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
use std::collections::HashSet;
use actix::{AsyncContext, WrapFuture};
use calimero_primitives::context::ContextId;
use tracing::{debug, error, info, warn};
use crate::NodeManager;
/// Cap on per-child `warn!` events emitted from the divergence dump.
/// A wide context (e.g. UnorderedMap with hundreds of entries) could
/// otherwise produce a log burst that overwhelms aggregation
/// pipelines on every divergence tick. The summary row below the
/// loop reports the full count regardless.
const MAX_DUMP_CHILDREN: usize = 64;
/// How many consecutive heartbeats the SAME same-DAG/different-root divergence
/// must survive before it is escalated from `warn!` to `error!` (and an active
/// recovery sync is kicked). Heartbeats fire every `HASH_HEARTBEAT_FREQUENCY_S`
/// (30s), so `2` means a divergence must persist *unchanged* for ≥1 full
/// interval (~30–60s) — long enough that the background periodic sync has had a
/// chance to heal it — before it's treated as a genuinely stuck split-brain.
/// A first observation, or one whose hashes are still moving (sync making
/// progress), stays at `warn!` and never trips a log-scanning CI gate.
const DIVERGENCE_PERSIST_THRESHOLD: u32 = 2;
pub(super) fn handle_hash_heartbeat(
manager: &mut NodeManager,
ctx: &mut actix::Context<NodeManager>,
source: libp2p::PeerId,
context_id: ContextId,
their_root_hash: calimero_primitives::hash::Hash,
their_dag_heads: Vec<[u8; 32]>,
) {
let context_client = manager.clients.context.clone();
if let Ok(Some(our_context)) = context_client.get_context(&context_id) {
let our_heads_set: HashSet<_> = our_context.dag_heads.iter().collect();
let their_heads_set: HashSet<_> = their_dag_heads.iter().collect();
if our_heads_set == their_heads_set && our_context.root_hash != their_root_hash {
// #2319 — before reporting a real divergence, reconcile the
// cached `ContextMeta.root_hash` with the live index. The
// cache is populated from the WASM-returned root_hash at the
// end of each method execution; a concurrent sync apply
// (HashComparison `EntityPush`, level-sync leaf push, etc.)
// can advance the actual index right after WASM returns but
// before the cache is written, leaving the cache pointing
// at a pre-recalc full_hash while the index has already
// moved on. The two nodes still converge in storage, but
// the heartbeat sees the stale caches and fires
// DIVERGENCE.
//
// Re-read the live full_hash from the index. If it matches
// either the peer's hash or our prior cache, refresh the
// cache and skip the false-positive divergence event; only
// a *post-refresh* hash mismatch is a real divergence.
//
// Residual TOCTOU: a concurrent apply landing between
// `compute_root_hash` and `force_root_hash` can leave the
// cache transiently stale again. This doesn't eliminate
// the false-positive class — it narrows the window from
// "until the next WASM execution" to "until the next
// heartbeat tick" (also when the next ContextMeta write
// happens). The next heartbeat self-heals via this same
// branch, so we accept the residual window rather than
// double-reading.
let our_hash = match context_client.compute_root_hash(&context_id) {
Ok(live) => {
let live_hash = calimero_primitives::hash::Hash::from(live);
if live_hash != our_context.root_hash {
debug!(
%context_id,
?source,
cached_hash = ?our_context.root_hash,
live_hash = ?live_hash,
"Heartbeat divergence: cache stale vs live index, refreshing"
);
if let Err(err) = context_client.force_root_hash(&context_id, live_hash) {
// `warn`, not `debug`: a single failure here
// is usually a benign concurrent context
// delete, but a *persistent* failure leaves
// the cache stale and produces a stream of
// false-positive DIVERGENCE events on every
// heartbeat. Surface it where ops can see it
// without log filtering.
warn!(
%context_id,
?source,
%err,
"Heartbeat divergence: failed to refresh cached root hash"
);
}
if live_hash == their_root_hash {
// Converged once the cache caught up — drop any
// streak so a later transient starts fresh at WARN.
let _ = manager.divergence_streak.remove(&(context_id, source));
return;
}
}
// Use live_hash as the authoritative "our hash" — the
// cached value may be stale (and the refresh above
// may itself have failed); the live index is the
// truth we want in the divergence log for triage.
live_hash
}
Err(err) => {
debug!(
%context_id,
?source,
%err,
"Heartbeat divergence: failed to read live root hash; using cache"
);
our_context.root_hash
}
};
// #2319: surface divergence as a metric (`sync_root_hash_divergence_detected_total`)
// so vmagent can alert on the rate without grepping logs —
// with the determinism fixes this should stay near zero. Counted on
// EVERY observation, independent of the log-level gating below, so
// the rate stays faithful.
let _new = manager.divergence_detected.inc();
// Persistence gate: a divergence is only escalated to ERROR once the
// SAME (our, their) hash pair has survived
// `DIVERGENCE_PERSIST_THRESHOLD` consecutive heartbeats. A first
// observation — or one whose hashes are still moving, i.e. sync is
// making progress — is almost always a concurrent sync apply that
// landed mid-flight and self-heals next tick; logging that at ERROR
// turns a benign, recoverable event into a hard failure for
// log-scanning CI on unrelated work. The streak resets on any hash
// change (progress) and is cleared on convergence.
let count = match manager.divergence_streak.get(&(context_id, source)) {
Some(prev) if prev.our_hash == our_hash && prev.their_hash == their_root_hash => {
prev.count.saturating_add(1)
}
_ => 1,
};
let _ = manager.divergence_streak.insert(
(context_id, source),
crate::manager::DivergenceMark {
our_hash,
their_hash: their_root_hash,
count,
},
);
if count < DIVERGENCE_PERSIST_THRESHOLD {
warn!(
%context_id,
?source,
our_hash = ?our_hash,
their_hash = ?their_root_hash,
count,
"Divergence detected (same DAG heads, different root) — transient: a \
sync apply is likely in flight; expecting periodic sync to reconcile"
);
return;
}
// Persisted unchanged across >= DIVERGENCE_PERSIST_THRESHOLD
// heartbeats: background sync has NOT healed it — a genuinely stuck
// split-brain worth alarming and the triage dump below.
error!(
%context_id,
?source,
our_hash = ?our_hash,
their_hash = ?their_root_hash,
count,
dag_heads = ?their_dag_heads,
"DIVERGENCE DETECTED: Same DAG heads but different root hash (persisted across heartbeats)!"
);
// #2319 triage aid — dump ROOT's self summary + children
// list so a future flake can be triaged by diffing the two
// peers' dumps. Without this, the only observable signal
// is the two opaque root hashes and the remaining
// investigation requires re-running with more logging.
// Bounded by the heartbeat cadence (one DIVERGENCE event
// per peer per heartbeat) and by MAX_DUMP_CHILDREN.
//
// Self summary is logged before children so the diff order
// matches the analysis flow: identical children +
// different own_hash points at ROOT-entity write-path
// divergence (the pattern we saw on PR #2472 attempt 1,
// all 20 children matched).
match context_client.dump_root(&context_id) {
Ok(Some((self_dump, children))) => {
warn!(
target: "sync::divergence_dump",
%context_id,
?source,
root_own_hash = %hex::encode(self_dump.own_hash),
root_full_hash = %hex::encode(self_dump.full_hash),
root_entry_bytes_hash = ?self_dump.entry_bytes_hash.map(hex::encode),
root_entry_bytes_len = self_dump.entry_bytes_len,
children_count = self_dump.children_count,
"DIVERGENCE DUMP: ROOT self"
);
let child_count = children.len();
// Emit one event per child so log search/filter
// tools can group by `entity_id`. Cap the per-child
// emission at MAX_DUMP_CHILDREN — the summary row
// below reports the full count regardless.
for c in children.iter().take(MAX_DUMP_CHILDREN) {
warn!(
target: "sync::divergence_dump",
%context_id,
?source,
entity_id = %hex::encode(c.id),
merkle_hash = %hex::encode(c.merkle_hash),
created_at = c.created_at,
updated_at = c.updated_at,
crdt_type = ?c.crdt_type,
field_name = ?c.field_name,
"DIVERGENCE DUMP: ROOT child entry"
);
}
if child_count > MAX_DUMP_CHILDREN {
warn!(
target: "sync::divergence_dump",
%context_id,
?source,
emitted = MAX_DUMP_CHILDREN,
total = child_count,
"DIVERGENCE DUMP: ROOT children truncated"
);
}
warn!(
target: "sync::divergence_dump",
%context_id,
?source,
child_count,
"DIVERGENCE DUMP: ROOT children list emitted"
);
}
Ok(None) => {
warn!(
target: "sync::divergence_dump",
%context_id,
?source,
"DIVERGENCE DUMP: ROOT — no index entry"
);
}
Err(e) => {
warn!(
target: "sync::divergence_dump",
%context_id,
?source,
error = %e,
"DIVERGENCE DUMP: failed to read ROOT"
);
}
}
// Active recovery on the FIRST escalation: kick a sync now rather
// than waiting for the next periodic tick. We can't trigger the
// signed anchor-pull (`reconcile_after_divergence`) from here — the
// peer's gossiped root hash is unauthenticated, so pulling canonical
// state off it would be a poisoning vector — but a plain HC sync is
// the same path periodic sync uses, just sooner. Only on the exact
// threshold tick so a still-stuck divergence keeps ERRORing without
// re-spawning a sync every heartbeat.
if count == DIVERGENCE_PERSIST_THRESHOLD {
let node_client = manager.clients.node.clone();
let _ignored = ctx.spawn(
async move {
if let Err(e) = node_client.sync(Some(&context_id), None).await {
warn!(%context_id, ?e, "Persistent-divergence recovery sync failed to start");
}
}
.into_actor(manager),
);
}
return;
}
// Reaching here means we are NOT in a same-DAG / different-root
// divergence with this peer (converged, or a plain behind/ahead head
// difference handled below) — drop any stale streak so a future
// transient starts fresh at WARN rather than inheriting an old count.
let _ = manager.divergence_streak.remove(&(context_id, source));
if our_context.root_hash != their_root_hash {
let heads_we_dont_have: Vec<_> = their_heads_set.difference(&our_heads_set).collect();
if heads_we_dont_have.is_empty() {
debug!(
%context_id,
?source,
our_heads_count = our_context.dag_heads.len(),
their_heads_count = their_dag_heads.len(),
"Different root hash (peer is behind or concurrent updates)"
);
return;
}
info!(
%context_id,
?source,
our_heads_count = our_context.dag_heads.len(),
their_heads_count = their_dag_heads.len(),
missing_count = heads_we_dont_have.len(),
"Peer has DAG heads we don't have - triggering sync"
);
let node_client = manager.clients.node.clone();
let _ignored = ctx.spawn(
async move {
if let Err(e) = node_client.sync(Some(&context_id), None).await {
warn!(%context_id, ?e, "Failed to trigger sync from heartbeat");
}
}
.into_actor(manager),
);
}
}
}