feat: add agent run queue controls

This commit is contained in:
John
2026-07-02 07:52:07 +08:00
parent 8e5094fcd9
commit 3ea973968e
9 changed files with 387 additions and 66 deletions
+124 -32
View File
@@ -4,6 +4,8 @@ const DEFAULT_RUN_RETRY_DELAYS_MS = [1500, 5000, 15000];
const TERMINAL_STATUSES = new Set(['succeeded', 'failed']);
const CODE_TOOL_MODES = new Set(['code', 'code-task', 'code_task', 'code-tool', 'code_tool', 'code_tool_task']);
const RUN_METADATA_KEY = 'memindRun';
const DEFAULT_MAX_CONCURRENT_RUNS = 1;
const DEFAULT_RUN_TIMEOUT_MS = 15 * 60 * 1000;
function nowMs() {
return Date.now();
@@ -21,6 +23,18 @@ function serializeMessage(message) {
return JSON.stringify(message ?? {});
}
function positiveInteger(value, fallback) {
const n = Number(value);
if (!Number.isFinite(n) || n <= 0) return fallback;
return Math.floor(n);
}
function timeoutError(ms) {
const err = new Error(`agent run timed out after ${ms}ms`);
err.code = 'AGENT_RUN_TIMEOUT';
return err;
}
export function normalizeAgentRunToolMode(value) {
const normalized = String(value ?? 'chat').trim().toLowerCase();
if (!normalized || normalized === 'chat') return 'chat';
@@ -88,8 +102,31 @@ export function createAgentRunGateway({
tkmindProxy,
retryDelaysMs = DEFAULT_RUN_RETRY_DELAYS_MS,
autoDispatch = true,
maxConcurrentRuns = positiveInteger(
process.env.MEMIND_AGENT_RUN_QUEUE_CONCURRENCY,
DEFAULT_MAX_CONCURRENT_RUNS,
),
runTimeoutMs = positiveInteger(
process.env.MEMIND_AGENT_RUN_TIMEOUT_MS,
DEFAULT_RUN_TIMEOUT_MS,
),
}) {
const inFlight = new Set();
const queuedDispatches = [];
const queuedDispatchSet = new Set();
function enqueueRun(runId) {
if (!runId || inFlight.has(runId) || queuedDispatchSet.has(runId)) return false;
queuedDispatchSet.add(runId);
queuedDispatches.push(runId);
return true;
}
function nextQueuedRunId() {
const runId = queuedDispatches.shift();
if (runId) queuedDispatchSet.delete(runId);
return runId ?? null;
}
async function appendEvent(runId, type, data = null) {
await pool.query(
@@ -193,6 +230,50 @@ export function createAgentRunGateway({
await appendEvent(runId, status, fields);
}
async function runWithTimeout(runId, task) {
let timer = null;
const timeout = new Promise((_, reject) => {
timer = setTimeout(() => reject(timeoutError(runTimeoutMs)), runTimeoutMs);
});
try {
return await Promise.race([task(), timeout]);
} finally {
if (timer) clearTimeout(timer);
}
}
async function executeRun(row, runId) {
const userMessage = safeJsonParse(row.user_message_json, {});
const runOptions = getRunOptionsFromMessage(userMessage);
let sessionId = row.agent_session_id ?? null;
if (!sessionId) {
const sessionOptions = {};
if (runOptions.toolMode === 'code' && userAuth?.getCodeAgentSessionPolicy) {
sessionOptions.sessionPolicy = await userAuth.getCodeAgentSessionPolicy(row.user_id);
}
const session = await tkmindProxy.startSessionForUser(row.user_id, sessionOptions);
sessionId = session.id;
await pool.query(
`UPDATE h5_agent_runs SET agent_session_id = ?, updated_at = ? WHERE id = ?`,
[sessionId, nowMs(), runId],
);
await appendEvent(runId, 'session_started', {
sessionId,
toolMode: runOptions.toolMode,
taskType: runOptions.taskType,
});
}
await tkmindProxy.submitSessionReplyForUser(
row.user_id,
sessionId,
row.request_id,
userMessage,
{ toolMode: runOptions.toolMode },
);
return { sessionId };
}
async function processRun(runId) {
const row = await getRunById(runId);
if (!row || TERMINAL_STATUSES.has(row.status)) return;
@@ -207,41 +288,18 @@ export function createAgentRunGateway({
await appendEvent(runId, 'running', { attempt: nextAttempt });
try {
const userMessage = safeJsonParse(row.user_message_json, {});
const runOptions = getRunOptionsFromMessage(userMessage);
let sessionId = row.agent_session_id ?? null;
if (!sessionId) {
const sessionOptions = {};
if (runOptions.toolMode === 'code' && userAuth?.getCodeAgentSessionPolicy) {
sessionOptions.sessionPolicy = await userAuth.getCodeAgentSessionPolicy(row.user_id);
}
const session = await tkmindProxy.startSessionForUser(row.user_id, sessionOptions);
sessionId = session.id;
await pool.query(
`UPDATE h5_agent_runs SET agent_session_id = ?, updated_at = ? WHERE id = ?`,
[sessionId, nowMs(), runId],
);
await appendEvent(runId, 'session_started', {
sessionId,
toolMode: runOptions.toolMode,
taskType: runOptions.taskType,
});
}
await tkmindProxy.submitSessionReplyForUser(
row.user_id,
sessionId,
row.request_id,
userMessage,
{ toolMode: runOptions.toolMode },
);
const { sessionId } = await runWithTimeout(runId, () => executeRun(row, runId));
await markRun(runId, 'succeeded', {
agent_session_id: sessionId,
completed_at: nowMs(),
});
} catch (err) {
const message = err instanceof Error ? err.message : String(err);
const retryable = nextAttempt < retryDelaysMs.length;
const timedOut = err?.code === 'AGENT_RUN_TIMEOUT';
if (timedOut) {
await appendEvent(runId, 'timeout', { timeoutMs: runTimeoutMs });
}
const retryable = !timedOut && nextAttempt < retryDelaysMs.length;
await markRun(runId, retryable ? 'retryable' : 'failed', {
error_message: message,
completed_at: retryable ? null : nowMs(),
@@ -252,15 +310,49 @@ export function createAgentRunGateway({
}
}
function drainQueue() {
while (inFlight.size < maxConcurrentRuns) {
const runId = nextQueuedRunId();
if (!runId) break;
inFlight.add(runId);
void processRun(runId)
.finally(() => {
inFlight.delete(runId);
drainQueue();
});
}
}
function dispatchRun(runId) {
if (!runId || inFlight.has(runId)) return;
inFlight.add(runId);
void processRun(runId).finally(() => inFlight.delete(runId));
if (!enqueueRun(runId)) return;
drainQueue();
}
async function getQueueStatus() {
const [rows] = await pool.query(
`SELECT status, COUNT(*) AS count
FROM h5_agent_runs
WHERE status IN ('queued', 'running', 'retryable')
GROUP BY status`,
);
const statusCounts = {};
for (const row of rows) {
statusCounts[row.status] = Number(row.count ?? 0);
}
return {
maxConcurrentRuns,
runTimeoutMs,
inFlight: inFlight.size,
pendingDispatches: queuedDispatches.length,
statusCounts,
terminalStatuses: [...TERMINAL_STATUSES],
};
}
return {
createRun,
getRunForUser,
dispatchRun,
getQueueStatus,
};
}