harden reconnect behaviour

.changeset/lukas_reconnect.md

@@ -0,0 +1,8 @@
+---
+livekit: patch
+livekit-api: patch
+livekit-ffi: patch
+livekit-uniffi: patch
+---
+
+harden reconnect behaviour - #1148 (@lukasIO)

livekit-api/src/signal_client/mod.rs

@@ -215,9 +215,29 @@ impl SignalClient {
                 if matches!(&err, SignalError::WsError(WsError::Http(e)) if e.status() != 403) {
                     log::error!("unexpected signal error: {}", err.to_string());
                 }
-                let urls = RegionUrlProvider::fetch_region_urls(url, token).await?;
-                let mut last_err = err;
 
+                // Fetching region URLs is best-effort. `fetch_region_urls`
+                // already returns an empty list for non-cloud (direct /
+                // self-hosted) URLs, so those skip the fallback entirely. If the
+                // fetch itself fails (e.g. the region endpoint is unreachable),
+                // that must NOT be fatal: log a warning and fall back to the
+                // original connection error rather than masking it with the
+                // fetch error.
+                let urls = match RegionUrlProvider::fetch_region_urls(url, token).await {
+                    Ok(urls) => urls,
+                    Err(region_err) => {
+                        log::warn!(
+                            "failed to fetch region urls: {region_err}; surfacing original connection error"
+                        );
+                        return Err(err);
+                    }
+                };
+
+                // With no region URLs to try, this surfaces the original error.
+                // Otherwise we keep the most recent region attempt error, so that
+                // if every region fails the caller sees why the last region
+                // connection failed.
+                let mut last_err = err;
                 for url in urls.iter() {
                     log::info!("fallback connection to: {}", url);
                     match SignalInner::connect(url, token, options.clone(), publisher_offer.clone())
@@ -226,7 +246,7 @@ impl SignalClient {
                         Ok((inner, join_response, stream_events)) => {
                             return Ok(handle_success(inner, join_response, stream_events))
                         }
-                        Err(err) => last_err = err,
+                        Err(region_conn_err) => last_err = region_conn_err,
                     }
                 }
 

livekit-ffi-node-bindings/proto/room_pb.d.ts

@@ -81,6 +81,15 @@ export declare enum SimulateScenarioKind {
    * @generated from enum value: SIMULATE_FULL_RECONNECT = 7;
    */
   SIMULATE_FULL_RECONNECT = 7,
+
+  /**
+   * Asks the server to drop the signalling connection during the next resume,
+   * then triggers a resume locally. The resume cannot complete, so the engine
+   * escalates to a full reconnect — exercising the resume→full escalation path.
+   *
+   * @generated from enum value: SIMULATE_DISCONNECT_SIGNAL_ON_RESUME = 8;
+   */
+  SIMULATE_DISCONNECT_SIGNAL_ON_RESUME = 8,
 }
 
 /**

livekit-ffi-node-bindings/proto/room_pb.js

@@ -49,6 +49,7 @@ const SimulateScenarioKind = /*@__PURE__*/ proto2.makeEnum(
     {no: 5, name: "SIMULATE_FORCE_TCP"},
     {no: 6, name: "SIMULATE_FORCE_TLS"},
     {no: 7, name: "SIMULATE_FULL_RECONNECT"},
+    {no: 8, name: "SIMULATE_DISCONNECT_SIGNAL_ON_RESUME"},
   ],
 );
 

livekit-ffi/protocol/room.proto

@@ -84,6 +84,10 @@ enum SimulateScenarioKind {
   // Asks the server to send `LeaveRequest{Reconnect}`, forcing a full
   // reconnect (new RtcSession; SDK republishes existing local tracks).
   SIMULATE_FULL_RECONNECT = 7;
+  // Asks the server to drop the signalling connection during the next resume,
+  // then triggers a resume locally. The resume cannot complete, so the engine
+  // escalates to a full reconnect — exercising the resume→full escalation path.
+  SIMULATE_DISCONNECT_SIGNAL_ON_RESUME = 8;
 }
 message SimulateScenarioRequest {
   required uint64 room_handle = 1;

livekit-ffi/src/server/requests.rs

@@ -99,6 +99,9 @@ fn on_simulate_scenario(
         proto::SimulateScenarioKind::SimulateForceTcp => SimulateScenario::ForceTcp,
         proto::SimulateScenarioKind::SimulateForceTls => SimulateScenario::ForceTls,
         proto::SimulateScenarioKind::SimulateFullReconnect => SimulateScenario::FullReconnect,
+        proto::SimulateScenarioKind::SimulateDisconnectSignalOnResume => {
+            SimulateScenario::DisconnectSignalOnResume
+        }
     };
 
     let ffi_room = server.retrieve_handle::<room::FfiRoom>(request.room_handle)?.clone();

livekit/Cargo.toml

@@ -59,3 +59,4 @@ anyhow = "1.0.99"
 test-log = "0.2.18"
 test-case = "3.3"
 serial_test = "3.0"
+http = "1.1"