notedthat_api_http/
state.rs

1//! Shared application state for the axum router.
2
3use notedthat_core::{
4    AccessPolicy, Authenticator, EventPublisher, EventSource, KbDetails, KbSlug, Storage,
5};
6use notedthat_indexer::{IndexHealth, Searcher};
7
8use crate::readiness::ReadinessReceiver;
9use std::collections::BTreeMap;
10use std::sync::Arc;
11
12/// Application state shared across all axum handlers.
13///
14/// This is cloned cheaply for each request (all fields are behind [`Arc`]).
15#[derive(Clone)]
16pub struct AppState {
17    /// The backing storage implementation (injected at startup).
18    pub storage: Arc<dyn Storage>,
19    /// Canonical map of slug string → [`KbSlug`] for declared knowledge bases.
20    pub declared_kbs: Arc<BTreeMap<String, KbSlug>>,
21    /// Startup snapshot of manifest access policies, keyed by KB slug.
22    ///
23    /// One `Arc` per policy so a request-scoped authorization handle can hold
24    /// its knowledge base's policy cheaply instead of borrowing the whole map.
25    pub access_policies: Arc<BTreeMap<String, Arc<AccessPolicy>>>,
26    /// Startup snapshot of each manifest's display name and description,
27    /// keyed by KB slug: what `GET /api/v1/knowledgebases` shows (#98).
28    pub kb_details: Arc<BTreeMap<String, KbDetails>>,
29    /// The credential rules every request's principal is resolved through.
30    pub authenticator: Arc<Authenticator>,
31    /// Maximum accepted PUT body size in bytes (16 MiB in M2).
32    pub max_body_size: u64,
33    /// Maximum object size eligible for patch operations, in bytes.
34    pub max_patchable_size: u64,
35    /// Sender half of the async indexing queue.
36    pub indexer_tx: tokio::sync::mpsc::Sender<notedthat_indexer::IndexEvent>,
37    /// The search implementation (injected at startup).
38    pub searcher: Arc<dyn Searcher>,
39    /// The object change event log, when `NOTEDTHAT_EVENTS_BACKEND` selects one.
40    /// `None` is the default: writes are not announced and the events route
41    /// answers 404.
42    pub events: Option<Arc<dyn EventPublisher>>,
43    /// The per-knowledge-base index health record, shared with the indexer
44    /// worker and every write path (#97).
45    pub index_health: Arc<IndexHealth>,
46    /// The latest background probe of the storage and vector backends.
47    /// `/readyz` reads it and never probes inline.
48    pub readiness: ReadinessReceiver,
49    /// What `POST …/index/reconcile` starts (D67): a comparison of one
50    /// knowledge base's storage against the index, run by the server. `None`
51    /// where the backend has no on-demand pass — the `fs` backend today, whose
52    /// watcher covers it — and the route answers `404`.
53    pub reconcile: Option<Arc<dyn ReconcileTrigger>>,
54}
55
56/// A knowledge base already has a pass running; a second one now would not see
57/// what the first is past, so the caller retries once `last_reconcile` moves.
58#[derive(Debug, Clone, Copy, PartialEq, Eq)]
59pub struct ReconcileBusy;
60
61/// Starts a reconciliation pass for one knowledge base (D67).
62///
63/// Synchronous on purpose: an implementation only claims the knowledge base's
64/// slot and spawns the pass, so the route answers before any bucket is listed.
65pub trait ReconcileTrigger: Send + Sync {
66    /// Start a pass, or report that one is already running.
67    ///
68    /// # Errors
69    ///
70    /// [`ReconcileBusy`] when the knowledge base's previous pass has not finished.
71    fn trigger(&self, kb: &KbSlug) -> Result<(), ReconcileBusy>;
72}
73
74impl AppState {
75    /// Where a write made through this surface is reported, attributed to
76    /// `source` — the HTTP API itself, or MCP when the request says so.
77    pub(crate) fn sinks(&self, source: EventSource) -> notedthat_write::WriteSinks<'_> {
78        notedthat_write::WriteSinks::new(
79            &self.indexer_tx,
80            self.events.as_deref(),
81            &self.index_health,
82            source,
83        )
84    }
85}
86
87#[cfg(test)]
88mod tests {
89    use super::*;
90    use crate::testing::InMemoryStorage;
91    use std::collections::BTreeMap;
92    use std::sync::Arc;
93
94    fn minimal_state(tx: tokio::sync::mpsc::Sender<notedthat_indexer::IndexEvent>) -> AppState {
95        AppState {
96            storage: Arc::new(InMemoryStorage::default()),
97            declared_kbs: Arc::new(BTreeMap::new()),
98            access_policies: Arc::new(BTreeMap::new()),
99            kb_details: Arc::new(BTreeMap::new()),
100            authenticator: Arc::new(Authenticator::new("token")),
101            max_body_size: 1024,
102            max_patchable_size: 1024,
103            indexer_tx: tx,
104            searcher: Arc::new(crate::testing::NoopSearcher),
105            events: None,
106            index_health: Arc::new(notedthat_indexer::IndexHealth::new()),
107            readiness: crate::testing::ready_receiver(),
108            reconcile: None,
109        }
110    }
111
112    #[tokio::test]
113    async fn clone_shares_same_channel() {
114        let (tx, mut rx) = tokio::sync::mpsc::channel(10);
115        let state = minimal_state(tx);
116        let cloned = state.clone();
117
118        let event = notedthat_indexer::IndexEvent::Tombstone {
119            kb: notedthat_core::KbSlug::try_new("test").expect("valid kb slug"),
120            object_key: notedthat_core::ObjectPath::try_from("a.md").expect("valid path"),
121        };
122        cloned
123            .indexer_tx
124            .send(event.clone())
125            .await
126            .expect("send on cloned tx");
127
128        let received = rx.recv().await.expect("receive on original rx");
129        assert_eq!(
130            received, event,
131            "cloned Sender must share the same underlying channel"
132        );
133    }
134}
notedthat_api_http/state.rs

notedthat_api_http/
state.rs
