chore(runtime): Do not expose etcd lease ID

Instead there is a `connection_id() -> u64` method on
`DistributedRuntime`, which is always present. Remove the `Lease` object
and the unused lease related methods.

Also delete legacy unused `DiscoveryClient`.

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

Author: grahamking

components/src/dynamo/sglang/publisher.py

@@ -123,10 +123,8 @@ def init_kv_event_publish(self) -> Optional[ZmqKvEventPublisher]:
             ep = kv_events.get("endpoint")
             zmq_ep = ep.replace("*", get_local_ip_auto()) if ep else None
 
-            lease_id = self.generate_endpoint.lease_id()
-
             zmq_config = ZmqKvEventPublisherConfig(
-                worker_id=lease_id,
+                worker_id=self.generate_endpoint.connection_id(),
                 kv_block_size=self.server_args.page_size,
                 zmq_endpoint=zmq_ep,
             )

components/src/dynamo/trtllm/main.py

@@ -421,7 +421,7 @@ async def init(runtime: DistributedRuntime, config: Config):
                 component,
                 engine,
                 kv_listener,
-                int(endpoint.lease_id()),
+                int(endpoint.connection_id()),
                 config.kv_block_size,
                 metrics_labels,
             ) as publisher:

components/src/dynamo/vllm/main.py

@@ -135,7 +135,7 @@ def setup_kv_event_publisher(
         ).replace("*", "127.0.0.1")
 
         zmq_config = ZmqKvEventPublisherConfig(
-            worker_id=generate_endpoint.lease_id(),
+            worker_id=generate_endpoint.connection_id(),
             kv_block_size=vllm_config.cache_config.block_size,
             zmq_endpoint=zmq_endpoint,
         )

examples/multimodal/components/worker.py

@@ -163,7 +163,7 @@ def setup_vllm_engine(self, component: Component, endpoint: Endpoint):
         ).replace("*", "127.0.0.1")
 
         zmq_config = ZmqKvEventPublisherConfig(
-            worker_id=endpoint.lease_id(),
+            worker_id=endpoint.connection_id(),
             kv_block_size=vllm_config.cache_config.block_size,
             zmq_endpoint=zmq_endpoint,
         )

lib/bindings/python/rust/lib.rs

@@ -773,12 +773,9 @@ impl Endpoint {
         })
     }
 
-    fn lease_id(&self) -> u64 {
-        self.inner
-            .drt()
-            .primary_lease()
-            .map(|l| l.id())
-            .unwrap_or(0)
+    // Opaque unique ID for this worker. May change over worker lifetime.
+    fn connection_id(&self) -> u64 {
+        self.inner.drt().connection_id()
     }
 
     /// Get a RuntimeMetrics helper for creating Prometheus metrics

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

@@ -991,14 +991,7 @@ async fn create_kv_router_from_endpoint(
     // Get component from endpoint
     let component = endpoint.inner.component();
 
-    // Verify we're not in static mode
-    if component.drt().primary_lease().is_none() {
-        return Err(PyErr::new::<pyo3::exceptions::PyRuntimeError, _>(
-            "Failed to get primary lease: Cannot KV route static workers",
-        ));
-    }
-
-    // Create ModelManager and use it to create KvRouter (ensures etcd registration)
+    // Create ModelManager and use it to create KvRouter (ensures registration)
     let model_manager = Arc::new(llm_rs::discovery::ModelManager::new());
     let kv_router = model_manager
         .kv_chooser_for(component, block_size as u32, kv_router_config)

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

@@ -157,9 +157,9 @@ class Endpoint:
         """
         ...
 
-    async def lease_id(self) -> int:
+    async def connection_id(self) -> int:
         """
-        Return primary lease id. Currently, cannot set a different lease id.
+        Opaque unique ID for this worker. May change over worker lifetime.
         """
         ...
 

lib/llm/src/kv_router.rs

@@ -224,13 +224,7 @@ impl KvRouter {
         consumer_uuid: String,
     ) -> Result<Self> {
         let kv_router_config = kv_router_config.unwrap_or_default();
-
-        let cancellation_token = component
-            .drt()
-            .primary_lease()
-            .expect("Cannot KV route static workers")
-            .primary_token();
-
+        let cancellation_token = component.drt().primary_token();
         let generate_endpoint = component.endpoint("generate");
         let client = generate_endpoint.client().await?;
 

lib/llm/src/kv_router/publisher.rs

@@ -784,15 +784,7 @@ impl WorkerMetricsPublisher {
     }
 
     pub async fn create_endpoint(&self, component: Component) -> Result<()> {
-        let worker_id = component
-            .drt()
-            .primary_lease()
-            .map(|lease| lease.id())
-            .unwrap_or_else(|| {
-                tracing::warn!("Component is static, assuming worker_id of 0");
-                0
-            });
-
+        let worker_id = component.drt().connection_id();
         self.start_nats_metrics_publishing(component.namespace().clone(), worker_id);
         Ok(())
     }

lib/llm/src/mocker/engine.rs

@@ -246,11 +246,7 @@ impl MockVllmEngine {
         tracing::debug!("Component found for KV events publishing");
 
         tracing::debug!("Getting worker_id");
-        let worker_id = comp
-            .drt()
-            .primary_lease()
-            .expect("Cannot publish KV events without lease") // ← This will PANIC on static!
-            .id();
+        let worker_id = comp.drt().connection_id();
         // let worker_id = 0;
         tracing::debug!("Worker_id set to: {worker_id}");