mirrors/pikasTech-agentrun
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: retry transient backend timeouts

Browse Source
This commit is contained in:
AgentRun Codex
2026-06-22 07:56:40 +08:00
parent ccfbf2a866
commit a393e49ed0
5 changed files with 129 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/mgr/kubernetes-runner-job.ts
+18
View File
@@ -50,6 +50,9 @@ export interface RunnerJobDefaults {
lane?: string;
runnerIdleTimeoutMs?: number;
missingTerminalAfterToolTimeoutMs?: number;
backendRetryMaxAttempts?: number;
backendRetryInitialBackoffMs?: number;
backendRetryMaxBackoffMs?: number;
kubectlCommand?: string;
unideskSshEndpointEnv?: JsonRecord;
retention?: RunnerRetentionOptions;
@@ -66,6 +69,9 @@ export interface CreateRunnerJobInput extends JsonRecord {
serviceAccountName?: string;
runnerIdleTimeoutMs?: number;
missingTerminalAfterToolTimeoutMs?: number;
backendRetryMaxAttempts?: number;
backendRetryInitialBackoffMs?: number;
backendRetryMaxBackoffMs?: number;
idempotencyKey?: string;
imageRef?: JsonRecord;
transientEnv?: JsonRecord[];
@@ -98,6 +104,9 @@ export async function createKubernetesRunnerJob(options: { store: AgentRunStore;
const runnerId = optionalString(options.input.runnerId);
const runnerIdleTimeoutMs = optionalPositiveInteger(options.input.runnerIdleTimeoutMs, "runnerIdleTimeoutMs") ?? options.defaults.runnerIdleTimeoutMs;
const missingTerminalAfterToolTimeoutMs = optionalPositiveInteger(options.input.missingTerminalAfterToolTimeoutMs, "missingTerminalAfterToolTimeoutMs") ?? options.defaults.missingTerminalAfterToolTimeoutMs;
const backendRetryMaxAttempts = optionalPositiveInteger(options.input.backendRetryMaxAttempts, "backendRetryMaxAttempts") ?? options.defaults.backendRetryMaxAttempts;
const backendRetryInitialBackoffMs = optionalPositiveInteger(options.input.backendRetryInitialBackoffMs, "backendRetryInitialBackoffMs") ?? options.defaults.backendRetryInitialBackoffMs;
const backendRetryMaxBackoffMs = optionalPositiveInteger(options.input.backendRetryMaxBackoffMs, "backendRetryMaxBackoffMs") ?? options.defaults.backendRetryMaxBackoffMs;
const transientEnvSecretName = transientEnv.length > 0 ? transientEnvSecretNameForRun(run.id, commandId, attemptId, jobNamePrefix) : null;
const renderTransientEnv = transientEnvSecretName ? transientEnvWithSecretRefs(transientEnv, transientEnvSecretName) : transientEnv;
const normalizedPayload = {
@@ -112,6 +121,9 @@ export async function createKubernetesRunnerJob(options: { store: AgentRunStore;
runnerId: optionalString(options.input.runnerId) ?? null,
runnerIdleTimeoutMs: runnerIdleTimeoutMs ?? null,
missingTerminalAfterToolTimeoutMs: missingTerminalAfterToolTimeoutMs ?? null,
backendRetryMaxAttempts: backendRetryMaxAttempts ?? null,
backendRetryInitialBackoffMs: backendRetryInitialBackoffMs ?? null,
backendRetryMaxBackoffMs: backendRetryMaxBackoffMs ?? null,
transientEnv: transientEnv.map((item) => ({ name: item.name, valueHash: stableHash(item.value), sensitive: true })),
};
const payloadHash = stableHash(normalizedPayload);
@@ -174,6 +186,9 @@ export async function createKubernetesRunnerJob(options: { store: AgentRunStore;
transientEnv: renderTransientEnv,
...(runnerIdleTimeoutMs !== undefined ? { runnerIdleTimeoutMs } : {}),
...(missingTerminalAfterToolTimeoutMs !== undefined ? { missingTerminalAfterToolTimeoutMs } : {}),
...(backendRetryMaxAttempts !== undefined ? { backendRetryMaxAttempts } : {}),
...(backendRetryInitialBackoffMs !== undefined ? { backendRetryInitialBackoffMs } : {}),
...(backendRetryMaxBackoffMs !== undefined ? { backendRetryMaxBackoffMs } : {}),
...(serviceAccountName ? { serviceAccountName } : {}),
...(jobNamePrefix ? { jobNamePrefix } : {}),
...(lane ? { lane } : {}),
@@ -244,6 +259,9 @@ export async function createKubernetesRunnerJob(options: { store: AgentRunStore;
ttlSecondsAfterFinished: render.ttlSecondsAfterFinished,
ttlPolicy: render.ttlPolicy,
runnerIdleTimeoutMs: render.runnerIdleTimeoutMs,
backendRetryMaxAttempts: render.backendRetryMaxAttempts,
backendRetryInitialBackoffMs: render.backendRetryInitialBackoffMs,
backendRetryMaxBackoffMs: render.backendRetryMaxBackoffMs,
preCreateCleanup: preCreateRetention,
},
pollActions: [
src/mgr/server.ts
+7
View File
@@ -57,6 +57,9 @@ function runnerJobDefaultsForRequest(defaults: ManagerServerOptions["runnerJobDe
lane,
...(defaults?.runnerIdleTimeoutMs !== undefined ? { runnerIdleTimeoutMs: defaults.runnerIdleTimeoutMs } : optionalPositiveIntegerRecord("runnerIdleTimeoutMs", process.env.AGENTRUN_RUNNER_IDLE_TIMEOUT_MS)),
...(defaults?.missingTerminalAfterToolTimeoutMs !== undefined ? { missingTerminalAfterToolTimeoutMs: defaults.missingTerminalAfterToolTimeoutMs } : optionalPositiveIntegerRecord("missingTerminalAfterToolTimeoutMs", process.env.AGENTRUN_RUNNER_MISSING_TERMINAL_AFTER_TOOL_TIMEOUT_MS)),
...(defaults?.backendRetryMaxAttempts !== undefined ? { backendRetryMaxAttempts: defaults.backendRetryMaxAttempts } : optionalPositiveIntegerRecord("backendRetryMaxAttempts", process.env.AGENTRUN_BACKEND_RETRY_MAX_ATTEMPTS)),
...(defaults?.backendRetryInitialBackoffMs !== undefined ? { backendRetryInitialBackoffMs: defaults.backendRetryInitialBackoffMs } : optionalPositiveIntegerRecord("backendRetryInitialBackoffMs", process.env.AGENTRUN_BACKEND_RETRY_INITIAL_BACKOFF_MS)),
...(defaults?.backendRetryMaxBackoffMs !== undefined ? { backendRetryMaxBackoffMs: defaults.backendRetryMaxBackoffMs } : optionalPositiveIntegerRecord("backendRetryMaxBackoffMs", process.env.AGENTRUN_BACKEND_RETRY_MAX_BACKOFF_MS)),
...(defaults?.kubectlCommand ? { kubectlCommand: defaults.kubectlCommand } : {}),
...(defaults?.unideskSshEndpointEnv ? { unideskSshEndpointEnv: defaults.unideskSshEndpointEnv } : {}),
...(retention ? { retention } : {}),
@@ -111,6 +114,10 @@ export interface ManagerServerOptions {
jobNamePrefix?: string;
lane?: string;
runnerIdleTimeoutMs?: number;
missingTerminalAfterToolTimeoutMs?: number;
backendRetryMaxAttempts?: number;
backendRetryInitialBackoffMs?: number;
backendRetryMaxBackoffMs?: number;
kubectlCommand?: string;
unideskSshEndpointEnv?: JsonRecord;
retention?: RunnerRetentionOptions;