artifactize_core/
store.rs

1//! The state database and storage maintenance.
2
3use std::path::{Path, PathBuf};
4
5use crate::{platform, workspace::canonical_target};
6
7mod blocked;
8pub mod definitions;
9mod executions;
10pub(crate) mod history;
11mod human;
12pub use human::HumanClaim;
13mod catalog;
14pub(crate) mod connection;
15pub mod prune;
16pub(crate) mod receipts;
17mod result;
18mod run_record;
19mod runs;
20mod snapshots;
21mod touched;
22mod unreadable;
23mod validation;
24pub use unreadable::{Unreadable, read_unreadable};
25mod lifecycle;
26pub(crate) mod migrations;
27pub(crate) mod run_control;
28mod run_read;
29pub use lifecycle::{ExecutionState, GreenResult, PendingRequest, RequestState, RunState};
30pub use result::{ExecutionResult, Field as ResultField, RuntimeResult};
31pub use run_control::read_run_repo;
32pub use run_read::read_run;
33pub(crate) use run_read::{read_new_results, read_run_record, read_run_requests_only};
34pub use snapshots::{
35    ArtifactValidation, Blocker, ChildIdentity, EvalValidation, HumanDefinition, Validation,
36    ValidationSnapshot,
37};
38mod wait_timeout;
39pub use catalog::{CatalogRun, Signoff, read_catalog};
40pub use runs::{RunSummary, read_runs, read_scoped_runs};
41pub(crate) mod reader;
42mod requests;
43pub(crate) use executions::read_profiled_executions;
44pub use executions::{
45    Capacity, Claim, Execution, ExecutionOptions, Origin, Producer, Provenance,
46    read_keyed_executions, read_latest_cached,
47};
48#[cfg(any(test, feature = "test-support"))]
49pub use receipts::OwnedReceipts;
50pub(crate) use receipts::regular_files as check_probe_files;
51pub use receipts::{
52    DATABASE, EARLIER_STATE, LastRequest, QueueReason, Receipts, Request, Run, RunView,
53    StoppedBackend, read_latest_requests, read_state_id, schema_error, state_schema,
54};
55pub use requests::read_request_repo;
56pub use requests::{
57    RequestView, read_original, read_request, read_request_batch, read_requests, read_requests_in,
58    read_run_requests, read_session_request, read_waiting, read_waiting_originals,
59};
60
61/// How long one database operation waits for the state other processes are writing before it
62/// fails. Several Runs, the monitor and MCP servers share one state; a Run's write waits behind
63/// theirs rather than failing the Run, while a lock that never frees still ends in an error.
64pub(crate) const SQLITE_BUSY_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(30);
65
66/// How often a busy operation tries again. SQLite's own busy timeout backs off to 100 ms
67/// between tries, so under steady writes from other connections, which try every 1–2 ms when
68/// they start waiting, a long waiter rarely finds the lock free and runs out its time while
69/// the others commit (#372). Trying at this pace keeps every waiter in the race.
70const SQLITE_BUSY_POLL: std::time::Duration = std::time::Duration::from_millis(1);
71
72/// Make `db` wait for a busy database, trying every [`SQLITE_BUSY_POLL`] for up to
73/// [`SQLITE_BUSY_TIMEOUT`].
74pub(crate) fn wait_when_busy(db: &rusqlite::Connection) -> rusqlite::Result<()> {
75    db.busy_handler(Some(busy))
76}
77
78/// SQLite's busy handler: `tries` counts the earlier calls for this one wait, which runs on
79/// the calling thread, so the wait's start is kept per thread.
80fn busy(tries: i32) -> bool {
81    thread_local! {
82        static SINCE: std::cell::Cell<std::time::Instant> = std::cell::Cell::new(std::time::Instant::now());
83    }
84    let now = std::time::Instant::now();
85    if tries == 0 {
86        SINCE.set(now);
87    }
88    if now.duration_since(SINCE.get()) >= SQLITE_BUSY_TIMEOUT {
89        return false;
90    }
91    std::thread::sleep(SQLITE_BUSY_POLL);
92    true
93}
94
95/// Allow four short backoffs for a pending-delete handle to close, bounding the extra
96/// latency of a real file-access failure to 80ms. This is a retry policy, not a test deadline.
97const TRANSIENT_ACCESS_RETRIES: u32 = 5;
98const TRANSIENT_ACCESS_BACKOFF: std::time::Duration = std::time::Duration::from_millis(20);
99
100/// Retry `attempt` while `transient` says its error can clear on its own, before giving up
101/// with its last error. `transient` decides alone: a non-transient error, or the bound run out,
102/// both stop the loop and return that error immediately, never retried further.
103fn retry<T, E>(
104    mut attempt: impl FnMut() -> Result<T, E>,
105    transient: impl Fn(&E) -> bool,
106    mut before_retry: impl FnMut(),
107) -> Result<T, E> {
108    for remaining in (0..TRANSIENT_ACCESS_RETRIES).rev() {
109        match attempt() {
110            Ok(value) => return Ok(value),
111            Err(error) if remaining > 0 && transient(&error) => {
112                before_retry();
113            }
114            Err(error) => return Err(error),
115        }
116    }
117    unreachable!("the loop above always returns by its last attempt")
118}
119
120/// Retry `attempt` past [`platform::transient_file_access`] (Windows transiently denying a
121/// metadata or open probe while another handle finishes closing a just-deleted WAL sidecar)
122/// before giving up with its last error. Every read-only state open reaches this through
123/// [`receipts::check_files`]/[`check_probe_files`], which every reader calls first, so this is
124/// the one place the retry needs to live.
125pub(crate) fn retry_transient_access<T>(
126    attempt: impl FnMut() -> std::io::Result<T>,
127    before_retry: impl FnMut(),
128) -> std::io::Result<T> {
129    retry(attempt, platform::transient_file_access, before_retry)
130}
131
132#[cfg(test)]
133mod retry_tests {
134    use std::{cell::Cell, io};
135
136    #[test]
137    fn file_probe_returns_success_and_permanent_errors_without_backoff() {
138        let calls = Cell::new(0);
139        let result = super::retry_transient_access(
140            || {
141                calls.set(calls.get() + 1);
142                Ok(7)
143            },
144            || panic!("successful probes need no backoff"),
145        );
146        assert_eq!(result.unwrap(), 7);
147        assert_eq!(calls.get(), 1);
148
149        let calls = Cell::new(0);
150        let result = super::retry_transient_access(
151            || {
152                calls.set(calls.get() + 1);
153                Err::<(), _>(io::Error::other("permanent probe failure"))
154            },
155            || panic!("permanent errors cannot be retried"),
156        );
157        assert_eq!(result.unwrap_err().to_string(), "permanent probe failure");
158        assert_eq!(calls.get(), 1);
159    }
160}
161
162/// Resolve the single state directory without creating it.
163pub fn state_dir(explicit: Option<&Path>) -> Result<PathBuf, String> {
164    let path = match explicit {
165        Some(path) => path.to_path_buf(),
166        None => state_home().map_err(|e| e.to_string())?,
167    };
168    canonical_target(&path).map_err(|e| e.to_string())
169}
170
171/// SQLite `user_version`; schema 9 keeps a Run's definitions in their own row, written once,
172/// and numbers every change of a Run, request or execution row; schema 10 adds the open
173/// WAITING_HUMAN Run and names a Run that ended with only sign-offs left YELLOW. Versions from
174/// schema 6 upgrade additively.
175pub const STATE_SCHEMA_VERSION: u32 = 10;
176
177#[derive(Debug, thiserror::Error)]
178#[error("cannot resolve state home: set {}", state_home_variables())]
179pub struct StateHomeError;
180
181/// The variables `state_home` reads, in order, for the error that names them.
182fn state_home_variables() -> String {
183    let names: Vec<&str> = std::iter::once("ARTIFACTIZE_STATE_HOME")
184        .chain(platform::STATE_VARIABLES.iter().copied())
185        .chain([platform::HOME_VARIABLE])
186        .collect();
187    let (last, rest) = names.split_last().expect("constant names");
188    format!("{}, or {last}", rest.join(", "))
189}
190
191/// Resolve the default state home without creating directories; empty variables are ignored.
192/// The platform's state variables come after `ARTIFACTIZE_STATE_HOME` and before the home.
193pub fn state_home() -> Result<PathBuf, StateHomeError> {
194    platform::state_directory().ok_or(StateHomeError)
195}
196
197#[cfg(test)]
198#[path = "store/file_tests.rs"]
199mod file_tests;
artifactize_core/store.rs

artifactize_core/
store.rs
