fix: harden typed metrics review findings

This commit is contained in:
2026-08-14 13:50:04 +03:00
parent 996a5461de
commit b7face0e94
30 changed files with 722 additions and 382 deletions
@@ -10,7 +10,7 @@ use crank_metrics::{
};
use metrics::Unit;
use crate::{MetricKind, MetricUnit, metric_schema};
use crate::{MetricKind, MetricUnit};
pub async fn record_http_request(request: Request, next: Next) -> Response {
let route = request
@@ -24,11 +24,7 @@ pub async fn record_http_request(request: Request, next: Next) -> Response {
let _inflight = InFlightGuard::http();
let response = next.run(request).await;
let exemplar = response
.headers()
.get("x-trace-id")
.and_then(|value| value.to_str().ok())
.and_then(ExemplarTraceId::parse);
let exemplar = response.extensions().get::<ExemplarTraceId>().copied();
crank_metrics::record_http_request_with_exemplar(
route,
method,
@@ -46,8 +42,8 @@ pub fn record_db_pool_connections(total: u32, idle: usize) {
crank_metrics::set_db_pool_connections(DbPoolState::Used, total.saturating_sub(idle));
}
pub(crate) fn register_metric_schema() {
for definition in metric_schema() {
pub(crate) fn register_metric_schema(service: crank_metrics::MetricService) {
for definition in crank_metrics::metric_schema_for_service(service) {
let unit = match definition.unit {
MetricUnit::Count => Unit::Count,
MetricUnit::Seconds => Unit::Seconds,
@@ -65,7 +61,7 @@ pub(crate) fn register_metric_schema() {
}
}
crank_metrics::initialize_gauges();
crank_metrics::initialize_gauges(service);
}
#[cfg(test)]
+3 -1
View File
@@ -47,8 +47,10 @@ impl ObservabilityLifecycle {
build_subscriber_with_tracer(config, io::stdout, Some(tracer))?
.try_init()
.map_err(|_| ObservabilityInitError::SubscriberAlreadyInitialized)?;
let metrics_service = crank_metrics::MetricService::parse(identity.service())
.ok_or(MetricsSurfaceError::RecorderConfiguration)?;
let metrics_handle = install_prometheus_recorder(&identity)?;
register_metric_schema();
register_metric_schema(metrics_service);
let sentry_guard = init_sentry(&identity, redaction_limits, sentry_config)?;
Ok(Self {
+164 -24
View File
@@ -1,4 +1,11 @@
use std::net::SocketAddr;
use std::{
collections::HashMap,
net::SocketAddr,
sync::{
Arc,
atomic::{AtomicUsize, Ordering},
},
};
use axum::{
Router,
@@ -23,6 +30,8 @@ use crank_metrics::{ExemplarObservation, MAX_EXPOSITION_BYTES, MetricService, ex
const PROMETHEUS_CONTENT_TYPE: &str = "text/plain; version=0.0.4; charset=utf-8";
const OPENMETRICS_CONTENT_TYPE: &str = "application/openmetrics-text; version=1.0.0; charset=utf-8";
const EXPOSITION_TOO_LARGE: &str = "metrics exposition exceeds configured bound\n";
const TOO_MANY_SCRAPES: &str = "metrics scrape concurrency limit exceeded\n";
const MAX_CONCURRENT_SCRAPES: usize = 2;
#[derive(Clone)]
pub struct MetricsConfig {
@@ -93,6 +102,7 @@ struct MetricsState {
handle: PrometheusHandle,
token_digest: Option<[u8; 32]>,
requires_authentication: bool,
active_scrapes: Arc<AtomicUsize>,
}
pub struct MetricsSurface {
@@ -108,6 +118,7 @@ impl MetricsSurface {
handle,
token_digest: config.token_digest,
requires_authentication: config.requires_authentication(),
active_scrapes: Arc::new(AtomicUsize::new(0)),
},
config,
_recorder: None,
@@ -206,20 +217,23 @@ fn prometheus_builder(
}
async fn render_metrics(State(state): State<MetricsState>, headers: HeaderMap) -> Response {
let Some(_permit) = ScrapePermit::try_acquire(&state.active_scrapes) else {
return (StatusCode::SERVICE_UNAVAILABLE, TOO_MANY_SCRAPES).into_response();
};
let legacy = state.handle.render();
let openmetrics = headers
.get(header::ACCEPT)
.and_then(|value| value.to_str().ok())
.is_some_and(|value| {
value
.split(',')
.any(|part| part.trim().starts_with("application/openmetrics-text"))
});
.is_some_and(prefers_openmetrics);
let body = if openmetrics {
render_openmetrics(&legacy, &exemplar_snapshot())
} else {
legacy
};
exposition_response(body, openmetrics)
}
fn exposition_response(body: String, openmetrics: bool) -> Response {
if body.len() > MAX_EXPOSITION_BYTES {
return (StatusCode::SERVICE_UNAVAILABLE, EXPOSITION_TOO_LARGE).into_response();
}
@@ -235,14 +249,15 @@ async fn render_metrics(State(state): State<MetricsState>, headers: HeaderMap) -
response
}
fn render_openmetrics(legacy: &str, exemplars: &[ExemplarObservation]) -> String {
fn render_openmetrics(legacy: &str, exemplars: &[Arc<ExemplarObservation>]) -> String {
let index: HashMap<String, &ExemplarObservation> = exemplars
.iter()
.map(|exemplar| (exemplar_key(exemplar), exemplar.as_ref()))
.collect();
let mut output = String::with_capacity(legacy.len() + exemplars.len().saturating_mul(96) + 6);
for line in legacy.lines() {
output.push_str(line);
if let Some(exemplar) = exemplars
.iter()
.find(|candidate| line_matches_exemplar(line, candidate))
{
if let Some(exemplar) = sample_key(line).and_then(|key| index.get(&key).copied()) {
output.push_str(" # {trace_id=\"");
output.push_str(exemplar.trace_id.as_str());
output.push_str("\"} ");
@@ -254,19 +269,79 @@ fn render_openmetrics(legacy: &str, exemplars: &[ExemplarObservation]) -> String
output
}
fn line_matches_exemplar(line: &str, exemplar: &ExemplarObservation) -> bool {
if !line.starts_with(exemplar.metric) || !line[exemplar.metric.len()..].starts_with("_bucket{")
{
return false;
}
let expected_bound = exemplar
fn exemplar_key(exemplar: &ExemplarObservation) -> String {
let mut labels = exemplar.labels.clone();
labels.sort_unstable();
let bound = exemplar
.bucket_upper_bound
.map_or_else(|| "+Inf".to_owned(), |value| value.to_string());
line.contains(&format!("le=\"{expected_bound}\""))
&& exemplar
.labels
.iter()
.all(|(key, value)| line.contains(&format!("{key}=\"{value}\"")))
format!("{}|{:?}|{bound}", exemplar.metric, labels)
}
fn sample_key(line: &str) -> Option<String> {
let (head, _) = line.split_once("} ")?;
let (metric, raw_labels) = head.split_once("_bucket{")?;
let mut labels = Vec::new();
let mut bound = None;
for item in raw_labels.split(',') {
let (name, value) = item.split_once("=\"")?;
let value = value.strip_suffix('"')?;
match name {
"le" => bound = Some(value),
"service" | "version" | "environment" => {}
_ => labels.push((name, value)),
}
}
labels.sort_unstable();
Some(format!("{metric}|{labels:?}|{}", bound?))
}
fn prefers_openmetrics(value: &str) -> bool {
let mut open_q = 0.0_f32;
let mut legacy_q = 0.0_f32;
for range in value.split(',') {
let mut parts = range.trim().split(';');
let media = parts.next().unwrap_or_default().trim();
let mut q = 1.0_f32;
let mut supported_version = true;
for parameter in parts {
let Some((name, raw)) = parameter.trim().split_once('=') else {
continue;
};
match name.trim() {
"q" => q = raw.trim().parse().unwrap_or(0.0),
"version" if media == "application/openmetrics-text" => {
supported_version = raw.trim().trim_matches('"') == "1.0.0";
}
_ => {}
}
}
if media == "application/openmetrics-text" && supported_version {
open_q = open_q.max(q.clamp(0.0, 1.0));
} else if media == "text/plain" || media == "*/*" {
legacy_q = legacy_q.max(q.clamp(0.0, 1.0));
}
}
open_q > 0.0 && open_q >= legacy_q
}
struct ScrapePermit(Arc<AtomicUsize>);
impl ScrapePermit {
fn try_acquire(active: &Arc<AtomicUsize>) -> Option<Self> {
active
.fetch_update(Ordering::AcqRel, Ordering::Acquire, |count| {
(count < MAX_CONCURRENT_SCRAPES).then_some(count + 1)
})
.ok()
.map(|_| Self(Arc::clone(active)))
}
}
impl Drop for ScrapePermit {
fn drop(&mut self) {
self.0.fetch_sub(1, Ordering::Release);
}
}
async fn metrics_health() -> impl IntoResponse {
@@ -313,9 +388,17 @@ fn token_digest(token: &[u8]) -> [u8; 32] {
#[cfg(test)]
mod rendering_tests {
use std::{
sync::{Arc, atomic::AtomicUsize},
time::{Duration, Instant},
};
use crank_metrics::{ExemplarObservation, ExemplarTraceId};
use super::render_openmetrics;
use super::{
MAX_CONCURRENT_SCRAPES, ScrapePermit, exposition_response, prefers_openmetrics,
render_openmetrics,
};
#[test]
fn openmetrics_adds_bounded_exemplar_without_changing_aggregate() {
@@ -327,9 +410,66 @@ mod rendering_tests {
value: 0.007,
trace_id: ExemplarTraceId::parse("0123456789abcdef0123456789abcdef").unwrap(),
};
let rendered = render_openmetrics(legacy, &[exemplar]);
let rendered = render_openmetrics(legacy, &[Arc::new(exemplar)]);
assert!(rendered.contains("# {trace_id=\"0123456789abcdef0123456789abcdef\"} 0.007"));
assert!(rendered.ends_with("# EOF\n"));
assert_eq!(rendered.matches(" 1").count(), legacy.matches(" 1").count());
}
#[test]
fn accept_negotiation_is_exact_and_honors_quality_and_version() {
assert!(prefers_openmetrics(
"application/openmetrics-text; version=1.0.0"
));
assert!(prefers_openmetrics("application/openmetrics-text"));
assert!(!prefers_openmetrics("application/openmetrics-text; q=0"));
assert!(!prefers_openmetrics("application/openmetrics-textual"));
assert!(!prefers_openmetrics(
"application/openmetrics-text; version=0.0.1"
));
assert!(!prefers_openmetrics(
"application/openmetrics-text;q=0.2,text/plain;q=0.8"
));
}
#[test]
fn output_and_concurrent_scrapes_fail_closed_at_their_bounds() {
assert_eq!(
exposition_response("x".repeat(crank_metrics::MAX_EXPOSITION_BYTES + 1), false)
.status(),
axum::http::StatusCode::SERVICE_UNAVAILABLE
);
let active = Arc::new(AtomicUsize::new(0));
let permits = (0..MAX_CONCURRENT_SCRAPES)
.map(|_| ScrapePermit::try_acquire(&active).unwrap())
.collect::<Vec<_>>();
assert!(ScrapePermit::try_acquire(&active).is_none());
drop(permits);
assert!(ScrapePermit::try_acquire(&active).is_some());
}
#[test]
fn maximum_fixture_renders_in_linear_bounded_time() {
let trace = ExemplarTraceId::parse("0123456789abcdef0123456789abcdef").unwrap();
let mut legacy = String::new();
let mut exemplars = Vec::new();
for index in 0..5_000 {
let value: &'static str = Box::leak(format!("route-{index}").into_boxed_str());
legacy.push_str(&format!(
"crank_http_request_duration_seconds_bucket{{route=\"{value}\",method=\"GET\",le=\"0.01\"}} 1\n"
));
exemplars.push(Arc::new(ExemplarObservation {
metric: "crank_http_request_duration_seconds",
labels: vec![("route", value), ("method", "GET")],
bucket_upper_bound: Some(0.01),
value: 0.007,
trace_id: trace,
}));
}
let started = Instant::now();
let rendered = render_openmetrics(&legacy, &exemplars);
assert_eq!(rendered.matches("# {trace_id=").count(), 5_000);
assert!(started.elapsed() < Duration::from_secs(3));
assert!(rendered.len() < crank_metrics::MAX_EXPOSITION_BYTES);
}
}
@@ -5,6 +5,7 @@ use axum::{
body::{Body, to_bytes},
http::{Request, StatusCode, header},
middleware,
response::IntoResponse,
routing::get,
};
use crank_observability::{
@@ -32,10 +33,12 @@ async fn http_metrics_use_matched_routes_and_closed_labels() {
.route(
"/api/admin/workspaces/{workspace_id}/operations/{operation_id}",
get(|| async {
(
StatusCode::NO_CONTENT,
[("x-trace-id", "0123456789abcdef0123456789abcdef")],
)
let mut response = StatusCode::NO_CONTENT.into_response();
response.extensions_mut().insert(
crank_metrics::ExemplarTraceId::parse("0123456789abcdef0123456789abcdef")
.unwrap(),
);
response
}),
)
.layer(middleware::from_fn(record_http_request));
@@ -28,6 +28,35 @@ fn loopback_is_allowed_without_a_token() {
assert!(!config.requires_authentication());
}
#[test]
fn disabled_config_remains_disabled_without_auth_side_effects() {
let config = MetricsConfig::new(
false,
SocketAddr::new(IpAddr::V4(Ipv4Addr::UNSPECIFIED), 9464),
None,
)
.expect("disabled listener does not require a token");
assert!(!config.enabled());
}
#[tokio::test]
async fn bind_collision_returns_a_typed_safe_error() {
let listener = tokio::net::TcpListener::bind((Ipv4Addr::LOCALHOST, 0))
.await
.unwrap();
let address = listener.local_addr().unwrap();
let config = MetricsConfig::new(true, address, None).unwrap();
let error = match MetricsSurface::for_test(config, identity())
.unwrap()
.bind()
.await
{
Ok(_) => panic!("occupied port must fail closed"),
Err(error) => error,
};
assert_eq!(error.to_string(), "failed to bind metrics listener");
}
#[tokio::test]
async fn default_service_ports_bind_and_serve_real_metrics_listeners() {
for (service, port) in [("admin-api", 9464), ("mcp-server", 9465)] {