perf: optimize overview/tasks/polling/microservice-cache and improve frontend progressive loading

This commit is contained in:
Codex
2026-05-11 08:08:17 +00:00
parent 5a198baf77
commit 688376abc4
6 changed files with 521 additions and 144 deletions
+301 -49
View File
@@ -117,6 +117,7 @@ type TaskTerminalWaiter = (task: RawTaskRow | null) => void;
interface MicroserviceProxyCacheEntry {
expiresAt: number;
staleExpiresAt: number;
status: number;
contentType: string;
bodyText: string;
@@ -140,6 +141,7 @@ const operationPerformanceSamples: OperationPerformanceSample[] = [];
const maxPerformanceSamples = 3000;
const taskTerminalWaiters = new Map<string, Set<TaskTerminalWaiter>>();
const microserviceProxyCache = new Map<string, MicroserviceProxyCacheEntry>();
const microserviceProxyRefreshes = new Map<string, Promise<void>>();
let lastTaskSweepAt = 0;
let taskSweepInFlight: Promise<void> | null = null;
const microserviceProxyMaxBodyTextLength = 8 * 1024 * 1024;
@@ -1062,8 +1064,169 @@ async function getEvents(limit: number): Promise<ApiEvent[]> {
}));
}
async function getTasks(limit: number, statusFilter = "all", lite = false): Promise<ApiTask[]> {
function rowString(row: Record<string, unknown>, key: string): string {
const value = row[key];
return typeof value === "string" ? value : value === null || value === undefined ? "" : String(value);
}
function rowNumber(row: Record<string, unknown>, key: string): number | null {
const value = row[key];
const parsed = typeof value === "number" ? value : typeof value === "string" ? Number(value) : NaN;
return Number.isFinite(parsed) ? parsed : null;
}
function taskJsonSummary(row: Record<string, unknown>, prefix: "payload" | "result"): JsonValue {
const type = rowString(row, `${prefix}_type`);
if (type.length === 0) return prefix === "result" ? null : {};
const summary: Record<string, JsonValue> = { summaryOnly: true, type };
const fields = prefix === "payload"
? ["source", "serviceId", "method", "path", "mode", "targetBaseUrl", "timeoutMs", "targetProviderGatewayVersion", "providerGatewayVersion"]
: ["error", "reason", "message", "status", "exitCode", "code", "signal", "timeoutMs", "previousStatus", "mode", "policy", "targetProviderGatewayVersion", "providerGatewayVersion", "updaterContainerId"];
for (const field of fields) {
const value = row[`${prefix}_${field}`];
if (value !== null && value !== undefined && String(value).length > 0) {
summary[field] = typeof value === "number" || typeof value === "boolean" ? value : String(value);
}
}
const bodyChars = rowNumber(row, `${prefix}_body_text_chars`);
if (bodyChars !== null) summary.bodyText = `<omitted:${bodyChars} chars>`;
return summary;
}
function taskSummaryFromRow(row: Record<string, unknown>): ApiTask {
return {
id: String(row.id),
providerId: String(row.provider_id),
command: String(row.command),
status: String(row.status),
payload: taskJsonSummary(row, "payload"),
result: taskJsonSummary(row, "result"),
createdAt: row.created_at instanceof Date ? row.created_at.toISOString() : String(row.created_at),
updatedAt: row.updated_at instanceof Date ? row.updated_at.toISOString() : String(row.updated_at),
_summaryOnly: true,
} as ApiTask;
}
async function getTask(taskId: string): Promise<ApiTask | null> {
const rows = await sql<Array<Record<string, unknown>>>`
SELECT
id,
provider_id,
command,
status,
CASE
WHEN payload ? 'bodyText' THEN jsonb_set(payload - 'bodyText', '{bodyText}', to_jsonb(('<omitted:' || length(payload->>'bodyText')::text || ' chars>')::text))
ELSE payload
END AS payload,
CASE
WHEN result IS NOT NULL AND result ? 'bodyText' THEN jsonb_set(result - 'bodyText', '{bodyText}', to_jsonb(('<omitted:' || length(result->>'bodyText')::text || ' chars>')::text))
ELSE result
END AS result,
created_at,
updated_at
FROM unidesk_tasks
WHERE id = ${taskId}
LIMIT 1
`;
const row = rows[0];
if (row === undefined) return null;
return {
id: String(row.id),
providerId: String(row.provider_id),
command: String(row.command),
status: String(row.status),
payload: compactJson(row.payload ?? {}),
result: compactJson(row.result ?? null),
createdAt: row.created_at instanceof Date ? row.created_at.toISOString() : String(row.created_at),
updatedAt: row.updated_at instanceof Date ? row.updated_at.toISOString() : String(row.updated_at),
};
}
async function getTasks(limit: number, statusFilter = "all", lite = false, summary = false): Promise<ApiTask[]> {
await maybeMarkStaleTasksFailed();
if (summary && !lite) {
const rows = statusFilter === "pending"
? await sql<Array<Record<string, unknown>>>`
SELECT
id,
provider_id,
command,
status,
jsonb_typeof(payload) AS payload_type,
payload->>'source' AS "payload_source",
payload->>'serviceId' AS "payload_serviceId",
payload->>'method' AS "payload_method",
payload->>'path' AS "payload_path",
payload->>'mode' AS "payload_mode",
payload->>'targetBaseUrl' AS "payload_targetBaseUrl",
payload->>'timeoutMs' AS "payload_timeoutMs",
payload->>'targetProviderGatewayVersion' AS "payload_targetProviderGatewayVersion",
payload->>'providerGatewayVersion' AS "payload_providerGatewayVersion",
CASE WHEN payload ? 'bodyText' THEN length(payload->>'bodyText') ELSE NULL END AS payload_body_text_chars,
jsonb_typeof(result) AS result_type,
result->>'error' AS "result_error",
result->>'reason' AS "result_reason",
result->>'message' AS "result_message",
result->>'status' AS "result_status",
result->>'exitCode' AS "result_exitCode",
result->>'code' AS "result_code",
result->>'signal' AS "result_signal",
result->>'timeoutMs' AS "result_timeoutMs",
result->>'previousStatus' AS "result_previousStatus",
result->>'mode' AS "result_mode",
COALESCE(result->>'policy', result #>> '{plan,policy}') AS "result_policy",
COALESCE(result->>'targetProviderGatewayVersion', result #>> '{plan,targetProviderGatewayVersion}') AS "result_targetProviderGatewayVersion",
COALESCE(result->>'providerGatewayVersion', result #>> '{plan,providerGatewayVersion}') AS "result_providerGatewayVersion",
result->>'updaterContainerId' AS "result_updaterContainerId",
CASE WHEN result ? 'bodyText' THEN length(result->>'bodyText') ELSE NULL END AS result_body_text_chars,
created_at,
updated_at
FROM unidesk_tasks
WHERE status IN ('queued', 'dispatched', 'running')
ORDER BY updated_at DESC
LIMIT ${limit}
`
: await sql<Array<Record<string, unknown>>>`
SELECT
id,
provider_id,
command,
status,
jsonb_typeof(payload) AS payload_type,
payload->>'source' AS "payload_source",
payload->>'serviceId' AS "payload_serviceId",
payload->>'method' AS "payload_method",
payload->>'path' AS "payload_path",
payload->>'mode' AS "payload_mode",
payload->>'targetBaseUrl' AS "payload_targetBaseUrl",
payload->>'timeoutMs' AS "payload_timeoutMs",
payload->>'targetProviderGatewayVersion' AS "payload_targetProviderGatewayVersion",
payload->>'providerGatewayVersion' AS "payload_providerGatewayVersion",
CASE WHEN payload ? 'bodyText' THEN length(payload->>'bodyText') ELSE NULL END AS payload_body_text_chars,
jsonb_typeof(result) AS result_type,
result->>'error' AS "result_error",
result->>'reason' AS "result_reason",
result->>'message' AS "result_message",
result->>'status' AS "result_status",
result->>'exitCode' AS "result_exitCode",
result->>'code' AS "result_code",
result->>'signal' AS "result_signal",
result->>'timeoutMs' AS "result_timeoutMs",
result->>'previousStatus' AS "result_previousStatus",
result->>'mode' AS "result_mode",
COALESCE(result->>'policy', result #>> '{plan,policy}') AS "result_policy",
COALESCE(result->>'targetProviderGatewayVersion', result #>> '{plan,targetProviderGatewayVersion}') AS "result_targetProviderGatewayVersion",
COALESCE(result->>'providerGatewayVersion', result #>> '{plan,providerGatewayVersion}') AS "result_providerGatewayVersion",
result->>'updaterContainerId' AS "result_updaterContainerId",
CASE WHEN result ? 'bodyText' THEN length(result->>'bodyText') ELSE NULL END AS result_body_text_chars,
created_at,
updated_at
FROM unidesk_tasks
ORDER BY updated_at DESC
LIMIT ${limit}
`;
return rows.map(taskSummaryFromRow);
}
const rows = statusFilter === "pending"
? lite
? await sql<Array<Record<string, unknown>>>`
@@ -1161,22 +1324,39 @@ async function getPgdataUsage(): Promise<JsonValue> {
}
async function getOverview(): Promise<JsonValue> {
const nodes = await getNodes();
const pendingTasks = await countPendingTasks();
const dockerStatuses = await getNodeDockerStatuses();
const systemStatuses = await getNodeSystemStatuses(1);
const pgdata = await getPgdataUsage();
const online = nodes.filter((node) => node.status === "online").length;
const [nodeRows, dockerRows, systemRows, pendingTasks, pgdata] = await Promise.all([
sql<Array<{ node_count: string | number; online_node_count: string | number }>>`
SELECT
count(*)::int AS node_count,
count(*) FILTER (WHERE status = 'online')::int AS online_node_count
FROM unidesk_nodes
`,
sql<Array<{ docker_status_node_count: string | number }>>`
SELECT count(*) FILTER (WHERE d.status IS NOT NULL)::int AS docker_status_node_count
FROM unidesk_nodes n
LEFT JOIN unidesk_node_docker_status d ON d.provider_id = n.provider_id
`,
sql<Array<{ system_status_node_count: string | number }>>`
SELECT count(*) FILTER (WHERE s.status IS NOT NULL)::int AS system_status_node_count
FROM unidesk_nodes n
LEFT JOIN unidesk_node_system_status s ON s.provider_id = n.provider_id
`,
countPendingTasks(),
getPgdataUsage(),
]);
const nodeRow = nodeRows[0];
const dockerRow = dockerRows[0];
const systemRow = systemRows[0];
return {
service: "unidesk-core",
ok: true,
dbReady,
pgdata,
uptimeSeconds: Math.floor((Date.now() - serviceStartedAt.getTime()) / 1000),
nodeCount: nodes.length,
onlineNodeCount: online,
dockerStatusNodeCount: dockerStatuses.filter((item) => item.dockerStatus !== null).length,
systemStatusNodeCount: systemStatuses.filter((item) => item.current !== null).length,
nodeCount: Number(nodeRow?.node_count ?? 0),
onlineNodeCount: Number(nodeRow?.online_node_count ?? 0),
dockerStatusNodeCount: Number(dockerRow?.docker_status_node_count ?? 0),
systemStatusNodeCount: Number(systemRow?.system_status_node_count ?? 0),
pendingTaskCount: pendingTasks,
taskPendingTimeoutMs: config.taskPendingTimeoutMs,
activeSocketCount: activeProviders.size,
@@ -1432,12 +1612,27 @@ function responseFromMicroserviceResult(task: Awaited<ReturnType<typeof rawTask>
}
function microserviceCacheTtlMs(serviceId: string, targetPath: string): number {
if (serviceId === "pipeline" && targetPath === "/api/snapshot") return 6_000;
if (serviceId === "pipeline" && targetPath.startsWith("/api/oa-event-flow/")) return 20_000;
if (serviceId === "pipeline" && targetPath.startsWith("/api/model-quota/")) return 60_000;
if (serviceId === "pipeline" && targetPath.startsWith("/api/node-control/runs/")) return 6_000;
if (serviceId === "pipeline" && targetPath.startsWith("/api/runs/")) return 6_000;
if (serviceId === "findjob" && (targetPath === "/api/summary" || targetPath === "/api/jobs" || targetPath === "/api/drafts")) return 8_000;
if (serviceId === "met-nonlinear" && (targetPath === "/api/images" || targetPath === "/api/projects")) return 15_000;
if (serviceId === "met-nonlinear" && (targetPath === "/api/queue" || targetPath === "/api/summary" || targetPath === "/api/history")) return 1_500;
if (serviceId === "met-nonlinear" && (targetPath === "/api/queue" || targetPath === "/api/summary" || targetPath === "/api/history")) return 5_000;
if (serviceId === "codex-queue" && targetPath.includes("/transcript")) return 1_000;
return 750;
}
function microserviceCacheStaleMs(serviceId: string, targetPath: string): number {
if (serviceId === "pipeline" && targetPath.startsWith("/api/model-quota/")) return 5 * 60_000;
if (serviceId === "pipeline") return 45_000;
if (serviceId === "findjob") return 60_000;
if (serviceId === "met-nonlinear" && (targetPath === "/api/images" || targetPath === "/api/projects")) return 5 * 60_000;
if (serviceId === "met-nonlinear") return 45_000;
return 5_000;
}
function providerMicroserviceCacheTtlMs(serviceId: string, targetPath: string): number {
if (serviceId === "met-nonlinear" && (targetPath === "/api/images" || targetPath === "/api/projects")) return 60_000;
if (serviceId === "met-nonlinear" && targetPath === "/api/history") return 10_000;
@@ -1451,22 +1646,38 @@ function microserviceCacheKey(service: MicroserviceConfig, method: string, targe
return JSON.stringify([service.id, method, targetPath, proxyOptions.query, proxyOptions.jsonArrayLimits]);
}
function readMicroserviceCache(key: string): Response | null {
const entry = microserviceProxyCache.get(key);
if (entry === undefined) return null;
if (entry.expiresAt <= Date.now()) {
microserviceProxyCache.delete(key);
return null;
}
function responseFromMicroserviceCache(entry: MicroserviceProxyCacheEntry, state: "hit" | "stale"): Response {
return new Response(entry.bodyText, {
status: entry.status,
headers: {
"content-type": entry.contentType,
"x-unidesk-cache": "hit",
"x-unidesk-cache": state,
},
});
}
function readMicroserviceCache(key: string): Response | null {
const entry = microserviceProxyCache.get(key);
if (entry === undefined) return null;
if (entry.staleExpiresAt <= Date.now()) {
microserviceProxyCache.delete(key);
return null;
}
if (entry.expiresAt <= Date.now()) return null;
return responseFromMicroserviceCache(entry, "hit");
}
function readStaleMicroserviceCache(key: string): Response | null {
const entry = microserviceProxyCache.get(key);
if (entry === undefined) return null;
if (entry.staleExpiresAt <= Date.now()) {
microserviceProxyCache.delete(key);
return null;
}
if (entry.expiresAt > Date.now()) return null;
return responseFromMicroserviceCache(entry, "stale");
}
async function cacheableResponseSnapshot(response: Response): Promise<MicroserviceProxyCacheEntry | null> {
if (response.status < 200 || response.status >= 300) return null;
if (response.headers.get("x-unidesk-response-truncated") === "true") return null;
@@ -1474,6 +1685,7 @@ async function cacheableResponseSnapshot(response: Response): Promise<Microservi
if (bodyText.length > 2 * 1024 * 1024) return null;
return {
expiresAt: 0,
staleExpiresAt: 0,
status: response.status,
contentType: response.headers.get("content-type") ?? "application/octet-stream",
bodyText,
@@ -1482,7 +1694,9 @@ async function cacheableResponseSnapshot(response: Response): Promise<Microservi
function rememberMicroserviceCache(key: string, ttlMs: number, entry: MicroserviceProxyCacheEntry | null): void {
if (entry === null || ttlMs <= 0) return;
const keyParts = JSON.parse(key) as string[];
entry.expiresAt = Date.now() + ttlMs;
entry.staleExpiresAt = entry.expiresAt + microserviceCacheStaleMs(String(keyParts[0] ?? ""), String(keyParts[2] ?? ""));
microserviceProxyCache.set(key, entry);
if (microserviceProxyCache.size > 300) {
const now = Date.now();
@@ -1550,6 +1764,56 @@ async function directMicroserviceResponse(
}
}
async function fetchMicroserviceUpstreamResponse(
service: MicroserviceConfig,
method: string,
targetPath: string,
proxyOptions: { query: string; jsonArrayLimits: Record<string, JsonValue> },
requestHeaders: Record<string, JsonValue>,
bodyText: string,
): Promise<Response> {
if (canDirectProxyMicroservice(service)) {
return directMicroserviceResponse(service, method, targetPath, proxyOptions, requestHeaders, bodyText);
}
if (!(await providerSupports(service.providerId, "microservice.http"))) {
return jsonResponse({ ok: false, error: `provider does not declare microservice.http capability: ${service.providerId}` }, 409);
}
const { taskId, providerOnline } = await createAndSendTask(service.providerId, "microservice.http", {
source: "microservice-frontend-proxy",
serviceId: service.id,
method,
targetBaseUrl: service.backend.nodeBaseUrl,
path: targetPath,
query: proxyOptions.query,
requestHeaders,
bodyText,
jsonArrayLimits: proxyOptions.jsonArrayLimits,
timeoutMs: service.backend.timeoutMs,
cacheTtlMs: providerMicroserviceCacheTtlMs(service.id, targetPath),
});
if (!providerOnline) return jsonResponse({ ok: false, error: `provider is offline: ${service.providerId}`, taskId }, 503);
const task = await waitForTaskTerminal(taskId, service.backend.timeoutMs + 3000);
return responseFromMicroserviceResult(task);
}
function refreshMicroserviceCacheInBackground(
cacheKey: string,
ttlMs: number,
fetchResponse: () => Promise<Response>,
): void {
if (microserviceProxyRefreshes.has(cacheKey)) return;
const refresh = fetchResponse()
.then((response) => cacheableResponseSnapshot(response))
.then((entry) => rememberMicroserviceCache(cacheKey, ttlMs, entry))
.catch((error) => {
logger("warn", "microservice_cache_revalidate_failed", { cacheKey, error: errorToJson(error) });
})
.finally(() => {
microserviceProxyRefreshes.delete(cacheKey);
});
microserviceProxyRefreshes.set(cacheKey, refresh);
}
async function microserviceRoute(req: Request, url: URL): Promise<Response> {
const rest = url.pathname.slice("/api/microservices/".length);
const slashIndex = rest.indexOf("/");
@@ -1580,12 +1844,9 @@ async function microserviceRoute(req: Request, url: URL): Promise<Response> {
if (!isMicroservicePathAllowed(service, targetPath)) {
return jsonResponse({ ok: false, error: "microservice path is not allowed", serviceId, targetPath }, 403);
}
const directProxy = canDirectProxyMicroservice(service);
if (!directProxy && !(await providerSupports(service.providerId, "microservice.http"))) {
return jsonResponse({ ok: false, error: `provider does not declare microservice.http capability: ${service.providerId}` }, 409);
}
const proxyOptions = readMicroserviceArrayLimits(url);
const cacheKey = microserviceCacheKey(service, method, targetPath, proxyOptions);
const cacheTtlMs = microserviceCacheTtlMs(service.id, targetPath);
if (method === "GET" || method === "HEAD") {
const cached = readMicroserviceCache(cacheKey);
if (cached !== null) return cached;
@@ -1599,32 +1860,15 @@ async function microserviceRoute(req: Request, url: URL): Promise<Response> {
const requestHeaders: Record<string, JsonValue> = {};
const contentType = req.headers.get("content-type");
if (contentType !== null) requestHeaders["content-type"] = contentType.slice(0, 200);
if (directProxy) {
const response = await directMicroserviceResponse(service, method, targetPath, proxyOptions, requestHeaders, bodyText);
if (method === "GET" || method === "HEAD") {
rememberMicroserviceCache(cacheKey, microserviceCacheTtlMs(service.id, targetPath), await cacheableResponseSnapshot(response));
}
return response;
}
const { taskId, providerOnline } = await createAndSendTask(service.providerId, "microservice.http", {
source: "microservice-frontend-proxy",
serviceId: service.id,
method,
targetBaseUrl: service.backend.nodeBaseUrl,
path: targetPath,
query: proxyOptions.query,
requestHeaders,
bodyText,
jsonArrayLimits: proxyOptions.jsonArrayLimits,
timeoutMs: service.backend.timeoutMs,
cacheTtlMs: providerMicroserviceCacheTtlMs(service.id, targetPath),
});
if (!providerOnline) return jsonResponse({ ok: false, error: `provider is offline: ${service.providerId}`, taskId }, 503);
const task = await waitForTaskTerminal(taskId, service.backend.timeoutMs + 3000);
const response = responseFromMicroserviceResult(task);
if (method === "GET" || method === "HEAD") {
rememberMicroserviceCache(cacheKey, microserviceCacheTtlMs(service.id, targetPath), await cacheableResponseSnapshot(response));
const stale = readStaleMicroserviceCache(cacheKey);
if (stale !== null) {
refreshMicroserviceCacheInBackground(cacheKey, cacheTtlMs, () => fetchMicroserviceUpstreamResponse(service, method, targetPath, proxyOptions, requestHeaders, bodyText));
return stale;
}
}
const response = await fetchMicroserviceUpstreamResponse(service, method, targetPath, proxyOptions, requestHeaders, bodyText);
if (method === "GET" || method === "HEAD") rememberMicroserviceCache(cacheKey, cacheTtlMs, await cacheableResponseSnapshot(response));
return response;
}
@@ -1939,9 +2183,17 @@ async function routeInner(req: Request, server: Server<WsData>): Promise<Respons
if (url.pathname === "/api/nodes/system-status") return jsonResponse({ ok: true, systemStatuses: await withPerformanceOperation("core", "node_system_status", url.search, () => getNodeSystemStatuses(readLimit(url, 60))) });
if (url.pathname === "/api/nodes/docker-status") return jsonResponse({ ok: true, dockerStatuses: await withPerformanceOperation("core", "node_docker_status", url.pathname, () => getNodeDockerStatuses()) });
if (url.pathname === "/api/events") return jsonResponse({ ok: true, events: await withPerformanceOperation("core", "events", url.search, () => getEvents(readLimit(url, 100))) });
if (url.pathname.startsWith("/api/tasks/") && req.method === "GET") {
const taskId = decodeURIComponent(url.pathname.slice("/api/tasks/".length));
if (taskId.length === 0 || taskId.includes("/")) return jsonResponse({ ok: false, error: "invalid task id" }, 400);
const task = await withPerformanceOperation("core", "task_detail", taskId, () => getTask(taskId));
if (task === null) return jsonResponse({ ok: false, error: `task not found: ${taskId}` }, 404);
return jsonResponse({ ok: true, task });
}
if (url.pathname === "/api/tasks") {
const lite = ["1", "true", "yes"].includes((url.searchParams.get("lite") ?? "").toLowerCase());
return jsonResponse({ ok: true, tasks: await withPerformanceOperation("core", "tasks", url.search, () => getTasks(readLimit(url, 100), url.searchParams.get("status") ?? "all", lite)) });
const summary = ["1", "true", "yes"].includes((url.searchParams.get("summary") ?? "").toLowerCase());
return jsonResponse({ ok: true, tasks: await withPerformanceOperation("core", "tasks", url.search, () => getTasks(readLimit(url, 100), url.searchParams.get("status") ?? "all", lite, summary)) });
}
if (url.pathname === "/api/microservices") return jsonResponse({ ok: true, microservices: await withPerformanceOperation("core", "microservices", url.pathname, () => getMicroservices()) });
if (url.pathname === "/api/performance") return jsonResponse(await getPerformance());