advertise_internal_ip: true on a use_external_ip SFU leaves some subscribers with zero downlink frames (v1.13.7)
维护者通常 1 天内回复
还没有人认领这个 Issue。
评估
- 难度
- 4/5
- 预计耗时
- 3-5 天
- 新手友好度
- 45/100
- Issue 类型
- 缺陷
- 描述清晰度
- 基本清楚
- 活跃度
- 活跃
- 技术栈
- aws, go, kubernetes
- 领域
- backend, networking
调研方向
Start by tracing how rtc.advertise_internal_ip and rtc.use_external_ip affect ICE candidate generation, then inspect candidate-pair logs for the SFU and Python SDK to compare affected and unaffected sessions. Reproduce with the listed v1.13.7 configuration and participant setup; done means subscribers receive downlink frames while the private candidate is advertised alongside the public one.
由索引模型根据 Issue 内容生成。
描述
Describe the bug
With rtc.advertise_internal_ip: true (new in the config package pinned by v1.13.7) on a use_external_ip: true SFU, some participants connect and publish normally but receive no downlink media at all: the SFU creates the downtrack and the publisher keeps sending, yet zero frames reach the subscriber. Removing the option restored downlink immediately. Not root-caused; reporting because it blocks the obvious workaround for Go SDK participants behind NAT (see https://github.com/livekit/livekit/issues/4931).
Server
- Version: v1.13.7
- Environment: Kubernetes on AWS,
hostNetworkSFU pods,rtc.use_external_ip: true,rtc.interfaces.includesset to the primary interface, fixedudp_port,tcp_port: 0. The only candidate-list change with the option is an extra privatehost 10.x.x.x:<udp_port>candidate on the same socket, appended next to the public one.
Client
- Python SDK 1.1.13: every session on the flagged SFU pods hit a 120 s no-frame timeout (171/171 sessions OK on identical pods without the option).
- JS SDK: clients on some networks were affected (session counts up 2x with much shorter sessions, consistent with retries); JS clients on other networks and all Rust-SDK (libwebrtc) participants were unaffected.
To Reproduce
- livekit-server v1.13.7 with
use_external_ip: trueandadvertise_internal_ip: true. - Publish from one participant, subscribe from a Python SDK 1.1.13 participant over the internet.
- Subscriber connects, publisher stats show frames sent, subscriber receives none.
Expected behavior
Participants that cannot reach the private candidate ignore it and keep receiving media over the public one.
Additional context
Observed on two independent clusters within minutes of the rollout, with fresh API keys; reverted the same day. We can collect ICE candidate pair logs from the SFU and the Python client if useful.
- 主要语言
- Go
- 星标
- 21.3k
- 派生
- 2.4k
- 平均合并
- 17 小时 59 分钟
- 30 天内合并 PR
- 74
环境准备
- 提供 Dockerfile 或 Docker Compose 文件
- 没有 Pull Request 模板
- 没有贡献指南
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
livekit/livekit 的其他 Issue
-
难度 4/5 3-5 天 新手友好度 44/100
livekit/livekit#4982 · 1 条评论 ·
维护者通常 1 天内回复
-
Updating a SIP dispatch rule or inbound trunk to new trunks or numbers skips the conflict check可能已有人在做 @RaphaelFakhri 于 1 天前认领。 未关闭
难度 3/5 1-2 天 新手友好度 25/100
维护者通常 1 天内回复
-
难度 4/5 3-5 天 新手友好度 12/100
维护者通常 1 天内回复
-
Abandoned pending tracks are never released, so long-lived participants permanently hit too many pending tracks可能已有人在做 @kevin-pan-skydio 于 2 天前认领。 未关闭
难度 4/5 3-5 天 新手友好度 22/100
维护者通常 1 天内回复
-
难度 4/5 3-5 天 新手友好度 52/100
livekit/livekit#4959 · 1 条评论 ·
维护者通常 1 天内回复
相似的 Issue
-
难度 2/5 1-3 小时 新手友好度 84/100
维护者通常 1 天内回复
-
accepted bug wg/agentic-context
难度 2/5 1-3 小时 新手友好度 82/100
vllm-project/semantic-router#4922 · 1 条评论 ·
维护者通常 1 天内回复
-
[Bug] Extraction-mode autogen returns pod templates in random order for map-keyed CRDs, so deny and report messages change between runs可能已有人在做 @ManojLamani 今天认领。 未关闭
难度 2/5 1-3 小时 新手友好度 75/100
kyverno/kyverno#17980 · 1 个 reaction ·
维护者通常 1 天内回复
-
raised-by:reviewer
难度 1/5 1-3 小时 新手友好度 84/100
medici-finance/assay#2681 ·
维护者通常 1 天内回复
-
bug triaged
难度 2/5 1-3 小时 新手友好度 78/100
fullsend-ai/fullsend#8353 · 4 条评论 · 1 个 reaction ·
维护者通常 1 天内回复