👻 ping-config: checking nodes for real

A group’s usual check is a HEAD to url. A node that answers the headers and then gets cut on the body (stalling after ~16–20 KB, say) passes it. ping-config adds a second check: after url the core asks destination, and the node is alive only when it answers in time. It is the counterpart of Xray’s burstObservatory.pingConfig.

proxy-groups:
  - name: auto
    type: url-test
    proxies: [node1, node2]
    url: https://www.gstatic.com/generate_204
    interval: 300
    timeout: 5000
    ping-config:
      destination: https://speed.cloudflare.com/__down?bytes=51200
      http-method: GET
      interval: 60s
      sampling: 1
      timeout: 3s
      connectivity: http://connectivitycheck.gstatic.com/generate_204

Field	What it does	Default
`destination`	the address of the second check	the group’s `url`
`http-method`	`HEAD` or `GET`; `GET` reads the whole body (up to 4 MiB)	`HEAD`
`interval`	how often checks run; `"15s"`, `"1m"` or a number of seconds; at least 10 s	the group’s `interval` → 300 s
`sampling`	how many times a cycle checks each node (1–32)	1
`timeout`	timeout of the `destination` request; `"3s"`, `"500ms"` or a number of milliseconds	the group’s `timeout` → 5000 ms
`connectivity`	an address checked DIRECT to see whether the device itself is online	off

Works in url-test, fallback, load-balance and smart. select makes no background checks: there the block only changes manual tests.

The field is Prizrak-Core’s only; mihomo does not have it.

How a check runs

The delay, as before: a HEAD to url, with the global unified-delay. The numbers keep their usual scale.
Then destination: the node is alive only when a 2xx/3xx answer comes within timeout and, with GET, the whole body arrives. A cut, a stall halfway or an error makes the node dead, even though its HEAD passed.
A destination on the address of url goes over the same connection; when the site closed it, over a new one. A HEAD to url itself is not sent twice.

One profile per url

ping-config belongs to the group’s url. Every check of that url follows it: from any group, from a provider health-check on the same url, from the API and the apps. When two groups share a url with different blocks, the one higher in the config wins; the other is ignored with a warning in the log.

Schedule

At start every node is checked at once.
Then a cycle lasts interval × sampling, and each node is checked sampling times at random moments of it — not all at once. At most 10 checks run together.
lazy and Prizrak-Box’s power saving on phones apply as to the usual checks.
A node in several groups or providers on the same url is checked once, not once for each.
A node just checked another way (a manual test, max-failed-times) is skipped once.
max-failed-times triggers an immediate check, by the profile too.

`sampling`

1 — the last check decides, as before.
> 1 — a window of two cycles decides: the node is alive if any check in it succeeded, and its delay is the mean of the successful ones.

`connectivity`

After a failed check the core makes one direct request past the VPN. With no answer the failure is not recorded, so a dropped Wi-Fi does not mark every node dead at once. The answer is cached for ~2 s.

API and apps

The group’s testUrl stays the plain url, so dashboards (zashboard and others) and Prizrak-Box work unchanged. The group JSON gains a pingConfig object when its url has a profile.