feat(otel): propagate session.id to spans and log records

Cargo.toml

@@ -71,8 +71,9 @@ url = "2.5.8"
 opentelemetry = "0.31"
 opentelemetry_sdk = { version = "0.31", features = ["metrics"] }
 opentelemetry-otlp = "0.31"
-opentelemetry-appender-tracing = "0.31"
+opentelemetry-appender-tracing = { version = "0.31", features = ["experimental_span_attributes"] }
 opentelemetry-stdout = { version = "0.31", features = ["trace", "metrics", "logs"] }
+tracing-futures = { version = "0.2", features = ["futures-03"] }
 tracing-opentelemetry = "0.32"
 
 rayon = "1.10"
@@ -88,4 +89,11 @@ tree-sitter-swift = "0.7"
 tree-sitter-typescript = "0.23"
 
 [patch.crates-io]
-v8 = { path = "vendor/v8" }
+v8 = { path = "vendor/v8" }
+# TODO: switch to released version in opentelemetry 0.32.0
+# https://github.com/open-telemetry/opentelemetry-rust/issues/3408
+opentelemetry = { git = "https://github.com/open-telemetry/opentelemetry-rust", rev = "345cd74a" }
+opentelemetry_sdk = { git = "https://github.com/open-telemetry/opentelemetry-rust", rev = "345cd74a" }
+opentelemetry-appender-tracing = { git = "https://github.com/open-telemetry/opentelemetry-rust", rev = "345cd74a" }
+opentelemetry-otlp = { git = "https://github.com/open-telemetry/opentelemetry-rust", rev = "345cd74a" }
+opentelemetry-stdout = { git = "https://github.com/open-telemetry/opentelemetry-rust", rev = "345cd74a" }

crates/goose-test-support/Cargo.toml

@@ -9,6 +9,8 @@ description.workspace = true
 
 [dependencies]
 axum = "0.7"
+env-lock = { workspace = true }
+opentelemetry = { workspace = true }
 rmcp = { workspace = true, features = ["server", "macros", "transport-streamable-http-server"] }
 serde_json.workspace = true
 tokio.workspace = true

crates/goose-test-support/src/lib.rs

@@ -1,4 +1,5 @@
 pub mod mcp;
+pub mod otel;
 pub mod session;
 
 pub use mcp::{McpFixture, FAKE_CODE, TEST_IMAGE_B64};

crates/goose-test-support/src/otel.rs

@@ -0,0 +1,62 @@
+use opentelemetry::global;
+use opentelemetry::metrics::{Meter, MeterProvider};
+use opentelemetry::InstrumentationScope;
+use std::env;
+use std::sync::Arc;
+
+struct SavedMeterProvider(Arc<dyn MeterProvider + Send + Sync>);
+
+impl MeterProvider for SavedMeterProvider {
+    fn meter_with_scope(&self, scope: InstrumentationScope) -> Meter {
+        self.0.meter_with_scope(scope)
+    }
+}
+
+pub struct OtelTestGuard {
+    pub _env: env_lock::EnvGuard<'static>,
+    prev_tracer: global::GlobalTracerProvider,
+    prev_meter: Arc<dyn MeterProvider + Send + Sync>,
+}
+
+impl Drop for OtelTestGuard {
+    fn drop(&mut self) {
+        global::set_tracer_provider(self.prev_tracer.clone());
+        global::set_meter_provider(SavedMeterProvider(self.prev_meter.clone()));
+    }
+}
+
+pub fn clear_otel_env(overrides: &[(&'static str, &'static str)]) -> OtelTestGuard {
+    let prev_tracer = global::tracer_provider();
+    let prev_meter = global::meter_provider();
+
+    let mut keys: Vec<&'static str> = vec![
+        "OTEL_EXPORTER_OTLP_ENDPOINT",
+        "OTEL_EXPORTER_OTLP_LOGS_ENDPOINT",
+        "OTEL_EXPORTER_OTLP_METRICS_ENDPOINT",
+        "OTEL_EXPORTER_OTLP_METRICS_TEMPORALITY_PREFERENCE",
+        "OTEL_EXPORTER_OTLP_TIMEOUT",
+        "OTEL_EXPORTER_OTLP_TRACES_ENDPOINT",
+        "OTEL_LOG_LEVEL",
+        "OTEL_LOGS_EXPORTER",
+        "OTEL_METRICS_EXPORTER",
+        "OTEL_RESOURCE_ATTRIBUTES",
+        "OTEL_SDK_DISABLED",
+        "OTEL_SERVICE_NAME",
+        "OTEL_TRACES_EXPORTER",
+    ];
+    for &(k, _) in overrides {
+        if !keys.contains(&k) {
+            keys.push(k);
+        }
+    }
+
+    let guard = env_lock::lock_env(keys.into_iter().map(|k| (k, None::<&str>)));
+    for &(k, v) in overrides {
+        env::set_var(k, v);
+    }
+    OtelTestGuard {
+        _env: guard,
+        prev_tracer,
+        prev_meter,
+    }
+}

crates/goose/Cargo.toml

@@ -54,10 +54,11 @@ webbrowser = { workspace = true }
 lazy_static = "1.5.0"
 tracing = { workspace = true }
 tracing-subscriber = { workspace = true }
+tracing-futures = { workspace = true }
 tracing-opentelemetry = { workspace = true }
 opentelemetry = { workspace = true }
-opentelemetry-appender-tracing = { workspace = true }
 opentelemetry_sdk = { workspace = true }
+opentelemetry-appender-tracing = { workspace = true }
 opentelemetry-otlp = { workspace = true }
 opentelemetry-stdout = { workspace = true }
 keyring = { version = "3.6.2", features = [
@@ -163,6 +164,7 @@ ctor = "0.2.9"
 test-case = { workspace = true }
 env-lock = { workspace = true }
 rmcp = { workspace = true, features = ["transport-streamable-http-server"] }
+opentelemetry_sdk = { workspace = true, features = ["testing"] }
 goose-test-support = { path = "../goose-test-support" }
 
 [[example]]

crates/goose/src/agents/agent.rs

@@ -7,6 +7,7 @@ use std::sync::Arc;
 use anyhow::{anyhow, Context, Result};
 use futures::stream::BoxStream;
 use futures::{stream, FutureExt, Stream, StreamExt, TryStreamExt};
+use tracing_futures::Instrument;
 use uuid::Uuid;
 
 use super::container::Container;
@@ -495,7 +496,7 @@ impl Agent {
     }
 
     /// Dispatch a single tool call to the appropriate client
-    #[instrument(skip(self, tool_call, request_id, session), fields(input, output, session_id = %session.id))]
+    #[instrument(skip(self, tool_call, request_id, cancellation_token, session), fields(input, output, session.id = %session.id))]
     pub async fn dispatch_tool_call(
         &self,
         tool_call: CallToolRequestParams,
@@ -874,8 +875,8 @@ impl Agent {
     }
 
     #[instrument(
-        skip(self, user_message, session_config),
-        fields(user_message, trace_input)
+        skip(self, user_message, session_config, cancel_token),
+        fields(user_message, trace_input, session.id = %session_config.id)
     )]
     pub async fn reply(
         &self,
@@ -1119,9 +1120,8 @@ impl Agent {
         }
 
         let working_dir = session.working_dir.clone();
-        Ok(Box::pin(async_stream::try_stream! {
-            let reply_stream_span = tracing::info_span!(target: "goose::agents::agent", "reply_stream");
-            let _stream_guard = reply_stream_span.enter();
+        let reply_stream_span = tracing::info_span!(target: "goose::agents::agent", "reply_stream", session.id = %session_config.id);
+        let inner = Box::pin(async_stream::try_stream! {
             let mut turns_taken = 0u32;
             let max_turns = session_config.max_turns.unwrap_or_else(|| {
                 Config::global()
@@ -1689,7 +1689,8 @@ impl Agent {
             if !last_assistant_text.is_empty() {
                 tracing::info!(target: "goose::agents::agent", trace_output = last_assistant_text.as_str());
             }
-        }))
+        }.instrument(reply_stream_span));
+        Ok(inner)
     }
 
     pub async fn extend_system_prompt(&self, key: String, instruction: String) {

crates/goose/src/agents/reply_parts.rs

@@ -205,8 +205,12 @@ impl Agent {
         Ok((tools, toolshim_tools, system_prompt))
     }
 
-    /// Stream a response from the LLM provider.
-    /// Handles toolshim transformations if needed
+    // Don't add gen_ai.request.model here — provider.get_model_config()
+    // returns the wrong model for LeadWorkerProvider.
+    #[tracing::instrument(
+        skip(provider, session_id, system_prompt, messages, tools, toolshim_tools),
+        fields(session.id = %session_id)
+    )]
     pub(crate) async fn stream_response_from_provider(
         provider: Arc<dyn Provider>,
         session_id: &str,