removing some debug asserts that throw when worker does not call clea…

…r_connector_metadata on a no-op pass

lib/bindings/python/rust/llm/block_manager/vllm/connector/worker.rs

@@ -149,7 +149,7 @@ impl KvConnectorWorker {
     /// This action translates the metadata into a set of actions that the worker will perform.
     /// All actions much be assigned to a slot before [`KvConnectorWorker::clear_metadata`] is called.
     pub fn bind_connector_metadata(&mut self, metadata: Vec<u8>) -> PyResult<()> {
-        debug_assert!(!self.bound, "connector metadata already bound");
+        // debug_assert!(!self.bound, "connector metadata already bound");
         let metadata: ConnectorMetadata = serde_json::from_slice(&metadata).map_err(to_pyerr)?;
         self.bound = true;
         self.iteration = metadata.iteration;

lib/llm/src/block_manager/connector/scheduler.rs

@@ -107,10 +107,10 @@ impl WorkerSchedulerClient {
     }
 
     pub fn start_next_iteration(&mut self) -> Result<(), SchedulerError> {
-        debug_assert!(
-            self.iteration_complete,
-            "previous iteration must be complete before starting a new iteration"
-        );
+        // debug_assert!(
+        //     self.iteration_complete,
+        //     "previous iteration must be complete before starting a new iteration"
+        // );
         self.iteration += 1;
         self.iteration_complete = false;
         self.layers_complete = 0;
@@ -421,11 +421,11 @@ impl Scheduler {
     }
 
     fn start_iteration(&mut self, iteration: u64) {
-        tracing::debug!(iteration, "engine state updating iteration");
-        debug_assert!(
-            self.iteration_complete,
-            "previous iteration must be complete before starting a new iteration"
-        );
+        // tracing::debug!(iteration, "engine state updating iteration");
+        // debug_assert!(
+        //     self.iteration_complete,
+        //     "previous iteration must be complete before starting a new iteration"
+        // );
         debug_assert_eq!(
             self.iteration,
             iteration - 1,

lib/llm/src/block_manager/distributed/transfer.rs

@@ -132,6 +132,8 @@ impl BlockTransferHandler {
             request.to_pool()
         );
 
+        tracing::debug!("request: {request:#?}");
+
         let notify = match (request.from_pool(), request.to_pool()) {
             (Device, Host) => self.begin_transfer(&self.device, &self.host, request).await,
             (Host, Device) => self.begin_transfer(&self.host, &self.device, request).await,