Files
UniversalTrain-ai/deploy/dds/config/mbus_config.xml
T
chenlin c32df70326
Checks / Layout, translation and Colab checks (push) Has been cancelled
Checks / Ruff (push) Has been cancelled
Checks / Framework tests (native backend, CPU) (push) Has been cancelled
Checks / Portable Rust controller (push) Has been cancelled
release: v0.1 generic robot training and application platform
2026-11-05 11:05:01 +08:00

150 lines
6.9 KiB
XML
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
<?xml version="1.0" encoding="UTF-8"?>
<CycloneDDS xmlns="https://cdds.io/config"
xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
xsi:schemaLocation="https://cdds.io/config https://raw.githubusercontent.com/eclipse-cyclonedds/cyclonedds/master/etc/cyclonedds.xsd">
<Domain Id="0">
<Compatibility>
</Compatibility>
<General>
<!-- 本域只走 lo:domain 0 是纯本机 IPC,跨机流量一律走 domain 2。
只绑 lo 的附带好处是本域彻底不受弱网影响,也不会把本机 topic
暴露到 wlan0/eth0 上。 -->
<Interfaces>
<NetworkInterface name="lo"/>
</Interfaces>
<!-- lo 在 Linux 上没有 IFF_MULTICAST,组播本来就用不了,如实写 false,
免得看配置的人以为本域还有组播发现这条路。 -->
<AllowMulticast>false</AllowMulticast>
</General>
<Discovery>
<!-- 组播在 lo 上不可用,因此这个 peer 是本域唯一的发现手段(不再是
"wlan0/eth0 没起来时的兜底")。PruneDelay=inf 让本机发现端口被
一直探测:否则晚于 InitialLocatorPruneDelay(30s) 启动的进程,
会因为 peer 已被剪掉而永远发现不了先启动的 participant。 -->
<Peers>
<Peer Address="127.0.0.1" PruneDelay="inf"/>
</Peers>
<LeaseDuration>10s</LeaseDuration>
</Discovery>
<Internal>
<HeartbeatInterval>2s</HeartbeatInterval>
</Internal>
<Sizing>
</Sizing>
<Threads>
</Threads>
<Tracing>
<!-- <Verbosity>finer</Verbosity>
<OutputFile>cyclonedds_${CYCLONEDDS_PID}.log</OutputFile> -->
</Tracing>
</Domain>
<!-- Domain 1: Video streaming — shared memory for local, UDP for remote -->
<Domain Id="1">
<SharedMemory>
<Enable>true</Enable>
<LogLevel>info</LogLevel>
</SharedMemory>
<General>
<Interfaces>
<NetworkInterface name="lo"/>
<!-- <NetworkInterface name="wlan0" presence_required="false"/> -->
<!-- <NetworkInterface name="eth0" presence_required="false"/> -->
</Interfaces>
<AllowMulticast>true</AllowMulticast>
</General>
<Discovery>
<!-- Probe local discovery ports forever: a process started before
wlan0/eth0 is up selects only lo, which has no multicast on Linux;
without a never-pruned localhost peer it can never discover
participants started more than InitialLocatorPruneDelay (30s) later. -->
<LeaseDuration>10s</LeaseDuration>
</Discovery>
<Internal>
<HeartbeatInterval>2s</HeartbeatInterval>
</Internal>
<Tracing>
</Tracing>
</Domain>
<!-- Domain 2: 跨机弱网域(只走 wlan0/eth0,无 lo)。robot_control 跑在这里,
QoS 是 RELIABLE + KEEP_ALL(见 mbus_qos.xml 的 RobotControl_Control)。
QoS 只声明"要可达",能不能在下一个控制周期内把丢的帧补回来,取决于本段。 -->
<Domain Id="2">
<Compatibility>
</Compatibility>
<General>
<Interfaces>
<NetworkInterface name="wlan0" presence_required="false"/>
<NetworkInterface name="eth0" presence_required="false"/>
</Interfaces>
<!-- spdp:组播只用于 SPDP 发现,SEDP 和用户数据一律走单播。
Wi-Fi 的组播帧以最低基础速率发送且没有 802.11 层 ARQ(收不到不重发),
弱信号下丢包率远高于单播;RELIABLE 虽然能靠 RTPS 重传补回来,但那是
用一个 NACK 往返去补一条本来就不该丢的帧。改成单播后 802.11 自己的
重传先兜一层,RTPS 重传只处理漏网的。
本域加这一段的初衷是"多播发现",spdp 正好只保留发现所需的组播。 -->
<AllowMulticast>spdp</AllowMulticast>
</General>
<Discovery>
<!-- Probe local discovery ports forever: a process started before
wlan0/eth0 is up selects only lo, which has no multicast on Linux;
without a never-pruned localhost peer it can never discover
participants started more than InitialLocatorPruneDelay (30s) later. -->
<LeaseDuration>10s</LeaseDuration>
</Discovery>
<Internal>
<!-- 弱网重传参数,三个默认值会直接让 RELIABLE 形同虚设:
- HeartbeatInterval:reader 靠 HEARTBEAT 才知道自己缺样本。数据流
连续时 HB 会搭在 DATA 上,但控制流一停(松手静止),最后几帧的
缺失就要等下一个周期心跳 —— 2s 意味着最坏 2s 后才补上,等于
丢了 200 个控制周期。回到 CycloneDDS 默认 100ms,并把周期心跳
收敛到 [20ms, 500ms]。
- NackDelay:reader 发现缺样本后延迟多久发 NACK,默认 100ms
= 10 个控制周期,压到 10ms。
- AutoReschedNackDelay:NACK 或其响应本身丢了之后的重试间隔,
默认 3s,弱网下这是很可能发生的,压到 500ms。
- Watermarks:KEEP_ALL 写端积压超过 WhcHigh 才背压。WhcAdaptive
默认开启,会在重传期间把 WhcHigh 一路收缩到 WhcLow(默认 1kB
≈ 4 帧),正好在丢包时把实时控制环卡在 max_blocking_time 上。
这里关掉自适应并给固定水位:一帧 Control 约 256B(含 RTPS 开销),
WhcHigh 64kB ≈ 250 帧 ≈ 2.5s 重传窗口,再长就是拿过期指令喂运控。
注意 WhcAdaptive=false 时 whc_high 恒等于 WhcHighInit,两者必须
一起设;且 WhcHighInit 必须落在 [WhcLow, WhcHigh] 内,否则报
"Invalid watermark settings",participant 直接建不出来。 -->
<HeartbeatInterval min="20ms" max="500ms">100ms</HeartbeatInterval>
<NackDelay>10ms</NackDelay>
<AutoReschedNackDelay>500ms</AutoReschedNackDelay>
<Watermarks>
<WhcLow>8 kB</WhcLow>
<WhcHigh>64 kB</WhcHigh>
<WhcHighInit>64 kB</WhcHighInit>
<WhcAdaptive>false</WhcAdaptive>
</Watermarks>
</Internal>
<Sizing>
</Sizing>
<Threads>
</Threads>
<Tracing>
<!-- <Verbosity>finer</Verbosity>
<OutputFile>cyclonedds_${CYCLONEDDS_PID}.log</OutputFile> -->
</Tracing>
</Domain>
</CycloneDDS>