quepaxa 0.3.3

QuePaxa consensus for state machine replication: a sans-IO, no_std replicated log that, unlike Raft or Multi-Paxos, never stalls on election timeouts
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
//! Replica unit tests that need crate-private wire and state types.

use super::*;
use crate::codec::reply_proposal;
use proptest::prelude::*;

fn replica() -> Replica {
    let cfg = Config::new([3; 16], NodeId(0), alloc::vec![NodeId(0), NodeId(1), NodeId(2)], [9; 32]);
    let mut r = Replica::new(cfg, [], Timestamp(0)).unwrap();
    let mut out = Vec::new();
    r.poll_actions(&mut out);
    r.handle(Timestamp(0), Input::Persisted(PersistId(0))).unwrap();
    r.poll_actions(&mut out);
    r
}

fn proposal(priority: u64, origin: u32, seq: u64) -> Proposal {
    Proposal {
        priority,
        key: ValueKey { origin: NodeId(origin), incarnation: 1, seq },
        value: batch::encode(&[]),
    }
}

fn arb_body() -> impl Strategy<Value = Body> {
    let p = (any::<u64>(), 0u32..4, 0u64..3).prop_map(|(pr, o, s)| proposal(pr, o, s));
    let slot = prop_oneof![0u64..4, Just(u64::MAX), any::<u64>()].prop_map(Slot);
    let step = prop_oneof![0u32..12, Just(u32::MAX), any::<u32>()];
    prop_oneof![
        (slot.clone(), step.clone(), p.clone()).prop_map(|(slot, step, proposal)| Body::RecordRequest {
            slot,
            step,
            proposal
        }),
        (slot.clone(), step.clone(), step, proptest::option::of(p.clone()), proptest::option::of(p.clone()))
            .prop_map(|(slot, req_step, step, f, a)| {
                let key = ValueKey { origin: NodeId(0), incarnation: 1, seq: 0 };
                Body::RecordReply {
                    slot,
                    req_step,
                    step,
                    first: f.map(|f| reply_proposal(&f, key)),
                    prev_agg: a.map(|a| reply_proposal(&a, key)),
                }
            }),
        (slot.clone(), p.clone()).prop_map(|(slot, proposal)| Body::Decided { slot, proposal }),
        slot.clone().prop_map(|from| Body::CatchUp { from }),
        (slot.clone(), proptest::collection::vec(p, 0..3))
            .prop_map(|(first, proposals)| Body::Entries { first, proposals }),
        slot.prop_map(|through| Body::Compacted { through }),
    ]
}

#[test]
fn lost_catch_up_request_is_retried_on_the_timer() {
    let mut r = replica();
    let mut out = Vec::new();
    let catch_ups = |out: &[Action]| {
        out.iter()
            .filter(|a| match a {
                Action::Send(o) => {
                    matches!(
                        Message::decode(&o.bytes, &r_wire()),
                        Ok(Message { body: Body::CatchUp { .. }, .. })
                    )
                }
                _ => false,
            })
            .count()
    };
    // A peer advertises ten decided slots; the request it triggers is lost.
    let env =
        Envelope { cluster_id: [3; 16], config_epoch: 0, from: NodeId(1), to: NodeId(0), commit_index: 10 };
    let msg = Message { env, body: Body::CatchUp { from: Slot(0) } }.encode();
    r.handle(Timestamp(1), Input::Message(msg)).unwrap();
    r.poll_actions(&mut out);
    assert_eq!(catch_ups(&out), 1);
    out.clear();
    let deadline = r.next_deadline().expect("a retry is scheduled");
    r.handle_timeout(deadline).unwrap();
    r.poll_actions(&mut out);
    assert_eq!(catch_ups(&out), 1, "{out:?}");
    assert_eq!(r.metrics().catch_up_retries, 1);
}

#[test]
fn misaddressed_and_out_of_range_messages_are_rejected_at_the_exact_bounds() {
    let req = |slot: u64, step| Body::RecordRequest { slot: Slot(slot), step, proposal: proposal(1, 1, 0) };
    let reply =
        |req_step, step| Body::RecordReply { slot: Slot(0), req_step, step, first: None, prev_agg: None };
    // The documented limits, spelled out so a changed constant fails here.
    let (max_step, max_slot) = (u32::MAX - 16, u64::MAX - (1 << 32));
    let cases = [
        (1, 2, Body::CatchUp { from: Slot(0) }, true),
        (0, 0, Body::CatchUp { from: Slot(0) }, true),
        // A replica outside every known epoch: its requests are served, its replies and
        // decisions are not.
        (7, 0, Body::CatchUp { from: Slot(0) }, false),
        (7, 0, reply(4, 4), true),
        (7, 0, Body::Decided { slot: Slot(0), proposal: proposal(1, 7, 0) }, true),
        (1, 0, Body::CatchUp { from: Slot(0) }, false),
        (1, 0, req(0, max_step + 1), true),
        (1, 0, req(0, max_step), false),
        (1, 0, req(max_slot + 1, 4), true),
        (1, 0, req(max_slot, 4), false),
        (1, 0, reply(5, 4), true),
        (1, 0, reply(4, max_step + 1), true),
        (1, 0, reply(4, max_step), false),
        (1, 0, reply(4, 4), false),
        (1, 0, Body::Forward { batch: Bytes::from_static(b"not a batch") }, true),
    ];
    for (from, to, body, rejected) in cases {
        let mut r = replica();
        let env = Envelope {
            cluster_id: [3; 16],
            config_epoch: 0,
            from: NodeId(from),
            to: NodeId(to),
            commit_index: 0,
        };
        let what = alloc::format!("{from}->{to} {body:?}");
        r.handle(Timestamp(1), Input::Message(Message { env, body }.encode())).unwrap();
        assert_eq!(r.metrics().messages_rejected, u64::from(rejected), "{what}");
    }
}

#[test]
fn a_decided_value_that_is_not_a_batch_commits_nothing_and_is_counted() {
    let mut r = replica();
    let env =
        Envelope { cluster_id: [3; 16], config_epoch: 0, from: NodeId(1), to: NodeId(0), commit_index: 0 };
    let proposal = Proposal { value: Bytes::from_static(b"garbage"), ..proposal(1, 1, 0) };
    let body = Body::Decided { slot: Slot(0), proposal };
    r.handle(Timestamp(1), Input::Message(Message { env, body }.encode())).unwrap();
    let mut out = Vec::new();
    r.poll_actions(&mut out);
    assert!(out.iter().any(|a| matches!(a, Action::Commit { slot: Slot(0), entries } if entries.is_empty())));
    assert_eq!(r.metrics().corrupt_batches, 1);
}

#[test]
fn decisions_are_learned_up_to_the_last_valid_slot() {
    for (slot, learned) in [(0, 1), (MAX_SLOT, 1), (MAX_SLOT + 1, 0)] {
        let mut r = replica();
        let env = Envelope {
            cluster_id: [3; 16],
            config_epoch: 0,
            from: NodeId(1),
            to: NodeId(0),
            commit_index: 0,
        };
        let body = Body::Decided { slot: Slot(slot), proposal: proposal(1, 1, 0) };
        r.handle(Timestamp(1), Input::Message(Message { env, body }.encode())).unwrap();
        assert_eq!(r.metrics().slots_decided, learned, "slot {slot}");
    }
}

#[test]
fn snapshots_may_cover_slots_up_to_the_last_valid_one() {
    let members = [NodeId(0), NodeId(1), NodeId(2)];
    let state = |next| {
        CoreState {
            next,
            leaders: members.to_vec(),
            schedule: Schedule::new(&members),
            dedup: Dedup::default(),
        }
        .encode()
    };
    assert_eq!(replica().install_snapshot(Timestamp(1), Slot(MAX_SLOT - 1), state(MAX_SLOT)), Ok(true));
    assert_eq!(
        replica().install_snapshot(Timestamp(1), Slot(MAX_SLOT), state(MAX_SLOT + 1)),
        Err(Error::InvalidCoreState)
    );
}

#[test]
fn requests_below_the_floor_or_beyond_the_window_get_no_vote() {
    let members = [NodeId(0), NodeId(1), NodeId(2)];
    let state = CoreState {
        next: 10,
        leaders: members.to_vec(),
        schedule: Schedule::new(&members),
        dedup: Dedup::default(),
    }
    .encode();
    let replies = |slot: u64| {
        let mut r = replica();
        assert_eq!(r.install_snapshot(Timestamp(1), Slot(9), state.clone()), Ok(true));
        let mut out = Vec::new();
        r.poll_actions(&mut out);
        let env = Envelope {
            cluster_id: [3; 16],
            config_epoch: 0,
            from: NodeId(1),
            to: NodeId(0),
            commit_index: 0,
        };
        let body = Body::RecordRequest { slot: Slot(slot), step: 4, proposal: proposal(1, 1, 0) };
        r.handle(Timestamp(2), Input::Message(Message { env, body }.encode())).unwrap();
        out.clear();
        r.poll_actions(&mut out);
        let window = 4 * r.window();
        let bodies: Vec<Body> = out
            .iter()
            .flat_map(|a| match a {
                Action::Send(o) => alloc::vec![o.clone()],
                Action::Persist { then, .. } => then.clone(),
                _ => Vec::new(),
            })
            .filter_map(|o| Message::decode(&o.bytes, &r_wire()).ok().map(|m| m.body))
            .collect();
        (bodies, window)
    };
    let (below, window) = replies(3);
    assert!(matches!(below[..], [Body::Compacted { through: Slot(9) }]), "{below:?}");
    assert!(matches!(replies(10 + window - 1).0[..], [Body::RecordReply { .. }]));
    assert!(replies(10 + window).0.is_empty());
}

fn r_wire() -> WireCtx {
    WireCtx { cluster_id: [3; 16], config_epoch: 0, max_message_bytes: 16 << 20, max_value_bytes: 1 << 20 }
}

#[test]
fn decisions_are_applied_before_their_local_record_is_durable() {
    let mut r = replica();
    let mut out = Vec::new();
    // Ten decisions not originated by the leader (node 0) rotate it. They are applied at
    // once (a majority already holds each durably), and core state covers them.
    for slot in 0..10 {
        let env = Envelope {
            cluster_id: [3; 16],
            config_epoch: 0,
            from: NodeId(1),
            to: NodeId(0),
            commit_index: 0,
        };
        let body = Body::Decided { slot: Slot(slot), proposal: proposal(1, 1, slot) };
        r.handle(Timestamp(1), Input::Message(Message { env, body }.encode())).unwrap();
        r.poll_actions(&mut out);
    }
    assert_eq!(r.emitted_index(), 10);
    assert_eq!(out.iter().filter(|a| matches!(a, Action::Commit { .. })).count(), 10);
    let state = CoreState::decode(&r.core_state(), &r.cfg.members).unwrap();
    assert_eq!(state.schedule, r.schedule);
    assert_ne!(r.schedule, Schedule::new(&[NodeId(0), NodeId(1), NodeId(2)]));
}

proptest! {
    // Well-formed messages with adversarial field values from members never panic or
    // violate the API contract.
    #[test]
    fn arbitrary_member_messages_are_handled(msgs in proptest::collection::vec((1u32..3, arb_body()), 1..40)) {
        let mut r = replica();
        r.propose(Timestamp(1), Bytes::from_static(b"x")).unwrap();
        let mut out = Vec::new();
        r.poll_actions(&mut out);
        for (i, (from, body)) in msgs.into_iter().enumerate() {
            let env = Envelope { cluster_id: [3; 16], config_epoch: 0, from: NodeId(from), to: NodeId(0), commit_index: i as u64 };
            let bytes = Message { env, body }.encode();
            prop_assert!(r.handle(Timestamp(2 + i as u64), Input::Message(bytes)).is_ok());
            r.poll_actions(&mut out);
            let _ = r.handle_timeout(Timestamp(1_000_000_000 * (i as u64 + 1)));
            r.poll_actions(&mut out);
        }
    }
}

#[test]
fn one_local_change_at_a_time_even_before_it_is_decided() {
    let mut r = replica();
    let members = || alloc::vec![NodeId(0), NodeId(1), NodeId(2)];
    assert!(r.propose_reconfiguration(Timestamp(1), members(), Vec::new()).is_ok());
    let mut out = Vec::new();
    r.poll_actions(&mut out);
    assert_eq!(
        r.propose_reconfiguration(Timestamp(1), members(), Vec::new()),
        Err(Error::ReconfigurationPending)
    );
}

#[test]
fn membership_changes_are_bounded_in_size() {
    let many = |n: usize| (0..n as u32).map(NodeId).collect::<Vec<_>>();
    let mut r = replica();
    let too_many =
        r.propose_reconfiguration(Timestamp(1), many(crate::schedule::MAX_MEMBERS + 1), Vec::new());
    assert!(matches!(too_many, Err(Error::InvalidConfig(_))));
    let mut out = Vec::new();
    r.poll_actions(&mut out);
    assert!(r.propose_reconfiguration(Timestamp(1), many(crate::schedule::MAX_MEMBERS), Vec::new()).is_ok());
}

#[test]
fn a_snapshot_is_asked_for_again_only_after_the_hedge_cap() {
    let mut r = replica();
    let hedge_max = r.cfg.hedge_max;
    let asks = |r: &mut Replica, at: u64| {
        let env = Envelope {
            cluster_id: [3; 16],
            config_epoch: 0,
            from: NodeId(1),
            to: NodeId(0),
            commit_index: 0,
        };
        let body = Body::Compacted { through: Slot(9) };
        r.handle(Timestamp(at), Input::Message(Message { env, body }.encode())).unwrap();
        let mut out = Vec::new();
        r.poll_actions(&mut out);
        out.iter().filter(|a| matches!(a, Action::SnapshotNeeded { through: Slot(9), .. })).count()
    };
    assert_eq!(asks(&mut r, 1), 1);
    assert_eq!(asks(&mut r, 2), 0, "the first ask may still be answered");
    assert_eq!(asks(&mut r, 1 + hedge_max), 1, "a lost snapshot is asked for again");
}

#[test]
fn catch_up_replies_wait_for_durability_and_are_sent_even_when_empty() {
    let mut r = replica();
    let mut out = Vec::new();
    let msg = |body| {
        let env = Envelope {
            cluster_id: [3; 16],
            config_epoch: 0,
            from: NodeId(1),
            to: NodeId(0),
            commit_index: 0,
        };
        Input::Message(Message { env, body }.encode())
    };
    let entries = |out: &[Action]| -> Vec<(u64, usize)> {
        out.iter()
            .flat_map(|a| match a {
                Action::Send(o) => alloc::vec![o],
                Action::Persist { then, .. } => then.iter().collect(),
                _ => Vec::new(),
            })
            .filter_map(|o| match Message::decode(&o.bytes, &r_wire()) {
                Ok(Message { env, body: Body::Entries { proposals, .. } }) => {
                    Some((env.commit_index, proposals.len()))
                }
                _ => None,
            })
            .collect()
    };
    // Nothing decided: the reply still goes out, reporting no progress.
    r.handle(Timestamp(1), msg(Body::CatchUp { from: Slot(0) })).unwrap();
    r.poll_actions(&mut out);
    assert_eq!(entries(&out), [(0, 0)]);
    out.clear();
    // Slot 0 is decided but its record is not durable yet: a reply reporting it waits for
    // that write.
    r.handle(Timestamp(1), msg(Body::Decided { slot: Slot(0), proposal: proposal(1, 1, 0) })).unwrap();
    r.poll_actions(&mut out);
    let persist = out.iter().find_map(|a| match a {
        Action::Persist { id, .. } => Some(*id),
        _ => None,
    });
    out.clear();
    r.handle(Timestamp(1), msg(Body::CatchUp { from: Slot(0) })).unwrap();
    r.poll_actions(&mut out);
    assert!(entries(&out).is_empty(), "{out:?}");
    r.handle(Timestamp(1), Input::Persisted(persist.expect("the decision is written"))).unwrap();
    r.poll_actions(&mut out);
    assert_eq!(entries(&out), [(1, 1)], "{out:?}");
}

#[test]
fn an_empty_catch_up_page_is_retried_only_on_the_timer() {
    let mut r = replica();
    let mut out = Vec::new();
    let from_peer = |commit_index, body| {
        let env =
            Envelope { cluster_id: [3; 16], config_epoch: 0, from: NodeId(1), to: NodeId(0), commit_index };
        Input::Message(Message { env, body }.encode())
    };
    // A peer advertises ten slots; the catch-up request comes back empty.
    r.handle(Timestamp(1), from_peer(10, Body::CatchUp { from: Slot(0) })).unwrap();
    r.poll_actions(&mut out);
    r.handle(Timestamp(2), from_peer(10, Body::Entries { first: Slot(0), proposals: Vec::new() })).unwrap();
    r.poll_actions(&mut out);
    let retry = r.next_deadline().expect("a retry is scheduled");
    assert!(retry >= Timestamp(2).after(r.cfg.retransmit_interval), "{retry:?}");
}

/// Replicas 0-2 are configured, 3 and 4 join, and the members change to `new`. The joiners
/// vote on the new epoch's slots but never learn a decision, so a slot the new epoch commits
/// (a retirement barrier among them) says nothing about who holds the old slots. Returns
/// the actions reported once the new epoch has committed past its first slot.
fn with_joiners_that_never_learn(new: &[u32]) -> Vec<(usize, Action)> {
    let members = alloc::vec![NodeId(0), NodeId(1), NodeId(2)];
    let mut rs: Vec<Replica> = (0..5u32)
        .map(|i| {
            let mut cfg = Config::new([5; 16], NodeId(i), members.clone(), [i as u8 + 1; 32]);
            // The widest window: recorders vote up to 256 slots past what they know, so the
            // joiners vote on the new epoch's first slots without holding the old ones.
            cfg.max_in_flight_slots = 64;
            Replica::new(cfg, [], Timestamp(0)).unwrap()
        })
        .collect();
    let wire = rs[0].wire;
    let learns = |b: &Body| matches!(b, Body::Decided { .. } | Body::Entries { .. } | Body::CatchUp { .. });
    let mut seen = Vec::new();
    let run = |rs: &mut Vec<Replica>, seen: &mut Vec<(usize, Action)>| {
        let mut queue = alloc::collections::VecDeque::new();
        loop {
            for (i, r) in rs.iter_mut().enumerate() {
                let mut out = Vec::new();
                r.poll_actions(&mut out);
                for a in out {
                    match a {
                        Action::Send(o) => queue.push_back((o.to.0 as usize, Input::Message(o.bytes))),
                        Action::Persist { id, then, .. } => {
                            queue
                                .extend(then.into_iter().map(|o| (o.to.0 as usize, Input::Message(o.bytes))));
                            queue.push_back((i, Input::Persisted(id)));
                        }
                        other => seen.push((i, other)),
                    }
                }
            }
            let Some((to, input)) = queue.pop_front() else { return };
            if to >= 3
                && let Input::Message(b) = &input
                && Message::decode(b, &wire).is_ok_and(|m| learns(&m.body))
            {
                continue; // the joiners never learn a decision
            }
            rs[to].handle(Timestamp(0), input).unwrap();
        }
    };
    run(&mut rs, &mut seen);
    rs[2]
        .propose_reconfiguration(Timestamp(0), new.iter().copied().map(NodeId).collect(), Vec::new())
        .unwrap();
    run(&mut rs, &mut seen);
    for k in 0..90u32 {
        rs[2].propose(Timestamp(0), Bytes::from(alloc::format!("w{k}"))).unwrap();
        run(&mut rs, &mut seen);
    }
    for t in 1..30u64 {
        for r in rs.iter_mut() {
            r.handle_timeout(Timestamp(t * 1_000_000_000)).unwrap();
        }
        run(&mut rs, &mut seen);
    }
    let from = seen
        .iter()
        .find_map(|(_, a)| match a {
            Action::Reconfigured { epoch: 1, from, .. } => Some(from.0),
            _ => None,
        })
        .expect("the change was applied");
    assert!(rs[2].commit_index() > from, "the new epoch commits with the joiners' votes");
    assert_eq!((rs[3].commit_index(), rs[4].commit_index()), (0, 0), "the joiners learned nothing");
    seen
}

#[test]
fn a_barrier_the_new_epoch_commits_does_not_retire_without_a_caught_up_majority() {
    let retired = |seen: Vec<(usize, Action)>| -> Vec<(usize, Action)> {
        seen.into_iter().filter(|(_, a)| matches!(a, Action::Retired { .. })).collect()
    };
    // Replicas 0 and 1 leave; of the new members only 2 holds the old slots.
    let r = retired(with_joiners_that_never_learn(&[2, 3, 4]));
    assert!(r.is_empty(), "retired with only one of three new members holding the old slots: {r:?}");
    // Replica 0 leaves; 1 and 2 hold the old slots, and half of four is not a majority.
    let r = retired(with_joiners_that_never_learn(&[1, 2, 3, 4]));
    assert!(r.is_empty(), "retired with only two of four new members holding the old slots: {r:?}");
}

#[test]
fn a_snapshot_resolving_the_retirement_barrier_still_reports_other_proposals() {
    // A snapshot that covers both an application proposal and the internal retirement
    // barrier reports the proposal and confirms (never reports) the barrier.
    let mut r = replica();
    let mut out = Vec::new();
    let app = r.propose(Timestamp(1), Bytes::from_static(b"x")).unwrap();
    let barrier = r.submit(Bytes::new(), Kind::Barrier).unwrap();
    r.retire_barrier = Some((1, barrier));
    r.poll_actions(&mut out);
    let mut dedup = Dedup::default();
    for id in [app, barrier] {
        dedup.first_time(NodeId(0), r.incarnation, id);
    }
    let members = [NodeId(0), NodeId(1), NodeId(2)];
    let state =
        CoreState { next: 10, leaders: members.to_vec(), schedule: Schedule::new(&members), dedup }.encode();
    assert_eq!(r.install_snapshot(Timestamp(2), Slot(9), state), Ok(true));
    out.clear();
    r.poll_actions(&mut out);
    assert!(out.iter().any(|a| matches!(a, Action::Committed { id, .. } if *id == app)), "{out:?}");
    assert!(!out.iter().any(|a| matches!(a, Action::ReadReady { id, .. } if *id == barrier)), "{out:?}");
    assert_eq!(r.retire_barrier, None, "the barrier was confirmed");
}