Skip to main content

switchyard_libsy/algorithms/
hierarchical.rs

1// SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
2// SPDX-License-Identifier: Apache-2.0
3
4//! Routing that stacks a judge over a stage router.
5//!
6//! The judge runs as a [`Processor`]: it sets configuration the stage router reads,
7//! and picks no target itself.
8
9use std::collections::HashMap;
10use std::sync::Arc;
11
12use async_trait::async_trait;
13use parking_lot::Mutex;
14
15use super::fall_through::FallThrough;
16use super::llm_class::{LlmClassifierConfig, LlmTaskClassifier, TaskClassifierConfig};
17use super::stage::{StageRouterConfig, build_stage_route};
18use super::util::affinity::{ClassifyTrigger, evict_if_full, has_new_user_turn, retention_key};
19use super::util::stage::{StageTargets, Tier, set_fall_open};
20use crate::core::algorithm::{Algorithm, Driver, RoutingIdentity};
21use crate::core::classifier::Classifier;
22use crate::core::processor::{Event, Processor};
23use crate::core::state::State;
24use crate::{LibsyError, Result};
25use switchyard_protocol::{ModelId, Request};
26
27const HIERARCHICAL: &str = "hierarchical";
28
29/// Sets the stage router's fall-open tier from a judge verdict.
30///
31/// Retains the tier per routing identity, so it survives requests that carry no
32/// session ID when `message_hash_fallback` is on. The retained tier is replayed
33/// into state on every request so the cascade below reads it.
34struct TierSetter {
35    judge: Arc<dyn Classifier<State>>,
36    targets: StageTargets,
37    trigger: ClassifyTrigger,
38    message_hash_fallback: bool,
39    tiers: Mutex<HashMap<RoutingIdentity, Tier>>,
40}
41
42impl TierSetter {
43    /// Two requests for one identity can both pass this and both judge, since a
44    /// judge call sits between here and [`retain`](Self::retain). The later wins.
45    fn is_due(&self, identity: Option<&RoutingIdentity>, request: &Request) -> bool {
46        match self.trigger {
47            ClassifyTrigger::UserTurn => has_new_user_turn(&request.llm_request.messages),
48            // Unkeyed requests cannot be told apart, so every one is a new session.
49            ClassifyTrigger::NewSession => {
50                identity.is_none_or(|identity| !self.tiers.lock().contains_key(identity))
51            }
52            // Rejected by the constructor, and only in the enum for the standalone route.
53            ClassifyTrigger::EveryRequest => true,
54        }
55    }
56
57    fn retain(&self, identity: RoutingIdentity, tier: Tier) {
58        let mut tiers = self.tiers.lock();
59        // A user turn re-decides, so it overwrites. A session keeps its first verdict,
60        // matching how affinity retains an assignment.
61        let writable = self.trigger == ClassifyTrigger::UserTurn || !tiers.contains_key(&identity);
62        if writable {
63            evict_if_full(&mut tiers);
64            tiers.insert(identity, tier);
65        }
66    }
67}
68
69#[async_trait]
70impl Processor<State> for TierSetter {
71    async fn process(&self, state: &mut State, event: Event<'_>) -> Result<()> {
72        let Event::Request { request, driver } = event else {
73            return Ok(());
74        };
75        let identity = retention_key(request, self.message_hash_fallback);
76        if self.is_due(identity.as_ref(), request) {
77            let (classification, _) = self.judge.score(state, request, driver).await?;
78            if let Some(winner) = classification.argmax(false)?
79                && let Some(tier) = self.targets.tier_for(&winner.target)
80            {
81                set_fall_open(state, tier);
82                if let Some(identity) = identity {
83                    self.retain(identity, tier);
84                }
85                return Ok(());
86            }
87        }
88        // Either not this request's turn or the judge had no verdict, so the last
89        // tier stands rather than dropping back to the picker default.
90        if let Some(tier) = identity.and_then(|identity| self.tiers.lock().get(&identity).copied())
91        {
92            set_fall_open(state, tier);
93        }
94        Ok(())
95    }
96}
97
98/// A judge stacked over a stage router.
99pub struct HierarchicalRouterConfig {
100    /// Target the judge is called through. Not a routing destination.
101    pub judge_target: ModelId,
102    /// Judge settings, including how often `classify_trigger` runs it.
103    pub judge: TaskClassifierConfig,
104    /// Serves the turns, with the tier the judge picked as its fall-open default.
105    pub stage: StageRouterConfig,
106}
107
108/// Runs a stage router with a tier the judge picks.
109pub struct HierarchicalRouter {
110    route: FallThrough<State>,
111}
112
113impl HierarchicalRouter {
114    /// Stacks the judge over a stage router across the same tier pair.
115    ///
116    /// Errors on a configuration either algorithm rejects and on `every_request`.
117    ///
118    /// A stage router carrying its own judge is allowed, but that judge sits ahead
119    /// of the fall-open tier and so answers most of the turns this one set a tier for.
120    pub fn new(
121        capable: ModelId,
122        efficient: ModelId,
123        config: HierarchicalRouterConfig,
124    ) -> Result<Self> {
125        if config.judge.classify_trigger == ClassifyTrigger::EveryRequest {
126            return Err(LibsyError::AlgorithmError {
127                message: "hierarchical: classify_trigger must be user_turn or new_session"
128                    .to_string(),
129            });
130        }
131        let trigger = config.judge.classify_trigger;
132        let message_hash_fallback = config.judge.message_hash_fallback;
133        // Only the judge's Classifier face is used, so its own affinity never runs.
134        // Leaving these set would apply the standalone route's pairing rules to a
135        // trigger this router implements itself.
136        let judge_config = TaskClassifierConfig {
137            classify_trigger: ClassifyTrigger::EveryRequest,
138            message_hash_fallback: false,
139            ..config.judge
140        };
141        let judge = LlmTaskClassifier::new(LlmClassifierConfig::Capability {
142            judge_target: config.judge_target,
143            efficient_target: efficient.clone(),
144            capable_target: capable.clone(),
145            config: judge_config,
146        })?;
147        let setter = TierSetter {
148            judge: Arc::new(judge),
149            targets: StageTargets::new(capable.clone(), efficient.clone()),
150            trigger,
151            message_hash_fallback,
152            tiers: Mutex::new(HashMap::new()),
153        };
154        let route = build_stage_route(capable, efficient, config.stage)?
155            .with_name(HIERARCHICAL)
156            .with_processor(Arc::new(setter));
157        Ok(Self { route })
158    }
159}
160
161#[async_trait]
162impl Algorithm for HierarchicalRouter {
163    fn name(&self) -> &str {
164        HIERARCHICAL
165    }
166
167    async fn route(
168        self: Arc<Self>,
169        driver: Driver,
170        request: Request,
171    ) -> Result<crate::RoutingOutcome> {
172        self.route.execute(driver, request).await
173    }
174}
175
176#[cfg(test)]
177mod tests {
178    use std::sync::Arc;
179
180    use switchyard_protocol::{Message, Role};
181
182    use super::*;
183    use crate::algorithms::util::stage::PickerMode;
184    use crate::algorithms::util::tier_fixtures::{JUDGE, Recorder, turn_request};
185    use crate::core::testing::test_drive;
186
187    fn user_turn_request() -> Request {
188        let mut request = turn_request(false);
189        request
190            .llm_request
191            .messages
192            .push(Message::text(Role::User, "now rewrite the parser"));
193        request
194    }
195
196    /// The same request shape with no session ID, so only the hash can key it.
197    fn unkeyed(mut request: Request) -> Request {
198        if let Some(metadata) = request.metadata.as_mut() {
199            metadata.session_id = None;
200        }
201        request
202    }
203
204    fn hash_keyed_router() -> Result<Arc<HierarchicalRouter>> {
205        Ok(Arc::new(HierarchicalRouter::new(
206            ModelId::from("strong"),
207            ModelId::from("weak"),
208            HierarchicalRouterConfig {
209                judge_target: ModelId::from(JUDGE),
210                judge: TaskClassifierConfig {
211                    base_threshold: 0.5,
212                    classify_trigger: ClassifyTrigger::UserTurn,
213                    message_hash_fallback: true,
214                    ..Default::default()
215                },
216                stage: StageRouterConfig::new(PickerMode::EfficientFirst, 0.5),
217            },
218        )?))
219    }
220
221    fn router() -> Result<Arc<HierarchicalRouter>> {
222        Ok(Arc::new(HierarchicalRouter::new(
223            ModelId::from("strong"),
224            ModelId::from("weak"),
225            HierarchicalRouterConfig {
226                judge_target: ModelId::from(JUDGE),
227                judge: TaskClassifierConfig {
228                    base_threshold: 0.5,
229                    classify_trigger: ClassifyTrigger::UserTurn,
230                    ..Default::default()
231                },
232                stage: StageRouterConfig::new(PickerMode::EfficientFirst, 0.5),
233            },
234        )?))
235    }
236
237    #[test]
238    fn rejects_every_request_as_a_trigger() {
239        let config = HierarchicalRouterConfig {
240            judge_target: ModelId::from(JUDGE),
241            judge: TaskClassifierConfig::default(),
242            stage: StageRouterConfig::new(PickerMode::EfficientFirst, 0.5),
243        };
244        assert!(matches!(
245            HierarchicalRouter::new(ModelId::from("strong"), ModelId::from("weak"), config),
246            Err(LibsyError::AlgorithmError { .. })
247        ));
248    }
249
250    #[tokio::test]
251    async fn a_session_without_an_id_keys_on_the_message_hash() -> Result<()> {
252        let recorder = Arc::new(Recorder::default());
253        *recorder.judge_p_solve.lock() = 0.1;
254        let router = hash_keyed_router()?;
255
256        test_drive(
257            router.clone(),
258            unkeyed(user_turn_request()),
259            recorder.serve(),
260        )
261        .await?;
262        test_drive(
263            router.clone(),
264            unkeyed(turn_request(false)),
265            recorder.serve(),
266        )
267        .await?;
268
269        assert_eq!(
270            recorder.judge_calls(),
271            1,
272            "a tool step is not a user turn, session id or not"
273        );
274        assert_eq!(
275            recorder.routed()[1].target,
276            "strong",
277            "and the tier survives the tool step"
278        );
279        Ok(())
280    }
281
282    #[tokio::test]
283    async fn the_judge_sets_the_tier_once_a_turn_and_the_signals_run_within_it() -> Result<()> {
284        let recorder = Arc::new(Recorder::default());
285        *recorder.judge_p_solve.lock() = 0.1;
286        let router = router()?;
287
288        test_drive(router.clone(), user_turn_request(), recorder.serve()).await?;
289        test_drive(router.clone(), turn_request(false), recorder.serve()).await?;
290
291        let routed = recorder.routed();
292        assert_eq!(
293            routed[0].target, "strong",
294            "a quiet turn falls open to the verdict"
295        );
296        assert_eq!(
297            routed[1].target, "strong",
298            "which holds across the tool steps after it"
299        );
300        assert_eq!(
301            recorder.judge_calls(),
302            1,
303            "a tool step is not a new user turn"
304        );
305        Ok(())
306    }
307}