Improve connection reliability (#18)

* Create minimal example

* Wip

* Connect signal

* Signal refactor

* Connect primary peer

* Remove dead code

* Use event queue approach

* Introduce timer

* Proper cleanup

* Implement reconnecting state

* Remove disconnecting state

* Reintroduce all features

* Add additional transitions

* Docs and organization

* Remove any state handler

* Change event memory model

* Store local participant ID

* Consolidate event detail cases

* Cleanup session state

* Engine refactor

- Handle all signal responses in state machine
- Helper functions for protocol messages

* Simplify state machine memory model

Always discard events

* Helper methods for data packets

* Allow engine to take ownership of data packet

* Process data packets via state machine

* Notify of engine state change

* Add documentation

* Allow handlers to take ownership of events

* Exponential backoff

* Create helper to enqueue events

* Zero initialize protobuf structs

* Ignore non-binary messages silently

* Make engine state internal

* Document

* Handle and expose failure reason

- Do not retry for client errors
- Expose failure reason through public API

* Extract into helpers

* Unify peer state change handling

* Free event if cannot enqueue

* URL build options struct

* Map disconnect reason to failure reason

* Send leave on user-initiated close

* Separate session state

* Reenable track subscription

* Decode tag on fail

* Ignore proto messages

* Add markers

* Fix state reporting bug

Optimizer causes issue where enum return value cannot be -1

* Improve signal state handling

- Signal state enum with failure reason
- Handle normal WS closure

* Remove unneeded includes

* Retry on peer disconnect in addition to fail

* Remove unneeded config

* Refactor signal client

- Handle ping timeout
- Use FreeRTOS timer instead of esp_timer (ms accuracy is sufficient)
- Ensure cleanup under abnormal closure conditions (e.g. WiFi disconnect)

* Cleanup signal client init

* Cleanup engine init

* Use protocol helper for encoding requests

* Use protocol helpers for encoding data packets

* Inline remaining protocol helpers

* Use consistent naming convention

* Make nanopb encode/decode includes private

* Simplify internal API

* Send data packets

* Zero only

* Remove state guards for sub audio

Audio stream can begin while still in connecting state

* Handle room and participant update in connecting state

Early participant update messages are necessary to subscribe to remote track

* Cleanup previous connection in backoff

Perform same cleanup steps when entering both disconnected and backoff

* Handle ICE servers

Reintroduce support for dynamically setting ICE servers, use hardcoded for now.

* Use SAFE_FREE macro

* Remove dead code

* Do not handle data packets in FSM

* Add basic connection time benchmarking

* Queue peer SDP messages

* Clean up

* Increase timer stack depth

* Fix typo

* Document how to get failure reason

* Data packet optimization

- Ignore types that are not supported yet
- Fixed length sender SID

* Signal response optimization

- Ignore types that are not supported yet

* Fix incorrect field type

* Fix voice agent example agent join

* Demonstrate getting failure reason in all examples

* Consistent sdkconfig defaults between examples

* Bump version

Author: ladvoc

components/livekit/CMakeLists.txt

@@ -8,7 +8,6 @@ idf_component_register(
     PRIV_REQUIRES
         esp_codec_dev
         esp_netif
-        esp_timer
         esp_websocket_client
         esp_webrtc
         json

components/livekit/Kconfig

@@ -0,0 +1,24 @@
+menu "LiveKit"
+    config LK_MAX_RETRIES
+        int "Maximum connection retries"
+        range 0 100
+        default 7
+    config LK_MAX_ICE_SERVERS
+        int "Maximum number of ICE servers"
+        default 3
+    config LK_BENCHMARK
+        bool "Benchmark connection time"
+        default n
+    config LK_ENGINE_QUEUE_SIZE
+        int "Number of engine events to queue"
+        default 32
+    config LK_PUB_INTERVAL_MS
+        int "How often to capture and send AV frames"
+        default 20
+    config LK_PUB_AUDIO_TRACK_NAME
+        string "Name of the published audio track"
+        default "Audio"
+    config LK_PUB_VIDEO_TRACK_NAME
+        string "Name of the published video track"
+        default "Video"
+endmenu

components/livekit/core/common.h

@@ -9,13 +9,13 @@
 extern "C" {
 #endif
 
-/// State of an engine component or the engine itself.
+/// State of a connection.
 typedef enum {
-    CONNECTION_STATE_DISCONNECTED = 0, /*!< Disconnected */
-    CONNECTION_STATE_CONNECTING   = 1, /*!< Establishing connection */
-    CONNECTION_STATE_CONNECTED    = 2, /*!< Connected */
-    CONNECTION_STATE_RECONNECTING = 3, /*!< Connection was previously established, but was lost */
-    CONNECTION_STATE_FAILED       = 4  /*!< Connection failed */
+    CONNECTION_STATE_DISCONNECTED = 0, /// Disconnected
+    CONNECTION_STATE_CONNECTING   = 1, /// Establishing connection
+    CONNECTION_STATE_CONNECTED    = 2, /// Connected
+    CONNECTION_STATE_RECONNECTING = 3, /// Connection was previously established, but was lost
+    CONNECTION_STATE_FAILED       = 4  /// Connection failed
 } connection_state_t;
 
 typedef struct {