mirror of
https://github.com/n8n-io/n8n.git
synced 2026-08-02 05:55:33 +02:00
263 lines
8.1 KiB
TypeScript
263 lines
8.1 KiB
TypeScript
import { Logger } from '@n8n/backend-common';
|
|
import { ExecutionRepository } from '@n8n/db';
|
|
import { OnLeaderStepdown, OnLeaderTakeover } from '@n8n/decorators';
|
|
import { Service } from '@n8n/di';
|
|
import { InstanceSettings } from 'n8n-core';
|
|
import { ensureError } from '@n8n/utils/errors/ensure-error';
|
|
import {
|
|
sleep,
|
|
UnexpectedError,
|
|
UserError,
|
|
type IRun,
|
|
type IWorkflowExecutionDataProcess,
|
|
type RelatedExecution,
|
|
} from 'n8n-workflow';
|
|
|
|
import { ActiveExecutions } from '@/active-executions';
|
|
import { ExecutionAlreadyResumingError } from '@/errors/execution-already-resuming.error';
|
|
import { ExecutionPersistence } from '@/executions/execution-persistence';
|
|
import { OwnershipService } from '@/services/ownership.service';
|
|
import { WorkflowRunner } from '@/workflow-runner';
|
|
|
|
import {
|
|
shouldRestartParentExecution,
|
|
updateParentExecutionWithChildResults,
|
|
} from './workflow-helpers';
|
|
|
|
/** How many times each parent-resume step is attempted before giving up. */
|
|
const MAX_PARENT_RESUME_ATTEMPTS = 3;
|
|
|
|
/**
|
|
* Whether a resume parent failure is worth retrying. Only `UserError` and
|
|
* `UnexpectedError` are not. Everything else is retried, including `OperationalError` (which
|
|
* by convention signals a transient issue) and raw database or Redis failures
|
|
*/
|
|
function isRetryableResumeError(error: unknown): boolean {
|
|
return !(error instanceof UserError || error instanceof UnexpectedError);
|
|
}
|
|
|
|
@Service()
|
|
export class WaitTracker {
|
|
private waitingExecutions: {
|
|
[key: string]: {
|
|
executionId: string;
|
|
timer: NodeJS.Timeout;
|
|
};
|
|
} = {};
|
|
|
|
mainTimer: NodeJS.Timeout;
|
|
|
|
constructor(
|
|
private readonly logger: Logger,
|
|
private readonly executionRepository: ExecutionRepository,
|
|
private readonly executionPersistence: ExecutionPersistence,
|
|
private readonly ownershipService: OwnershipService,
|
|
private readonly activeExecutions: ActiveExecutions,
|
|
private readonly workflowRunner: WorkflowRunner,
|
|
private readonly instanceSettings: InstanceSettings,
|
|
) {
|
|
this.logger = this.logger.scoped('waiting-executions');
|
|
}
|
|
|
|
has(executionId: string) {
|
|
return this.waitingExecutions[executionId] !== undefined;
|
|
}
|
|
|
|
init() {
|
|
if (this.instanceSettings.isLeader) this.startTracking();
|
|
}
|
|
|
|
@OnLeaderTakeover()
|
|
private startTracking() {
|
|
// Poll every 60 seconds a list of upcoming executions
|
|
this.mainTimer = setInterval(() => {
|
|
void this.getWaitingExecutions();
|
|
}, 60000);
|
|
|
|
void this.getWaitingExecutions();
|
|
|
|
this.logger.debug('Started tracking waiting executions');
|
|
}
|
|
|
|
async getWaitingExecutions() {
|
|
this.logger.debug('Querying database for waiting executions');
|
|
|
|
const executions = await this.executionRepository.getWaitingExecutions();
|
|
|
|
if (executions.length === 0) {
|
|
return;
|
|
}
|
|
|
|
const executionIds = executions.map((execution) => execution.id).join(', ');
|
|
this.logger.debug(
|
|
`Found ${executions.length} executions. Setting timer for IDs: ${executionIds}`,
|
|
);
|
|
|
|
// Add timers for each waiting execution that they get started at the correct time
|
|
|
|
for (const execution of executions) {
|
|
const executionId = execution.id;
|
|
if (this.waitingExecutions[executionId] === undefined) {
|
|
const triggerTime = execution.waitTill!.getTime() - new Date().getTime();
|
|
this.waitingExecutions[executionId] = {
|
|
executionId,
|
|
timer: setTimeout(() => {
|
|
void this.startExecution(executionId);
|
|
}, triggerTime),
|
|
};
|
|
}
|
|
}
|
|
}
|
|
|
|
stopExecution(executionId: string) {
|
|
if (!this.waitingExecutions[executionId]) return;
|
|
|
|
clearTimeout(this.waitingExecutions[executionId].timer);
|
|
|
|
delete this.waitingExecutions[executionId];
|
|
}
|
|
|
|
async startExecution(executionId: string) {
|
|
this.logger.debug(`Resuming execution ${executionId}`, { executionId });
|
|
delete this.waitingExecutions[executionId];
|
|
|
|
// Get the data to execute
|
|
const fullExecutionData = await this.executionPersistence.findSingleExecution(executionId, {
|
|
includeData: true,
|
|
unflattenData: true,
|
|
});
|
|
|
|
if (!fullExecutionData) {
|
|
throw new UnexpectedError('Execution does not exist.', { extra: { executionId } });
|
|
}
|
|
if (fullExecutionData.finished) {
|
|
throw new UnexpectedError('The execution did succeed and can so not be started again.');
|
|
}
|
|
|
|
if (!fullExecutionData.workflowData.id) {
|
|
throw new UnexpectedError('Only saved workflows can be resumed.');
|
|
}
|
|
|
|
const workflowId = fullExecutionData.workflowData.id;
|
|
const project = await this.ownershipService.getWorkflowProjectCached(workflowId);
|
|
|
|
const data: IWorkflowExecutionDataProcess = {
|
|
executionMode: fullExecutionData.mode,
|
|
executionData: fullExecutionData.data,
|
|
workflowData: fullExecutionData.workflowData,
|
|
projectId: project.id,
|
|
pushRef: fullExecutionData.data.pushRef,
|
|
startedAt: fullExecutionData.startedAt,
|
|
};
|
|
|
|
// Start the execution again
|
|
try {
|
|
await this.workflowRunner.run(data, false, false, executionId);
|
|
} catch (error) {
|
|
if (error instanceof ExecutionAlreadyResumingError) {
|
|
// This execution is already being resumed by another child execution
|
|
// This is expected in "run once for each item" mode when multiple children complete
|
|
this.logger.debug(
|
|
`Execution ${executionId} is already being resumed, skipping duplicate resume`,
|
|
{ executionId },
|
|
);
|
|
return;
|
|
}
|
|
// Rethrow any other errors
|
|
throw error;
|
|
}
|
|
|
|
const { parentExecution } = fullExecutionData.data;
|
|
if (shouldRestartParentExecution(parentExecution)) {
|
|
// on child execution completion, resume parent execution
|
|
void this.resumeParentExecution(
|
|
parentExecution,
|
|
this.activeExecutions.getPostExecutePromise(executionId),
|
|
{ executionId, workflowId },
|
|
);
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Resume a parent execution once its child execution has completed.
|
|
*
|
|
* The resume crosses several async boundaries (DB write to patch the parent,
|
|
* then resuming the parent). Each step is retried up to `MAX_PARENT_RESUME_ATTEMPTS`
|
|
* so a transient failure recovers and the parent resumes.
|
|
* If every attempt fails, the error is caught and logged below; the parent stays in `waiting`, but the
|
|
* failure is now visible and attributable instead of lost.
|
|
* This never rejects, so callers can invoke it as fire and forget.
|
|
*/
|
|
async resumeParentExecution(
|
|
parentExecution: RelatedExecution,
|
|
executePromise: Promise<IRun | undefined>,
|
|
childExecution?: RelatedExecution,
|
|
): Promise<void> {
|
|
try {
|
|
const subworkflowResults = await executePromise;
|
|
if (!subworkflowResults) return;
|
|
if (subworkflowResults.status === 'waiting') return; // The child execution is waiting, not completing.
|
|
|
|
await this.withRetry(
|
|
async () => {
|
|
await updateParentExecutionWithChildResults(
|
|
parentExecution.executionId,
|
|
subworkflowResults,
|
|
childExecution,
|
|
);
|
|
},
|
|
MAX_PARENT_RESUME_ATTEMPTS,
|
|
isRetryableResumeError,
|
|
);
|
|
|
|
await this.withRetry(
|
|
async () => {
|
|
await this.startExecution(parentExecution.executionId);
|
|
},
|
|
MAX_PARENT_RESUME_ATTEMPTS,
|
|
isRetryableResumeError,
|
|
);
|
|
} catch (error) {
|
|
this.logger.error('Failed to resume parent execution after sub-workflow completed', {
|
|
parentExecutionId: parentExecution.executionId,
|
|
error: ensureError(error).message,
|
|
});
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Run an operation up to `maxAttempts` times with exponential backoff, returning
|
|
* on the first success and rethrowing the last error if they all fail. Generic
|
|
* (not specific to parent resume) — the caller passes the attempt count and an
|
|
* optional `shouldRetry` predicate; an error it rejects is rethrown immediately
|
|
* instead of being retried.
|
|
*/
|
|
private async withRetry(
|
|
operation: () => Promise<void>,
|
|
maxAttempts: number,
|
|
shouldRetry: (error: unknown) => boolean = () => true,
|
|
): Promise<void> {
|
|
for (let attempt = 1; ; attempt++) {
|
|
try {
|
|
await operation();
|
|
return;
|
|
} catch (error) {
|
|
if (attempt >= maxAttempts || !shouldRetry(error)) throw error;
|
|
await sleep(100 * 2 ** (attempt - 1));
|
|
}
|
|
}
|
|
}
|
|
|
|
@OnLeaderStepdown()
|
|
stopTracking() {
|
|
if (!this.mainTimer) return;
|
|
|
|
clearInterval(this.mainTimer);
|
|
Object.keys(this.waitingExecutions).forEach((executionId) => {
|
|
clearTimeout(this.waitingExecutions[executionId].timer);
|
|
});
|
|
|
|
this.logger.debug('Stopped tracking waiting executions');
|
|
}
|
|
}
|