hpx 2.5.24

High Performance HTTP Client
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355

//! Middleware for retrying requests.

mod classify;
mod scope;

use std::{error::Error as StdError, future::Ready, sync::Arc, time::Duration};

use http::{Request, Response};
use tower::retry::{
    Policy,
    budget::{Budget, TpsBudget},
};

pub(crate) use self::{
    classify::{Action, Classifier, ClassifyFn, ReqRep},
    scope::{ScopeFn, Scoped},
};
use crate::{Body, ClientResponseBody, error::BoxError, retry};

/// A retry policy for HTTP requests.
#[derive(Clone)]
pub(crate) struct RetryPolicy {
    budget: Option<Arc<TpsBudget>>,
    classifier: Classifier,
    max_retries_per_request: u32,
    retry_cnt: u32,
    scope: Scoped,
}

impl RetryPolicy {
    /// Create a new `RetryPolicy`.
    #[inline]
    pub(crate) fn new(policy: retry::Policy) -> Self {
        Self {
            budget: policy
                .budget
                .map(|budget| Arc::new(TpsBudget::new(Duration::from_secs(10), 10, budget))),
            classifier: policy.classifier,
            max_retries_per_request: policy.max_retries_per_request,
            retry_cnt: 0,
            scope: policy.scope,
        }
    }
}

type Req = Request<Body>;

type Res = Response<ClientResponseBody>;

pub(crate) fn clone_http_request(req: &Req) -> Option<Req> {
    let body = req.body().try_clone()?;
    let mut new = http::Request::new(body);
    *new.method_mut() = req.method().clone();
    *new.uri_mut() = req.uri().clone();
    *new.version_mut() = req.version();
    *new.headers_mut() = req.headers().clone();
    *new.extensions_mut() = req.extensions().clone();
    Some(new)
}

impl Policy<Req, Res, BoxError> for RetryPolicy {
    type Future = Ready<()>;

    fn retry(&mut self, req: &mut Req, result: &mut Result<Res, BoxError>) -> Option<Self::Future> {
        match self.classifier.classify(req, result) {
            Action::Success => {
                trace!(
                    "Request successful, no retry needed: {} {}",
                    req.method(),
                    req.uri()
                );

                if let Some(ref budget) = self.budget {
                    budget.deposit();
                    trace!("Token deposited back to retry budget");
                }
                None
            }
            Action::Retryable => {
                trace!(
                    "Retrying request ({} attempts so far): {} {} - {}",
                    self.retry_cnt,
                    req.method(),
                    req.uri(),
                    match result {
                        Ok(res) => format!("HTTP {}", res.status()),
                        Err(e) => format!("Error: {}", e),
                    }
                );

                Some(std::future::ready(()))
            }
        }
    }

    fn clone_request(&mut self, req: &Req) -> Option<Req> {
        if self.retry_cnt > 0 && !self.scope.applies_to(req) {
            trace!("not in scope, not retrying");
            return None;
        }

        if self.retry_cnt >= self.max_retries_per_request {
            trace!("max_retries_per_request hit");
            return None;
        }

        // Withdraw budget token only when we're about to actually retry.
        if !self.budget.as_ref().is_none_or(|b| b.withdraw()) {
            debug!(
                "Request is retryable but retry budget exhausted: {} {}",
                req.method(),
                req.uri()
            );
            return None;
        }

        self.retry_cnt += 1;
        if let Some(cloned) = clone_http_request(req) {
            Some(cloned)
        } else {
            // Clone failed — deposit the token back.
            if let Some(ref budget) = self.budget {
                budget.deposit();
            }
            None
        }
    }
}

/// Determines whether the given error/method pair is safe to retry.
///
/// The default classifier retries:
/// - HTTP/2 GOAWAY (graceful shutdown) and REFUSED_STREAM resets (RFC 9113);
/// - connection-establishment failures — the request was never delivered to
///   the server, so retrying is safe for every method;
/// - transport timeouts and connection resets, but only for idempotent
///   methods (GET/HEAD/PUT/DELETE/OPTIONS/TRACE) — after a timeout/reset the
///   server may or may not have processed the request, so retrying a
///   non-idempotent method could cause duplicate side effects (e.g. a
///   re-placed order).
fn is_retryable_error(err: &(dyn StdError + 'static), method: &http::Method) -> bool {
    let mut source = Some(err);

    while let Some(err) = source {
        #[cfg(feature = "http2")]
        if let Some(h2_err) = err.downcast_ref::<http2::Error>() {
            // They sent us a graceful shutdown, try with a new connection!
            if h2_err.is_go_away()
                && h2_err.is_remote()
                && h2_err.reason() == Some(http2::Reason::NO_ERROR)
            {
                return true;
            }

            // REFUSED_STREAM was sent from the server, which is safe to retry.
            // https://www.rfc-editor.org/rfc/rfc9113.html#section-8.7-3.2
            if h2_err.is_reset()
                && h2_err.is_remote()
                && h2_err.reason() == Some(http2::Reason::REFUSED_STREAM)
            {
                return true;
            }
        }

        // Connection-establishment failures: the request was never delivered
        // to the server, so retrying is safe regardless of the method.
        if err
            .downcast_ref::<crate::Error>()
            .is_some_and(|e| e.is_connect())
            || err
                .downcast_ref::<crate::client::Error>()
                .is_some_and(|e| e.is_connect())
            || matches!(
                err.downcast_ref::<std::io::Error>().map(|e| e.kind()),
                Some(std::io::ErrorKind::ConnectionRefused | std::io::ErrorKind::ConnectionAborted)
            )
        {
            return true;
        }

        // Timeout / connection reset on a (likely reused) connection. The
        // server may or may not have processed the request — only retry
        // idempotent methods to avoid duplicate side effects.
        if method_is_idempotent(method) && (is_timeout_error(err) || is_connection_reset_error(err))
        {
            return true;
        }

        source = err.source();
    }

    false
}

/// True for methods that a server may execute multiple times without
/// additional side effects.
fn method_is_idempotent(method: &http::Method) -> bool {
    matches!(
        *method,
        http::Method::GET
            | http::Method::HEAD
            | http::Method::PUT
            | http::Method::DELETE
            | http::Method::OPTIONS
            | http::Method::TRACE
    )
}

/// True if the error is a timeout, checking the current layer of the source
/// chain. Mirrors [`crate::Error::is_timeout`] so the retry classifier
/// recognizes the same conditions as the public error API. This includes the
/// `Error::request(TimedOut)` shape produced by the outer request `Timeout`
/// layer, so a timed-out attempt can be retried.
fn is_timeout_error(err: &(dyn StdError + 'static)) -> bool {
    err.is::<crate::error::TimedOut>()
        || err
            .downcast_ref::<crate::Error>()
            .is_some_and(|e| e.is_timeout())
        || err
            .downcast_ref::<crate::client::CoreError>()
            .is_some_and(|e| e.is_timeout())
        || err
            .downcast_ref::<std::io::Error>()
            .is_some_and(|e| e.kind() == std::io::ErrorKind::TimedOut)
}

/// True if the error is a connection reset (peer closed the socket abruptly).
fn is_connection_reset_error(err: &(dyn StdError + 'static)) -> bool {
    err.downcast_ref::<std::io::Error>()
        .is_some_and(|e| e.kind() == std::io::ErrorKind::ConnectionReset)
}

#[cfg(test)]
mod tests {
    use std::error::Error as StdError;

    use http::Request;
    use http_body_util::BodyExt;
    use tower::retry::Policy as _;

    use super::{RetryPolicy, is_retryable_error};
    use crate::{Body, retry};

    fn make_retry_policy(max_retries: u32) -> RetryPolicy {
        let policy = retry::Policy {
            budget: Some(0.2),
            classifier: crate::client::layer::retry::Classifier::Never,
            max_retries_per_request: max_retries,
            scope: crate::client::layer::retry::Scoped::Unscoped,
        };
        RetryPolicy::new(policy)
    }

    #[test]
    fn budget_preserved_when_clone_fails() {
        let mut policy = make_retry_policy(2);
        // Create a request with a non-clonable body (streaming/boxed).
        let body: Body = http_body_util::Empty::new()
            .map_err(|e| -> crate::error::BoxError { match e {} })
            .boxed()
            .into();
        let req = Request::new(body);

        // clone_request should return None because the body is not clonable.
        let result = policy.clone_request(&req);
        assert!(result.is_none(), "non-clonable body should return None");

        // Verify the budget still has tokens by cloning a clonable request.
        let clonable_req = Request::new(Body::from("hello"));
        let result2 = policy.clone_request(&clonable_req);
        assert!(
            result2.is_some(),
            "budget should still have tokens after failed clone"
        );
    }

    // -----------------------------------------------------------------------
    // Default classifier: network-level retry classification
    // -----------------------------------------------------------------------

    fn io_err(kind: std::io::ErrorKind) -> Box<dyn StdError + Send + Sync> {
        Box::new(std::io::Error::new(kind, "synthetic io error"))
    }

    /// Coerce a concrete error into a trait object for `is_retryable_error`.
    fn to_dyn<'a>(err: &'a (dyn StdError + 'static)) -> &'a (dyn StdError + 'static) {
        err
    }

    #[test]
    fn connect_failure_retryable_for_all_methods() {
        let err = crate::Error::request(io_err(std::io::ErrorKind::ConnectionRefused));
        // Connection establishment failed: the request was never delivered,
        // so retrying is safe even for state-changing methods.
        assert!(is_retryable_error(to_dyn(&err), &http::Method::GET));
        assert!(is_retryable_error(to_dyn(&err), &http::Method::POST));

        let aborted = crate::Error::request(io_err(std::io::ErrorKind::ConnectionAborted));
        assert!(is_retryable_error(to_dyn(&aborted), &http::Method::GET));
        assert!(is_retryable_error(to_dyn(&aborted), &http::Method::POST));
    }

    #[test]
    fn timeout_retryable_only_for_idempotent_methods() {
        let err = crate::Error::request(io_err(std::io::ErrorKind::TimedOut));
        assert!(is_retryable_error(to_dyn(&err), &http::Method::GET));
        assert!(is_retryable_error(to_dyn(&err), &http::Method::DELETE));
        // A timeout leaves ambiguous whether the server processed the request;
        // never retry non-idempotent methods (e.g. order placement).
        assert!(!is_retryable_error(to_dyn(&err), &http::Method::POST));
        assert!(!is_retryable_error(to_dyn(&err), &http::Method::PATCH));
    }

    #[test]
    fn connection_reset_retryable_only_for_idempotent_methods() {
        let err = crate::Error::request(io_err(std::io::ErrorKind::ConnectionReset));
        assert!(is_retryable_error(to_dyn(&err), &http::Method::GET));
        assert!(is_retryable_error(to_dyn(&err), &http::Method::PUT));
        assert!(!is_retryable_error(to_dyn(&err), &http::Method::POST));
    }

    #[test]
    fn timeout_detected_deep_in_source_chain() {
        // io::Error(ConnectionReset) -> crate::Error -> io::Error(other) wrap.
        let inner = crate::Error::request(io_err(std::io::ErrorKind::TimedOut));
        let wrapped = std::io::Error::other(inner);
        assert!(is_retryable_error(
            &wrapped as &(dyn StdError + 'static),
            &http::Method::GET,
        ));
        assert!(!is_retryable_error(
            &wrapped as &(dyn StdError + 'static),
            &http::Method::POST,
        ));
    }

    #[test]
    fn public_error_request_timedout_retryable_for_idempotent_methods() {
        // The public `Error::request(TimedOut)` shape produced by the outer
        // request `Timeout` layer must be recognized by the classifier so a
        // timed-out attempt can be retried.
        let err = crate::Error::request(crate::error::TimedOut);
        assert!(is_retryable_error(to_dyn(&err), &http::Method::GET));
        assert!(is_retryable_error(to_dyn(&err), &http::Method::DELETE));
        assert!(!is_retryable_error(to_dyn(&err), &http::Method::POST));
        assert!(!is_retryable_error(to_dyn(&err), &http::Method::PATCH));
    }

    #[test]
    fn unrelated_errors_are_not_retryable() {
        let err = crate::Error::request(io_err(std::io::ErrorKind::Other));
        assert!(!is_retryable_error(to_dyn(&err), &http::Method::GET));
        assert!(!is_retryable_error(to_dyn(&err), &http::Method::POST));
    }
}