lib.rs 1.79 KB
Newer Older
1
2
3
4
5
6
7
8
9
// SPDX-FileCopyrightText: Copyright (c) 2024-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
// SPDX-License-Identifier: Apache-2.0

//! KV Router - Radix tree data structures for LLM KV cache routing.
//!
//! This crate provides the core radix tree implementation and protocols for
//! efficient KV cache lookup and routing in distributed LLM inference systems.

pub mod approx;
Yan Ru Pei's avatar
Yan Ru Pei committed
10
pub mod concurrent_radix_tree;
11
pub mod indexer;
12
13
#[cfg(feature = "bench")]
pub mod naive_indexers;
Yan Ru Pei's avatar
Yan Ru Pei committed
14
pub mod nested_map;
15
pub mod protocols;
16
pub mod radix_tree;
17
18
pub mod scheduling;
pub mod sequences;
19
pub mod zmq_wire;
20
21
22
23
24
25
26

// Backward-compat re-exports: preserve old module paths for external consumers
pub use scheduling::config;
pub use scheduling::queue;
pub use scheduling::selector;
pub use sequences::multi_worker as multi_worker_sequence;
pub use sequences::single as sequence;
27

28
29
#[cfg(any(test, feature = "bench"))]
pub mod test_utils;
Yan Ru Pei's avatar
Yan Ru Pei committed
30

31
// Re-export key types for convenience
32
pub use self::multi_worker_sequence::{
33
34
35
    ActiveSequencesMultiWorker, SequenceError, SequencePublisher, SequenceRequest,
    SequenceSubscriber,
};
36
37
38
39
pub use self::sequence::{ActiveSequences, RequestId};
pub use concurrent_radix_tree::ConcurrentRadixTree;
pub use config::{KvRouterConfig, RouterConfigOverride};
pub use indexer::{MaybeError, SyncIndexer, ThreadPoolIndexer};
40
41
#[cfg(feature = "bench")]
pub use naive_indexers::{InvertedIndex, NaiveNestedMap};
Yan Ru Pei's avatar
Yan Ru Pei committed
42
pub use nested_map::PositionalIndexer;
43
pub use protocols::{
44
    KvCacheEventError, LocalBlockHash, OverlapScores, RouterEvent, WorkerConfigLike, WorkerId,
45
46
    compute_block_hash_for_seq,
};
47
pub use queue::SchedulerQueue;
48
pub use radix_tree::RadixTree;
49
50
pub use scheduling::{KvSchedulerError, PotentialLoad, SchedulingRequest, SchedulingResponse};
pub use selector::{DefaultWorkerSelector, WorkerSelector};