true, a remote execution crash makes the worker exit its own container after returning the error. Default: `true`.
 - `COMFY_MODAL_AUTO_DEPLOY`: Automatically deploy the Modal app on first remote invocation when lookup fails. Default: `true`.
 - `COMFY_MODAL_ALLOW_EPHEMERAL_FALLBACK`: Re-enable slow `app.run()` fallback in remote mode. Default: `false`.

docs/LEARNINGS.md

@@ -134,3 +134,4 @@
 - Static mapped sub-runs cannot reuse the same "suppress status chatter" rule as per-item mapped lanes for numeric node progress. Those static phases usually do not have a lane id at all, so the local relay has to keep lane-less `node_progress` events for execute/boundary nodes or upstream samplers like `KSampler` will run remotely without ever showing a bar in the UI.
 - Once those lane-less static progress events are forwarded, they also need their own explicit clear path. Suppressed static streams do not forward the terminal `execution_success` status event into the frontend, so relying on phase completion to clear ordinary progress leaves the last `100%` sampler bar stuck onscreen.
 - Distributed node-cache restore and persist should not be treated as unrelated phases. If one prompt run restores `NC_<hash>` into the live `PromptExecutor` cache and then persists the same node/key again unconditionally, Modal Dict churns on pure cache hits for no benefit; the persist pass should recognize restored hits and skip rewriting the same key.
+- A distributed node-cache size cap that is generous enough for medium images can still be too expensive in practice. In these logs, repeatedly serializing and writing ~25 MiB cache entries dominated the post-execution phase, so the default cap should bias toward smaller transport-safe artifacts unless there is a clear reason to keep large intermediates.

settings.py

@@ -32,7 +32,7 @@ class ModalSyncSettings:
     custom_nodes_dir: Path | None
     interrupt_dict_name: str = "comfy-modal-sync-interrupts"
     node_output_cache_dict_name: str = "comfy-modal-sync-node-cache"
-    node_output_cache_max_bytes: int = 50 * 1024 * 1024
+    node_output_cache_max_bytes: int = 5 * 1024 * 1024
     terminate_container_on_error: bool = True
     modal_gpu: str = "A100"
     scaledown_window_seconds: int = 600
@@ -179,7 +179,7 @@ def get_settings() -> ModalSyncSettings:
         ),
         node_output_cache_max_bytes=_read_int_env(
             "COMFY_MODAL_NODE_CACHE_MAX_BYTES",
-            50 * 1024 * 1024,
+            5 * 1024 * 1024,
         ),
         terminate_container_on_error=_read_bool_env("COMFY_MODAL_TERMINATE_CONTAINER_ON_ERROR")
         is not False,

tests/test_settings.py

@@ -138,6 +138,21 @@ def test_settings_defaults_node_cache_dict_name_from_app_name(
     assert settings.node_output_cache_dict_name == "my-modal-app-node-cache"
 
 
+def test_settings_defaults_node_cache_max_bytes_to_five_mib(
+    settings_module: Any,
+    monkeypatch: Any,
+) -> None:
+    """The distributed node-cache size cap should default to 5 MiB."""
+    monkeypatch.delenv("COMFY_MODAL_NODE_CACHE_MAX_BYTES", raising=False)
+    settings_module.get_settings.cache_clear()
+    try:
+        settings = settings_module.get_settings()
+    finally:
+        settings_module.get_settings.cache_clear()
+
+    assert settings.node_output_cache_max_bytes == 5 * 1024 * 1024
+
+
 def test_settings_reads_node_cache_overrides(
     settings_module: Any,
     monkeypatch: Any,