fix: 0-segment retry loop, event bus isolation, and dashboard snapshot serve by sheepweb · Pull Request #117 · RikaCelery/XhRec

sheepweb · 2026-06-21T12:30:29Z

Summary

Clean PR from the sheepweb/XhRec fork. Cherry-picked 7 business commits from origin/main, excluding CCG tooling commits.

Commits

#	SHA	Description
1	`a9bc9be`	prevent zombie re-arm loop for rooms producing 0 segments
2	`b86b8b6`	replace never-retry with exponential backoff for 0-segment failures
3	`526f725`	use fixed 30s retry interval instead of exponential backoff
4	`2e04b21`	indefinite 30s retry, stop when room deactivated
5	`eb5027f`	keep 30s retries until room becomes unrecordable
6	`ec81671`	keep request bus commands off lossy event stream
7	`c7ceaf1`	serve dashboard from component snapshots

Changes

9 source files, +118 / −16 — all under src/:

SchedulerComponent.kt — 0-segment retry lifecycle (5 iterative fixes)
SessionComponent.kt — zombie loop guard
Events.kt — event type refinements
EventBus.kt + EventBusTest.kt — request bus command isolation from lossy event stream
Main.kt, HttpServerComponent.kt, MetricComponent.kt, RoomComponent.kt — dashboard snapshot serving

Verification

Branch from upstream/main (3a30f4a) — 7 commits cherry-picked cleanly
CCG commits (0b302a2, 60a4b5b) and .ccg/ directory excluded (zero .ccg paths in diff)
PR check: runs ./gradlew test — EventBusTest included in this change set

RecordingStopped now carries segmentsDispatched count from pollingLoop. SchedulerComponent only re-arms when >0 segments were actually dispatched. Rooms that return 404 before any download (index=-1 / 0 segments) wait for RoomStatusChanged instead of entering an infinite start-stop cycle. Also fixes silent session discard: startSession no longer silently returns when stuck in Fetching state — it cancels the stale pollingJob and allows re-entry. configureSession failure now publishes RecordingStopped(0) to properly notify the scheduler.

0-segment RecordingStopped now retries with backoff (30s/60s/120s/240s/480s) instead of permanently giving up. Max 5 attempts, then waits for RoomStatusChanged. This prevents rooms that are genuinely public but experience transient CDN unavailability from being stuck forever, while still avoiding the infinite zombie loop.

…0-segment failures

…tivated Removed max retry limit and exponential backoff. 0-segment RecordingStopped now retries every 30s indefinitely. The loop terminates naturally when the room is deactivated or the system shuts down - no artificial cap needed.

Track rooms currently in a recordable state (public, or groupShow with autopay). 0-segment retries now continue every 30s while the room remains recordable and stop automatically once RoomStatusChanged reports an unrecordable state or the room is deactivated.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

RikaCelery

有些问题，感觉没有找到核心原因，再考虑一下

RikaCelery · 2026-06-28T02:36:55Z

        val client = ClientManager.getProxiedClient("api")
        val response = withRetry(3) {
-            client.get("https://stripchat.com/api/front/v3/config/initial") {
+            client.get("https://poplive.xyz/api/front/v3/config/initial") {


自己验证一下API端点通不通，需不需要额外验证

RikaCelery · 2026-06-28T02:38:13Z

            try {
                val client = ClientManager.getProxiedClient("event")
-                client.webSocket("wss://websocket-v6.xhamsterlive.com/connection/websocket") {
+                client.webSocket("wss://websocket-v6.poplive.xyz/connection/websocket") {


连不上，再确认一下poplive和xhamsterlive到底是不是一样，尤其是直播间内消息和流地址

RikaCelery · 2026-06-28T02:41:44Z

    }

+    fun snapshotRooms(): List<Room> = rooms.values.map { it.copy() }
+


这种不要单独开一个函数然后注入HttpComponent，eventBus的存在就是尽可能降低组件之间来回注入的。
如果觉得会导致Bus卡住应该在命令收发侧去调试, 或者设计全异步的命令处理流程

RikaCelery · 2026-06-28T02:42:10Z

 ) : Actor<SchedulerMsg>("SchedulerComponent", eventBus, parentScope) {

    private val armed = ConcurrentHashMap<Long, ArmedRoom>()
+    private val recordableRooms = ConcurrentHashMap.newKeySet<Long>()


加这个的作用是什么？

看了一圈这个变量除了记录当前状态还有别的做用吗。

Scheduler不需要缓存当前房间是否可以录制，只需要在每次尝试启动Session之前判断是否可以录制即可

RikaCelery · 2026-06-28T02:43:03Z

        val client = ClientManager.getProxiedClient("api")
        val response = withRetry(3, stopIf = { false }) {
-            client.post("https://stripchat.com/api/front/show/models/$roomId/groupShows/${user.userId}") {
+            client.post("https://poplive.xyz/api/front/show/models/$roomId/groupShows/${user.userId}") {


RikaCelery · 2026-06-28T02:45:36Z

        return sb.toString()
    }
+
+    fun snapshotRoomDetailedStatus(): Map<Long, Map<String, Any>> =


再考虑一下，不要绕开EventBus，这点消息量完全处理的过来

RikaCelery · 2026-06-28T02:55:46Z

                    }
                } else {
-                    logger.debug("Recording stopped for room {}", event.roomId)
+                    logger.info(


这里当时设计的是让Scheduler启动一次录制会话（Session），只要Scheduler没说停止录制Session就应该一直尝试去录制。

也就是说，Session要自己维护一个状态机，处理好启动失败时候的重试。如果session决定不再重试了，要告诉Scheduler，不然Scheduler会认为Session依然在录制。

RikaCelery · 2026-06-28T02:58:51Z

        if (existing != null) {
            val blocked = when (existing.state) {
-                SessionState.Fetching, SessionState.Recording -> true
+                SessionState.Fetching -> {


这算一个防止重复启动的逻辑，并不确定上层是否会同时启动两次。

这里只需要确保一次startSession中，state一定会从Fetching迁移到Recording或者Closing即可，不需要在第二次调用的时候停止第一次的

RikaCelery · 2026-06-28T02:59:51Z


                    try {
-                        const res = await fetch(`https://zh.xhamsterlive.com/api/front/v1/broadcasts/${modelName}`);
+                        const res = await fetch(`https://poplive.xyz/api/front/v1/broadcasts/${modelName}`);


跟上面的一样，再确认一下

RikaCelery · 2026-06-28T03:01:28Z

-# https://stripchat.com/modelA q:720p limit:120
-; https://stripchat.com/modelB q:240p
-https://stripchat.com/modelC q:highest
+# https://poplive.xyz/modelA q:720p limit:120


如果改了这里，注意Bootstrap组件加载房间的时候会不会有问题

但本来项目就是为了xhamster写的，我觉得不改README比较好

sheepweb and others added 9 commits June 21, 2026 20:29

fix: use fixed 30s retry interval instead of exponential backoff for …

801adeb

…0-segment failures

fix: keep request bus commands off lossy event stream

eddfa77

fix: serve dashboard from component snapshots

d82fdae

style: remove trailing blank line in events

881cc2f

chore: replace stripchat.com / xhamsterlive.com with poplive.xyz

142fc0b

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

This comment was marked as outdated.

Sign in to view

RikaCelery self-requested a review June 28, 2026 02:34

RikaCelery requested changes Jun 28, 2026

View reviewed changes

sheepweb closed this Jul 15, 2026

		}

		fun snapshotRooms(): List<Room> = rooms.values.map { it.copy() }

Conversation

sheepweb commented Jun 21, 2026

Summary

Commits

Changes

Verification

Uh oh!

This comment was marked as outdated.

Uh oh!

RikaCelery left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants