fix: 持久化 runner 空闲保活策略

src/mgr/kubernetes-runner-job.ts

@@ -44,6 +44,7 @@ export interface RunnerJobDefaults {
   envIdentity?: string;
   artifactCatalogFile?: string;
   serviceAccountName?: string;
+  runnerIdleTimeoutMs?: number;
   kubectlCommand?: string;
   unideskSshEndpointEnv?: JsonRecord;
 }
@@ -57,6 +58,7 @@ export interface CreateRunnerJobInput extends JsonRecord {
   runnerId?: string;
   sourceCommit?: string;
   serviceAccountName?: string;
+  runnerIdleTimeoutMs?: number;
   idempotencyKey?: string;
   imageRef?: JsonRecord;
   transientEnv?: JsonRecord[];
@@ -85,6 +87,7 @@ export async function createKubernetesRunnerJob(options: { store: AgentRunStore;
   const transientEnv = assembleToolContextTransientEnv(run.executionPolicy, transientEnvField(options.input.transientEnv), options.defaults);
   const attemptId = optionalString(options.input.attemptId) ?? `attempt_${Date.now().toString(36)}`;
   const runnerId = optionalString(options.input.runnerId);
+  const runnerIdleTimeoutMs = optionalPositiveInteger(options.input.runnerIdleTimeoutMs, "runnerIdleTimeoutMs") ?? options.defaults.runnerIdleTimeoutMs;
   const transientEnvSecretName = transientEnv.length > 0 ? transientEnvSecretNameForRun(run.id, commandId, attemptId) : null;
   const renderTransientEnv = transientEnvSecretName ? transientEnvWithSecretRefs(transientEnv, transientEnvSecretName) : transientEnv;
   const normalizedPayload = {
@@ -97,6 +100,7 @@ export async function createKubernetesRunnerJob(options: { store: AgentRunStore;
     serviceAccountName: serviceAccountName ?? null,
     attemptId: optionalString(options.input.attemptId) ?? null,
     runnerId: optionalString(options.input.runnerId) ?? null,
+    runnerIdleTimeoutMs: runnerIdleTimeoutMs ?? null,
     transientEnv: transientEnv.map((item) => ({ name: item.name, valueHash: stableHash(item.value), sensitive: true })),
   };
   const payloadHash = stableHash(normalizedPayload);
@@ -150,6 +154,7 @@ export async function createKubernetesRunnerJob(options: { store: AgentRunStore;
     namespace,
     sourceCommit,
     transientEnv: renderTransientEnv,
+    ...(runnerIdleTimeoutMs !== undefined ? { runnerIdleTimeoutMs } : {}),
     ...(serviceAccountName ? { serviceAccountName } : {}),
     ...(sessionPvc ? { sessionPvc } : {}),
   };
@@ -214,6 +219,7 @@ export async function createKubernetesRunnerJob(options: { store: AgentRunStore;
     workReady: staticWorkReadyCapabilitySummary(),
     retention: {
       ttlSecondsAfterFinished: render.ttlSecondsAfterFinished,
+      runnerIdleTimeoutMs: render.runnerIdleTimeoutMs,
     },
     pollActions: [
       runnerJobActionDescriptor({ action: "inspect-run", operation: "describe", resourceKind: "run", resourceName: run.id, runId: run.id }),
@@ -479,6 +485,12 @@ function optionalString(value: unknown): string | undefined {
   return typeof value === "string" && value.trim().length > 0 ? value.trim() : undefined;
 }
 
+function optionalPositiveInteger(value: unknown, key: string): number | undefined {
+  if (value === undefined || value === null) return undefined;
+  if (!Number.isInteger(value) || Number(value) <= 0) throw new AgentRunError("schema-invalid", `${key} must be a positive integer`, { httpStatus: 400 });
+  return Number(value);
+}
+
 function objectPath(value: unknown, path: string[]): string | null {
   let current: unknown = value;
   for (const key of path) {

src/mgr/server.ts

@@ -46,6 +46,7 @@ function runnerJobDefaultsForRequest(defaults: ManagerServerOptions["runnerJobDe
     ...optionalStringRecord("envIdentity", defaults?.envIdentity ?? process.env.AGENTRUN_ENV_IDENTITY),
     ...optionalStringRecord("artifactCatalogFile", defaults?.artifactCatalogFile ?? process.env.AGENTRUN_ARTIFACT_CATALOG_FILE),
     serviceAccountName: defaults?.serviceAccountName ?? process.env.AGENTRUN_RUNNER_SERVICE_ACCOUNT ?? "agentrun-v01-runner",
+    ...(defaults?.runnerIdleTimeoutMs !== undefined ? { runnerIdleTimeoutMs: defaults.runnerIdleTimeoutMs } : optionalPositiveIntegerRecord("runnerIdleTimeoutMs", process.env.AGENTRUN_RUNNER_IDLE_TIMEOUT_MS)),
     ...(defaults?.kubectlCommand ? { kubectlCommand: defaults.kubectlCommand } : {}),
     ...(defaults?.unideskSshEndpointEnv ? { unideskSshEndpointEnv: defaults.unideskSshEndpointEnv } : {}),
   };
@@ -64,6 +65,7 @@ export interface ManagerServerOptions {
     envIdentity?: string;
     artifactCatalogFile?: string;
     serviceAccountName?: string;
+    runnerIdleTimeoutMs?: number;
     kubectlCommand?: string;
     unideskSshEndpointEnv?: JsonRecord;
   };
@@ -947,6 +949,13 @@ function optionalStringRecord(key: string, value: unknown): JsonRecord {
   return typeof value === "string" && value.trim().length > 0 ? { [key]: value.trim() } : {};
 }
 
+function optionalPositiveIntegerRecord(key: string, value: unknown): JsonRecord {
+  if (value === undefined || value === null || value === "") return {};
+  const parsed = Number(value);
+  if (!Number.isInteger(parsed) || parsed <= 0) throw new AgentRunError("schema-invalid", `${key} must be a positive integer`, { httpStatus: 400 });
+  return { [key]: parsed };
+}
+
 function normalizeError(error: unknown): AgentRunError {
   if (error instanceof AgentRunError) return error;
   return new AgentRunError("infra-failed", error instanceof Error ? error.message : String(error), { httpStatus: 500 });