feat: harden community production foundation through story 1.5

This commit is contained in:
2026-08-14 00:21:59 +03:00
parent c30461cc92
commit f6fc2e5c9b
161 changed files with 16758 additions and 2515 deletions
@@ -1,7 +1,7 @@
use std::{
borrow::Cow,
collections::BTreeMap,
env, fmt,
fmt,
future::Future,
time::{Duration, SystemTime},
};
@@ -13,11 +13,8 @@ use sentry::{
};
use thiserror::Error;
use crate::{
RedactionLimits, ServiceIdentity, propagation::current_trace_id, redaction::truncate_string,
};
use crate::{RedactionLimits, ServiceIdentity, propagation::current_trace_id};
const SENTRY_DSN_ENV: &str = "CRANK_SENTRY_DSN";
const CRITICAL_ERROR_MESSAGE: &str = "critical error";
const SENTRY_SHUTDOWN_TIMEOUT: Duration = Duration::from_secs(2);
// Sentry serializes SystemTime as a finite f64; this keeps conservative fixed headroom.
@@ -25,6 +22,7 @@ const MAX_SERIALIZED_TIMESTAMP_BYTES: usize = 32;
tokio::task_local! {
static REQUEST_ID: String;
static TRACE_ID: String;
}
pub struct SentryConfig {
@@ -43,14 +41,6 @@ impl SentryConfig {
Ok(Self { dsn: Some(dsn) })
}
pub fn from_env() -> Result<Self, SentryConfigError> {
match env::var(SENTRY_DSN_ENV) {
Ok(value) => Self::parse(Some(&value)),
Err(env::VarError::NotPresent) => Self::parse(None),
Err(env::VarError::NotUnicode(_)) => Err(SentryConfigError::InvalidEnvironmentEncoding),
}
}
pub fn enabled(&self) -> bool {
self.dsn.is_some()
}
@@ -69,8 +59,6 @@ impl fmt::Debug for SentryConfig {
pub enum SentryConfigError {
#[error("CRANK_SENTRY_DSN is not a valid Sentry DSN")]
InvalidDsn,
#[error("CRANK_SENTRY_DSN is not valid UTF-8")]
InvalidEnvironmentEncoding,
#[error("critical error event budget cannot hold the required fields")]
EventBudgetTooSmall,
}
@@ -123,11 +111,43 @@ pub fn capture_critical_error(category: CriticalErrorCategory) {
});
}
pub async fn with_request_correlation<F>(request_id: String, future: F) -> F::Output
pub async fn with_request_correlation<F>(
request_id: String,
trace_id: String,
future: F,
) -> F::Output
where
F: Future,
{
REQUEST_ID.scope(request_id, future).await
if !valid_request_id(&request_id) || !valid_trace_id(&trace_id) {
return future.await;
}
REQUEST_ID
.scope(request_id, TRACE_ID.scope(trace_id, future))
.await
}
fn valid_request_id(value: &str) -> bool {
!value.is_empty()
&& value.len() <= 128
&& value
.bytes()
.all(|byte| (0x21..=0x7e).contains(&byte) && byte != b',' && byte != b';')
}
fn valid_trace_id(value: &str) -> bool {
value.len() == 32
&& value != "00000000000000000000000000000000"
&& value
.bytes()
.all(|byte| byte.is_ascii_digit() || (b'a'..=b'f').contains(&byte))
}
pub fn current_request_correlation() -> (Option<String>, Option<String>) {
(
REQUEST_ID.try_with(Clone::clone).ok(),
TRACE_ID.try_with(Clone::clone).ok(),
)
}
pub(crate) fn init_sentry(
@@ -184,16 +204,7 @@ fn sanitize_event(
CriticalErrorCategory::Panic
}
});
let mut tags = correlation_tags()
.into_iter()
.map(|(key, value)| (key, truncate_string(&value, limits.max_string_bytes)))
.collect::<BTreeMap<_, _>>();
for key in ["request_id", "trace_id"] {
if let Some(value) = event.tags.get(key) {
tags.entry(key.to_owned())
.or_insert_with(|| truncate_string(value, limits.max_string_bytes));
}
}
let mut tags = correlation_tags().into_iter().collect::<BTreeMap<_, _>>();
tags.insert("service".to_owned(), identity.service().to_owned());
tags.insert("category".to_owned(), category.as_str().to_owned());
@@ -222,19 +233,21 @@ fn correlation_tags() -> BTreeMap<String, String> {
if let Ok(request_id) = REQUEST_ID.try_with(Clone::clone) {
tags.insert("request_id".to_owned(), request_id);
}
if let Some(trace_id) = current_trace_id() {
if let Ok(trace_id) = TRACE_ID.try_with(Clone::clone) {
tags.insert("trace_id".to_owned(), trace_id);
} else if let Some(trace_id) = current_trace_id() {
tags.insert("trace_id".to_owned(), trace_id);
}
tags
}
fn enforce_event_budget(mut event: Event<'static>, max_event_bytes: usize) -> Event<'static> {
fn enforce_event_budget(event: Event<'static>, max_event_bytes: usize) -> Event<'static> {
if serialized_event_len(&event) <= max_event_bytes {
return event;
}
event.tags.remove("request_id");
event.tags.remove("trace_id");
// Startup validation reserves enough room for maximum canonical IDs. They
// are never evicted from a support event to satisfy a byte budget.
debug_assert!(serialized_event_len(&event) <= max_event_bytes);
event
}
@@ -257,7 +270,7 @@ fn required_critical_event_budget(identity: &ServiceIdentity, limits: RedactionL
CriticalErrorCategory::ALL
.into_iter()
.map(|category| {
let event = sanitize_event(
let mut event = sanitize_event(
Event {
tags: BTreeMap::from([("category".to_owned(), category.as_str().to_owned())]),
timestamp: SystemTime::UNIX_EPOCH,
@@ -266,6 +279,8 @@ fn required_critical_event_budget(identity: &ServiceIdentity, limits: RedactionL
identity,
unbounded_limits,
);
event.tags.insert("request_id".to_owned(), "r".repeat(128));
event.tags.insert("trace_id".to_owned(), "a".repeat(32));
serialized_event_len(&event)
.saturating_add(MAX_SERIALIZED_TIMESTAMP_BYTES.saturating_sub(1))
})
@@ -439,11 +454,15 @@ mod tests {
let events = sentry::test::with_captured_events_options(
|| {
tracing::dispatcher::with_default(&dispatch, || {
runtime.block_on(with_request_correlation("request-123".to_owned(), async {
let span = tracing::info_span!(target: "crank::trace", "http.request");
let _span_guard = span.enter();
capture_critical_error(CriticalErrorCategory::DataIntegrity);
}));
runtime.block_on(with_request_correlation(
"request-123".to_owned(),
"0af7651916cd43dd8448eb211c80319c".to_owned(),
async {
let span = tracing::info_span!(target: "crank::trace", "http.request");
let _span_guard = span.enter();
capture_critical_error(CriticalErrorCategory::DataIntegrity);
},
));
});
},
options,
@@ -496,6 +515,7 @@ mod tests {
tracing::dispatcher::with_default(&dispatch, || {
runtime.block_on(with_request_correlation(
"panic-request-123".to_owned(),
"0af7651916cd43dd8448eb211c80319c".to_owned(),
async {
let span =
tracing::info_span!(target: "crank::trace", "http.request");