import { formatEmbeddedAgentExecutionPhase } from "../../agents/pi-embedded-runner/execution-phase.js";

import { readSessionEntry } from "../../config/sessions/store-load.js";

import type { SessionEntry } from "../../config/sessions/types.js";

import type { CronConfig, CronRetryOn } from "../../config/types.cron.js";

import type { CronConfig } from "../../config/types.cron.js";

import type { HeartbeatRunResult } from "../../infra/heartbeat-wake.js";

import {

HEARTBEAT_SKIP_CRON_IN_PROGRESS,

import { clearCronJobActive, markCronJobActive } from "../active-jobs.js";

import { resolveCronDeliveryPlan, resolveFailureDestination } from "../delivery-plan.js";

import { resolveCronAgentSessionKey } from "../isolated-agent/session-key.js";

import { resolveCronExecutionRetryHint } from "../retry-hint.js";

import {

createCronRunDiagnosticsFromError,

normalizeCronRunDiagnostics,

/** Default max retries for one-shot jobs on transient errors (#24355). */

const DEFAULT_MAX_TRANSIENT_RETRIES = 3;

const TRANSIENT_PATTERNS: Record<string, RegExp> = {

rate_limit:

/(rate[_ ]limit|too many requests|429|resource has been exhausted|cloudflare|tokens per day)/i,

overloaded:

/\b529\b|\boverloaded(?:_error)?\b|high demand|temporar(?:ily|y) overloaded|capacity exceeded/i,

network: /(network|econnreset|econnrefused|fetch failed|socket)/i,

timeout: /(timeout|etimedout)/i,

server_error: /\b5\d{2}\b/,

};

function isTransientCronError(error: string | undefined, retryOn?: CronRetryOn[]): boolean {

if (!error || typeof error !== "string") {

return false;

}

const keys = retryOn?.length ? retryOn : (Object.keys(TRANSIENT_PATTERNS) as CronRetryOn[]);

const classified = resolveFailoverReasonFromError(error);

if (classified && keys.includes(classified as CronRetryOn)) {

return true;

}

return keys.some((k) => TRANSIENT_PATTERNS[k]?.test(error));

}

function resolveCronNextRunWithLowerBound(params: {

state: CronServiceState;

job: CronJob;

job.state.nextRunAtMs = undefined;

} else if (result.status === "error") {

const retryConfig = resolveRetryConfig(state.deps.cronConfig);

const transient = isTransientCronError(result.error, retryConfig.retryOn);

const retryHint = resolveCronExecutionRetryHint(

result.error,

retryConfig.retryOn,

job.state.lastErrorReason,

);

// consecutiveErrors is always set to ≥1 by the increment block above.

const consecutive = job.state.consecutiveErrors;

if (transient && consecutive <= retryConfig.maxAttempts) {

if (retryHint.retryable && consecutive <= retryConfig.maxAttempts) {

// Schedule retry with backoff (#24355).

const backoff = errorBackoffMs(consecutive, retryConfig.backoffMs);

job.state.nextRunAtMs = result.endedAt + backoff;

jobName: job.name,

consecutiveErrors: consecutive,

error: result.error,

reason: transient ? "max retries exhausted" : "permanent error",

reason: retryHint.retryable ? "max retries exhausted" : "permanent error",

retryCategory: retryHint.category,

},

"cron: disabling one-shot job after error",

);