Two review-pass nits from the v1.1.0-foundation review:
- Blueprint §6 Tech Stack table still listed the database as
"PostgreSQL + hstore" with an hstore-for-KV rationale — directly
contradicting the §8.1 KV rewrite that explicitly rejected hstore
in favour of JSONB. Updates the row so the high-level summary
matches the §8.1 reasoning.
- LocalExecutorClient::execute now documents the permit-vs-timeout
interaction: when tokio::time::timeout fires the future drops and
the permit returns, but the detached spawn_blocking thread keeps
running until the Rhai script winds down. In-use blocking threads
can briefly exceed the gate's permit count after a timeout. Calling
it out so future readers don't read the implementation as buggy.
No behaviour change.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds a single global concurrency cap on the data-plane dispatch path:
- orchestrator-core::gate::ExecutionGate wraps tokio::Semaphore.
Non-blocking try_acquire — no queue. PICLOUD_MAX_CONCURRENT_EXECUTIONS
env var (default 32) sets the cap.
- LocalExecutorClient acquires a permit before spawn_blocking; the
permit drops with the future so the slot returns automatically.
- On refusal, ExecError::Overloaded { retry_after_secs: 1 } surfaces
upward. ApiError::IntoResponse already maps that to 503 with a
Retry-After header (landed in the previous commit alongside the
variant itself).
- picloud binary constructs the gate once at build_app and shares it
with LocalExecutorClient.
The cap exists so a Rhai script storm can't drain the blocking-thread
pool — pushing back hard beats letting requests pile up against a
finite worker count. Per-app / per-script caps stay deferred until a
real workload demands them.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Brings the MVP feature set online: upload a Rhai script, get an HTTP
endpoint that runs it sandboxed in-process, list/update/delete it, and
have invalid sources rejected at upload time. Verified live through
Caddy with a full lifecycle (`create → list → get → execute → update
→ delete`) plus error paths (syntax error, duplicate name, deleted).
Layout — every concern lands behind the trait seam its layer owns, so
cluster-mode in v1.3+ is a swap of two impls, not a rewrite:
* shared::ScriptValidator — manager calls into validation without
a hard dep on executor-core; executor-core impls the trait on
`Engine`. Pinned in shared so neither crate has to know about
the other.
* executor-core::Engine — real Rhai engine: sandbox limits (max
operations / string size / map size / call depth), disabled
`print`, blocked `import` (DummyModuleResolver), `log::trace
/info/warn/error` registered as a static module with shared
log-capture buffer (no `log::debug` because `debug` is a Rhai
reserved keyword — `log::trace` covers the same need).
- `ctx` is pushed as a Scope constant exposing
execution_id, script_id, script_name, request_id,
invocation_type, request.{path,headers,body}.
- Response convention: a Map with `statusCode` is the
structured shape (`{statusCode, headers?, body}`); any
other return value is a 200 with the value as the body.
- Engine::execute is now synchronous (pure compute); the
async wrapper + wall-clock timeout live in
LocalExecutorClient, which spawns_blocking and applies a
300s hard ceiling regardless of per-script config.
- 10 unit tests cover validate, exec, structured response,
ctx exposure, log capture, op-budget enforcement, runtime
errors, blocked imports, JSON round-tripping.
* manager-core::repo — full sqlx CRUD over the `scripts` table,
with proper unique-violation handling for duplicate names.
Embedded migrations via `sqlx::migrate!` (one initial
`0001_init.sql` for pgcrypto + scripts + execution_logs).
* manager-core::api — `admin_router` mounts `/scripts` and
`/scripts/{id}`. Create + Update validate source through the
injected `ScriptValidator` before persistence. Returns proper
422/409/404 status codes via `ApiError::IntoResponse`.
* orchestrator-core::api — `data_plane_router` mounts
`/execute/{id}`: resolves the script through `ScriptResolver`,
constructs the `ExecRequest` from headers+body, awaits
`ExecutorClient::execute(..., timeout)`, translates the
`ExecResponse` to an axum `Response` with header passthrough.
Maps `ExecError` variants to 422/504/502/507.
* picloud all-in-one — opens the pool, runs migrations, builds
one engine, nests both routers under `/api/admin` and `/api`,
enables structured JSON tracing and graceful shutdown on
SIGTERM. Single `PostgresScriptRepository` Arc is shared by
the admin router (writes) and the resolver (reads).
Other changes:
* Workspace axum bump 0.7 → 0.8 for the `{id}` path syntax
matching the route definitions.
* Workspace clippy: allow `needless_pass_by_value` and
`boxed_local` to keep API ergonomics over pedantic noise.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sets up the PiCloud monorepo as a Cargo workspace organised around the
three-service architecture (manager / orchestrator / executor), each
backed by a *-core library crate so the same logic powers both the MVP
all-in-one `picloud` binary and the future split-process cluster mode.
* crates/shared, executor-core, orchestrator-core, manager-core
define the library surface and trait seams between the three
services (`ExecutorClient`, `ScriptResolver`, `ScriptRepository`).
* crates/picloud is the MVP entrypoint; serves /healthz on 8080
(override via PICLOUD_BIND).
* crates/picloud-{manager,orchestrator,executor} are skeleton
binaries that keep the crate boundaries honest until cluster
mode is built out in v1.3+.
* docs/git-workflow.md defines the trunk-based workflow:
short-lived branches, Conventional Commits, separate hotfix
flow with mandatory reproduction tests.
* CLAUDE.md captures the working rules for future Claude sessions.
Workspace passes `cargo fmt`, `cargo clippy -D warnings` (with
pedantic enabled), and `cargo test --workspace`. The all-in-one
binary responds on `/healthz` and `/`.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
