switchyard_runner/
runner.rs

1// SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
2// SPDX-License-Identifier: Apache-2.0
3
4//! Named route table and server-facing route metadata.
5
6use std::collections::BTreeMap;
7use std::path::Path;
8use std::sync::Arc;
9use std::time::Duration;
10
11use libsy::RoutingOutcome;
12use serde_json::Value;
13use switchyard_llm_client::TranslatingLlmClient;
14use switchyard_protocol::{ModelId, WireFormat};
15
16use crate::config;
17use crate::{ModelCapabilities, Route, RunnerError};
18
19/// Immutable named route table.
20pub struct Runner {
21    routes: Vec<(ModelId, Route)>,
22    fallback_base_url: Option<String>,
23    provider_api_keys: Vec<String>,
24    model_properties_probes: BTreeMap<ModelId, ModelPropertiesProbe>,
25}
26
27/// Borrowed model metadata returned while listing routes.
28pub struct ModelInfo<'a> {
29    pub id: &'a ModelId,
30    pub algorithm: &'a str,
31    pub capabilities: ModelCapabilities,
32}
33
34/// Fully resolved routing decision.
35pub struct DecisionDescription {
36    pub selected: DecisionTarget,
37    pub fallbacks: Vec<DecisionTarget>,
38    /// Metadata the algorithm attached to the routing outcome, if any.
39    pub metadata: Option<libsy::OutcomeMetadata>,
40}
41
42/// Non-secret configured target details returned by the decision endpoint.
43#[derive(Clone)]
44pub struct DecisionTarget {
45    pub target: String,
46    pub model: ModelId,
47    pub format: WireFormat,
48    pub base_url: String,
49    pub extra_body: BTreeMap<String, Value>,
50    /// Effort the target forces, so a caller that dispatches the call itself can apply it.
51    pub reasoning_effort: Option<String>,
52}
53
54pub(crate) struct ModelPropertiesProbe {
55    pub(crate) model: ModelId,
56    pub(crate) format: WireFormat,
57    pub(crate) client: Arc<TranslatingLlmClient>,
58}
59
60impl Runner {
61    /// Loads and validates a version-1 deployment TOML file.
62    pub fn load(path: impl AsRef<Path>) -> Result<Self, RunnerError> {
63        config::load_runner(path)
64    }
65
66    /// Loads and validates a version-1 deployment TOML document.
67    ///
68    /// Use [`Self::load`] when the deployment is stored in a file.
69    pub fn from_toml(source: &str) -> Result<Self, RunnerError> {
70        config::runner_from_toml(source)
71    }
72
73    /// Builds a runner from named routes in caller-provided order.
74    /// Pre-condition: There must be at least one route.
75    pub fn new(routes: Vec<(ModelId, Route)>) -> Self {
76        Self {
77            routes,
78            fallback_base_url: None,
79            provider_api_keys: Vec::new(),
80            model_properties_probes: BTreeMap::new(),
81        }
82    }
83
84    /// Registers deployment-owned API keys for serving-surface output redaction.
85    /// TOML loading registers these automatically; programmatic hosts must supply them.
86    pub fn with_provider_api_keys(mut self, keys: Vec<String>) -> Self {
87        self.provider_api_keys = keys;
88        self
89    }
90
91    /// Returns deployment-owned secrets for serving-surface output redactors.
92    pub fn provider_api_keys(&self) -> &[String] {
93        &self.provider_api_keys
94    }
95
96    pub(crate) fn with_fallback_url(mut self, fallback_base_url: Option<String>) -> Self {
97        self.fallback_base_url = fallback_base_url;
98        self
99    }
100
101    pub(crate) fn with_model_properties_probes(
102        mut self,
103        probes: BTreeMap<ModelId, ModelPropertiesProbe>,
104    ) -> Self {
105        self.model_properties_probes = probes;
106        self
107    }
108
109    /// Returns the route registered for a model.
110    pub fn route(&self, model: &str) -> Option<&Route> {
111        self.routes
112            .iter()
113            .find(|(id, _)| id.as_str() == model)
114            .map(|(_, route)| route)
115    }
116
117    /// Iterates over configured routes in caller-provided order.
118    pub fn models(&self) -> impl Iterator<Item = ModelInfo<'_>> {
119        self.routes.iter().map(|(id, route)| ModelInfo {
120            id,
121            algorithm: route.algorithm_name(),
122            capabilities: route.capabilities(),
123        })
124    }
125
126    /// Reads one route's configured local backend properties without exposing its credentials.
127    pub async fn model_properties(&self, model: &str, timeout: Duration) -> Option<Value> {
128        let probe = self.model_properties_probes.get(model)?;
129        probe
130            .client
131            .get_model_properties(&probe.model, probe.format, timeout)
132            .await
133            .ok()
134    }
135
136    /// Returns the validated API root used for unmatched HTTP requests.
137    pub fn fallback_base_url(&self) -> Option<&str> {
138        self.fallback_base_url.as_deref()
139    }
140
141    /// Resolves an outcome to configured target names and non-secret client settings.
142    pub fn describe_decision(
143        &self,
144        model: &ModelId,
145        outcome: &RoutingOutcome,
146    ) -> Option<DecisionDescription> {
147        let route = self.route(model.as_str())?;
148        let resolve = |selected: &ModelId| {
149            let mut target = route.decision_target(selected)?;
150            let mut url = reqwest::Url::parse(&target.base_url).ok()?;
151            let query: Vec<_> = url
152                .query_pairs()
153                .filter(|(name, _)| !matches!(name.as_ref(), "key" | "api_key"))
154                .map(|(name, value)| (name.into_owned(), value.into_owned()))
155                .collect();
156            if query.len() != url.query_pairs().count() {
157                url.set_query(None);
158                if !query.is_empty() {
159                    url.query_pairs_mut().extend_pairs(query);
160                }
161                // Only the returned metadata changes; inference still needs its credentials.
162                target.base_url = url.into();
163            }
164            Some(target)
165        };
166        let mut model_ids = outcome.selected_model_ids.iter();
167        Some(DecisionDescription {
168            selected: resolve(model_ids.next()?)?,
169            fallbacks: model_ids.map(resolve).collect::<Option<Vec<_>>>()?,
170            metadata: outcome.metadata.clone(),
171        })
172    }
173}
switchyard_runner/runner.rs

switchyard_runner/
runner.rs
