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
use crate::execute_delay;
use crate::options::Options;
use crate::pretty::PrettyPrint;
use crate::PORT_COUNTERS;
use crate::REQUESTS;
use crate::REQUEST_BYTES;
use crate::RESPONSES;
use crate::RESPONSE_BYTES;
use anyhow::Result;
use futures::stream::{select_all, StreamExt};
use hyper::Response;
use owo_colors::OwoColorize;
use std::net::SocketAddr;
use std::sync::Arc;
use std::time::Duration;
use tracing::error;
use crate::create_listener;
use crate::ServerConfig;
pub fn run_thread(
id: usize,
addrs: Vec<SocketAddr>,
config: Arc<ServerConfig>,
opts: &Options,
) -> Result<()> {
// tokio io_uring implementation for Linux
use tracing::info;
let num_entries = opts.uring_entries.next_power_of_two();
let cqsize = num_entries * 2;
let delay = opts.delay;
let mut uring = tokio_uring::uring_builder();
uring.setup_single_issuer().setup_cqsize(cqsize);
if let Some(idle) = opts.uring_sqpoll {
uring.setup_sqpoll(idle);
} else {
uring.setup_coop_taskrun().setup_taskrun_flag();
}
let meter = opts.meter;
let verbose = opts.verbose;
tokio_uring::builder()
.entries(num_entries) // Large ring size is critical for throughput
.uring_builder(&uring)
.start(async move {
// Create multiple listeners for each address
let mut listeners = Vec::new();
for addr in &addrs {
let std_listener = match create_listener(*addr, opts) {
Ok(l) => l,
Err(e) => {
error!("Failed to create listener for {}: {}", addr, e);
return;
}
};
listeners.push(tokio_uring::net::TcpListener::from_std(std_listener));
}
info!(
"Thread {} listening on {:?} (tokio_uring, entries: {}, sqpoll: {:?})",
id, addrs, opts.uring_entries, opts.uring_sqpoll
);
// Convert each listener into a stream using unfold
let streams: Vec<_> = listeners
.into_iter()
.map(|listener| {
let port = listener.local_addr().unwrap().port();
Box::pin(futures::stream::unfold(listener, move |l| async move {
match l.accept().await {
Ok((tcp_stream, _)) => Some((Ok((tcp_stream, port)), l)),
Err(e) => {
error!("Accept error: {}", e);
// Continue accepting despite errors
Some((Err(e), l))
}
}
}))
})
.collect();
let mut all_listeners = select_all(streams);
loop {
match all_listeners.next().await {
Some(Ok((tcp_stream, port))) => {
let config = config.clone();
// Spawn task to handle the connection with io_uring
tokio_uring::spawn(async move {
if let Err(e) = handle_connection_uring(
tcp_stream, port, config, false, meter, delay, verbose,
)
.await
{
error!("Error handling tokio_uring connection: {}", e);
}
});
}
Some(Err(e)) => {
error!("Thread {} accept error: {}", id, e);
}
None => {
error!("Thread {} all listeners closed", id);
break;
}
}
}
});
Ok(())
}
#[cfg(all(target_os = "linux", feature = "tokio_uring"))]
async fn handle_connection_uring(
stream: tokio_uring::net::TcpStream,
port: u16,
config: Arc<ServerConfig>,
http2: bool,
meter: bool,
delay: Option<Duration>,
verbose: u8,
) -> Result<usize> {
use http_wire::WireDecode;
use std::mem::MaybeUninit;
use crate::response::build_response;
if http2 {
return Err(anyhow::anyhow!("HTTP/2 not supported with tokio_uring"));
}
let response_bytes = build_response(&config)?;
let mut response_buf = response_bytes; // Reuse to avoid allocations on each write
// Single buffer strategy for maximum performance:
// - In the common case (complete request in one read), we parse directly with zero copies
// - Only compact when we have leftover data from incomplete requests (rare case)
let mut buf: Vec<u8> = Vec::with_capacity(8192);
let mut parsed = 0; // Number of bytes already parsed/consumed
let mut requests_served = 0;
loop {
// FAST PATH: If we've consumed all data, reset the buffer (zero-cost operation)
if parsed == buf.len() && parsed > 0 {
unsafe {
buf.set_len(0);
}
parsed = 0;
}
// SLOW PATH: If we have many dead bytes at the beginning, compact the buffer
// This only happens when requests are split across multiple reads (rare)
else if parsed > 4096 {
buf.copy_within(parsed.., 0);
unsafe {
buf.set_len(buf.len() - parsed);
}
parsed = 0;
}
// Ensure we have enough spare capacity for reading
// Only reserve if needed to avoid unnecessary allocations in the common case
let current_len = buf.len();
if buf.capacity() - current_len < 4096 {
buf.reserve(4096);
}
// TRICK: Create a temporary Vec from the spare capacity for tokio_uring's ownership-based API
// This allows us to read directly into the buffer's unused capacity without copying.
// tokio_uring takes ownership of the buffer, passes it to io_uring, and returns it.
let spare_cap = buf.capacity() - current_len;
let temp_buf =
unsafe { Vec::from_raw_parts(buf.as_mut_ptr().add(current_len), 0, spare_cap) };
let (result, temp_buf) = stream.read(temp_buf).await;
std::mem::forget(temp_buf);
let n = match result {
Ok(0) => {
break; // Connection closed (EOF)
}
Ok(n) => n,
Err(e) => {
return Err(e.into());
}
};
// Update the main buffer's length and forget temp_buf to avoid double-free
// (temp_buf points to the same memory as buf's spare capacity)
unsafe {
buf.set_len(current_len + n);
}
// Parse all complete requests in the buffer
// This handles HTTP pipelining: multiple requests in a single read
let mut headers = [const { MaybeUninit::uninit() }; 128];
loop {
match http_wire::request::FullRequest::decode_uninit(&buf[parsed..], &mut headers) {
Ok((req, req_len)) => {
requests_served += 1;
parsed += req_len;
if verbose > 0 {
println!("↩ {}:\n{}", "request".bold(), req.pretty(verbose));
}
if meter {
REQUESTS.add(1);
REQUEST_BYTES.add(req_len);
let entry = PORT_COUNTERS.entry(port).or_default();
entry.requests.add(1);
entry.request_bytes.add(req_len);
}
if let Some(d) = delay {
execute_delay(d).await;
}
if verbose > 0 {
let mut print_builder = Response::builder().status(config.status);
for (k, v) in &config.headers {
print_builder = print_builder.header(k, v);
}
let print_resp = print_builder.body(config.body.clone()).unwrap();
println!("↪ {}:\n{}", "response".bold(), print_resp.pretty(verbose));
}
// Reuse response_buf (tokio_uring returns it after the write)
let (res, buf_back) = stream.write_all(response_buf).await;
response_buf = buf_back;
res?;
if meter {
RESPONSES.add(1);
RESPONSE_BYTES.add(response_buf.len());
let entry = PORT_COUNTERS.entry(port).or_default();
entry.responses.add(1);
entry.response_bytes.add(response_buf.len());
}
}
Err(_) => break, // Incomplete request or end of batch
}
}
// Loop continues: common case is parsed == buf.len(), so fast path activates on next iteration
}
Ok(requests_served)
}