feat(core): A.6.1 — use OS connectivity signals to drive reconnect

Extends the A.6 supervisor with an OS-level connectivity hint so a
returning network triggers a redial immediately instead of waiting
out the current backoff slot (up to 60 s). The watchdog remains the
authoritative loss detector — the OS signal is advisory.

* `chanora_core::NetworkState` (Unknown / Online / Offline) is owned
  by `ChanoraSession` via a `tokio::sync::watch::Sender`.
  `set_network_state()` / `network_state()` are the public accessors.
* The supervisor's watch-phase `select!` gains a `network_rx`
  branch: Offline pre-charges watchdog misses (capped at
  `MAX_MISSES - 1`) so the next probe failure trips immediately;
  Online clears stale misses. This shrinks UI-banner latency on a
  Wi-Fi drop from ~15 s to ~5 s.
* The reconnect-loop's backoff sleep races against Online: a
  transition cuts the sleep short and resets the attempt counter so
  future losses start at the smallest backoff window again.
* `chanora_bridge` adds `BridgeNetworkState` (mirror enum) and a
  sync `set_network_state(state)` function. On platforms with no
  signal wired the supervisor stays at Unknown and falls back to
  pure watchdog/backoff — no behavioural regression vs A.6.
* Flutter adds `connectivity_plus ^6.1.0` and wires
  `_wireConnectivity()` in `main()`: seeds with `checkConnectivity()`
  then forwards every `onConnectivityChanged` to the bridge,
  mapping any non-`none` transport to Online.

Verified on Moto G Stylus 5G (Android 14): `svc wifi disable && svc
data disable` for ~40 s — reconnect banner appeared promptly
because the watchdog was pre-charged. After `svc wifi enable && svc
data enable` the supervisor woke from its 15 s backoff slot and
reconnected within seconds; the channel tree re-rendered without
user action.
This commit is contained in:
EdisonJwa
2026-05-15 01:07:21 +08:00
parent 0bef61aea2
commit f52d702e27
10 changed files with 485 additions and 63 deletions
+107 -16
View File
@@ -41,7 +41,7 @@ use std::sync::Arc;
use std::time::Duration;
use thiserror::Error;
use tokio::sync::{broadcast, oneshot, Mutex};
use tokio::sync::{broadcast, oneshot, watch, Mutex};
use tokio::task::JoinHandle;
use tracing::{info, warn};
@@ -121,6 +121,20 @@ pub enum SessionEvent {
AudioStopped,
}
/// Coarse OS-reported network state. Populated by the Flutter side
/// via `connectivity_plus`; on platforms where no signal is wired
/// we stay at `Unknown` forever and the supervisor falls back to
/// pure watchdog/backoff behaviour.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum NetworkState {
/// No signal seen yet — treat as ambiguous; don't change behaviour.
Unknown,
/// OS reports at least one network with internet capability.
Online,
/// OS reports no networks available.
Offline,
}
/// Channel capacity for the broadcast events. Generous because
/// reconnect cycles emit several events per attempt; if subscribers
/// fall behind we'd rather skip than block the supervisor.
@@ -161,18 +175,39 @@ struct ConnectedState {
pub struct ChanoraSession {
inner: Arc<Mutex<Option<ConnectedState>>>,
events_tx: broadcast::Sender<SessionEvent>,
/// OS-reported network state. Updated by the bridge from
/// `connectivity_plus` callbacks. Supervisor observes via
/// [`watch::Receiver`].
network_tx: watch::Sender<NetworkState>,
}
impl ChanoraSession {
/// Construct an empty session. Performs no I/O.
pub fn new() -> Self {
let (events_tx, _) = broadcast::channel(EVENT_CHANNEL_CAPACITY);
let (network_tx, _) = watch::channel(NetworkState::Unknown);
Self {
inner: Arc::new(Mutex::new(None)),
events_tx,
network_tx,
}
}
/// Push an OS connectivity update. Called by the bridge when
/// `connectivity_plus` fires. Safe to call from any thread.
pub fn set_network_state(&self, state: NetworkState) {
// `send_if_modified` would suppress duplicate sends, but
// `watch::Sender::send` already drops sends with no
// subscribers gracefully. Using `send_replace` so the value
// is updated even before any subscriber attaches.
let _ = self.network_tx.send_replace(state);
}
/// Read the latest OS connectivity state. Useful for diagnostics.
pub fn network_state(&self) -> NetworkState {
*self.network_tx.borrow()
}
/// Subscribe to lifecycle events. The returned receiver fires
/// on connect / lost / reconnecting / disconnected /
/// audio-started / audio-stopped transitions. Multiple
@@ -213,6 +248,7 @@ impl ChanoraSession {
probe,
cancel_rx,
sup_inner.clone(),
self.network_tx.subscribe(),
));
let _ = self.events_tx.send(SessionEvent::Connected {
@@ -355,6 +391,7 @@ async fn supervisor_loop(
initial_probe: chanora_protocol::SnapshotProbe,
mut cancel_rx: oneshot::Receiver<()>,
sup_inner: Arc<Mutex<SupervisorInner>>,
mut network_rx: watch::Receiver<NetworkState>,
) {
let mut lost_rx = initial_lost_rx;
let mut probe = initial_probe;
@@ -390,6 +427,39 @@ async fn supervisor_loop(
}
}
}
changed = network_rx.changed() => {
if changed.is_err() {
// The session was dropped; nothing more to do.
return;
}
let new_state = *network_rx.borrow_and_update();
if new_state == NetworkState::Offline {
// OS says we're offline. Don't fabricate a
// loss outright — captive portals and
// transient flicker can produce false
// negatives — but pre-charge the watchdog
// so the next probe failure trips it
// immediately. This shrinks UI latency from
// ~15s to ~5s in the common case.
misses = misses.saturating_add(1).min(WATCHDOG_MAX_MISSES - 1);
info!(
target: "chanora_core",
misses,
"connectivity: OS reports offline; pre-charging watchdog"
);
} else if new_state == NetworkState::Online {
// Reset on confirmed online so we don't
// carry stale Offline charges into the
// healthy state.
if misses > 0 {
info!(
target: "chanora_core",
"connectivity: OS reports online; clearing watchdog misses"
);
}
misses = 0;
}
}
_ = watchdog.tick() => {
match tokio::time::timeout(WATCHDOG_PROBE_TIMEOUT, probe.probe()).await {
Ok(Ok(_)) => {
@@ -426,14 +496,6 @@ async fn supervisor_loop(
misses,
"watchdog: declaring connection lost"
);
// Replace the dead protocol client with a
// dropped slot so the reconnect path below
// doesn't accidentally keep using it. We
// also stop audio here (the reconnect path
// does this too, but doing it now ensures
// the mic / playback engine stops talking
// to a stale voice_out_tx as quickly as
// possible).
break chanora_protocol::DisconnectReason::Error(
"watchdog: server stopped responding".to_string(),
);
@@ -490,17 +552,46 @@ async fn supervisor_loop(
"reconnect: sleeping before next attempt"
);
// Sleep with cancellation support.
let slept = tokio::select! {
// Sleep with cancellation support. An OS
// "online" notification short-circuits the
// sleep and resets the attempt counter so the
// next outage starts with the smallest backoff
// window again.
enum SleepOutcome { Elapsed, NetworkUp, Cancelled }
let outcome = tokio::select! {
biased;
_ = &mut cancel_rx => {
_ = &mut cancel_rx => SleepOutcome::Cancelled,
changed = network_rx.changed() => {
if changed.is_err() {
SleepOutcome::Cancelled
} else if *network_rx.borrow_and_update() == NetworkState::Online {
SleepOutcome::NetworkUp
} else {
// Offline / Unknown transition — keep waiting the rest of the slot.
tokio::select! {
_ = &mut cancel_rx => SleepOutcome::Cancelled,
_ = tokio::time::sleep(Duration::from_secs(delay_secs as u64)) => SleepOutcome::Elapsed,
}
}
}
_ = tokio::time::sleep(Duration::from_secs(delay_secs as u64)) => SleepOutcome::Elapsed,
};
match outcome {
SleepOutcome::Cancelled => {
info!(target: "chanora_core", "supervisor cancelled during backoff");
return;
}
_ = tokio::time::sleep(Duration::from_secs(delay_secs as u64)) => true,
};
if !slept {
return;
SleepOutcome::NetworkUp => {
info!(
target: "chanora_core",
attempt,
"connectivity: OS reports online; redialling immediately"
);
// Reset attempt counter so future losses
// start the backoff schedule fresh.
attempt = 0;
}
SleepOutcome::Elapsed => {}
}
info!(target: "chanora_core", attempt, "reconnect: dialling");