Files
Sybil-2/server/src/llm/model-catalog.ts

101 lines
3.0 KiB
TypeScript
Raw Normal View History

2026-02-14 21:00:30 -08:00
import type { FastifyBaseLogger } from "fastify";
2026-06-13 12:02:22 -07:00
import {
fetchProviderCatalogModels,
getProviderCatalogFallbackModels,
listModelCatalogProviders,
} from "./provider-adapters.js";
2026-02-14 21:00:30 -08:00
import type { Provider } from "./types.js";
export type ProviderModelSnapshot = {
models: string[];
loadedAt: string | null;
error: string | null;
};
2026-05-04 21:52:39 -07:00
export type ModelCatalogSnapshot = Partial<Record<Provider, ProviderModelSnapshot>>;
2026-02-14 21:00:30 -08:00
const MODEL_FETCH_TIMEOUT_MS = 15000;
2026-05-20 22:08:45 -07:00
const MODEL_CATALOG_REFRESH_INTERVAL_MS = 24 * 60 * 60 * 1000;
2026-02-14 21:00:30 -08:00
2026-06-13 12:02:22 -07:00
const modelCatalog: ModelCatalogSnapshot = {};
2026-02-14 21:00:30 -08:00
2026-05-20 22:08:45 -07:00
let catalogRefreshPromise: Promise<void> | null = null;
2026-02-14 21:00:30 -08:00
async function withTimeout<T>(promise: Promise<T>, timeoutMs: number, label: string) {
let timeoutId: NodeJS.Timeout | null = null;
try {
return await Promise.race([
promise,
new Promise<T>((_, reject) => {
timeoutId = setTimeout(() => {
reject(new Error(`${label} timed out after ${timeoutMs}ms`));
}, timeoutMs);
}),
]);
} finally {
if (timeoutId) clearTimeout(timeoutId);
}
}
async function refreshProviderModels(provider: Provider, logger?: FastifyBaseLogger) {
try {
2026-06-13 12:02:22 -07:00
const models = await withTimeout(fetchProviderCatalogModels(provider), MODEL_FETCH_TIMEOUT_MS, `${provider} model fetch`);
2026-02-14 21:00:30 -08:00
modelCatalog[provider] = {
models,
loadedAt: new Date().toISOString(),
error: null,
};
logger?.info({ provider, modelCount: models.length }, "model catalog loaded");
} catch (err: any) {
const message = err?.message ?? String(err);
2026-05-20 22:08:45 -07:00
const previous = modelCatalog[provider];
2026-06-13 12:02:22 -07:00
const fallbackModels = getProviderCatalogFallbackModels(provider);
2026-02-14 21:00:30 -08:00
modelCatalog[provider] = {
2026-05-20 22:08:45 -07:00
models: previous?.models.length ? previous.models : fallbackModels,
loadedAt: previous?.loadedAt ?? null,
2026-02-14 21:00:30 -08:00
error: message,
};
logger?.warn({ provider, err: message }, "failed to load provider model catalog");
}
}
2026-05-20 22:08:45 -07:00
export async function refreshModelCatalog(logger?: FastifyBaseLogger) {
if (catalogRefreshPromise) return catalogRefreshPromise;
2026-06-13 12:02:22 -07:00
catalogRefreshPromise = Promise.all(listModelCatalogProviders().map((provider) => refreshProviderModels(provider, logger)))
2026-05-20 22:08:45 -07:00
.then(() => undefined)
.finally(() => {
catalogRefreshPromise = null;
});
return catalogRefreshPromise;
}
2026-02-14 21:00:30 -08:00
export async function warmModelCatalog(logger?: FastifyBaseLogger) {
2026-05-20 22:08:45 -07:00
await refreshModelCatalog(logger);
}
export function startModelCatalogRefreshLoop(logger?: FastifyBaseLogger) {
const timer = setInterval(() => {
void refreshModelCatalog(logger);
}, MODEL_CATALOG_REFRESH_INTERVAL_MS);
timer.unref?.();
return () => {
clearInterval(timer);
};
2026-02-14 21:00:30 -08:00
}
export function getModelCatalogSnapshot(): ModelCatalogSnapshot {
2026-05-04 21:52:39 -07:00
const snapshot: ModelCatalogSnapshot = {};
2026-06-13 12:02:22 -07:00
for (const provider of listModelCatalogProviders()) {
2026-05-04 21:52:39 -07:00
const entry = modelCatalog[provider] ?? { models: [], loadedAt: null, error: null };
snapshot[provider] = {
models: [...entry.models],
loadedAt: entry.loadedAt,
error: entry.error,
};
}
return snapshot;
2026-02-14 21:00:30 -08:00
}