chore: Remove old DisaggregatedRouter, making etcd presence optional (#4011)

Signed-off-by: Graham King <[email protected]>

Author: grahamking

lib/bindings/python/rust/lib.rs

@@ -150,7 +150,6 @@ fn _core(m: &Bound<'_, PyModule>) -> PyResult<()> {
     m.add_class::<Endpoint>()?;
     m.add_class::<Client>()?;
     m.add_class::<AsyncResponseStream>()?;
-    m.add_class::<llm::disagg_router::DisaggregatedRouter>()?;
     m.add_class::<llm::entrypoint::EntrypointArgs>()?;
     m.add_class::<llm::entrypoint::EngineConfig>()?;
     m.add_class::<llm::entrypoint::EngineType>()?;

lib/bindings/python/rust/llm.rs

@@ -27,7 +27,6 @@
 use super::*;
 
 pub mod backend;
-pub mod disagg_router;
 pub mod entrypoint;
 pub mod kv;
 pub mod local_model;

lib/bindings/python/rust/llm/disagg_router.rs

@@ -217,58 +217,6 @@ class Client:
         """
         ...
 
-class DisaggregatedRouter:
-    """
-    A router that determines whether to perform prefill locally or remotely based on
-    sequence length thresholds.
-    """
-
-    def __init__(
-        self,
-        drt: DistributedRuntime,
-        model_name: str,
-        default_max_local_prefill_length: int,
-    ) -> None:
-        """
-        Create a `DisaggregatedRouter` object.
-
-        Args:
-            drt: The distributed runtime instance
-            model_name: Name of the model
-            default_max_local_prefill_length: Default maximum sequence length that can be processed locally
-        """
-        ...
-
-    def prefill_remote(self, prefill_length: int, prefix_hit_length: int) -> bool:
-        """
-        Determine if prefill should be performed remotely based on sequence lengths.
-
-        Args:
-            prefill_length: Total length of the sequence to prefill
-            prefix_hit_length: Length of the prefix that was already processed
-
-        Returns:
-            True if prefill should be performed remotely, False otherwise
-        """
-        ...
-
-    def update_value(self, max_local_prefill_length: int) -> None:
-        """
-        Update the maximum local prefill length threshold.
-
-        Args:
-            max_local_prefill_length: New maximum sequence length that can be processed locally
-        """
-        ...
-
-    def get_model_name(self) -> str:
-        """
-        Get the name of the model associated with this router.
-
-        Returns:
-            The model name as a string
-        """
-        ...
 
 def compute_block_hash_for_seq_py(tokens: List[int], kv_block_size: int) -> List[int]:
     """

lib/bindings/python/src/dynamo/_core.pyi

@@ -6,7 +6,6 @@
 import logging
 
 from dynamo._core import ApproxKvIndexer as ApproxKvIndexer
-from dynamo._core import DisaggregatedRouter as DisaggregatedRouter
 from dynamo._core import EngineType
 from dynamo._core import EntrypointArgs as EntrypointArgs
 from dynamo._core import ForwardPassMetrics as ForwardPassMetrics