1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258//! Provider health probing — lightweight HTTP checks for local LLM providers.
//!
//! Probes local providers (Ollama, vLLM, LM Studio) for reachability and
//! dynamically discovers which models they currently serve.
use std::time::Instant;
/// Result of probing a provider endpoint.
#[derive(Debug, Clone, Default)]
pub struct ProbeResult {
/// Whether the provider responded successfully.
pub reachable: bool,
/// Round-trip latency in milliseconds.
pub latency_ms: u64,
/// Model IDs discovered from the provider's listing endpoint.
pub discovered_models: Vec<String>,
/// Error message if the probe failed.
pub error: Option<String>,
}
/// Check if a provider is a local provider (no key required, localhost URL).
///
/// Returns true for `"ollama"`, `"vllm"`, `"lmstudio"`.
pub fn is_local_provider(provider: &str) -> bool {
matches!(
provider.to_lowercase().as_str(),
"ollama" | "vllm" | "lmstudio"
)
}
/// Probe timeout for local provider health checks.
const PROBE_TIMEOUT_SECS: u64 = 5;
/// Probe a provider's health by hitting its model listing endpoint.
///
/// - **Ollama**: `GET {base_url_root}/api/tags` → parses `.models[].name`
/// - **OpenAI-compat** (vLLM, LM Studio): `GET {base_url}/models` → parses `.data[].id`
///
/// `base_url` should be the provider's base URL from the catalog (e.g.,
/// `http://localhost:11434/v1` for Ollama, `http://localhost:8000/v1` for vLLM).
pub async fn probe_provider(provider: &str, base_url: &str) -> ProbeResult {
let start = Instant::now();
let client = match reqwest::Client::builder()
.timeout(std::time::Duration::from_secs(PROBE_TIMEOUT_SECS))
.build()
{
Ok(c) => c,
Err(e) => {
return ProbeResult {
error: Some(format!("Failed to build HTTP client: {e}")),
..Default::default()
};
}
};
let lower = provider.to_lowercase();
// Ollama uses a non-OpenAI endpoint for model listing
let (url, is_ollama) = if lower == "ollama" {
// base_url is typically "http://localhost:11434/v1" — strip /v1 for the tags endpoint
let root = base_url
.trim_end_matches('/')
.trim_end_matches("/v1")
.trim_end_matches("/v1/");
(format!("{root}/api/tags"), true)
} else {
// OpenAI-compatible: GET {base_url}/models
let trimmed = base_url.trim_end_matches('/');
(format!("{trimmed}/models"), false)
};
let resp = match client.get(&url).send().await {
Ok(r) => r,
Err(e) => {
return ProbeResult {
latency_ms: start.elapsed().as_millis() as u64,
error: Some(format!("{e}")),
..Default::default()
};
}
};
if !resp.status().is_success() {
return ProbeResult {
latency_ms: start.elapsed().as_millis() as u64,
error: Some(format!("HTTP {}", resp.status())),
..Default::default()
};
}
let body: serde_json::Value = match resp.json().await {
Ok(v) => v,
Err(e) => {
return ProbeResult {
reachable: true, // server responded, just bad JSON
latency_ms: start.elapsed().as_millis() as u64,
error: Some(format!("Invalid JSON: {e}")),
..Default::default()
};
}
};
let latency_ms = start.elapsed().as_millis() as u64;
// Parse model names
let models = if is_ollama {
// Ollama: { "models": [ { "name": "llama3.2:latest", ... }, ... ] }
body.get("models")
.and_then(|v| v.as_array())
.map(|arr| {
arr.iter()
.filter_map(|m| {
m.get("name")
.and_then(|n| n.as_str())
.map(|s| s.to_string())
})
.collect()
})
.unwrap_or_default()
} else {
// OpenAI-compatible: { "data": [ { "id": "model-name", ... }, ... ] }
body.get("data")
.and_then(|v| v.as_array())
.map(|arr| {
arr.iter()
.filter_map(|m| m.get("id").and_then(|n| n.as_str()).map(|s| s.to_string()))
.collect()
})
.unwrap_or_default()
};
ProbeResult {
reachable: true,
latency_ms,
discovered_models: models,
error: None,
}
}
/// Lightweight model probe -- sends a minimal completion request to verify a model is responsive.
///
/// Unlike `probe_provider` which checks the listing endpoint, this actually sends
/// a tiny prompt ("Hi") to verify the model can generate completions. Used by the
/// circuit breaker to re-test a provider during cooldown.
///
/// Returns `Ok(latency_ms)` if the model responds, or `Err(error_message)` if it fails.
pub async fn probe_model(
provider: &str,
base_url: &str,
model: &str,
api_key: Option<&str>,
) -> Result<u64, String> {
let start = Instant::now();
let client = reqwest::Client::builder()
.timeout(std::time::Duration::from_secs(10))
.build()
.map_err(|e| format!("HTTP client error: {e}"))?;
let url = format!("{}/chat/completions", base_url.trim_end_matches('/'));
let body = serde_json::json!({
"model": model,
"messages": [{"role": "user", "content": "Hi"}],
"max_tokens": 1,
"temperature": 0.0
});
let mut req = client.post(&url).json(&body);
if let Some(key) = api_key {
// Detect provider to set correct auth header
let lower = provider.to_lowercase();
if lower == "gemini" {
req = req.header("x-goog-api-key", key);
} else {
req = req.header("Authorization", format!("Bearer {key}"));
}
}
let resp = req.send().await.map_err(|e| format!("{e}"))?;
let latency = start.elapsed().as_millis() as u64;
if resp.status().is_success() {
Ok(latency)
} else {
let status = resp.status().as_u16();
let body = resp.text().await.unwrap_or_default();
Err(format!("HTTP {status}: {}", &body[..body.len().min(200)]))
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn test_is_local_provider_true_for_ollama() {
assert!(is_local_provider("ollama"));
assert!(is_local_provider("Ollama"));
assert!(is_local_provider("OLLAMA"));
assert!(is_local_provider("vllm"));
assert!(is_local_provider("lmstudio"));
}
#[test]
fn test_is_local_provider_false_for_openai() {
assert!(!is_local_provider("openai"));
assert!(!is_local_provider("anthropic"));
assert!(!is_local_provider("gemini"));
assert!(!is_local_provider("groq"));
}
#[test]
fn test_probe_result_default() {
let result = ProbeResult::default();
assert!(!result.reachable);
assert_eq!(result.latency_ms, 0);
assert!(result.discovered_models.is_empty());
assert!(result.error.is_none());
}
#[tokio::test]
async fn test_probe_unreachable_returns_error() {
// Probe a port that's almost certainly not running a server
let result = probe_provider("ollama", "http://127.0.0.1:19999").await;
assert!(!result.reachable);
assert!(result.error.is_some());
}
#[test]
fn test_probe_timeout_value() {
assert_eq!(PROBE_TIMEOUT_SECS, 5);
}
#[test]
fn test_probe_model_url_construction() {
// Verify the URL format logic used inside probe_model.
let url = format!(
"{}/chat/completions",
"http://localhost:8000/v1".trim_end_matches('/')
);
assert_eq!(url, "http://localhost:8000/v1/chat/completions");
let url2 = format!(
"{}/chat/completions",
"http://localhost:8000/v1/".trim_end_matches('/')
);
assert_eq!(url2, "http://localhost:8000/v1/chat/completions");
}
#[tokio::test]
async fn test_probe_model_unreachable() {
let result = probe_model("test", "http://127.0.0.1:19998/v1", "test-model", None).await;
assert!(result.is_err());
}
}