Merge branch 'main' of https://github.com/triton-inference-server/core into spolisetty/tri-26-triton-dali-ensemble-model-memory-issue

Author: pskiran1

src/infer_request.cc

@@ -481,11 +481,13 @@ InferenceRequest::Release(
   // object.
   auto& release_callbacks = request->release_callbacks_;
   for (int i = release_callbacks.size() - 1; i >= 0; --i) {
-    auto [release_fn, is_internal] = release_callbacks[i];
-    if (is_internal) {
+    // Callbacks must be invoked before erasing
+    RETURN_IF_ERROR(release_callbacks[i].first(request, release_flags));
+    if (release_callbacks[i].second) {
+      // Erase internal callbacks to avoid duplicate callbacks in case of
+      // reusing the InferenceRequest object
       release_callbacks.erase(release_callbacks.begin() + i);
     }
-    release_fn(request, release_flags);
     if (request == nullptr) {
       return Status::Success;
     }

src/sequence_batch_scheduler/sequence_utils.h

@@ -1,4 +1,4 @@
-// Copyright 2023, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+// Copyright 2023-2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 //
 // Redistribution and use in source and binary forms, with or without
 // modification, are permitted provided that the following conditions
@@ -93,7 +93,10 @@ class IterativeSequencer : public Sequencer {
       std::unique_ptr<InferenceRequest>& irequest,
       InferenceRequest::InternalReleaseFn&& callback) override
   {
-    if (irequest->Flags() & TRITONSERVER_REQUEST_FLAG_SEQUENCE_START) {
+    // Internal release callbacks are removed after getting invoked in
+    // InferenceRequest::Release. Make sure internal release callback is added
+    // for each iterative sequence request.
+    if (!(irequest->Flags() & TRITONSERVER_REQUEST_FLAG_SEQUENCE_END)) {
       irequest->AddInternalReleaseCallback(std::move(callback));
     }
   }