feat(serve): graceful shutdown + connection registry (v0.2 chunk 2)

Shutdown is drain-then-force-stop: flag.store(true) -> sup_h.join() (the
no-more-accepts barrier) -> cast BeginDrain -> poll ConnCount every 50ms
until 0 or drain_timeout; past the deadline each tick casts ForceStopConns
and re-sweeps, so conns that registered after the deadline are still
caught.

Listener shutdown is redesigned vs chunk 1: Restart::Permanent ->
Transient, wait-failure return -> panic (Transient restarts it), and a
shared Arc<AtomicBool> flag checked at loop top with a 250ms
wait_readable_timeout tick instead of request_stop — no request_stop on
the supervisor or listeners anywhere. Historical note: the redesign was
originally forced by smarm's then-lossy stop against QUEUED actors (fixed
upstream in 7bab4d2); it is kept because flag-based listener shutdown is
simpler and stop-semantics-free.

Conns self-register with the registry (ConnStarted from the conn actor,
not a listener cast): per-sender FIFO ordering is the only thing that
prevents ConnEnded overtaking ConnStarted; see conn_registry module docs.

conn_actor: in the catch_unwind(pipeline.run) Err branch,
smarm::preempt::check_cancelled() runs BEFORE composing the 500 — smarm's
stop is an undowncastable panic_any(StopSentinel), so the catch_unwind
would otherwise swallow a stop and 500-and-keep-running. The stop flag is
persistent, so check_cancelled re-raises cleanly.

Handle.shutdown() wakes the root via a 100ms try_recv poll
(SHUTDOWN_POLL): a cross-thread Sender::send's unpark is a
try_with_runtime no-op without runtime TLS on the sending thread (still
true on smarm 8e5b754; cross-thread unpark is a recorded smarm roadmap
candidate — this poll dies with it).

Two smarm bugs were found during this chunk and fixed upstream: lossy
stop against QUEUED actors (7bab4d2) and the terminal-wake shutdown
stall/hang (eddf3fe); post-mortem in artefact smarm-bug-terminal-wake.md.
This commit is contained in:
Claude
2026-06-11 21:45:11 +00:00
parent 5fe696992a
commit c658ac06b2
5 changed files with 564 additions and 36 deletions
+167
View File
@@ -0,0 +1,167 @@
//! Connection registry — the shutdown coordinator (v0.2 chunk 2).
//!
//! A `gen_server` tracking live connection pids and their busy/idle
//! state. Connection actors self-register as their first action and
//! self-deregister via a drop guard (so panic unwinds deregister too);
//! both casts come from the same sender, so Started always precedes
//! Ended in the inbox. (The roadmap sketched the *listener* casting
//! `{Started, pid}`, but then a short-lived conn's Ended could overtake
//! its Started and leak a dead pid into the set forever. Self-
//! registration makes the order a per-sender FIFO guarantee instead of
//! a race.)
//!
//! Listeners are deliberately NOT tracked here: they shut down via a
//! shared flag + timed accept-waits (see `serve`), never via
//! `request_stop` — stopping pids that announce themselves is racy (the
//! spawn-to-registration gap), and smarm's `request_stop` is lossy
//! against an actor that is QUEUED and then parks without passing an
//! observation point (see the v0.2 shutdown notes in the commit
//! message). Connections don't suffer this in practice: every stop the
//! registry issues targets a pid that just sent us a cast (so it is
//! running or parked, both covered), and the force-stop path re-sweeps
//! until the set empties.
//!
//! Drain protocol: `BeginDrain` stops every idle connection immediately
//! and flips the registry into draining mode, in which any connection
//! that *becomes* idle (finishes its in-flight request) is stopped on the
//! spot. Busy connections are left to finish; `ForceStopConns` (sent by
//! `serve` at the drain deadline) stops whatever remains. `request_stop`
//! unwinds a conn actor parked in `wait_readable` safely (smarm's 06-10
//! io fix) and `OwnedFd::drop` closes its socket on the way out.
//!
//! This server is also the planned introspection point for ws/channels
//! (roadmap v0.4+), which is why it exists as its own module rather than
//! being inlined into `serve`.
use smarm::{GenServer, Pid, ServerBuilder, ServerRef};
use std::collections::HashMap;
// ---------------------------------------------------------------------------
// Messages
// ---------------------------------------------------------------------------
pub enum Cast {
/// A connection actor started (self-registered, initially idle: it has
/// not parsed a request head yet).
ConnStarted(Pid),
/// Parsed a request head; a response is now owed.
ConnBusy(Pid),
/// Response written; parked (or about to park) waiting for the next
/// keep-alive request.
ConnIdle(Pid),
ConnEnded(Pid),
/// Stop idle conns now and stop each remaining conn as it goes idle.
BeginDrain,
/// Drain deadline passed: stop every remaining conn.
ForceStopConns,
}
pub enum Call {
ConnCount,
}
pub enum Reply {
ConnCount(usize),
}
// ---------------------------------------------------------------------------
// Server
// ---------------------------------------------------------------------------
#[derive(Clone, Copy, PartialEq, Eq)]
enum ConnState {
Busy,
Idle,
}
#[derive(Default)]
pub struct ConnRegistry {
conns: HashMap<Pid, ConnState>,
draining: bool,
}
impl GenServer for ConnRegistry {
type Call = Call;
type Reply = Reply;
type Cast = Cast;
type Info = ();
fn handle_call(&mut self, request: Call) -> Reply {
match request {
Call::ConnCount => Reply::ConnCount(self.conns.len()),
}
}
fn handle_cast(&mut self, request: Cast) {
match request {
Cast::ConnStarted(pid) => {
self.conns.insert(pid, ConnState::Idle);
if self.draining {
// Listener-stop race: this conn was accepted just
// before its listener died. Drain means no new work.
smarm::request_stop(pid);
}
}
Cast::ConnBusy(pid) => {
if let Some(s) = self.conns.get_mut(&pid) {
*s = ConnState::Busy;
}
}
Cast::ConnIdle(pid) => {
if let Some(s) = self.conns.get_mut(&pid) {
*s = ConnState::Idle;
if self.draining {
// Finished its in-flight request; nothing more is
// owed. The pid leaves the map via its drop
// guard's ConnEnded once the unwind completes.
smarm::request_stop(pid);
}
}
}
Cast::ConnEnded(pid) => { self.conns.remove(&pid); }
Cast::BeginDrain => {
self.draining = true;
for (pid, state) in &self.conns {
if *state == ConnState::Idle {
smarm::request_stop(*pid);
}
}
}
Cast::ForceStopConns => {
for pid in self.conns.keys() {
smarm::request_stop(*pid);
}
}
}
}
}
pub fn start() -> ServerRef<ConnRegistry> {
ServerBuilder::new(ConnRegistry::default()).start()
}
// ---------------------------------------------------------------------------
// Drop guard — self-deregistration on any exit path.
// ---------------------------------------------------------------------------
/// Casts `make(pid)` on drop. Runs on normal return, `request_stop`
/// unwind, and panic unwind alike; the cast is infallible from the
/// guard's perspective (a dead registry just returns an ignored Err).
pub struct DeregisterGuard {
registry: ServerRef<ConnRegistry>,
pid: Pid,
make: fn(Pid) -> Cast,
}
impl DeregisterGuard {
pub fn new(registry: ServerRef<ConnRegistry>, pid: Pid, make: fn(Pid) -> Cast) -> Self {
Self { registry, pid, make }
}
}
impl Drop for DeregisterGuard {
fn drop(&mut self) {
let _ = self.registry.cast((self.make)(self.pid));
}
}