| // Per-provider HTTP timeout overrides (issue #547, reworked from PR #509). | |
| // | |
| // PROVIDER_TIMEOUT_<PLATFORM>=<milliseconds> overrides the platform's built-in | |
| // chat timeout (e.g. PROVIDER_TIMEOUT_NVIDIA=300000). 0 disables the timeout | |
| // entirely. Values are read when the provider is constructed — for the built-in | |
| // platforms that means process start, so a change requires a restart. | |
| // | |
| // PROVIDER_STREAM_STALL_TIMEOUT_MS is the streaming counterpart for the | |
| // mid-stream inactivity watchdog (issue #553): how long an SSE stream may go | |
| // without a single byte before the gateway gives up on it. | |
| // PROVIDER_STREAM_STALL_TIMEOUT_<PLATFORM> (issue #584) overrides it for one | |
| // platform; precedence is per-platform > global > built-in default (90s). | |
| // | |
| // The stall watchdog does NOT bound time-to-first-byte: the FIRST read of a | |
| // stream gets a grace budget derived from the platform's chat timeout (the | |
| // PROVIDER_TIMEOUT_<PLATFORM> knob above), floored at the stall budget — see | |
| // BaseProvider.readSseStream. Before #584 the stall budget also bounded the | |
| // first byte, which killed healthy long prefills (NVIDIA NIM sends SSE headers | |
| // instantly, then prefills 100k-token prompts for minutes) at 90s. | |
| const warned = new Set<string>(); | |
| function warnOnce(name: string, message: string): void { | |
| if (warned.has(name)) return; | |
| warned.add(name); | |
| console.warn(message); | |
| } | |
| function parseTimeoutEnv(name: string, defaultMs: number): number { | |
| const raw = process.env[name]; | |
| if (raw === undefined || raw.trim() === '') return defaultMs; | |
| const parsed = Number(raw.trim()); | |
| if (!Number.isInteger(parsed) || parsed < 0) { | |
| warnOnce(name, `[config] Ignoring ${name}="${raw}": expected a non-negative integer of milliseconds (0 disables the timeout). Using the default ${defaultMs}ms.`); | |
| return defaultMs; | |
| } | |
| if (parsed > 0 && parsed < 1000) { | |
| warnOnce(name, `[config] ${name}=${parsed}ms is under a second; nearly every request to this provider will abort. Was this meant to be seconds?`); | |
| } | |
| return parsed; | |
| } | |
| export function providerTimeoutEnvName(platform: string): string { | |
| return `PROVIDER_TIMEOUT_${platform.toUpperCase().replace(/[^A-Z0-9]/g, '_')}`; | |
| } | |
| /** Effective chat timeout for a platform: the PROVIDER_TIMEOUT_<PLATFORM> env | |
| * override when set and valid, else the built-in default. Returns 0 to mean | |
| * "no timeout" — callers must skip their abort timer for 0, never schedule | |
| * setTimeout(0). */ | |
| export function providerTimeoutMs(platform: string, defaultMs: number): number { | |
| return parseTimeoutEnv(providerTimeoutEnvName(platform), defaultMs); | |
| } | |
| export const DEFAULT_STREAM_STALL_TIMEOUT_MS = 90_000; | |
| export function streamStallTimeoutEnvName(platform: string): string { | |
| return `PROVIDER_STREAM_STALL_TIMEOUT_${platform.toUpperCase().replace(/[^A-Z0-9]/g, '_')}`; | |
| } | |
| /** Effective mid-stream inactivity timeout, 0 disables. Resolved per call so | |
| * tests can vary the env. Precedence (#584): the per-platform | |
| * PROVIDER_STREAM_STALL_TIMEOUT_<PLATFORM> when a platform is given and the | |
| * var is set and valid, else the global PROVIDER_STREAM_STALL_TIMEOUT_MS, | |
| * else the built-in 90s default. */ | |
| export function streamStallTimeoutMs(platform?: string): number { | |
| const globalMs = parseTimeoutEnv('PROVIDER_STREAM_STALL_TIMEOUT_MS', DEFAULT_STREAM_STALL_TIMEOUT_MS); | |
| if (!platform) return globalMs; | |
| return parseTimeoutEnv(streamStallTimeoutEnvName(platform), globalMs); | |
| } | |
| /** Test hook: forget which malformed values have already been warned about. */ | |
| export function resetTimeoutWarnings(): void { | |
| warned.clear(); | |
| } | |