fix(api): share one probe-target resolution between both proxy health checks (#10657)

Merged — locally validated (15/15 focused tests, typecheck:core clean, file-size/changelog gates green) after resolving base-drift against #10654 (both landed today, same file — combined import block, no logical conflict). Thanks!
This commit is contained in:
Dizzle
2026-08-20 15:32:38 +02:00
committed by GitHub
parent b43ad73166
commit f4772500bc
6 changed files with 289 additions and 11 deletions

View File

@@ -1988,6 +1988,13 @@ APP_LOG_TO_FILE=true
# Reachability probe target for the scheduler and the auto-test endpoint.
# Point it at an internal/self-hosted URL to avoid the public default.
# PROXY_HEALTH_TEST_URL=https://httpbin.org/ip
# Probes started at once per batch, for the scheduler and the auto-test endpoint.
# Floored at 1 and capped at 50. Default: 10.
# PROXY_HEALTH_TEST_CONCURRENCY=10
# Delay in ms between two probe departures inside a batch. Without it the whole batch
# leaves at once and a shared egress IP can trip a rate-limited target. 0 disables the
# spacing; capped at 5000. Default: 100.
# PROXY_HEALTH_TEST_STAGGER_MS=100
# Set "true" to let the scheduler auto-remove proxies after repeated failures.
# PROXY_AUTO_REMOVE=false
# Consecutive failures before an auto-remove fires. Default: 3.

View File

@@ -1032,7 +1032,9 @@ Anthropic-compatible provider instead.
| `PROXY_HEALTH_UNHEALTHY_CACHE_TTL_MS` | `2000` | `src/lib/proxyHealth.ts` | Cache TTL for failed proxy health probes. Keep this shorter than `PROXY_HEALTH_CACHE_TTL_MS` so transient proxy timeouts under high concurrency retry quickly without disabling fast-fail for truly dead proxies. |
| `PROXY_HEALTH_ENABLED` | `true` | `src/lib/proxyHealth/scheduler.ts` | Set `false` to disable the background proxy health scheduler that periodically probes registered proxies. |
| `PROXY_HEALTH_INTERVAL_MS` | `600000` | `src/lib/proxyHealth/scheduler.ts` | Background health-scheduler sweep interval in ms (minimum `60000`). |
| `PROXY_HEALTH_TEST_URL` | `https://httpbin.org/ip` | `src/lib/proxyHealth/scheduler.ts` | Reachability probe target used by the scheduler and the `/api/settings/proxies/auto-test` endpoint. Point it at an internal/self-hosted URL to avoid the public default. |
| `PROXY_HEALTH_TEST_URL` | `https://httpbin.org/ip` | `src/lib/proxyHealth/probeTarget.ts` | Reachability probe target used by the scheduler and the `/api/settings/proxies/auto-test` endpoint. Point it at an internal/self-hosted URL to avoid the public default. |
| `PROXY_HEALTH_TEST_CONCURRENCY` | `10` | `src/lib/proxyHealth/probeTarget.ts` | Probes started at once per batch, shared by the scheduler and the `/api/settings/proxies/auto-test` endpoint. Floored at 1 and capped at 50. |
| `PROXY_HEALTH_TEST_STAGGER_MS` | `100` | `src/lib/proxyHealth/probeTarget.ts` | Delay in ms between two probe departures inside a batch. Without it the whole batch leaves at the same moment and a shared egress IP can trip a rate-limited target. Set to `0` to disable the spacing; capped at 5000. |
| `PROXY_HEALTH_AUTO_DEACTIVATE` | `false` | `src/lib/proxyHealth/statusPolicy.ts` | When `false` (default), automated reachability probes (the scheduler + the `/api/settings/proxies/auto-test` "Test All" button) are **read-only** and never write a proxy's status — only the operator sets active/inactive, so a flaky probe can't strand an assigned proxy (#6246). Set `true` to restore the legacy test-and-set behaviour. |
| `PROXY_AUTO_REMOVE` | `false` | `src/lib/proxyHealth/scheduler.ts` | Set `true` to let the scheduler auto-remove proxies after repeated consecutive failures. |
| `PROXY_AUTO_REMOVE_AFTER` | `3` | `src/lib/proxyHealth/scheduler.ts` | Consecutive failures before the scheduler auto-removes a proxy (when `PROXY_AUTO_REMOVE=true`). |

View File

@@ -6,14 +6,20 @@ import { createProxyDispatcher, proxyConfigToUrl } from "@omniroute/open-sse/uti
import { fetch as undiciFetch } from "undici";
import { classifyProbeStatus } from "@/lib/proxyHealth/decision";
import { resolveHealthCheckStatusWrite } from "@/lib/proxyHealth/statusPolicy";
import {
resolveProbeConcurrency,
resolveProbeStaggerMs,
resolveProbeTarget,
waitForProbeSlot,
} from "@/lib/proxyHealth/probeTarget";
import { isValidationFailure, validateBody } from "@/shared/validation/helpers";
import { createErrorResponse } from "@/lib/api/errorResponse";
const TEST_TIMEOUT_MS = 5000;
// Reachability probe target. Configurable so operators can point it at an
// internal/self-hosted endpoint instead of the public default.
const TEST_URL = process.env.PROXY_HEALTH_TEST_URL || "https://httpbin.org/ip";
const CONCURRENCY = 10;
// Shared with the background sweep — see src/lib/proxyHealth/probeTarget.ts.
const TEST_URL = resolveProbeTarget();
const CONCURRENCY = resolveProbeConcurrency();
const STAGGER_MS = resolveProbeStaggerMs();
const autoTestSchema = z.object({
ids: z.array(z.string()).optional(),
@@ -149,7 +155,14 @@ export async function POST(request: Request) {
const results: TestResult[] = [];
for (let i = 0; i < proxiesToTest.length; i += CONCURRENCY) {
const batch = proxiesToTest.slice(i, i + CONCURRENCY);
const batchResults = await Promise.allSettled(batch.map((proxy) => testSingleProxy(proxy)));
const batchResults = await Promise.allSettled(
batch.map(async (proxy, indexInBatch) => {
// Same intra-batch spacing as the background sweep: "Test All" fires the whole
// batch at once too, so it is just as capable of tripping a rate-limited target.
await waitForProbeSlot(indexInBatch, STAGGER_MS);
return testSingleProxy(proxy);
})
);
for (const result of batchResults) {
if (result.status === "fulfilled") results.push(result.value);
}

View File

@@ -0,0 +1,84 @@
/**
* Shared resolution of the reachability-probe parameters (#8411).
*
* The scheduler sweep and the bulk "Test All" endpoint each carried their own copy of the
* probe target and batch size. Both now resolve through here, so an operator tunes one
* surface instead of two that can silently drift apart.
*
* The resolvers are pure and take the environment as a parameter, so tests never have to
* mutate `process.env`.
*/
import { sleep } from "@omniroute/open-sse/utils/sleep";
export const DEFAULT_PROBE_TARGET = "https://httpbin.org/ip";
export const DEFAULT_PROBE_CONCURRENCY = 10;
export const DEFAULT_PROBE_STAGGER_MS = 100;
/**
* Upper bounds. Making the batch size configurable without a ceiling would let a single
* env var recreate the very probe storm this module exists to damp.
*/
export const MAX_PROBE_CONCURRENCY = 50;
export const MAX_PROBE_STAGGER_MS = 5000;
type ProbeEnv = Record<string, string | undefined>;
function resolveBoundedInt(raw: string | undefined, fallback: number, min: number, max: number) {
const parsed = parseInt(raw ?? "", 10);
if (!Number.isFinite(parsed)) return fallback;
return Math.min(Math.max(parsed, min), max);
}
/**
* Deliberately keeps the historical `||` semantics: a target that is empty falls back to the
* default, but one made only of whitespace is passed through untouched. Trimming it here would
* silently change how an existing deployment behaves.
*/
export function resolveProbeTarget(env: ProbeEnv = process.env): string {
return env.PROXY_HEALTH_TEST_URL || DEFAULT_PROBE_TARGET;
}
/** Floored at 1: a zero batch size would make the `i += concurrency` loop never advance. */
export function resolveProbeConcurrency(env: ProbeEnv = process.env): number {
return resolveBoundedInt(
env.PROXY_HEALTH_TEST_CONCURRENCY,
DEFAULT_PROBE_CONCURRENCY,
1,
MAX_PROBE_CONCURRENCY
);
}
export function resolveProbeStaggerMs(env: ProbeEnv = process.env): number {
return resolveBoundedInt(
env.PROXY_HEALTH_TEST_STAGGER_MS,
DEFAULT_PROBE_STAGGER_MS,
0,
MAX_PROBE_STAGGER_MS
);
}
/**
* Delay before the Nth probe of a batch starts.
*
* A batch fires `Promise.allSettled(batch.map(...))`, so without this every probe leaves at the
* same tick and a shared egress IP hits the target with `concurrency` simultaneous requests.
* Spacing the departures is what removes that spike; the delay is a plain multiple of the index
* rather than a random jitter so a sweep stays reproducible and exactly testable.
*
* The first probe of a batch always returns 0 — no batch is ever slowed down at its head.
*/
export function staggerDelayMs(indexInBatch: number, stepMs: number): number {
if (indexInBatch <= 0 || stepMs <= 0) return 0;
return indexInBatch * stepMs;
}
/**
* Hold a probe back until its slot in the batch. Call this from the batch `map`, before the
* probe itself: both call sites arm their timeout inside their own test function, so waiting
* out here is what keeps every probe's timeout budget whole.
*/
export async function waitForProbeSlot(indexInBatch: number, stepMs: number): Promise<void> {
const delay = staggerDelayMs(indexInBatch, stepMs);
if (delay > 0) await sleep(delay);
}

View File

@@ -34,16 +34,24 @@ import {
decideProxyHealthAction,
type ProxyProbeOutcome,
} from "./decision.ts";
import {
resolveProbeConcurrency,
resolveProbeStaggerMs,
resolveProbeTarget,
waitForProbeSlot,
} from "./probeTarget.ts";
// #6246: a HEAD to the public probe target through a legit (often loaded) proxy
// can exceed a few seconds; the old 5s ceiling produced false negatives that
// flipped healthy proxies to inactive. Raise it and treat our own timeout as
// inconclusive (see testOneProxy) rather than a proxy failure.
const TEST_TIMEOUT_MS = 15000;
// Reachability probe target for proxy health checks. Configurable so operators
// can point it at an internal/self-hosted endpoint instead of the public default.
const TEST_URL = process.env.PROXY_HEALTH_TEST_URL || "https://httpbin.org/ip";
const CONCURRENCY = 10;
// Probe target, batch size and intra-batch spacing come from probeTarget.ts, which the
// auto-test endpoint reads too — one surface to tune instead of two that can drift apart.
// Resolved at module load, as these constants always were.
const TEST_URL = resolveProbeTarget();
const CONCURRENCY = resolveProbeConcurrency();
const STAGGER_MS = resolveProbeStaggerMs();
const INITIAL_DELAY_MS = 60_000;
const DEFAULT_INTERVAL_MS = 600_000;
const DEFAULT_REMOVE_AFTER = 3;
@@ -160,7 +168,10 @@ async function sweep(): Promise<void> {
for (let i = 0; i < proxies.length; i += CONCURRENCY) {
const batch = proxies.slice(i, i + CONCURRENCY);
const results = await Promise.allSettled(
batch.map(async (proxy) => {
batch.map(async (proxy, indexInBatch) => {
// Spread the departures: without this the whole batch leaves at the same tick and a
// shared egress IP hits the target with CONCURRENCY simultaneous requests.
await waitForProbeSlot(indexInBatch, STAGGER_MS);
const outcome = await testOneProxy(proxy);
return { id: proxy.id, outcome };
})

View File

@@ -0,0 +1,161 @@
/**
* Guards for the shared probe-target resolution (#8411).
*
* The scheduler sweep and the "Test All" endpoint used to carry their own copy of the probe
* target and batch size. They now share src/lib/proxyHealth/probeTarget.ts, so these tests pin
* two things: that the defaults still match what both call sites hardcoded before, and that no
* environment value can produce a batch size that would hang the sweep loop.
*/
import test from "node:test";
import assert from "node:assert/strict";
import {
DEFAULT_PROBE_CONCURRENCY,
DEFAULT_PROBE_STAGGER_MS,
DEFAULT_PROBE_TARGET,
MAX_PROBE_CONCURRENCY,
MAX_PROBE_STAGGER_MS,
resolveProbeConcurrency,
resolveProbeStaggerMs,
resolveProbeTarget,
staggerDelayMs,
waitForProbeSlot,
} from "../../src/lib/proxyHealth/probeTarget.ts";
// ─── the loop-safety invariant ─────────────────────────
test("no environment value can yield a batch size below 1", () => {
// `for (i = 0; i < n; i += concurrency)` never advances at 0 and walks backwards below it,
// so this is the one property that turns a bad config into a hung sweep rather than a slow one.
const hostile = [
"0",
"-1",
"-9999",
"",
" ",
"abc",
"NaN",
"Infinity",
"-Infinity",
"1e-9",
"0.4",
"null",
undefined,
];
for (const raw of hostile) {
const resolved = resolveProbeConcurrency({ PROXY_HEALTH_TEST_CONCURRENCY: raw });
assert.ok(
Number.isInteger(resolved) && resolved >= 1,
`concurrency ${JSON.stringify(raw)} resolved to ${resolved}`
);
}
});
// ─── defaults: what both call sites hardcoded before ───
test("the defaults reproduce the previous hardcoded values", () => {
assert.equal(resolveProbeTarget({}), DEFAULT_PROBE_TARGET);
assert.equal(DEFAULT_PROBE_TARGET, "https://httpbin.org/ip");
assert.equal(resolveProbeConcurrency({}), DEFAULT_PROBE_CONCURRENCY);
assert.equal(DEFAULT_PROBE_CONCURRENCY, 10);
assert.equal(resolveProbeStaggerMs({}), DEFAULT_PROBE_STAGGER_MS);
});
// ─── target ────────────────────────────────────────────
test("an operator-supplied target wins", () => {
assert.equal(
resolveProbeTarget({ PROXY_HEALTH_TEST_URL: "https://probe.internal/ping" }),
"https://probe.internal/ping"
);
});
test("an empty target falls back, a whitespace one is left untouched", () => {
assert.equal(resolveProbeTarget({ PROXY_HEALTH_TEST_URL: "" }), DEFAULT_PROBE_TARGET);
// Whitespace is truthy, so it reached the probe before this refactor. Trimming it here would
// silently repair a broken deployment — a behavior change this change is not entitled to make.
assert.equal(resolveProbeTarget({ PROXY_HEALTH_TEST_URL: " " }), " ");
});
// ─── concurrency ───────────────────────────────────────
test("concurrency honours a sane value and is capped", () => {
assert.equal(resolveProbeConcurrency({ PROXY_HEALTH_TEST_CONCURRENCY: "3" }), 3);
assert.equal(
resolveProbeConcurrency({ PROXY_HEALTH_TEST_CONCURRENCY: "10000" }),
MAX_PROBE_CONCURRENCY
);
});
test("an unparseable concurrency falls back instead of yielding NaN", () => {
assert.equal(
resolveProbeConcurrency({ PROXY_HEALTH_TEST_CONCURRENCY: "abc" }),
DEFAULT_PROBE_CONCURRENCY
);
});
// ─── stagger ───────────────────────────────────────────
test("the stagger step is clamped to its bounds", () => {
assert.equal(resolveProbeStaggerMs({ PROXY_HEALTH_TEST_STAGGER_MS: "250" }), 250);
assert.equal(resolveProbeStaggerMs({ PROXY_HEALTH_TEST_STAGGER_MS: "0" }), 0);
assert.equal(resolveProbeStaggerMs({ PROXY_HEALTH_TEST_STAGGER_MS: "-50" }), 0);
assert.equal(
resolveProbeStaggerMs({ PROXY_HEALTH_TEST_STAGGER_MS: "999999" }),
MAX_PROBE_STAGGER_MS
);
});
test("an unparseable stagger falls back instead of yielding NaN", () => {
const resolved = resolveProbeStaggerMs({ PROXY_HEALTH_TEST_STAGGER_MS: "later" });
assert.equal(resolved, DEFAULT_PROBE_STAGGER_MS);
assert.ok(Number.isFinite(resolved));
});
// ─── delay computation ─────────────────────────────────
test("the head of a batch is never delayed", () => {
assert.equal(staggerDelayMs(0, 100), 0);
});
test("delays grow strictly with the position in the batch", () => {
const step = 100;
for (let i = 1; i < 10; i++) {
assert.equal(staggerDelayMs(i, step), i * step);
assert.ok(staggerDelayMs(i, step) > staggerDelayMs(i - 1, step));
}
});
test("a zero or negative step inserts no wait at all", () => {
for (const step of [0, -1]) {
for (let i = 0; i < 5; i++) {
assert.equal(staggerDelayMs(i, step), 0);
}
}
});
test("a negative index cannot produce a negative wait", () => {
assert.equal(staggerDelayMs(-1, 100), 0);
});
// ─── the wait itself ───────────────────────────────────
test("the head of a batch is not held back at all", async () => {
const before = Date.now();
await waitForProbeSlot(0, 100);
// No timer is armed for slot 0, so this resolves on the microtask queue. The assertion is
// deliberately loose: it proves no ~100ms timer fired, not a precise scheduler timing.
assert.ok(Date.now() - before < 50, "slot 0 must not wait");
});
test("a later slot is actually held back", async () => {
const before = Date.now();
await waitForProbeSlot(2, 30);
assert.ok(Date.now() - before >= 55, "slot 2 must wait about two steps");
});
test("a disabled step holds nobody back", async () => {
const before = Date.now();
await Promise.all([waitForProbeSlot(5, 0), waitForProbeSlot(9, 0)]);
assert.ok(Date.now() - before < 50, "a zero step must insert no wait");
});