feat(interceptors): per-interceptor wall-clock timeout (§9.4 M5)

A [[interceptors]] marker can set timeout_ms (migration 0075, CHECK > 0); a
runaway guard (loop {}) is interrupted and the op DENIED (fail closed) within
budget rather than hanging the write path. The effective deadline is
min(caller-remaining, now + timeout) — a hook can only tighten, never extend,
its caller's deadline. NULL uses PICLOUD_INTERCEPTOR_TIMEOUT_MS (default 5s).

Wiring: run_resolved_blocking splits into a core + a _with_timeout variant that
computes the effective deadline from engine::ambient_deadline() and runs via
execute_ast_with_deadline; run_one_hook passes the marker's timeout_ms (or the
env default). timeout_ms threaded end to end — manifest, plan, BundleInterceptor,
the reconcile diff (part of the mutable body: a timeout change is an Update),
insert_interceptor_tx, resolve_chain/list_for_owner/list_on_app_chain +
SealedInterceptor/InterceptorMarker, and interceptor_service → ResolvedInterceptor.
Schema snapshot re-blessed. Pinned by a journey: a loop{} guard with
timeout_ms=100 is denied and its write does not persist.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
MechaCat02
2026-07-16 20:18:03 +02:00
parent 5592f1c97e
commit 8b70d52d43
12 changed files with 244 additions and 47 deletions

View File

@@ -126,6 +126,10 @@ pub struct BundleInterceptor {
#[serde(default = "default_before_phase")]
pub phase: String,
pub script: String,
/// §9.4 M5: per-hook timeout in ms (`None` = instance default). `default`
/// so a pre-M5 CLI (which omits it) still deserializes.
#[serde(default)]
pub timeout_ms: Option<i32>,
}
fn default_before_phase() -> String {
@@ -828,8 +832,8 @@ pub struct CurrentState {
/// v1.2 Workflows owned directly by this node (app-owned; empty for a group).
pub workflows: Vec<crate::workflow_repo::Workflow>,
/// §9.4 interceptor markers declared directly at this node, as
/// `(service, op, phase, script)` tuples.
pub interceptors: Vec<(String, String, String, String)>,
/// `(service, op, phase, script, timeout_ms)` tuples.
pub interceptors: Vec<(String, String, String, String, Option<i32>)>,
}
/// One row of the read-only extension-point report (§5.5).
@@ -1620,6 +1624,7 @@ impl ApplyService {
&bi.op,
&bi.phase,
&bi.script,
bi.timeout_ms,
)
.await
.map_err(|e| ApplyError::Backend(e.to_string()))?;
@@ -4263,7 +4268,7 @@ impl ApplyService {
.await
.map_err(|e| ApplyError::Backend(e.to_string()))?
.into_iter()
.map(|m| (m.service, m.op, m.phase, m.script))
.map(|m| (m.service, m.op, m.phase, m.script, m.timeout_ms))
.collect();
Ok(CurrentState {
scripts,
@@ -4352,24 +4357,30 @@ fn compute_diff_with_names(
/// declaring is a `Delete` (applied only under `--prune`).
fn diff_interceptors(current: &CurrentState, bundle: &Bundle) -> Vec<ResourceChange> {
let key = |service: &str, op: &str, phase: &str| format!("{service}/{op}/{phase}");
let live: HashMap<String, &str> = current
// The reconcile identity is `(service, op, phase)`; the script AND the
// timeout are the mutable body, so a change to either is an Update.
let live: HashMap<String, (&str, Option<i32>)> = current
.interceptors
.iter()
.map(|(s, o, p, script)| (key(s, o, p), script.as_str()))
.map(|(s, o, p, script, timeout)| (key(s, o, p), (script.as_str(), *timeout)))
.collect();
let mut out = Vec::new();
for bi in &bundle.interceptors {
let k = key(&bi.service, &bi.op, &bi.phase);
match live.get(&k) {
Some(cur) if *cur == bi.script => out.push(ResourceChange {
op: Op::NoOp,
key: k,
detail: None,
}),
Some((cur_script, cur_timeout))
if *cur_script == bi.script && *cur_timeout == bi.timeout_ms =>
{
out.push(ResourceChange {
op: Op::NoOp,
key: k,
detail: None,
});
}
Some(_) => out.push(ResourceChange {
op: Op::Update,
key: k,
detail: Some("interceptor script changed".into()),
detail: Some("interceptor script or timeout changed".into()),
}),
None => out.push(ResourceChange {
op: Op::Create,
@@ -4378,7 +4389,7 @@ fn diff_interceptors(current: &CurrentState, bundle: &Bundle) -> Vec<ResourceCha
}),
}
}
for (s, o, p, _) in &current.interceptors {
for (s, o, p, _, _) in &current.interceptors {
let present = bundle
.interceptors
.iter()

View File

@@ -23,6 +23,8 @@ pub struct InterceptorMarker {
/// `before` (allow/deny + transform) or `after` (observe; §9.4 M3).
pub phase: String,
pub script: String,
/// §9.4 M5: per-hook timeout, or `None` for the instance default.
pub timeout_ms: Option<i32>,
}
/// List the markers declared **directly at** `owner` (not inherited), ordered
@@ -31,10 +33,10 @@ pub async fn list_for_owner(
pool: &PgPool,
owner: ScriptOwner,
) -> Result<Vec<InterceptorMarker>, sqlx::Error> {
let rows: Vec<(String, String, String, String)> = match owner {
let rows: Vec<(String, String, String, String, Option<i32>)> = match owner {
ScriptOwner::App(a) => {
sqlx::query_as(
"SELECT service, op, phase, interceptor_script FROM interceptors \
"SELECT service, op, phase, interceptor_script, timeout_ms FROM interceptors \
WHERE app_id = $1 ORDER BY service, op, phase",
)
.bind(a.into_inner())
@@ -43,7 +45,7 @@ pub async fn list_for_owner(
}
ScriptOwner::Group(g) => {
sqlx::query_as(
"SELECT service, op, phase, interceptor_script FROM interceptors \
"SELECT service, op, phase, interceptor_script, timeout_ms FROM interceptors \
WHERE group_id = $1 ORDER BY service, op, phase",
)
.bind(g.into_inner())
@@ -53,12 +55,15 @@ pub async fn list_for_owner(
};
Ok(rows
.into_iter()
.map(|(service, op, phase, script)| InterceptorMarker {
service,
op,
phase,
script,
})
.map(
|(service, op, phase, script, timeout_ms)| InterceptorMarker {
service,
op,
phase,
script,
timeout_ms,
},
)
.collect())
}
@@ -69,10 +74,10 @@ pub async fn list_on_app_chain(
pool: &PgPool,
app_id: AppId,
) -> Result<Vec<InterceptorMarker>, sqlx::Error> {
let rows: Vec<(String, String, String, String)> = sqlx::query_as(&format!(
let rows: Vec<(String, String, String, String, Option<i32>)> = sqlx::query_as(&format!(
"{CHAIN_LEVELS_CTE} \
SELECT DISTINCT ON (i.service, i.op, i.phase) \
i.service, i.op, i.phase, i.interceptor_script \
i.service, i.op, i.phase, i.interceptor_script, i.timeout_ms \
FROM chain c \
JOIN interceptors i ON (i.app_id = c.app_owner OR i.group_id = c.group_owner) \
ORDER BY i.service, i.op, i.phase, c.depth ASC",
@@ -82,12 +87,15 @@ pub async fn list_on_app_chain(
.await?;
Ok(rows
.into_iter()
.map(|(service, op, phase, script)| InterceptorMarker {
service,
op,
phase,
script,
})
.map(
|(service, op, phase, script, timeout_ms)| InterceptorMarker {
service,
op,
phase,
script,
timeout_ms,
},
)
.collect())
}
@@ -100,6 +108,8 @@ pub struct SealedInterceptor {
pub marker_app: Option<Uuid>,
pub marker_group: Option<Uuid>,
pub script_name: String,
/// §9.4 M5: per-hook wall-clock timeout, or `None` for the instance default.
pub timeout_ms: Option<i32>,
/// `(script_id, source, updated_at)` of the sealed script, or `None` if it
/// is missing/disabled at the declaring owner.
pub script: Option<(Uuid, String, DateTime<Utc>)>,
@@ -157,6 +167,7 @@ pub async fn resolve_chain(
Option<Uuid>, // marker_group
String, // script_name
String, // phase
Option<i32>, // timeout_ms
i32, // depth (0 = app, larger = ancestor)
Option<Uuid>, // script id
Option<String>, // script source
@@ -166,12 +177,12 @@ pub async fn resolve_chain(
markers AS ( \
SELECT i.interceptor_script AS script_name, \
i.app_id AS marker_app, i.group_id AS marker_group, \
i.phase AS phase, c.depth AS depth \
i.phase AS phase, i.timeout_ms AS timeout_ms, c.depth AS depth \
FROM chain c \
JOIN interceptors i ON (i.app_id = c.app_owner OR i.group_id = c.group_owner) \
WHERE i.service = $2 AND i.op = $3 \
) \
SELECT m.marker_app, m.marker_group, m.script_name, m.phase, m.depth, \
SELECT m.marker_app, m.marker_group, m.script_name, m.phase, m.timeout_ms, m.depth, \
s.id, s.source, s.updated_at \
FROM markers m \
LEFT JOIN scripts s ON ( \
@@ -191,11 +202,12 @@ pub async fn resolve_chain(
// depth = ancestor runs first); `after` app→ancestor = depth ASC.
let mut before: Vec<(i32, SealedInterceptor)> = Vec::new();
let mut after: Vec<(i32, SealedInterceptor)> = Vec::new();
for (marker_app, marker_group, script_name, phase, depth, sid, src, upd) in rows {
for (marker_app, marker_group, script_name, phase, timeout_ms, depth, sid, src, upd) in rows {
let sealed = SealedInterceptor {
marker_app,
marker_group,
script_name,
timeout_ms,
script: match (sid, src, upd) {
(Some(id), Some(source), Some(updated_at)) => Some((id, source, updated_at)),
_ => None,
@@ -225,37 +237,43 @@ pub async fn insert_interceptor_tx(
op: &str,
phase: &str,
script: &str,
timeout_ms: Option<i32>,
) -> Result<(), sqlx::Error> {
match owner {
ScriptOwner::App(a) => {
sqlx::query(
// The conflict arbiter includes `phase` to match the
// per-(owner, service, op, phase) index (§9.4 M3).
"INSERT INTO interceptors (app_id, service, op, phase, interceptor_script) \
VALUES ($1, $2, $3, $4, $5) \
// per-(owner, service, op, phase) index (§9.4 M3). A re-apply
// that changes only the script or timeout updates in place.
"INSERT INTO interceptors (app_id, service, op, phase, interceptor_script, timeout_ms) \
VALUES ($1, $2, $3, $4, $5, $6) \
ON CONFLICT (app_id, service, op, phase) WHERE app_id IS NOT NULL \
DO UPDATE SET interceptor_script = EXCLUDED.interceptor_script, updated_at = NOW()",
DO UPDATE SET interceptor_script = EXCLUDED.interceptor_script, \
timeout_ms = EXCLUDED.timeout_ms, updated_at = NOW()",
)
.bind(a.into_inner())
.bind(service)
.bind(op)
.bind(phase)
.bind(script)
.bind(timeout_ms)
.execute(&mut **tx)
.await?;
}
ScriptOwner::Group(g) => {
sqlx::query(
"INSERT INTO interceptors (group_id, service, op, phase, interceptor_script) \
VALUES ($1, $2, $3, $4, $5) \
"INSERT INTO interceptors (group_id, service, op, phase, interceptor_script, timeout_ms) \
VALUES ($1, $2, $3, $4, $5, $6) \
ON CONFLICT (group_id, service, op, phase) WHERE group_id IS NOT NULL \
DO UPDATE SET interceptor_script = EXCLUDED.interceptor_script, updated_at = NOW()",
DO UPDATE SET interceptor_script = EXCLUDED.interceptor_script, \
timeout_ms = EXCLUDED.timeout_ms, updated_at = NOW()",
)
.bind(g.into_inner())
.bind(service)
.bind(op)
.bind(phase)
.bind(script)
.bind(timeout_ms)
.execute(&mut **tx)
.await?;
}

View File

@@ -33,6 +33,9 @@ impl InterceptorServiceImpl {
/// (`cx.app_id`); only the `owner` is the sealing owner.
fn map_entry(app_id: picloud_shared::AppId, sealed: SealedInterceptor) -> InterceptorEntry {
let owner = sealed.sealing_owner();
// A negative/oversized DB value is impossible (CHECK > 0), but clamp
// defensively to `None` (→ instance default) rather than panic.
let timeout_ms = sealed.timeout_ms.and_then(|v| u32::try_from(v).ok());
match sealed.script {
Some((sid, source, updated_at)) => InterceptorEntry::Run(ResolvedInterceptor {
script: Box::new(ResolvedScript {
@@ -43,7 +46,7 @@ impl InterceptorServiceImpl {
updated_at,
name: sealed.script_name,
}),
timeout_ms: None,
timeout_ms,
}),
None => InterceptorEntry::Dangling(sealed.script_name),
}