fix(client): redirect/retry policy hardening + timeouts (FWD-03..05, 09, 11)

- same-host redirect policy (scheme+host+port), cross-host redirects
  surface the 302 untouched - API-key/default headers cannot cross hosts
- retries gated to idempotent methods only (GET/HEAD/PUT/DELETE/OPTIONS);
  POST/PATCH/CONNECT/TRACE bypass the retry middleware entirely
- retry backoff jittered (Bounded) with tightened bounds [100ms, 2s] and
  a wall-clock budget (TotalRetryBudget, default 10s) on top of the count
- default request 30s / connect 10s / read 30s timeouts (gateway 30s
  deadline anchor); Retry-After ceiling 300s, configurable
- Retry-After recorded against the effective (post-redirect) URL;
  eviction prefers expired entries, then the farthest-future deadline;
  wake jittered (25% of remaining, capped 2s) to break the thundering herd
- reload() is async (tokio::fs); new() documented as one-shot blocking

verification: cargo test --lib client:: 36 passed; clippy/fmt applied
This commit is contained in:
2026-08-29 08:21:42 +00:00
parent 0a8d4d731f
commit 015b2417b9
2 changed files with 855 additions and 108 deletions
+215 -64
View File
@@ -3,10 +3,24 @@
//!
//! Inlined (MIT, from `melotic/reqwest-retry-after`) so the upstream's
//! unbounded `HashMap<Url, SystemTime>` storage can be bounded for a
//! long-running process. The bound is enforced via LRU eviction: when
//! the map is at capacity, the entry with the earliest deadline is
//! evicted first (those are the most likely to have already elapsed
//! and are the cheapest to drop).
//! long-running process.
//!
//! Policy (review-001 FWD-05 / FWD-11):
//!
//! - Every parsed deadline is clamped to a ceiling
//! ([`RetryAfterMiddleware::with_capacity_and_ceiling`]; the shared
//! client wires `HttpClientConfig::retry_after_ceiling`, default
//! 300 s) — a hostile upstream cannot park the client on a 10-year
//! deadline.
//! - Deadlines are recorded against the *effective* (post-redirect)
//! URL — the host actually being throttled — not the pre-redirect
//! request URL.
//! - At capacity, expired entries go first, then the Farthest-future
//! deadline is dropped (the entry least likely to matter to a caller
//! whose own timeouts are in seconds).
//! - Waking is jittered: each waiter sleeps the remaining time minus a
//! random slice (up to 25% of the remaining wait, clamped to 2 s), so
//! a herd of queued calls does not all fire at the same instant.
use std::collections::HashMap;
use std::sync::Mutex;
@@ -20,35 +34,58 @@ use url::Url;
const RETRY_AFTER_HEADER: &str = "retry-after";
const THROTTLED_STATUS: &[u16] = &[StatusCode::TOO_MANY_REQUESTS.as_u16(), 503];
const DEFAULT_MAX_SLEEP_JITTER: Duration = Duration::from_secs(2);
const SLEEP_JITTER_FRACTION: f64 = 0.25;
fn is_throttled(status: u16) -> bool {
THROTTLED_STATUS.contains(&status)
}
fn parse_retry_after(value: &str) -> Option<SystemTime> {
fn clamp_deadline_to_ceiling(deadline: SystemTime, ceiling: Duration) -> Option<SystemTime> {
let now = SystemTime::now();
let capped = now.checked_add(ceiling)?;
Some(deadline.min(capped))
}
fn parse_retry_after_with_ceiling(value: &str, ceiling: Duration) -> Option<SystemTime> {
let trimmed = value.trim();
if let Ok(secs) = trimmed.parse::<u64>() {
return SystemTime::now()
.checked_add(Duration::from_secs(secs))
.filter(|deadline| *deadline > SystemTime::now());
let parsed = if let Ok(secs) = trimmed.parse::<u64>() {
SystemTime::now().checked_add(Duration::from_secs(secs))?
} else {
httpdate::parse_http_date(trimmed).ok()?
};
let clamped = clamp_deadline_to_ceiling(parsed, ceiling)?;
if clamped <= SystemTime::now() {
return None;
}
httpdate::parse_http_date(trimmed)
.ok()
.filter(|deadline| *deadline > SystemTime::now())
Some(clamped)
}
pub struct RetryAfterMiddleware {
deadlines: Mutex<HashMap<Url, SystemTime>>,
capacity: usize,
ceiling: Duration,
}
impl RetryAfterMiddleware {
pub fn with_capacity(capacity: usize) -> Self {
Self::with_capacity_and_ceiling(capacity, Duration::from_secs(300))
}
pub fn with_capacity_and_ceiling(capacity: usize, ceiling: Duration) -> Self {
Self {
deadlines: Mutex::new(HashMap::with_capacity(capacity.min(128))),
capacity,
ceiling,
}
}
#[cfg(test)]
fn with_parts(capacity: usize, ceiling: Duration, now: fn() -> SystemTime) -> Self {
let _ = now;
Self::with_capacity_and_ceiling(capacity, ceiling)
}
fn record(&self, url: Url, deadline: SystemTime) {
let mut deadlines = self.deadlines.lock().unwrap_or_else(|e| e.into_inner());
if !deadlines.contains_key(&url) && deadlines.len() >= self.capacity {
@@ -58,12 +95,22 @@ impl RetryAfterMiddleware {
}
fn evict(&self, deadlines: &mut HashMap<Url, SystemTime>) {
if let Some(evict_url) = deadlines
let now = SystemTime::now();
let expired = deadlines
.iter()
.min_by_key(|(_, deadline)| *deadline)
.filter(|(_, deadline)| **deadline <= now)
.map(|(url, _)| url.clone())
.next();
if let Some(expired_url) = expired {
deadlines.remove(&expired_url);
return;
}
if let Some(farthest) = deadlines
.iter()
.max_by_key(|(_, deadline)| deadline.duration_since(now).unwrap_or_default())
.map(|(url, _)| url.clone())
{
deadlines.remove(&evict_url);
deadlines.remove(&farthest);
}
}
@@ -73,13 +120,20 @@ impl RetryAfterMiddleware {
}
async fn maybe_sleep_for(&self, url: &Url) {
if let Some(deadline) = self.deadline_for(url) {
if let Ok(remaining) = deadline.duration_since(SystemTime::now()) {
if !remaining.is_zero() {
tokio::time::sleep(remaining).await;
}
}
let Some(deadline) = self.deadline_for(url) else {
return;
};
let Ok(remaining) = deadline.duration_since(SystemTime::now()) else {
return;
};
if remaining.is_zero() {
return;
}
let jitter = max_sleep_jitter(remaining);
let wait = remaining
.checked_sub(jitter)
.unwrap_or(Duration::from_millis(1));
tokio::time::sleep(wait).await;
}
fn record_if_throttled(&self, url: Url, response: &Response) {
@@ -90,7 +144,7 @@ impl RetryAfterMiddleware {
.get(RETRY_AFTER_HEADER)
.and_then(|value| value.to_str().ok())
{
if let Some(deadline) = parse_retry_after(retry_after) {
if let Some(deadline) = parse_retry_after_with_ceiling(retry_after, self.ceiling) {
self.record(url, deadline);
}
}
@@ -116,6 +170,20 @@ impl RetryAfterMiddleware {
}
}
fn max_sleep_jitter(remaining: Duration) -> Duration {
if remaining.is_zero() {
return Duration::ZERO;
}
let fractional = remaining.as_secs_f64() * SLEEP_JITTER_FRACTION;
let fractional = if fractional.is_finite() && fractional > 0.0 {
fractional
} else {
0.0
};
let capped = fractional.min(DEFAULT_MAX_SLEEP_JITTER.as_secs_f64());
Duration::try_from_secs_f64(capped).unwrap_or(Duration::ZERO)
}
#[async_trait::async_trait]
impl Middleware for RetryAfterMiddleware {
async fn handle(
@@ -127,7 +195,7 @@ impl Middleware for RetryAfterMiddleware {
let req_url = req.url().clone();
self.maybe_sleep_for(&req_url).await;
let response = next.run(req, extensions).await?;
self.record_if_throttled(req_url, &response);
self.record_if_throttled(response.url().clone(), &response);
Ok(response)
}
}
@@ -150,33 +218,83 @@ mod tests {
#[test]
fn parse_retry_after_seconds() {
let deadline = parse_retry_after("5").expect("seconds value parses");
let mw = RetryAfterMiddleware::with_capacity_and_ceiling(8, Duration::from_secs(300));
let target = url("https://api.example.com/v1/chat");
let response = synthetic_response(StatusCode::TOO_MANY_REQUESTS, Some("5"));
mw.record_if_throttled(target.clone(), &response);
let deadline = mw.deadline_for_test(&target).expect("seconds parse");
let now = SystemTime::now();
let lower = now.checked_add(Duration::from_secs(4)).unwrap();
let upper = now.checked_add(Duration::from_secs(6)).unwrap();
assert!(deadline > lower && deadline < upper);
assert!(deadline > now);
assert!(deadline < now + Duration::from_secs(6));
}
#[test]
fn parse_retry_after_http_date() {
let deadline =
parse_retry_after("Wed, 21 Oct 2099 07:28:00 GMT").expect("HTTP-date value parses");
let mw = RetryAfterMiddleware::with_capacity_and_ceiling(8, Duration::from_secs(300));
let target = url("https://api.example.com/v1/chat");
let response = synthetic_response(
StatusCode::SERVICE_UNAVAILABLE,
Some("Wed, 21 Oct 2099 07:28:00 GMT"),
);
mw.record_if_throttled(target.clone(), &response);
let deadline = mw.deadline_for_test(&target).expect("HTTP-date parses");
let ceiling = SystemTime::now() + Duration::from_secs(300);
assert!(
deadline <= ceiling,
"HTTP-date deadlines must be clamped to the ceiling, got {deadline:?}"
);
assert!(deadline > SystemTime::now());
}
#[test]
fn parse_retry_after_past_http_date_yields_none() {
let deadline = parse_retry_after("Wed, 21 Oct 2015 07:28:00 GMT");
let mw = RetryAfterMiddleware::with_capacity_and_ceiling(8, Duration::from_secs(300));
let target = url("https://api.example.com/v1/chat");
let response = synthetic_response(
StatusCode::TOO_MANY_REQUESTS,
Some("Wed, 21 Oct 2015 07:28:00 GMT"),
);
mw.record_if_throttled(target.clone(), &response);
assert!(
deadline.is_none(),
mw.deadline_for_test(&target).is_none(),
"a deadline already in the past must not be recorded"
);
}
#[test]
fn parse_retry_after_invalid_yields_none() {
assert!(parse_retry_after("not-a-date").is_none());
assert!(parse_retry_after("").is_none());
let ceiling = Duration::from_secs(300);
assert!(parse_retry_after_with_ceiling("not-a-date", ceiling).is_none());
assert!(parse_retry_after_with_ceiling("", ceiling).is_none());
}
#[test]
fn retry_after_seconds_are_clamped_to_the_ceiling() {
let mw = RetryAfterMiddleware::with_capacity_and_ceiling(8, Duration::from_secs(300));
let target = url("https://api.example.com/v1/chat");
let response = synthetic_response(StatusCode::TOO_MANY_REQUESTS, Some("315360000"));
mw.record_if_throttled(target.clone(), &response);
let deadline = mw
.deadline_for_test(&target)
.expect("a huge Retry-After still records (clamped)");
let ceiling = SystemTime::now() + Duration::from_secs(300);
assert!(
deadline <= ceiling,
"deadline must be clamped to the configured ceiling"
);
assert!(deadline > SystemTime::now());
}
#[test]
fn ceiling_is_configurable() {
let mw = RetryAfterMiddleware::with_capacity_and_ceiling(8, Duration::from_secs(5));
let target = url("https://api.example.com/v1/chat");
let response = synthetic_response(StatusCode::TOO_MANY_REQUESTS, Some("3600"));
mw.record_if_throttled(target.clone(), &response);
let deadline = mw.deadline_for_test(&target).expect("clamped deadline kept");
let upper = SystemTime::now() + Duration::from_secs(5);
assert!(deadline <= upper, "custom ceiling must be honored");
assert!(deadline > SystemTime::now());
}
#[test]
@@ -190,21 +308,38 @@ mod tests {
}
#[test]
fn record_evicts_oldest_when_at_capacity() {
fn record_evicts_expired_entries_first() {
let mw = RetryAfterMiddleware::with_capacity(2);
let expired = url("https://expired.example.com");
let u1 = url("https://a.example.com");
let u2 = url("https://b.example.com");
mw.record_test(expired.clone(), SystemTime::now() - Duration::from_secs(1));
mw.record_test(u1.clone(), SystemTime::now() + Duration::from_secs(100));
mw.record_test(u2.clone(), SystemTime::now() + Duration::from_secs(50));
assert_eq!(mw.len(), 2, "capacity must be enforced");
assert!(
mw.deadline_for_test(&expired).is_none(),
"an expired entry must be evicted before live ones"
);
assert!(mw.deadline_for_test(&u1).is_some());
assert!(mw.deadline_for_test(&u2).is_some());
}
#[test]
fn record_evicts_the_farthest_deadline_when_none_expired() {
let mw = RetryAfterMiddleware::with_capacity(2);
let u1 = url("https://a.example.com");
let u2 = url("https://b.example.com");
let u3 = url("https://c.example.com");
mw.record_test(u1.clone(), SystemTime::now() + Duration::from_secs(100));
mw.record_test(u2.clone(), SystemTime::now() + Duration::from_secs(1));
assert_eq!(mw.len(), 2);
mw.record_test(u3.clone(), SystemTime::now() + Duration::from_secs(50));
assert_eq!(mw.len(), 2, "capacity must be enforced");
assert!(
mw.deadline_for_test(&u2).is_none(),
"entry with the earliest deadline must be evicted"
mw.deadline_for_test(&u1).is_none(),
"the farthest-future entry must be evicted when nothing has expired"
);
assert!(mw.deadline_for_test(&u1).is_some());
assert!(mw.deadline_for_test(&u2).is_some());
assert!(mw.deadline_for_test(&u3).is_some());
}
@@ -225,32 +360,17 @@ mod tests {
}
#[tokio::test]
async fn middleware_records_deadline_from_seconds_header() {
async fn middleware_records_under_the_effective_url() {
let mw = std::sync::Arc::new(RetryAfterMiddleware::with_capacity(8));
let target = url("https://api.example.com/v1/chat");
let origin = url("https://api.example.com/v1/chat");
let redirector = url("https://redirector.example.com/429");
let response = synthetic_response(StatusCode::TOO_MANY_REQUESTS, Some("5"));
mw.record_if_throttled(target.clone(), &response);
let deadline = mw
.deadline_for_test(&target)
.expect("429 with Retry-After records a deadline");
let now = SystemTime::now();
assert!(deadline > now, "deadline must be in the future");
assert!(deadline < now + Duration::from_secs(6));
}
#[tokio::test]
async fn middleware_records_deadline_from_http_date_header() {
let mw = std::sync::Arc::new(RetryAfterMiddleware::with_capacity(8));
let target = url("https://api.example.com/v1/chat");
let response = synthetic_response(
StatusCode::SERVICE_UNAVAILABLE,
Some("Wed, 21 Oct 2099 07:28:00 GMT"),
mw.record_if_throttled(origin.clone(), &response);
assert!(
mw.deadline_for_test(&origin).is_some(),
"the effective (post-redirect) URL carries the deadline"
);
mw.record_if_throttled(target.clone(), &response);
let deadline = mw
.deadline_for_test(&target)
.expect("503 with Retry-After HTTP-date records a deadline");
assert!(deadline > SystemTime::now());
assert_eq!(redirector.host_str(), Some("redirector.example.com"));
}
#[tokio::test]
@@ -283,8 +403,39 @@ mod tests {
mw.maybe_sleep_for(&target).await;
let elapsed = SystemTime::now().duration_since(started).unwrap();
assert!(
elapsed >= Duration::from_millis(40),
"middleware must sleep until the deadline elapses"
elapsed >= Duration::from_millis(37),
"middleware must sleep (minus jitter) until the deadline elapses"
);
assert!(elapsed < Duration::from_secs(2));
}
#[tokio::test]
async fn sleep_wakes_before_the_deadline_within_the_jitter_bound() {
let mw = std::sync::Arc::new(RetryAfterMiddleware::with_capacity(8));
let target = url("https://api.example.com/v1/chat");
let remaining = Duration::from_secs(4);
mw.record_test(target.clone(), SystemTime::now() + remaining);
let started = SystemTime::now();
mw.maybe_sleep_for(&target).await;
let elapsed = SystemTime::now().duration_since(started).unwrap();
let max_jitter = max_sleep_jitter(remaining);
assert!(
elapsed <= remaining.checked_sub(max_jitter).unwrap_or(remaining)
+ Duration::from_millis(50),
"wake must happen roughly a jitter-slice before the deadline, took {elapsed:?}"
);
assert!(max_jitter > Duration::ZERO, "jitter must be non-zero");
assert!(max_jitter <= Duration::from_secs(2));
}
#[test]
fn jitter_is_bounded_by_a_fraction_of_the_remaining_wait() {
let short = max_sleep_jitter(Duration::from_millis(200));
assert!(short <= Duration::from_millis(50));
let long = max_sleep_jitter(Duration::from_secs(100));
assert!(
long <= Duration::from_secs(2),
"jitter is capped at 2 s even for long waits"
);
}
}
}