Files
llmx/codex-rs/otel/src/otel_event_manager.rs
vishnu-oai 04c1782e52 OpenTelemetry events (#2103)
### Title

## otel

Codex can emit [OpenTelemetry](https://opentelemetry.io/) **log events**
that
describe each run: outbound API requests, streamed responses, user
input,
tool-approval decisions, and the result of every tool invocation. Export
is
**disabled by default** so local runs remain self-contained. Opt in by
adding an
`[otel]` table and choosing an exporter.

```toml
[otel]
environment = "staging"   # defaults to "dev"
exporter = "none"          # defaults to "none"; set to otlp-http or otlp-grpc to send events
log_user_prompt = false    # defaults to false; redact prompt text unless explicitly enabled
```

Codex tags every exported event with `service.name = "codex-cli"`, the
CLI
version, and an `env` attribute so downstream collectors can distinguish
dev/staging/prod traffic. Only telemetry produced inside the
`codex_otel`
crate—the events listed below—is forwarded to the exporter.

### Event catalog

Every event shares a common set of metadata fields: `event.timestamp`,
`conversation.id`, `app.version`, `auth_mode` (when available),
`user.account_id` (when available), `terminal.type`, `model`, and
`slug`.

With OTEL enabled Codex emits the following event types (in addition to
the
metadata above):

- `codex.api_request`
  - `cf_ray` (optional)
  - `attempt`
  - `duration_ms`
  - `http.response.status_code` (optional)
  - `error.message` (failures)
- `codex.sse_event`
  - `event.kind`
  - `duration_ms`
  - `error.message` (failures)
  - `input_token_count` (completion only)
  - `output_token_count` (completion only)
  - `cached_token_count` (completion only, optional)
  - `reasoning_token_count` (completion only, optional)
  - `tool_token_count` (completion only)
- `codex.user_prompt`
  - `prompt_length`
  - `prompt` (redacted unless `log_user_prompt = true`)
- `codex.tool_decision`
  - `tool_name`
  - `call_id`
- `decision` (`approved`, `approved_for_session`, `denied`, or `abort`)
  - `source` (`config` or `user`)
- `codex.tool_result`
  - `tool_name`
  - `call_id`
  - `arguments`
  - `duration_ms` (execution time for the tool)
  - `success` (`"true"` or `"false"`)
  - `output`

### Choosing an exporter

Set `otel.exporter` to control where events go:

- `none` – leaves instrumentation active but skips exporting. This is
the
  default.
- `otlp-http` – posts OTLP log records to an OTLP/HTTP collector.
Specify the
  endpoint, protocol, and headers your collector expects:

  ```toml
  [otel]
  exporter = { otlp-http = {
    endpoint = "https://otel.example.com/v1/logs",
    protocol = "binary",
    headers = { "x-otlp-api-key" = "${OTLP_TOKEN}" }
  }}
  ```

- `otlp-grpc` – streams OTLP log records over gRPC. Provide the endpoint
and any
  metadata headers:

  ```toml
  [otel]
  exporter = { otlp-grpc = {
    endpoint = "https://otel.example.com:4317",
    headers = { "x-otlp-meta" = "abc123" }
  }}
  ```

If the exporter is `none` nothing is written anywhere; otherwise you
must run or point to your
own collector. All exporters run on a background batch worker that is
flushed on
shutdown.

If you build Codex from source the OTEL crate is still behind an `otel`
feature
flag; the official prebuilt binaries ship with the feature enabled. When
the
feature is disabled the telemetry hooks become no-ops so the CLI
continues to
function without the extra dependencies.

---------

Co-authored-by: Anton Panasenko <apanasenko@openai.com>
2025-09-29 11:30:55 -07:00

460 lines
16 KiB
Rust

use chrono::SecondsFormat;
use chrono::Utc;
use codex_protocol::config_types::ReasoningEffort;
use codex_protocol::config_types::ReasoningSummary;
use codex_protocol::mcp_protocol::AuthMode;
use codex_protocol::mcp_protocol::ConversationId;
use codex_protocol::models::ResponseItem;
use codex_protocol::protocol::AskForApproval;
use codex_protocol::protocol::InputItem;
use codex_protocol::protocol::ReviewDecision;
use codex_protocol::protocol::SandboxPolicy;
use eventsource_stream::Event as StreamEvent;
use eventsource_stream::EventStreamError as StreamError;
use reqwest::Error;
use reqwest::Response;
use serde::Serialize;
use std::fmt::Display;
use std::time::Duration;
use std::time::Instant;
use strum_macros::Display;
use tokio::time::error::Elapsed;
#[derive(Debug, Clone, Serialize, Display)]
#[serde(rename_all = "snake_case")]
pub enum ToolDecisionSource {
Config,
User,
}
#[derive(Debug, Clone)]
pub struct OtelEventMetadata {
conversation_id: ConversationId,
auth_mode: Option<String>,
account_id: Option<String>,
model: String,
slug: String,
log_user_prompts: bool,
app_version: &'static str,
terminal_type: String,
}
#[derive(Debug, Clone)]
pub struct OtelEventManager {
metadata: OtelEventMetadata,
}
impl OtelEventManager {
pub fn new(
conversation_id: ConversationId,
model: &str,
slug: &str,
account_id: Option<String>,
auth_mode: Option<AuthMode>,
log_user_prompts: bool,
terminal_type: String,
) -> OtelEventManager {
Self {
metadata: OtelEventMetadata {
conversation_id,
auth_mode: auth_mode.map(|m| m.to_string()),
account_id,
model: model.to_owned(),
slug: slug.to_owned(),
log_user_prompts,
app_version: env!("CARGO_PKG_VERSION"),
terminal_type,
},
}
}
pub fn with_model(&self, model: &str, slug: &str) -> Self {
let mut manager = self.clone();
manager.metadata.model = model.to_owned();
manager.metadata.slug = slug.to_owned();
manager
}
#[allow(clippy::too_many_arguments)]
pub fn conversation_starts(
&self,
provider_name: &str,
reasoning_effort: Option<ReasoningEffort>,
reasoning_summary: ReasoningSummary,
context_window: Option<u64>,
max_output_tokens: Option<u64>,
auto_compact_token_limit: Option<i64>,
approval_policy: AskForApproval,
sandbox_policy: SandboxPolicy,
mcp_servers: Vec<&str>,
active_profile: Option<String>,
) {
tracing::event!(
tracing::Level::INFO,
event.name = "codex.conversation_starts",
event.timestamp = %timestamp(),
conversation.id = %self.metadata.conversation_id,
app.version = %self.metadata.app_version,
auth_mode = self.metadata.auth_mode,
user.account_id = self.metadata.account_id,
terminal.type = %self.metadata.terminal_type,
model = %self.metadata.model,
slug = %self.metadata.slug,
provider_name = %provider_name,
reasoning_effort = reasoning_effort.map(|e| e.to_string()),
reasoning_summary = %reasoning_summary,
context_window = context_window,
max_output_tokens = max_output_tokens,
auto_compact_token_limit = auto_compact_token_limit,
approval_policy = %approval_policy,
sandbox_policy = %sandbox_policy,
mcp_servers = mcp_servers.join(", "),
active_profile = active_profile,
)
}
pub async fn log_request<F, Fut>(&self, attempt: u64, f: F) -> Result<Response, Error>
where
F: FnOnce() -> Fut,
Fut: Future<Output = Result<Response, Error>>,
{
let start = std::time::Instant::now();
let response = f().await;
let duration = start.elapsed();
let (status, error) = match &response {
Ok(response) => (Some(response.status().as_u16()), None),
Err(error) => (error.status().map(|s| s.as_u16()), Some(error.to_string())),
};
tracing::event!(
tracing::Level::INFO,
event.name = "codex.api_request",
event.timestamp = %timestamp(),
conversation.id = %self.metadata.conversation_id,
app.version = %self.metadata.app_version,
auth_mode = self.metadata.auth_mode,
user.account_id = self.metadata.account_id,
terminal.type = %self.metadata.terminal_type,
model = %self.metadata.model,
slug = %self.metadata.slug,
duration_ms = %duration.as_millis(),
http.response.status_code = status,
error.message = error,
attempt = attempt,
);
response
}
pub async fn log_sse_event<Next, Fut, E>(
&self,
next: Next,
) -> Result<Option<Result<StreamEvent, StreamError<E>>>, Elapsed>
where
Next: FnOnce() -> Fut,
Fut: Future<Output = Result<Option<Result<StreamEvent, StreamError<E>>>, Elapsed>>,
E: Display,
{
let start = std::time::Instant::now();
let response = next().await;
let duration = start.elapsed();
match response {
Ok(Some(Ok(ref sse))) => {
if sse.data.trim() == "[DONE]" {
self.sse_event(&sse.event, duration);
} else {
match serde_json::from_str::<serde_json::Value>(&sse.data) {
Ok(error) if sse.event == "response.failed" => {
self.sse_event_failed(Some(&sse.event), duration, &error);
}
Ok(content) if sse.event == "response.output_item.done" => {
match serde_json::from_value::<ResponseItem>(content) {
Ok(_) => self.sse_event(&sse.event, duration),
Err(_) => {
self.sse_event_failed(
Some(&sse.event),
duration,
&"failed to parse response.output_item.done",
);
}
};
}
Ok(_) => {
self.sse_event(&sse.event, duration);
}
Err(error) => {
self.sse_event_failed(Some(&sse.event), duration, &error);
}
}
}
}
Ok(Some(Err(ref error))) => {
self.sse_event_failed(None, duration, error);
}
Ok(None) => {}
Err(_) => {
self.sse_event_failed(None, duration, &"idle timeout waiting for SSE");
}
}
response
}
fn sse_event(&self, kind: &str, duration: Duration) {
tracing::event!(
tracing::Level::INFO,
event.name = "codex.sse_event",
event.timestamp = %timestamp(),
event.kind = %kind,
conversation.id = %self.metadata.conversation_id,
app.version = %self.metadata.app_version,
auth_mode = self.metadata.auth_mode,
user.account_id = self.metadata.account_id,
terminal.type = %self.metadata.terminal_type,
model = %self.metadata.model,
slug = %self.metadata.slug,
duration_ms = %duration.as_millis(),
);
}
pub fn sse_event_failed<T>(&self, kind: Option<&String>, duration: Duration, error: &T)
where
T: Display,
{
match kind {
Some(kind) => tracing::event!(
tracing::Level::INFO,
event.name = "codex.sse_event",
event.timestamp = %timestamp(),
event.kind = %kind,
conversation.id = %self.metadata.conversation_id,
app.version = %self.metadata.app_version,
auth_mode = self.metadata.auth_mode,
user.account_id = self.metadata.account_id,
terminal.type = %self.metadata.terminal_type,
model = %self.metadata.model,
slug = %self.metadata.slug,
duration_ms = %duration.as_millis(),
error.message = %error,
),
None => tracing::event!(
tracing::Level::INFO,
event.name = "codex.sse_event",
event.timestamp = %timestamp(),
conversation.id = %self.metadata.conversation_id,
app.version = %self.metadata.app_version,
auth_mode = self.metadata.auth_mode,
user.account_id = self.metadata.account_id,
terminal.type = %self.metadata.terminal_type,
model = %self.metadata.model,
slug = %self.metadata.slug,
duration_ms = %duration.as_millis(),
error.message = %error,
),
}
}
pub fn see_event_completed_failed<T>(&self, error: &T)
where
T: Display,
{
tracing::event!(
tracing::Level::INFO,
event.name = "codex.sse_event",
event.kind = %"response.completed",
event.timestamp = %timestamp(),
conversation.id = %self.metadata.conversation_id,
app.version = %self.metadata.app_version,
auth_mode = self.metadata.auth_mode,
user.account_id = self.metadata.account_id,
terminal.type = %self.metadata.terminal_type,
model = %self.metadata.model,
slug = %self.metadata.slug,
error.message = %error,
)
}
pub fn sse_event_completed(
&self,
input_token_count: u64,
output_token_count: u64,
cached_token_count: Option<u64>,
reasoning_token_count: Option<u64>,
tool_token_count: u64,
) {
tracing::event!(
tracing::Level::INFO,
event.name = "codex.sse_event",
event.timestamp = %timestamp(),
event.kind = %"response.completed",
conversation.id = %self.metadata.conversation_id,
app.version = %self.metadata.app_version,
auth_mode = self.metadata.auth_mode,
user.account_id = self.metadata.account_id,
terminal.type = %self.metadata.terminal_type,
model = %self.metadata.model,
slug = %self.metadata.slug,
input_token_count = %input_token_count,
output_token_count = %output_token_count,
cached_token_count = cached_token_count,
reasoning_token_count = reasoning_token_count,
tool_token_count = %tool_token_count,
);
}
pub fn user_prompt(&self, items: &[InputItem]) {
let prompt = items
.iter()
.flat_map(|item| match item {
InputItem::Text { text } => Some(text.as_str()),
_ => None,
})
.collect::<String>();
let prompt_to_log = if self.metadata.log_user_prompts {
prompt.as_str()
} else {
"[REDACTED]"
};
tracing::event!(
tracing::Level::INFO,
event.name = "codex.user_prompt",
event.timestamp = %timestamp(),
conversation.id = %self.metadata.conversation_id,
app.version = %self.metadata.app_version,
auth_mode = self.metadata.auth_mode,
user.account_id = self.metadata.account_id,
terminal.type = %self.metadata.terminal_type,
model = %self.metadata.model,
slug = %self.metadata.slug,
prompt_length = %prompt.chars().count(),
prompt = %prompt_to_log,
);
}
pub fn tool_decision(
&self,
tool_name: &str,
call_id: &str,
decision: ReviewDecision,
source: ToolDecisionSource,
) {
tracing::event!(
tracing::Level::INFO,
event.name = "codex.tool_decision",
event.timestamp = %timestamp(),
conversation.id = %self.metadata.conversation_id,
app.version = %self.metadata.app_version,
auth_mode = self.metadata.auth_mode,
user.account_id = self.metadata.account_id,
terminal.type = %self.metadata.terminal_type,
model = %self.metadata.model,
slug = %self.metadata.slug,
tool_name = %tool_name,
call_id = %call_id,
decision = %decision.to_string().to_lowercase(),
source = %source.to_string(),
);
}
pub async fn log_tool_result<F, Fut, E>(
&self,
tool_name: &str,
call_id: &str,
arguments: &str,
f: F,
) -> Result<String, E>
where
F: FnOnce() -> Fut,
Fut: Future<Output = Result<String, E>>,
E: Display,
{
let start = Instant::now();
let result = f().await;
let duration = start.elapsed();
let (output, success) = match &result {
Ok(content) => (content, true),
Err(error) => (&error.to_string(), false),
};
tracing::event!(
tracing::Level::INFO,
event.name = "codex.tool_result",
event.timestamp = %timestamp(),
conversation.id = %self.metadata.conversation_id,
app.version = %self.metadata.app_version,
auth_mode = self.metadata.auth_mode,
user.account_id = self.metadata.account_id,
terminal.type = %self.metadata.terminal_type,
model = %self.metadata.model,
slug = %self.metadata.slug,
tool_name = %tool_name,
call_id = %call_id,
arguments = %arguments,
duration_ms = %duration.as_millis(),
success = %success,
output = %output,
);
result
}
pub fn log_tool_failed(&self, tool_name: &str, error: &str) {
tracing::event!(
tracing::Level::INFO,
event.name = "codex.tool_result",
event.timestamp = %timestamp(),
conversation.id = %self.metadata.conversation_id,
app.version = %self.metadata.app_version,
auth_mode = self.metadata.auth_mode,
user.account_id = self.metadata.account_id,
terminal.type = %self.metadata.terminal_type,
model = %self.metadata.model,
slug = %self.metadata.slug,
tool_name = %tool_name,
duration_ms = %Duration::ZERO.as_millis(),
success = %false,
output = %error,
);
}
pub fn tool_result(
&self,
tool_name: &str,
call_id: &str,
arguments: &str,
duration: Duration,
success: bool,
output: &str,
) {
let success_str = if success { "true" } else { "false" };
tracing::event!(
tracing::Level::INFO,
event.name = "codex.tool_result",
event.timestamp = %timestamp(),
conversation.id = %self.metadata.conversation_id,
app.version = %self.metadata.app_version,
auth_mode = self.metadata.auth_mode,
user.account_id = self.metadata.account_id,
terminal.type = %self.metadata.terminal_type,
model = %self.metadata.model,
slug = %self.metadata.slug,
tool_name = %tool_name,
call_id = %call_id,
arguments = %arguments,
duration_ms = %duration.as_millis(),
success = %success_str,
output = %output,
);
}
}
fn timestamp() -> String {
Utc::now().to_rfc3339_opts(SecondsFormat::Millis, true)
}