synapse/
observability.rs

1//! OpenLLMetry `gen_ai.*` span attributes; metrics are recorded through [`GatewayMetrics`].
2//! Both lanes emit the same shape so native-Vertex calls are not blind.
3
4use crate::routing::classify::Lane;
5use crate::routing::executor::Completion;
6use crate::telemetry::GatewayMetrics;
7
8/// Attributes describing one served request, used to build a tracing span and
9/// to emit metrics. `provider`/`model` reflect the leg that actually served.
10#[derive(Debug, Clone)]
11pub struct GenAiSpan {
12    pub system: &'static str, // gen_ai.system: vertexai|openai|dashscope|oai_compat
13    pub request_model: String,
14    pub response_model: String,
15    pub lane: Lane,
16    pub route: String,
17    pub tenant: String,
18    pub workspace: Option<String>,
19    pub legs_attempted: u32,
20    pub stream: bool,
21    pub input_tokens: u64,
22    pub output_tokens: u64,
23}
24
25/// Map a provider id to its OpenLLMetry `gen_ai.system` value.
26pub fn system_for(provider: &str) -> &'static str {
27    match provider {
28        "vertex" => "vertexai",
29        "openai" => "openai",
30        "qwen" => "dashscope",
31        _ => "oai_compat",
32    }
33}
34
35impl GenAiSpan {
36    pub fn from_completion(
37        c: &Completion,
38        lane: Lane,
39        route: &str,
40        tenant: &str,
41        workspace: Option<&str>,
42        legs_attempted: u32,
43        stream: bool,
44    ) -> Self {
45        Self {
46            system: system_for(&c.provider),
47            request_model: route.to_string(),
48            response_model: c.model.clone(),
49            lane,
50            route: route.to_string(),
51            tenant: tenant.to_string(),
52            workspace: workspace.map(str::to_string),
53            legs_attempted,
54            stream,
55            input_tokens: c.input_tokens,
56            output_tokens: c.output_tokens,
57        }
58    }
59
60    /// Record this request on `metrics`.
61    pub fn emit_metrics(&self, metrics: &GatewayMetrics, latency_secs: f64) {
62        metrics.request(self, latency_secs);
63    }
64}
65
66#[cfg(test)]
67mod tests {
68    use super::*;
69
70    #[test]
71    fn maps_provider_to_gen_ai_system() {
72        assert_eq!(system_for("vertex"), "vertexai");
73        assert_eq!(system_for("qwen"), "dashscope");
74        assert_eq!(system_for("openai"), "openai");
75        assert_eq!(system_for("oai_compat"), "oai_compat");
76    }
77
78    #[test]
79    fn from_completion_carries_actual_served_model() {
80        let c = Completion {
81            provider: "vertex".into(),
82            model: "gemini-3-flash".into(),
83            content: "x".into(),
84            tool_calls: Vec::new(),
85            finish_reason: crate::routing::stream::FinishReason::Stop,
86            input_tokens: 1,
87            output_tokens: 2,
88        };
89        let span =
90            GenAiSpan::from_completion(&c, Lane::Standard, "fast", "acme", Some("ws1"), 2, false);
91        assert_eq!(span.system, "vertexai");
92        assert_eq!(span.response_model, "gemini-3-flash");
93        assert_eq!(span.workspace.as_deref(), Some("ws1"));
94        assert_eq!(span.input_tokens, 1);
95    }
96}
synapse/observability.rs

synapse/
observability.rs
