Files
MechaCat02 966b209d2e feat(quota): add per-app storage ceilings
An app's OWN store had no ceiling of any kind — only the group-SHARED
collections did, which is the inversion of what you'd expect. And
`authz::script_gate` returns `Ok(())` when there is no principal, so a
PUBLIC UNAUTHENTICATED route could reach `files::collection(..).create(..)`
and write blobs in a loop until the disk filled. `PICLOUD_FILES_MAX_FILE_SIZE_BYTES`
caps ONE file at 100 MB; nothing capped the count. Same for KV keys and docs
against Postgres.

Ceilings: `PICLOUD_APP_KV_MAX_ROWS` / `PICLOUD_APP_DOCS_MAX_ROWS` (100k) and
`PICLOUD_APP_FILES_MAX_TOTAL_BYTES` (10 GiB).

They are enforced DIFFERENTLY from the group ones, on purpose — porting the
group design as-is would have been a serious regression:

  * KV/docs check a ROW count, on INSERT only, with NO lock. `kv::set` is the
    hottest write path in the system; taking a per-app advisory lock on every
    set would serialize an app's entire data plane, and a `SUM(...)` scan would
    make write cost grow with the app's stored size. What the lock buys is
    small — unlocked overshoot is bounded by write concurrency (32) against a
    ceiling of 100k, ~0.03%. These are anti-DoS rails, not billing. An update
    adds no row and is bounded by the per-value cap, so it pays nothing at all.
  * No per-app BYTE ceiling for KV/docs: every value is already capped at
    `PICLOUD_KV_MAX_VALUE_BYTES`, so `max_rows x max_value_bytes` is ALREADY a
    finite bound. A second scan would buy nothing.
  * FILES are the exception and DO lock + sum: one blob may be 100 MB, so 32
    racing uploads could overshoot by gigabytes of real disk, and a file write
    is heavy enough that the lock and the SUM are lost in the noise. Checked on
    create AND update — an update that skipped the ceiling would be a free
    bypass (the same bug just fixed for group files: grow a 1-byte file to
    100 MB, repeat).

`group_quota` is renamed `quota`: it now serves both owners, and the module
doc is where the two enforcement strategies are contrasted.

Pinned by tests/atomic_write.rs: the key ceiling refuses a new key while
still allowing an update at the ceiling (rejecting updates would brick an app
the moment it filled up — worse than the DoS the rail exists to stop), and 10
concurrent uploads cannot exceed the disk ceiling, nor can an update grow past it.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-14 20:32:32 +02:00

171 lines
5.9 KiB
Rust

//! `KvService` — the v1.1.1 key-value store contract.
//!
//! Lives in `picloud-shared` (not `executor-core`) so the Rhai bridge,
//! the manager-core Postgres impl, and any future in-memory test impl
//! can all depend on the same trait without dragging
//! `executor-core` into `manager-core`'s dep graph.
//!
//! Implementations MUST derive every storage `app_id` from `cx.app_id`
//! — never from a script-passed argument. That is the cross-app
//! isolation boundary; see `docs/sdk-shape.md`.
use async_trait::async_trait;
use thiserror::Error;
use crate::SdkCallCx;
/// `KvService` is collection-scoped. Scripts get a handle via
/// `kv::collection(name)` and call `get`/`set`/`has`/`delete`/`list`
/// on it. The trait surface accepts the collection by name so the
/// Postgres impl can avoid an extra round-trip to materialize the
/// collection (collections are namespaces, not first-class rows).
#[async_trait]
pub trait KvService: Send + Sync {
async fn get(
&self,
cx: &SdkCallCx,
collection: &str,
key: &str,
) -> Result<Option<serde_json::Value>, KvError>;
async fn set(
&self,
cx: &SdkCallCx,
collection: &str,
key: &str,
value: serde_json::Value,
) -> Result<(), KvError>;
/// Compare-and-swap. Atomically writes `new` only if the current value
/// equals `expected` — or, when `expected` is `None`, only if the key is
/// ABSENT (insert-if-absent). Returns `true` if the swap happened, `false`
/// if the precondition failed (value differed / key already present).
/// Default: unsupported, so only storage-backed impls opt in.
async fn set_if(
&self,
cx: &SdkCallCx,
collection: &str,
key: &str,
expected: Option<serde_json::Value>,
new: serde_json::Value,
) -> Result<bool, KvError> {
let _ = (cx, collection, key, expected, new);
Err(KvError::Backend(
"set_if is not supported by this kv implementation".into(),
))
}
async fn delete(&self, cx: &SdkCallCx, collection: &str, key: &str) -> Result<bool, KvError>;
async fn has(&self, cx: &SdkCallCx, collection: &str, key: &str) -> Result<bool, KvError>;
/// Cursor-style pagination. `cursor` is opaque to the caller;
/// implementations encode the resume key inside. `None` cursor
/// starts from the beginning. Implementations cap `limit` at a
/// reasonable ceiling internally (script can't request an unbounded
/// page).
async fn list(
&self,
cx: &SdkCallCx,
collection: &str,
cursor: Option<&str>,
limit: u32,
) -> Result<KvListPage, KvError>;
}
/// One page of keys from `KvService::list`. `next_cursor` is `Some`
/// when more pages exist, `None` when exhausted. The cursor encoding
/// is implementation-defined (the Postgres impl base64-encodes the
/// last key).
#[derive(Debug, Clone)]
pub struct KvListPage {
pub keys: Vec<String>,
pub next_cursor: Option<String>,
}
/// Stub used by the test harness so executor-core integration tests
/// (which don't touch KV) can construct a `Services` bundle without
/// spinning up Postgres. Every call returns
/// `KvError::Backend("...")` so accidental KV use surfaces clearly.
#[derive(Debug, Default, Clone, Copy)]
pub struct NoopKvService;
#[async_trait]
impl KvService for NoopKvService {
async fn get(
&self,
_cx: &SdkCallCx,
_collection: &str,
_key: &str,
) -> Result<Option<serde_json::Value>, KvError> {
Err(KvError::Backend("kv is not wired in".into()))
}
async fn set(
&self,
_cx: &SdkCallCx,
_collection: &str,
_key: &str,
_value: serde_json::Value,
) -> Result<(), KvError> {
Err(KvError::Backend("kv is not wired in".into()))
}
async fn delete(
&self,
_cx: &SdkCallCx,
_collection: &str,
_key: &str,
) -> Result<bool, KvError> {
Err(KvError::Backend("kv is not wired in".into()))
}
async fn has(&self, _cx: &SdkCallCx, _collection: &str, _key: &str) -> Result<bool, KvError> {
Err(KvError::Backend("kv is not wired in".into()))
}
async fn list(
&self,
_cx: &SdkCallCx,
_collection: &str,
_cursor: Option<&str>,
_limit: u32,
) -> Result<KvListPage, KvError> {
Err(KvError::Backend("kv is not wired in".into()))
}
}
/// Failure modes surfaced to the Rhai bridge. The bridge converts each
/// to a Rhai runtime error string; the discriminants exist so internal
/// callers (admin endpoints, tests, GC) can react more precisely.
#[derive(Debug, Error)]
pub enum KvError {
/// Empty collection name; rejected at the SDK boundary per
/// `docs/sdk-shape.md`.
#[error("collection name must not be empty")]
InvalidCollection,
/// Caller principal lacked the required capability. Only raised
/// when `cx.principal.is_some()` — scripts running with
/// `principal: None` (public HTTP) operate under script-as-gate
/// semantics and skip the capability check.
#[error("forbidden")]
Forbidden,
/// The app is at its per-app row ceiling (`PICLOUD_APP_KV_MAX_ROWS`) and
/// this write would add a key. An update to an existing key is exempt.
#[error("kv: app key quota exceeded ({actual} keys; limit {limit})")]
QuotaExceeded { limit: u64, actual: u64 },
/// JSON-encoded value exceeded the per-app `max_value_bytes` cap.
/// Defends Postgres JSONB columns from anonymous-public-script DoS;
/// the limit is configurable via `PICLOUD_KV_MAX_VALUE_BYTES`.
#[error("kv: value too large ({actual} bytes; limit {limit})")]
ValueTooLarge { limit: usize, actual: usize },
/// Anything else — Postgres unavailable, serialization failure,
/// etc. The string is safe to surface to a script.
#[error("kv backend error: {0}")]
Backend(String),
}