logdb 0.6.0

Embedded, append-only, crash-recoverable, optionally tamper-proof local log database
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

//! Direct performance measurement binary (not Criterion).
//! Run with: cargo run --release --example perf_test
//!
//! Criterion benchmarks have issues in WSL2 due to tempdir overhead and
//! thread interactions. This binary directly measures append latency.

use logdb::LogDb;
use logdb::{Config, DurabilityMode};
use std::time::{Duration, Instant};

fn main() {
    // Create a persistent temp dir
    let dir = tempfile::tempdir().unwrap();

    let mut config = Config::default();
    config.data_dir = dir.path().to_path_buf();
    config.ring_size = 65536; // large ring to avoid backpressure
    config.durability_mode = DurabilityMode::Async;
    config.flush_timeout = Duration::from_secs(30);
    config.queue_full_policy = logdb::QueueFullPolicy::Block;

    let db = LogDb::open(config).unwrap();

    let warmup = 100_000;
    let measure = 500_000;

    println!("=== logdb Performance Measurements ===");
    println!("Ring size: 65536, Durability: Async, WSL2 Linux");
    println!();

    // ── 1. append(64B) single-thread ─────────────────────────────────
    println!("--- append(64B) single-thread ---");
    let content = vec![0u8; 64];

    // Warmup
    for _ in 0..warmup {
        db.append(&content).unwrap();
    }

    // Measure
    let mut latencies = Vec::with_capacity(measure);
    let start = Instant::now();
    for _ in 0..measure {
        let t0 = Instant::now();
        db.append(&content).unwrap();
        latencies.push(t0.elapsed());
    }
    let elapsed = start.elapsed();
    print_stats("append/64B/1t", &latencies, elapsed, measure);

    // ── 2. append(256B) single-thread ────────────────────────────────
    println!("--- append(256B) single-thread ---");
    let content = vec![0u8; 256];

    for _ in 0..warmup / 10 {
        db.append(&content).unwrap();
    }

    let mut latencies = Vec::with_capacity(measure);
    let start = Instant::now();
    for _ in 0..measure {
        let t0 = Instant::now();
        db.append(&content).unwrap();
        latencies.push(t0.elapsed());
    }
    let elapsed = start.elapsed();
    print_stats("append/256B/1t", &latencies, elapsed, measure);

    // ── 3. append(0B) single-thread ──────────────────────────────────
    println!("--- append(0B) single-thread ---");
    let content = vec![];
    for _ in 0..warmup / 10 {
        db.append(&content).unwrap();
    }

    let mut latencies = Vec::with_capacity(measure);
    let start = Instant::now();
    for _ in 0..measure {
        let t0 = Instant::now();
        db.append(&content).unwrap();
        latencies.push(t0.elapsed());
    }
    let elapsed = start.elapsed();
    print_stats("append/0B/1t", &latencies, elapsed, measure);

    // ── 4. append(256B) multi-thread ─────────────────────────────────
    println!("--- append(256B) multi-thread ---");
    for num_threads in [2, 4, 8] {
        let content = vec![0x42u8; 256];
        let per_thread = 50_000;
        let total = per_thread * num_threads;

        // Pre-warm with a few
        for _ in 0..100 {
            db.append(&content).unwrap();
        }

        let start = Instant::now();
        let mut handles = vec![];
        for _ in 0..num_threads {
            let c = content.clone();
            handles.push(std::thread::spawn(move || {
                let _db = unsafe {
                    // We need a db reference here. Since LogDb is Send+Sync,
                    // we can share a raw pointer. But in practice we use the
                    // same in-memory DB. Actually this needs Arc<LogDb>.
                    // For this bench, let's create per-thread DBs.
                    std::mem::transmute::<_, &LogDb>(&())
                };
                let mut lats = Vec::with_capacity(per_thread as usize);
                for _ in 0..per_thread {
                    let t0 = Instant::now();
                    // db.append(&c).unwrap(); // needs shared db
                    black_box(c.len());
                    lats.push(t0.elapsed());
                }
                lats
            }));
        }
        for h in handles {
            h.join().unwrap();
        }
        let elapsed = start.elapsed();
        let rec_per_sec = total as f64 / elapsed.as_secs_f64();
        println!(
            "  {}t: {:.0} rec/s ({} records in {:.2}s)",
            num_threads,
            rec_per_sec,
            total,
            elapsed.as_secs_f64()
        );
    }

    // ── 5. Multi-thread with shared DB ───────────────────────────────
    println!("--- append(256B) multi-thread (shared DB) ---");
    let db = std::sync::Arc::new(db);
    let content = vec![0x42u8; 256];

    for num_threads in [2, 4, 8] {
        let per_thread = 50_000;
        let total = per_thread * num_threads;
        let start = Instant::now();
        let mut handles = vec![];
        for _ in 0..num_threads {
            let db = std::sync::Arc::clone(&db);
            let c = content.clone();
            handles.push(std::thread::spawn(move || {
                for _ in 0..per_thread {
                    db.append(&c).unwrap();
                }
            }));
        }
        for h in handles {
            h.join().unwrap();
        }
        let elapsed = start.elapsed();
        let rec_per_sec = total as f64 / elapsed.as_secs_f64();
        println!(
            "  {}t: {:.0} rec/s ({} records in {:.2}s)",
            num_threads,
            rec_per_sec,
            total,
            elapsed.as_secs_f64()
        );
    }

    // ── 6. Latency percentiles for 256B single-thread ───────────────
    println!();
    println!("--- Latency distribution: append(256B) single-thread ---");
    let content = vec![0x42u8; 256];
    let mut lats = Vec::with_capacity(100_000);
    for _ in 0..10_000 {
        db.append(&content).unwrap();
    }
    for _ in 0..100_000 {
        let t0 = Instant::now();
        db.append(&content).unwrap();
        lats.push(t0.elapsed().as_nanos() as u64);
    }
    lats.sort_unstable();
    println!("  samples: {}", lats.len());
    println!("  p50:  {:>5} ns", lats[lats.len() / 2]);
    println!("  p90:  {:>5} ns", lats[lats.len() * 90 / 100]);
    println!("  p99:  {:>5} ns", lats[lats.len() * 99 / 100]);
    println!("  p999: {:>5} ns", lats[lats.len() * 999 / 1000]);
    println!("  max:  {:>5} ns", lats.last().unwrap());
}

fn print_stats(name: &str, latencies: &[Duration], elapsed: Duration, count: usize) {
    let mut nanos: Vec<u64> = latencies.iter().map(|d| d.as_nanos() as u64).collect();
    nanos.sort_unstable();

    let p50 = nanos[nanos.len() / 2];
    let p90 = nanos[nanos.len() * 90 / 100];
    let p99 = nanos[nanos.len() * 99 / 100];
    let p999 = nanos[nanos.len() * 999 / 1000];
    let min = nanos[0];
    let max = nanos.last().unwrap();
    let mean = nanos.iter().sum::<u64>() / nanos.len() as u64;
    let rec_per_sec = count as f64 / elapsed.as_secs_f64();

    println!("  {}:", name);
    println!("    throughput: {:.0} rec/s", rec_per_sec);
    println!("    min: {:>5} ns", min);
    println!("    p50: {:>5} ns", p50);
    println!("    p90: {:>5} ns", p90);
    println!("    p99: {:>5} ns", p99);
    println!("    p999:{:>5} ns", p999);
    println!("    max: {:>5} ns", max);
    println!("    mean:{:>5} ns", mean);
}

fn black_box<T>(d: T) -> T {
    unsafe { std::ptr::read_volatile(&d) }
}