Skip to content

feat: Add support for max_inflight_requests parameter to prevent unbounded memory growth in ensemble models #141

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 10 commits into
base: main
Choose a base branch
from
Open

feat: Add support for max_inflight_requests parameter to prevent unbounded memory growth in ensemble models #141

Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
08365ea
Update
pskiran1 Oct 17, 2025
50bf543
Update copyright
pskiran1 Oct 17, 2025
b08533e
Update field name
pskiran1 Oct 24, 2025
a9d4f7e
Update protobuf/model_config.proto
pskiran1 Oct 27, 2025
8dbb656
Fix pre-commit
pskiran1 Oct 27, 2025
2b8fe03
Fix pre-commit
pskiran1 Oct 27, 2025
0881524
Update protobuf/model_config.proto
pskiran1 Oct 31, 2025
c1f4434
Update
pskiran1 Oct 31, 2025
0b5f2d1
Update protobuf/model_config.proto
pskiran1 Oct 31, 2025
efd890a
Update description
pskiran1 Oct 31, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 16 additions & 1 deletion protobuf/model_config.proto
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// Copyright 2018-2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
// Copyright 2018-2025, 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
Expand Down Expand Up @@ -1659,6 +1659,21 @@ message ModelEnsembling
//@@ The models and the input / output mappings used within the ensemble.
//@@
repeated Step step = 1;

//@@ .. cpp:var:: uint32 max_inflight_requests
//@@
//@@ The maximum number of concurrent inflight requests allowed at each
//@@ ensemble step per inference request. This limit prevents unbounded
//@@ memory growth when ensemble steps produce responses faster than
//@@ downstream steps can consume, e.g. decoupled models.
//@@ Default value is 0, which indicates that no limit is enforced.
//@@
//@@ Note: Applying this limit may block upstream steps while they wait
//@@ for downstream capacity. This blocking does not cancel or internally
//@@ time out intermediate requests, but clients may experience increased
//@@ end-to-end latency.
//@@
uint32 max_inflight_requests = 2;
}

//@@
Expand Down