harald/zeroclaw
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 3

feat: add verbose logging and complete observability (#251)

Browse source
This commit is contained in:
mai1015 2026-02-16 05:59:07 -05:00 • committed by GitHub
parent 6d56a040ce
commit 50f508766f
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
8 changed files with 316 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

72
src/observability/otel.rs
View file

@ -15,6 +15,8 @@ pub struct OtelObserver {
// Metrics instruments
agent_starts: Counter<u64>,
agent_duration: Histogram<f64>,
llm_calls: Counter<u64>,
llm_duration: Histogram<f64>,
tool_calls: Counter<u64>,
tool_duration: Histogram<f64>,
channel_messages: Counter<u64>,
@ -89,6 +91,17 @@ impl OtelObserver {
.with_unit("s")
.build();
let llm_calls = meter
.u64_counter("zeroclaw.llm.calls")
.with_description("Total LLM provider calls")
.build();
let llm_duration = meter
.f64_histogram("zeroclaw.llm.duration")
.with_description("LLM provider call duration in seconds")
.with_unit("s")
.build();
let tool_calls = meter
.u64_counter("zeroclaw.tool.calls")
.with_description("Total tool calls")
@ -141,6 +154,8 @@ impl OtelObserver {
meter_provider: meter_provider_clone,
agent_starts,
agent_duration,
llm_calls,
llm_duration,
tool_calls,
tool_duration,
channel_messages,
@ -168,6 +183,45 @@ impl Observer for OtelObserver {
],
);
}
ObserverEvent::LlmRequest { .. } => {}
ObserverEvent::LlmResponse {
provider,
model,
duration,
success,
error_message: _,
} => {
let secs = duration.as_secs_f64();
let attrs = [
KeyValue::new("provider", provider.clone()),
KeyValue::new("model", model.clone()),
KeyValue::new("success", success.to_string()),
];
self.llm_calls.add(1, &attrs);
self.llm_duration.record(secs, &attrs);
// Create a completed span for visibility in trace backends.
let start_time = SystemTime::now()
.checked_sub(*duration)
.unwrap_or(SystemTime::now());
let mut span = tracer.build(
opentelemetry::trace::SpanBuilder::from_name("llm.call")
.with_kind(SpanKind::Internal)
.with_start_time(start_time)
.with_attributes(vec![
KeyValue::new("provider", provider.clone()),
KeyValue::new("model", model.clone()),
KeyValue::new("success", *success),
KeyValue::new("duration_s", secs),
]),
);
if *success {
span.set_status(Status::Ok);
} else {
span.set_status(Status::error(""));
}
span.end();
}
ObserverEvent::AgentEnd {
duration,
tokens_used,
@ -193,6 +247,7 @@ impl Observer for OtelObserver {
// Note: tokens are recorded via record_metric(TokensUsed) to avoid
// double-counting. AgentEnd only records duration.
}
ObserverEvent::ToolCallStart { .. } => {}
ObserverEvent::ToolCall {
tool,
duration,
@ -230,6 +285,7 @@ impl Observer for OtelObserver {
self.tool_duration
.record(secs, &[KeyValue::new("tool", tool.clone())]);
}
ObserverEvent::TurnComplete => {}
ObserverEvent::ChannelMessage { channel, direction } => {
self.channel_messages.add(
1,
@ -323,6 +379,18 @@ mod tests {
provider: "openrouter".into(),
model: "claude-sonnet".into(),
});
obs.record_event(&ObserverEvent::LlmRequest {
provider: "openrouter".into(),
model: "claude-sonnet".into(),
messages_count: 2,
});
obs.record_event(&ObserverEvent::LlmResponse {
provider: "openrouter".into(),
model: "claude-sonnet".into(),
duration: Duration::from_millis(250),
success: true,
error_message: None,
});
obs.record_event(&ObserverEvent::AgentEnd {
duration: Duration::from_millis(500),
tokens_used: Some(100),
@ -331,6 +399,9 @@ mod tests {
duration: Duration::ZERO,
tokens_used: None,
});
obs.record_event(&ObserverEvent::ToolCallStart {
tool: "shell".into(),
});
obs.record_event(&ObserverEvent::ToolCall {
tool: "shell".into(),
duration: Duration::from_millis(10),
@ -341,6 +412,7 @@ mod tests {
duration: Duration::from_millis(5),
success: false,
});
obs.record_event(&ObserverEvent::TurnComplete);
obs.record_event(&ObserverEvent::ChannelMessage {
channel: "telegram".into(),
direction: "inbound".into(),