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
use std::time::Duration;
/// Thread-local xorshift64 PRNG — fast, no external deps, no locks.
/// Each thread gets independent state seeded from system time + thread ID,
/// so concurrent crawlers produce different sequences.
#[inline]
fn thread_rng_u64(bound: u64) -> u64 {
use std::cell::Cell;
thread_local! {
static STATE: Cell<u64> = Cell::new({
let nanos = std::time::SystemTime::now()
.duration_since(std::time::UNIX_EPOCH)
.unwrap_or_default()
.subsec_nanos() as u64;
// Use stack address as thread-unique entropy (different per thread).
let stack_addr = &nanos as *const u64 as u64;
nanos ^ stack_addr.wrapping_mul(0x9E3779B97F4A7C15)
});
}
if bound == 0 {
return 0;
}
STATE.with(|cell| {
let mut x = cell.get();
// xorshift64
x ^= x << 13;
x ^= x >> 7;
x ^= x << 17;
cell.set(x);
x % bound
})
}
/// Compute exponential backoff with full jitter.
///
/// The delay is a uniform random value in `[0, cap]` where
/// `cap = min(base_ms * 2^attempt, max_ms)`.
///
/// This gives maximum spread between concurrent retriers at every
/// tick level — no two crawlers will cluster on the same delay.
///
/// Uses checked arithmetic throughout — no panics on overflow.
#[inline]
pub fn backoff_delay(attempt: u32, base_ms: u64, max_ms: u64) -> Duration {
let exp = base_ms.saturating_mul(1u64.checked_shl(attempt).unwrap_or(u64::MAX));
let cap = exp.min(max_ms);
// Full jitter: uniform in [0, cap]
let delay = thread_rng_u64(cap.saturating_add(1));
Duration::from_millis(delay)
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn attempt_zero_within_cap() {
// Full jitter: attempt=0, base=100 → cap=100, delay in [0, 100]
for _ in 0..100 {
let d = backoff_delay(0, 100, 10_000);
assert!(d.as_millis() <= 100);
}
}
#[test]
fn higher_attempt_has_higher_cap() {
// attempt=3, base=100 → cap=800, delay in [0, 800]
// attempt=0 → cap=100, delay in [0, 100]
// Over many samples, attempt=3 average (~400) >> attempt=0 average (~50)
let mut sum_low: u128 = 0;
let mut sum_high: u128 = 0;
let n = 200;
for _ in 0..n {
sum_low += backoff_delay(0, 100, 100_000).as_millis();
sum_high += backoff_delay(3, 100, 100_000).as_millis();
}
assert!(
sum_high > sum_low,
"attempt 3 average should exceed attempt 0 average"
);
}
#[test]
fn respects_max_ms_cap() {
for _ in 0..100 {
let d = backoff_delay(30, 1000, 5000);
assert!(d.as_millis() <= 5000);
}
}
#[test]
fn full_jitter_produces_spread() {
// Full jitter should produce different values across calls.
// With cap=10000, getting 100 identical values is astronomically unlikely.
let mut seen = std::collections::HashSet::new();
for _ in 0..100 {
seen.insert(backoff_delay(3, 1000, 100_000).as_millis());
}
assert!(
seen.len() > 10,
"full jitter should produce spread, got only {} distinct values",
seen.len()
);
}
#[test]
fn huge_attempt_saturates_without_panic() {
let d = backoff_delay(u32::MAX, 1000, 60_000);
assert!(d.as_millis() <= 60_000);
}
#[test]
fn zero_base_ms_does_not_panic() {
let d = backoff_delay(5, 0, 10_000);
assert_eq!(d.as_millis(), 0);
}
#[test]
fn zero_max_ms_returns_zero() {
let d = backoff_delay(3, 100, 0);
assert_eq!(d.as_millis(), 0);
}
/// Verify backoff_delay is safe under concurrent access — no panics,
/// no deadlocks, all values within expected bounds.
/// Each thread gets independent PRNG state, so no contention.
#[test]
fn concurrent_safety_no_panic_or_deadlock() {
use std::sync::{Arc, Barrier};
let threads = 32;
let iterations = 500;
let barrier = Arc::new(Barrier::new(threads));
let handles: Vec<_> = (0..threads)
.map(|t| {
let barrier = Arc::clone(&barrier);
std::thread::spawn(move || {
barrier.wait();
for i in 0..iterations {
let attempt = ((t * iterations + i) % 20) as u32;
let d = backoff_delay(attempt, 200, 15_000);
assert!(
d.as_millis() <= 15_000,
"thread {t} iter {i}: delay {}ms exceeds cap",
d.as_millis()
);
}
})
})
.collect();
for h in handles {
h.join().expect("thread must not panic");
}
}
/// Concurrent threads should produce divergent sequences (thread-local seed
/// includes stack address), preventing thundering herd.
#[test]
fn concurrent_threads_diverge() {
use std::sync::{Arc, Barrier};
let threads = 8;
let barrier = Arc::new(Barrier::new(threads));
let handles: Vec<_> = (0..threads)
.map(|_| {
let barrier = Arc::clone(&barrier);
std::thread::spawn(move || {
barrier.wait();
let seq: Vec<u128> = (0..20)
.map(|i| backoff_delay(i, 1000, 60_000).as_millis())
.collect();
seq
})
})
.collect();
let seqs: Vec<Vec<u128>> = handles.into_iter().map(|h| h.join().unwrap()).collect();
// Count how many pairs of threads have identical sequences
let mut identical_pairs = 0;
for i in 0..seqs.len() {
for j in (i + 1)..seqs.len() {
if seqs[i] == seqs[j] {
identical_pairs += 1;
}
}
}
assert_eq!(
identical_pairs, 0,
"threads should produce different delay sequences"
);
}
#[test]
fn base_ms_one_stays_bounded() {
for attempt in 0..20 {
let d = backoff_delay(attempt, 1, 60_000);
let cap = 1u64.checked_shl(attempt).unwrap_or(u64::MAX).min(60_000);
assert!(d.as_millis() <= cap as u128);
}
}
}