netring 0.18.0

High-performance zero-copy packet I/O for Linux (AF_PACKET TPACKET_V3 + AF_XDP)
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
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
# netring

High-performance zero-copy packet I/O for Linux, **async-first**.

`netring` provides packet capture and injection via AF_PACKET (TPACKET_V3
block-based mmap ring buffers) and AF_XDP (kernel-bypass via XDP sockets).
The recommended API is async/tokio; sync types are first-class but
mostly used as the underlying source for the async wrappers.

## Quick start (async, recommended)

```toml
[dependencies]
netring = { version = "0.16", features = ["tokio"] }
```

```rust,no_run
// Capture: zero-copy borrowed batches via AsyncFd.
# async fn _ex() -> Result<(), netring::Error> {
let mut cap = netring::AsyncCapture::open("eth0")?;
loop {
    let mut guard = cap.readable().await?;
    if let Some(batch) = guard.next_batch() {
        for pkt in &batch {
            handle(pkt.data()).await;
        }
    }
}
# async fn handle(_: &[u8]) {}
# }
```

```rust,ignore
// Stream-style consumption with futures::StreamExt
// (add `futures = "0.3"` to your Cargo.toml):
use futures::StreamExt;

let mut stream = netring::AsyncCapture::open("eth0")?.into_stream();
while let Some(batch) = stream.next().await {
    for pkt in batch? {
        let _ = pkt.data;
    }
}
```

```rust,no_run
// Inject with backpressure (awaits POLLOUT when ring is full):
# async fn _ex() -> Result<(), netring::Error> {
let mut tx = netring::AsyncInjector::open("eth0")?;
tx.send(&[0xff; 64]).await?;
tx.flush().await?;
# Ok(()) }
```

```rust,no_run
// AF_XDP (kernel bypass, 10M+ pps) — same shape as AsyncCapture:
# #[cfg(feature = "af-xdp")]
# async fn _ex() -> Result<(), netring::Error> {
let mut xdp = netring::AsyncXdpSocket::open("eth0")?;
let batch = xdp.try_recv_batch().await?;
for pkt in &batch {
    let _ = pkt.data();
}
# Ok(()) }
```

See [docs/ASYNC_GUIDE.md](docs/ASYNC_GUIDE.md) for the full async story —
patterns, trade-offs, when to use which entry point, and `Send`/`!Send`
considerations.

## Flow & session tracking

```toml
[dependencies]
netring = { version = "0.16", features = ["tokio", "flow"] }
futures = "0.3"
```

```rust,ignore
use futures::StreamExt;
use netring::AsyncCapture;
use netring::flow::extract::FiveTuple;
use netring::flow::FlowEvent;

let cap = AsyncCapture::open("eth0")?;
let mut stream = cap.flow_stream(FiveTuple::bidirectional());
while let Some(evt) = stream.next().await {
    match evt? {
        FlowEvent::Started { key, .. } => println!("+ {} <-> {}", key.a, key.b),
        FlowEvent::Ended { key, history, .. } => println!("- {} <-> {}  hist={history}", key.a, key.b),
        _ => {}
    }
}
```

Pluggable flow keys (5-tuple, IpPair, MacPair, VLAN/MPLS/VXLAN/GTP-U
decap combinators, custom extractors), bidirectional sessions, TCP
state machine with Zeek-style history string, idle-timeout sweep,
LRU eviction, optional TCP reassembly hook (sync `Reassembler` or
async `AsyncReassembler` with `channel_factory` for backpressure).

The flow types live in a separate cross-platform crate
[`flowscope`](https://github.com/p13marc/flowscope) (no Linux, no
tokio, no async runtime — usable with pcap, tun-tap, embedded).
`netring` is the Linux capture integration; the underlying flow API
works on any source of `&[u8]` frames.

`flowscope` also ships feature-gated L7 modules: `http` (HTTP/1.x),
`tls` (TLS handshake observation, optional JA3), `dns` (DNS-over-UDP
parser + correlator), `icmp` (ICMPv4 + ICMPv6 with `IcmpInner`
cross-protocol correlation), and `pcap` (offline replay).

## Multi-protocol monitor + anomaly correlation

For the common case of watching one interface for several
protocols simultaneously, `ProtocolMonitorBuilder` collapses the
hand-rolled `tokio::select!` choreography to a single declarative
call:

```rust,ignore
use futures::StreamExt;
use netring::flow::extract::FiveTuple;
use netring::protocol::{ProtocolEvent, ProtocolMessage, ProtocolMonitorBuilder};

let mut monitor = ProtocolMonitorBuilder::new()
    .interface("eth0")
    .flow()        // ICMP/TCP/UDP lifecycle
    .http()        // TCP/80,8080 → HttpParser
    .dns()         // UDP/53 → DnsUdpParser::with_correlation()
    .tls()         // TCP/443,8443 → TlsParser
    .icmp()        // ICMPv4 + ICMPv6 → IcmpParser
    .build(FiveTuple::bidirectional())?;

while let Some(evt) = monitor.next().await {
    match evt? {
        ProtocolEvent::Flow(_) => {}
        ProtocolEvent::Message { message: ProtocolMessage::Http(_), .. } => {}
        ProtocolEvent::Message { message: ProtocolMessage::Dns(_), .. } => {}
        _ => {}
    }
}
```

The monitor opens one filtered `AsyncCapture` per enabled
protocol (each with a narrow kernel-side BPF filter) and
round-robin polls them into a unified `Stream<Item =
Result<ProtocolEvent<K>, Error>>` — one chatty protocol won't
starve the others.

**Anomaly correlation** sits on top of `ProtocolEvent` as a
small typed-rule harness. Each detector is an `impl AnomalyRule<K>`
of ~30 LoC; `AnomalyMonitor` fans every event through every rule:

```rust,ignore
use netring::anomaly::{Anomaly, AnomalyMonitor, AnomalyRule, FlowAnomalyRule, Severity};
use netring::correlate::TimeBucketedCounter;
use netring::flow::extract::FiveTupleKey;
use netring::protocol::{ProtocolEvent, ProtocolMessage};
use flowscope::dns::DnsMessage;
use std::net::IpAddr;
use std::time::Duration;

struct DnsBurstRule {
    counts: TimeBucketedCounter<IpAddr>,
    threshold: u64,
}
impl AnomalyRule<FiveTupleKey> for DnsBurstRule {
    fn name(&self) -> &'static str { "DnsBurst" }
    fn observe(&mut self, evt: &ProtocolEvent<FiveTupleKey>, emit: &mut Vec<Anomaly<FiveTupleKey>>) {
        let ProtocolEvent::Message { message: ProtocolMessage::Dns(DnsMessage::Query(_)), key, ts, .. } = evt
        else { return };
        self.counts.bump(key.a.ip(), *ts);
        if self.counts.count(&key.a.ip(), *ts) > self.threshold {
            emit.push(Anomaly::new(self.name(), Severity::Warning, *ts).with_key(*key));
        }
    }
}

let mut rules = AnomalyMonitor::<FiveTupleKey>::new()
    .with_rule(DnsBurstRule { counts: TimeBucketedCounter::new(Duration::from_secs(10), Duration::from_secs(1)), threshold: 50 })
    .with_rule(FlowAnomalyRule::default());  // lifts flowscope's own anomalies in too
// monitor.next().await → rules.observe(&evt) → Vec<Anomaly<FiveTupleKey>>
```

`Anomaly<K>` impls `Display` for one-line greppable output and
`to_json_line()` for production-pipeline JSON (no `serde` dep —
escaping is hand-rolled to RFC 8259 §7). Severity tiers
(`Info/Warning/Error/Critical`) port directly to flowscope's
`AnomalyKind::severity()` via a `From` impl. Eight reference
detectors live under `examples/anomaly/`: `dns_query_burst`,
`dns_resolved_no_connection`, `anomaly_monitor_demo`,
`slow_tls_handshake`, `lateral_movement`, `icmp_explained_drop`,
`pcap_replay_anomaly`, `tls_to_unresolved_ip` (3-protocol). Set
`NETRING_JSON=1` to switch the showcase to JSON output. Pair
with `cargo run --example synthetic_traffic` to demo on `lo`
without `CAP_NET_RAW`.

See [docs/WRITING_DETECTORS.md](docs/WRITING_DETECTORS.md) for
the full tutorial — anatomy of an `AnomalyRule`, state-primitive
decision table, `observe` vs `on_tick`, cross-protocol patterns,
testing, production deployment, and MITRE ATT&CK mapping.

## Stream observability

Every async stream type — `FlowStream`, `SessionStream`,
`DatagramStream`, `DedupStream` — implements the sealed
[`StreamCapture`] trait. That gives uniform access to kernel ring
stats and the underlying `AsyncCapture` even after the stream has
consumed it:

```rust,ignore
use netring::{AsyncCapture, BpfFilter, StreamCapture};
use netring::flow::extract::FiveTuple;

let cap = AsyncCapture::open("eth0")?;
let stream = cap.flow_stream(FiveTuple::bidirectional());

// Kernel ring stats while the stream runs:
let stats = stream.capture_stats()?;
println!("ring drops: {}", stats.drops);

// Atomic BPF filter swap on a running stream:
let new_filter = BpfFilter::builder().tcp().dst_port(443).build()?;
stream.capture().set_filter(&new_filter)?;
```

Pair with `with_pcap_tap(writer)` on any of the four stream types
to record raw frames **before** the flow tracker processes them —
decoded events and a wire-faithful capture file from one invocation:

```rust,ignore
use std::fs::File;
use std::io::BufWriter;
use netring::pcap::CaptureWriter;

let writer = CaptureWriter::create(BufWriter::new(File::create("trace.pcap")?))?;
let stream = cap
    .flow_stream(FiveTuple::bidirectional())
    .with_pcap_tap(writer);
```

`TapErrorPolicy::{Continue (default), DropTap, FailStream}`
controls disk-full / I/O-glitch handling.

## BPF filter ergonomics

`AsyncCapture::open_with_filter` is the one-call sugar for the
common case:

```rust,no_run
use netring::{AsyncCapture, BpfFilter};

let filter = BpfFilter::builder().tcp().dst_port(15987).build().unwrap();
let _cap = AsyncCapture::open_with_filter("eth0", filter).unwrap();
```

For runtime filter swaps without tearing down the kernel ring:

```rust,no_run
use netring::{AsyncCapture, BpfFilter};

let cap = AsyncCapture::open("eth0").unwrap();
let new = BpfFilter::builder().tcp().dst_port(8443).build().unwrap();
cap.set_filter(&new).unwrap();  // atomic in-kernel replacement
```

`set_filter` is gated to AF_PACKET-backed captures via the
`PacketSetFilter` trait; `AsyncCapture<XdpSocket>` doesn't expose
it (XDP filtering belongs in the XDP program).

## Multi-source capture

`AsyncMultiCapture` fans in N captures of two shapes — multiple
interfaces, or one interface with a fanout-group worker pool — into
a single tagged stream:

```rust,ignore
use futures::StreamExt;
use netring::AsyncMultiCapture;
use netring::flow::extract::FiveTuple;

// Multi-interface gateway:
let multi = AsyncMultiCapture::open(["eth0", "eth1"])?;
let mut stream = multi.flow_stream(FiveTuple::bidirectional());
while let Some(tagged) = stream.next().await {
    let evt = tagged?;
    let iface = stream.label(evt.source_idx).unwrap_or("?");
    println!("[{iface}] {:?}", evt.event);
}

// Worker pool scaling (FanoutMode::Cpu by default):
let workers = AsyncMultiCapture::open_workers("eth0", 4, 0xDE57)?;
```

Per-source breakdown and aggregate stats:

```rust,ignore
let agg = stream.capture_stats();
for (label, stats) in stream.per_source_capture_stats() { /* ... */ }
```

See [`docs/scaling.md`](docs/scaling.md) for the canonical multi-core
recipe, the `FanoutMode` decision matrix, and 7 anti-patterns
(including the `FANOUT_HASH`-on-skewed-traffic and `PACKET_FANOUT`
-on-`lo` gotchas).

## Offline replay

`AsyncPcapSource` reads PCAP and PCAPNG files asynchronously (format
auto-detected at open) and yields `OwnedPacket`s through a tokio
`Stream`. The companion `PcapFlowStream` bridges to the same
flowscope `FlowTracker` used by live capture, so the same downstream
code runs both live and offline:

```rust,ignore
use futures::StreamExt;
use netring::AsyncPcapSource;
use netring::flow::extract::FiveTuple;

let source = AsyncPcapSource::open("trace.pcap").await?;
let mut events = source.flow_events(FiveTuple::bidirectional());
while let Some(evt) = events.next().await {
    let _ = evt?;
}
```

`AsyncPcapConfig` controls pacing (`replay_speed = 1.0` for wire
rate, `2.0` for double speed, `0.0` for as-fast-as-possible) and
`loop_at_eof` for stress testing.

[`StreamCapture`]: https://docs.rs/netring/latest/netring/trait.StreamCapture.html

## BPF filtering

netring ships a typed classic-BPF builder — no shelling out to
`tcpdump -dd`, no native-library deps:

```rust,no_run
use netring::{BpfFilter, Capture};

let filter = BpfFilter::builder()
    .tcp()
    .dst_port(443)
    .or(|b| b.udp().dst_port(53))
    .build()
    .unwrap();

let cap = Capture::builder()
    .interface("eth0")
    .bpf_filter(filter)
    .build()
    .unwrap();
```

Vocabulary: `eth_type` / `ipv4` / `ipv6` / `arp`, `vlan` / `vlan_id`,
`ip_proto` / `tcp` / `udp` / `icmp`, `src_host` / `dst_host` / `host`,
`src_net` / `dst_net` / `net`, `src_port` / `dst_port` / `port`,
plus `negate()` and `or(|b| ...)`. See
[`examples/bpf_filter.rs`](examples/bpf_filter.rs) for a runnable
demo. The escape hatch `BpfFilter::new(insns)` still accepts raw
bytecode from `tcpdump -dd` or any other source.

`BpfFilter::matches(&[u8]) -> bool` runs the bytecode in pure Rust
for offline validation against pcap data.

## Sync API

The sync types power the async wrappers and are also usable directly:

```rust,no_run
// Flat iterator — simplest path.
let mut cap = netring::Capture::open("eth0").unwrap();
for pkt in cap.packets().take(100) {
    println!("[{}] {} bytes", pkt.timestamp(), pkt.len());
}
```

```rust,no_run
// Batch processing with sequence-gap detection.
use netring::Capture;
use std::time::Duration;

let mut cap = Capture::builder()
    .interface("eth0")
    .block_size(1 << 22)
    .build()
    .unwrap();

while let Some(batch) = cap.next_batch_blocking(Duration::from_millis(100)).unwrap() {
    for pkt in &batch {
        let _ = pkt.data();
    }
}
```

## Features

| Feature | Default | Description |
|---------|---------|-------------|
| `tokio` | off | Async wrappers (`AsyncCapture`, `AsyncInjector`, `AsyncXdpSocket`, `PacketStream`) |
| `af-xdp` | off | AF_XDP kernel-bypass packet I/O (pure Rust, no native deps) |
| `xdp-loader` | off | Built-in redirect-all XDP program loader for AF_XDP via [`aya`]https://crates.io/crates/aya. Implies `af-xdp`. See [`async_xdp_self_loaded`]netring/examples/async_xdp_self_loaded.rs example. |
| `channel` | off | Thread + bounded channel adapter (runtime-agnostic) |
| `parse` | off | Packet header parsing via `etherparse` |
| `pcap` | off | Stream packets to PCAP files |
| `metrics` | off | `metrics` crate counters (`netring_capture_*_total`) |
| `flow` | off | Pluggable flow & session tracking (pulls `flowscope`, see [Flow & session tracking]#flow--session-tracking above) |

## Public API

| Concept | Sync type | Async wrapper |
|---------|-----------|---------------|
| AF_PACKET RX | `Capture` | `AsyncCapture<Capture>` |
| AF_PACKET TX | `Injector` | `AsyncInjector` |
| AF_XDP (RX + TX) | `XdpSocket` | `AsyncXdpSocket` |
| Bridge two interfaces | `Bridge` | `Bridge::run_async` |
| Channel adapter || `ChannelCapture` (sync threads) |

Every type has a `::open(iface)` shortcut for the simple case and a
`::builder()` for full configuration.

## Default Configuration

| Parameter | Default | Description |
|-----------|---------|-------------|
| `block_size` | 4 MiB | Ring buffer block size |
| `block_count` | 64 | Number of blocks (256 MiB total) |
| `frame_size` | 2048 | Minimum frame size |
| `block_timeout_ms` | 60 | Block retirement timeout |
| `fill_rxhash` | true | Kernel fills RX flow hash |

## Performance Tuning

| Profile | block_size | block_count | timeout_ms | Notes |
|---------|-----------|-------------|------------|-------|
| High throughput | 4 MiB | 128–256 | 60 | + `FanoutMode::Cpu` + thread pinning |
| Low latency | 256 KiB | 64 | 1–10 | + `busy_poll_us(50).prefer_busy_poll(true).busy_poll_budget(64)` (kernel ≥ 5.11) |
| Memory-constrained | 1 MiB | 16 | 100 | 16 MiB total ring |
| Jumbo frames | 4 MiB | 64 | 60 | `frame_size(65536)` |

See [docs/TUNING_GUIDE.md](docs/TUNING_GUIDE.md) for detailed tuning advice.

## Fanout Modes

Distribute packets across multiple sockets for multi-threaded capture:

| Mode | Strategy |
|------|----------|
| `Hash` | Flow hash (same flow → same socket) |
| `Cpu` | Route to CPU that received the NIC interrupt |
| `LoadBalance` | Round-robin |
| `Rollover` | Fill one socket, overflow to next |
| `Random` | Random distribution |
| `QueueMapping` | NIC hardware queue mapping |

```rust,no_run
use netring::{Capture, FanoutMode, FanoutFlags};

let cap = Capture::builder()
    .interface("eth0")
    .fanout(FanoutMode::Cpu, 42)
    .fanout_flags(FanoutFlags::ROLLOVER | FanoutFlags::DEFRAG)
    .build()
    .unwrap();
```

## Statistics

```rust,no_run
# let cap = netring::Capture::open("lo").unwrap();
let stats = cap.stats().unwrap();
println!("received: {}, dropped: {}, frozen: {}",
    stats.packets, stats.drops, stats.freeze_count);
```

Reading stats resets the kernel counters — call periodically for rate calculation.

## System Requirements

- **Linux** kernel 3.2+ (for TPACKET_V3), 5.4+ (for AF_XDP)
- **Rust** 1.95+ (edition 2024)

### Capabilities

| Capability | Required For |
|------------|-------------|
| `CAP_NET_RAW` | Creating AF_PACKET / AF_XDP sockets |
| `CAP_IPC_LOCK` | `MAP_LOCKED` (or sufficient `RLIMIT_MEMLOCK`) |
| `CAP_NET_ADMIN` | Promiscuous mode |

```bash
# Recommended: use justfile (sudo only once for setcap)
just setcap          # grants CAP_NET_RAW on all binaries
just test            # runs without sudo
just capture eth0    # runs without sudo

# Manual alternative
sudo setcap cap_net_raw+ep target/release/examples/capture
```

## Examples

```bash
just setcap                  # grant capabilities once (needs sudo)
just capture eth0            # basic packet capture
just batch eth0              # low-level batch API with sequence gap detection
just fanout eth0 4           # multi-threaded fanout capture
just inject lo               # packet injection
just stats eth0              # live statistics monitor (pkt/s, drops)
just low-latency eth0        # low-latency tuning demo
just dpi eth0                # deep packet inspection (HTTP/TLS/DNS/SSH detection)
just channel eth0            # channel adapter (runtime-agnostic)
just async eth0              # async capture with tokio (readable() pattern)
just async-stream eth0       # async capture as a futures::Stream
just async-inject lo 1000    # async TX with backpressure (AsyncInjector)
just async-signal eth0       # async capture with Ctrl-C graceful shutdown
just async-pipeline eth0 4   # async capture → tokio::mpsc → 4 worker tasks
just async-bridge eth0 eth1  # async transparent bridge (Bridge::run_async)
just ebpf                    # eBPF/aya integration demo (AsFd verification)
cargo run --example xdp_send --features af-xdp -- lo  # AF_XDP TX-only (uses XdpMode::Tx)

# 0.13.0 — stream observability, BPF ergonomics, multi-source, offline replay:
cargo run --example async_flow_with_tap   --features "tokio,flow,parse,pcap" -- eth0 out.pcap
cargo run --example async_filter          --features "tokio,flow,parse" -- eth0 80
cargo run --example async_fanout_workers  --features "tokio,flow,parse" -- eth0 4
cargo run --example async_multi_interface --features "tokio,flow,parse" -- lo eth0
cargo run --example async_pcap_replay     --features "tokio,flow,parse,pcap" -- trace.pcap 1.0
# 0.13.1 — async sibling of stats_monitor (StreamCapture::capture_stats demo):
cargo run --example async_stats_monitor   --features "tokio,flow,parse" -- eth0 30
# 0.14.0 — flowscope 0.4 ergonomics: one-step pcap-to-sessions + on_tick parsers:
cargo run --example async_pcap_sessions   --features "tokio,flow,parse,pcap" -- trace.pcap
cargo run --example async_on_tick         --features "tokio,flow,parse" -- lo 30
# 0.15.0+ — real-life L7 monitors using flowscope's HTTP / DNS parsers:
cargo run --example multi_protocol_monitor --features "tokio,flow,parse"  -- eth0 30
cargo run --example http_session           --features "tokio,http"        -- eth0 60
cargo run --example dns_lookups            --features "tokio,dns"         -- eth0 60
cargo run --example full_monitor           --features "tokio,http,dns"    -- eth0 60
```

Examples are organized by topic under
[`examples/`](examples/README.md) — `basic/`, `async_basics/`,
`filter/`, `scaling/`, `xdp/`, `flow/`, `l7/`, `pcap/`. See
[`examples/README.md`](examples/README.md) for a per-category
index with the right `--features` flags.

## Documentation

- [Scaling capture across cores]docs/scaling.md`FanoutMode`
  decision matrix, multi-worker recipe, anti-patterns (added 0.13.0)
- [Architecture]docs/ARCHITECTURE.md — system design, lifetime model, ring layout
- [API Overview]docs/API_OVERVIEW.md — all types, methods, and configuration
- [Tuning Guide]docs/TUNING_GUIDE.md — performance profiles, system tuning, monitoring
- [Troubleshooting]docs/TROUBLESHOOTING.md — common errors and fixes

## License

Licensed under either of [Apache License, Version 2.0](LICENSE-APACHE)
or [MIT License](LICENSE-MIT) at your option.