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
use super::*;
use std::time::Duration;
fn generation_advanced_after_subscribe(baseline: u64, post_subscribe: u64) -> bool {
post_subscribe > baseline
}
/// Opaque, detach-on-drop handle to the background loop spawned by
/// [`Store::react_loop`].
///
/// This legacy reactor is *fire-and-detach*: the loop runs the
/// subscribe→react→append cycle for the life of the process and has no
/// cooperative stop signal. The background worker also holds its own
/// [`Arc<Store>`](std::sync::Arc) clone (it appends reactions through it), so
/// the loop cannot be joined to completion from the outside — dropping every
/// *other* `Arc<Store>` does not break that self-reference. The handle is
/// therefore deliberately join-free: it exists only to keep the concrete
/// background-thread type out of the public API and to make the detach explicit
/// at the call site (drop = detach).
///
/// When you need cooperative shutdown — a `stop()`/`join()` lifecycle, restart
/// policy, decode-error surfacing, or at-least-once checkpointing — use the
/// typed reactors ([`Store::react_loop_typed`], [`Store::react_loop_multi`],
/// [`Store::react_loop_multi_raw`]) and their
/// [`TypedReactorHandle`](crate::store::TypedReactorHandle) instead.
#[must_use = "dropping a ReactLoopHandle detaches (does not stop) the background reactor loop"]
pub struct ReactLoopHandle {
_inner: std::thread::JoinHandle<()>,
}
impl std::fmt::Debug for ReactLoopHandle {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_struct("ReactLoopHandle").finish_non_exhaustive()
}
}
impl Store<Open> {
/// SUBSCRIBE: push-based, lossy.
pub fn subscribe_lossy(&self, region: &Region) -> Subscription {
// justifies: INV-TYPESTATE-OPEN-HAS-WRITER; Store<Open> typestate guarantees writer presence at
// construction (see Store::open_with_cache in src/store/lifecycle.rs — it fails the open
// instead of yielding Store<Open> if the writer cannot be spawned).
// The expect here documents an invariant, it does not recover from
// one: observing None means the store is mid-drop and every public
// path through Store<Open> is already invalid.
let rx = self
.writer_ref()
.subscribers
.subscribe_with_region(self.config.broadcast_capacity, region.clone());
Subscription::new(rx, region.clone())
}
/// REACT: spawn a background thread running the subscribe→react→append loop.
/// Returns an opaque [`ReactLoopHandle`]. The loop runs until the store is
/// dropped (subscription closes).
///
/// # Errors
/// Returns `StoreError::Io` if the background thread cannot be spawned.
pub fn react_loop<R>(
self: &Arc<Self>,
region: &Region,
reactor: R,
) -> Result<ReactLoopHandle, StoreError>
where
R: crate::event::sourcing::Reactive<serde_json::Value> + Send + 'static,
{
let store = Arc::clone(self);
let region = region.clone();
let sub = self
.writer_ref()
.reactor_subscribers
.subscribe_with_region(self.config.broadcast_capacity, region.clone());
std::thread::Builder::new()
.name("batpak-reactor".into())
.spawn(move || {
use crate::id::EntityIdType;
while let Ok(envelope) = sub.recv() {
let notif = envelope.notification;
for (coord, kind, payload) in reactor.react(&envelope.stored.event) {
if let Err(e) = store.append_reaction(
&coord,
kind,
&payload,
crate::id::CorrelationId::from(notif.correlation_id),
crate::id::CausationId::from(notif.event_id.as_u128()),
) {
tracing::warn!("react_loop: failed to append reaction: {e}");
}
}
}
})
.map(|_inner| ReactLoopHandle { _inner })
.map_err(StoreError::Io)
}
/// WATCH: reactive projection subscription. Returns a `ProjectionWatcher`
/// that re-projects `T` when new events arrive for `entity`.
///
/// Internally subscribes to entity events, then re-projects on each notification.
/// The watcher is pull-based: the caller drives the loop via
/// [`ProjectionWatcher::recv`], which returns
/// `Result<(u64, Option<T>), WatcherError>` — see the method docs for the
/// full three-way return taxonomy (materialized state, empty fold, store
/// closed, or watcher closure after the lossy/prunable subscription is
/// dropped). The returned generation is persisted honestly: redundant
/// wakeups for an already-delivered generation are suppressed, but an
/// append that advances the entity generation can still yield the same
/// folded state if `T::relevant_event_kinds()` filters it out.
///
/// Requires `Arc<Store>` because the watcher outlives the borrow.
pub fn watch_projection<T>(
self: &Arc<Self>,
entity: &str,
freshness: Freshness,
) -> ProjectionWatcher<T>
where
T: EventSourced + serde::Serialize + serde::de::DeserializeOwned + Send + 'static,
{
let baseline_generation = self.entity_generation(entity).unwrap_or(0);
let sub = self.subscribe_lossy(&Region::entity(entity));
let post_subscribe_generation = self.entity_generation(entity).unwrap_or(0);
let store = Arc::clone(self);
let entity_owned = entity.to_owned();
ProjectionWatcher::new(
sub,
store,
entity_owned,
freshness,
baseline_generation,
generation_advanced_after_subscribe(baseline_generation, post_subscribe_generation),
)
}
/// WATCH: cursor-backed reactive projection subscription.
///
/// This returns a guaranteed-delivery watcher over an ordered cursor
/// instead of a lossy subscription. A cursor-backed watcher cannot be
/// pruned under fanout backpressure; its `recv()` method therefore
/// returns `CursorWatcherError`, which has no subscription-pruned branch.
///
/// When `checkpoint_id` is provided, the cursor resumes from its durable
/// checkpoint. Without a checkpoint it starts from the beginning of the
/// current in-memory index, so the first `recv()` can materialize already
/// committed entity state instead of relying on a startup catch-up probe.
///
/// # Errors
/// Returns [`StoreError`] when the checkpoint-bound cursor cannot be
/// constructed, for example because an existing checkpoint belongs to a
/// different region or is corrupt.
pub fn watch_projection_with_cursor<T>(
self: &Arc<Self>,
entity: &str,
freshness: Freshness,
checkpoint_id: Option<CheckpointId>,
) -> Result<ProjectionWatcher<T, Cursor>, StoreError>
where
T: EventSourced + serde::Serialize + serde::de::DeserializeOwned + Send + 'static,
{
let region = Region::entity(entity);
let cursor = match checkpoint_id {
Some(id) => Cursor::new_with_checkpoint(
region,
Arc::clone(&self.index),
&self.config.data_dir,
&id,
)?,
None => self.cursor_guaranteed(®ion),
};
Ok(ProjectionWatcher::new_cursor(
cursor,
Arc::clone(self),
entity.to_owned(),
freshness,
0,
Duration::from_millis(10),
))
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn generation_advanced_after_subscribe_is_strictly_forward() {
assert!(
!generation_advanced_after_subscribe(7, 7),
"PROPERTY: equal baseline/post-subscribe generations must not trigger an initial watcher catch-up"
);
assert!(
generation_advanced_after_subscribe(7, 8),
"PROPERTY: a post-subscribe generation above baseline must trigger the initial watcher catch-up"
);
assert!(
!generation_advanced_after_subscribe(8, 7),
"PROPERTY: older post-subscribe observations must never trigger an initial watcher catch-up"
);
}
}