harmos_runtime/state/
snapshot.rs1use std::collections::{HashMap, VecDeque};
18use std::future::{Future, poll_fn};
19use std::num::{NonZeroU64, NonZeroUsize};
20use std::panic::{AssertUnwindSafe, catch_unwind};
21use std::pin::pin;
22use std::sync::{
23 Arc,
24 atomic::{AtomicU64, Ordering},
25};
26use std::task::Poll;
27use serde::{Serialize, de::DeserializeOwned};
29use super::snapshot_codec;
31use crate::state::{Tail, TrackedState, Undos, dispatch::Dispatch};
32use crate::storage::{SchemaKey, StorageManager};
33use crate::{
34 Application, Error, Result, runtime::Incompatible, state::At, state::Entry, state::Position, state::Receipt, state::Recorded,
35 state::RequestKey, storage::Payload,
36};
37
38type Encoding<A> =
45 fn(&SnapshotPolicy, &At<<A as Application>::State>, &Undos<A>, &Tail<A>, &HashMap<RequestKey, Receipt>) -> Result<Vec<u8>>;
46type Decoding<A> = fn(&[u8], &SnapshotPolicy, &Dispatch<A, Payload<A>>, usize) -> std::result::Result<Restored<A>, Incompatible>;
47
48#[derive(Clone, Debug)]
51pub struct SnapshotPolicy {
52 pub schema: SchemaKey,
53 pub every: NonZeroU64,
54 pub max_bytes: NonZeroUsize,
55}
56
57pub(crate) struct Snapshots<A: Application> {
58 pub(crate) policy: SnapshotPolicy,
59 encode: Encoding<A>,
60 pub(crate) decode: Decoding<A>,
61 last: AtomicU64,
62}
63
64pub(crate) struct Restored<A: Application> {
65 pub(crate) state: At<A::State>,
66 pub(crate) undos: Undos<A>,
67 pub(crate) receipts: HashMap<RequestKey, Receipt>,
68 pub(crate) tail: VecDeque<Arc<Entry<Recorded<A>>>>,
69}
70
71impl<A: Application> Snapshots<A> {
78 pub(crate) fn new(policy: SnapshotPolicy) -> Self
79 where
80 A::State: Serialize + DeserializeOwned,
81 {
82 Self {
83 policy,
84 encode: snapshot_codec::encode::<A>,
85 decode: snapshot_codec::decode::<A>,
86 last: AtomicU64::new(0),
87 }
88 }
89
90 pub(crate) fn resumed(&self, position: Position) {
91 self.last.store(position.0, Ordering::Relaxed);
92 }
93
94 #[tracing::instrument(target = "harmos::state", name = "state.capture", level = "debug", skip_all)]
95 pub(crate) async fn capture(
96 &self, storage: &StorageManager<crate::storage::ErasedCatalog>, tracked: &TrackedState<A>, undos: &Undos<A>,
97 receipts: &HashMap<RequestKey, Receipt>, final_save: bool,
98 ) {
99 if tracked.poisoned() {
100 return;
101 }
102 let position = tracked.state.read().await.position;
103 let last = self.last.load(Ordering::Relaxed);
104 if position.0 <= last || (!final_save && position.0 - last < self.policy.every.get()) {
105 return;
106 }
107 let result = async {
110 if final_save {
111 storage.flush().await?;
113 storage.flush_recovery().await?;
114 }
115 let mut sealing = tracked.sealing.subscribe();
116 let reservation = storage.reserve_snapshot(self.policy.max_bytes.get());
117 let mut permit =
118 if final_save {
119 reservation.await?
120 } else {
121 if *sealing.borrow() {
122 return Ok(());
123 }
124 let mut reservation = pin!(reservation);
125 let mut changed = pin!(sealing.changed());
126 let Some(permit) = poll_fn(|cx| {
127 if changed.as_mut().poll(cx).is_ready() { Poll::Ready(None) } else { reservation.as_mut().poll(cx).map(Some) }
128 })
129 .await
130 else {
131 return Ok(());
132 };
133 permit?
134 };
135 let state = tracked.state.read().await;
136 let tail = tracked.tail.read().await;
137 self.last.store(position.0, Ordering::Relaxed);
138 let payload =
139 catch_unwind(AssertUnwindSafe(|| (self.encode)(&self.policy, &state, undos, &tail, receipts))).map_err(|_| {
140 Error::InvalidInput {
141 field: "state snapshot",
142 reason: "serializer panicked",
143 }
144 })??;
145 permit.snapshot_payload(payload)?;
146 permit.commit(position);
147 Ok::<(), Error>(())
148 }
149 .await;
150 if let Err(error) = result {
151 let error = harmos_domain::failure::Report(&error);
152 tracing::warn!(%error, position = position.0, "snapshot skipped; recovery retains the journal");
153 }
154 }
155}
156
157