ComfyUI 连接检测
Is the other ComfyUI even awake? This node asks before you waste a run
- 是否可连接
- 状态码
- 消息
- 耗时_ms
- 地址
The whole idea behind ComfyUI-RemoteCall is that one ComfyUI instance drives another. You build a small "controller" workflow on machine A, and it uploads files, submits jobs, and pulls results from machine B across the network. Before you do any of that, you want to know B is actually up - the address isn't a typo, the port isn't closed, the box didn't fall asleep. That's this node's entire job, and it does it with one HTTP GET.
The display name is 连接检测 (Connection Check). It's the pack's health probe, and it's the node you'll drop in first when you're standing up a two-instance rig, because it costs you one tiny request and it can save you a long confusing debugging session.
How it works
Under the hood it's embarrassingly simple, and that's the point. You give it the target address and a timeout; it normalizes the URL (adds http:// if you left it off, strips trailing slashes), then hits GET {address}/system_stats with a ComfyUI-RemoteCall user-agent. Any real ComfyUI answers that with HTTP 200. Then it hands you the verdict plus the numbers to prove it.
The inputs that matter
- 地址 (address) - the other instance's base URL, like
http://192.168.1.20:8188or a Tailscale URL. There's no default, so the first thing you do is type one in. - 超时_ms (timeout) - default 3000, adjustable from 500 to 20000 ms. If the box hangs, a short timeout means you fail fast instead of staring at a spinner. For a remote box on a flaky link, bump it up.
The outputs
- 是否可连接 (BOOLEAN) - the verdict, and it's a real boolean, so you can branch on it.
- 状态码 (INT) - the HTTP status code, or
-1when the address is invalid or unreachable. - 消息 (STRING) -
连接成功,连接超时, or无法连接: .... - 耗时_ms (INT) - how long the probe took, useful when you're comparing hosts.
- 地址 (STRING) - the cleaned, normalized URL. This is the sneaky useful one: wire it straight into the 地址 socket on 上传提交 (Upload Submit) and 查询执行结果 (Query Result), and you only type the address once per workflow instead of three times.
Installing it
It ships in the ComfyUI-RemoteCall pack. Easiest route: ComfyUI Manager → Install Custom Nodes → search ComfyUI-RemoteCall (or the Chinese name 远程调用), then restart. Manual route:
cd ComfyUI/custom_nodes
git clone https://github.com/playboy-dongan/ComfyUI-RemoteCall
Dependencies are just requests and av (the README and requirements.txt agree; both usually ride along with a normal ComfyUI install). There are no model files to download - it's pure plumbing.
When it's wrong
A 200 on /system_stats is the whole success test. Failure shows up as 状态码 -1 with a 无法连接 message, which covers a wrong port, a firewall, the target not started, or an instance that's only listening on localhost while you point it at a LAN address. One thing to know: if the target runs an auth layer (like ComfyUI's --listen plus an authentication middleware), the probe comes back with a 401 or 403 - the instance is up, but your controller can't get in, which is a different and useful answer to have.
It's not a glamorous node, and it shouldn't be. It answers one question and answers it cheaply, which is exactly what you want in the first five minutes of a remote-call setup.
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| 地址 | STRING | — | |
| 超时_ms | INT | 3000500–20000 | — |
Outputs (5)
| Name | Type | Description |
|---|---|---|
| 是否可连接 | BOOLEAN | — |
| 状态码 | INT | — |
| 消息 | STRING | — |
| 耗时_ms | INT | — |
| 地址 | STRING | — |