n8n/packages/cli/commands/worker.ts

270 lines
8.5 KiB
TypeScript
Raw Normal View History

:sparkles: Unify execution id + Queue system (#1340) * Unify execution ID across executions * Fix indentation and improved comments * WIP: saving data after each node execution * Added on/off to save data after each step, saving initial data and retries working * Fixing lint issues * Fixing more lint issues * :sparkles: Add bull to execute workflows * :shirt: Fix lint issue * :zap: Add graceful shutdown to worker * :zap: Add loading staticData to worker * :shirt: Fix lint issue * :zap: Fix import * Changed tables metadata to add nullable to stoppedAt * Reload database on migration run * Fixed reloading database schema for sqlite by reconnecting and fixing postgres migration * Added checks to Redis and exiting process if connection is unavailable * Fixing error with new installations * Fix issue with data not being sent back to browser on manual executions with defined destination * Merging bull and unify execution id branch fixes * Main process will now get execution success from database instead of redis * Omit execution duration if execution did not stop * Fix issue with execution list displaying inconsistant information information while a workflow is running * Remove unused hooks to clarify for developers that these wont run in queue mode * Added active pooling to help recover from Redis crashes * Lint issues * Changing default polling interval to 60 seconds * Removed unnecessary attributes from bull job * :zap: Improved output on worker job start Co-authored-by: Jan Oberhauser <jan.oberhauser@gmail.com>
2021-02-08 23:59:32 -08:00
import * as PCancelable from 'p-cancelable';
import { Command, flags } from '@oclif/command';
import {
UserSettings,
WorkflowExecute,
} from 'n8n-core';
import {
IDataObject,
INodeTypes,
IRun,
IWorkflowExecuteHooks,
Workflow,
WorkflowHooks,
} from 'n8n-workflow';
import {
FindOneOptions,
} from 'typeorm';
import {
ActiveExecutions,
CredentialsOverwrites,
CredentialTypes,
Db,
ExternalHooks,
GenericHelpers,
IBullJobData,
IBullJobResponse,
IExecutionFlattedDb,
IExecutionResponse,
LoadNodesAndCredentials,
NodeTypes,
ResponseHelper,
WorkflowCredentials,
WorkflowExecuteAdditionalData,
} from "../src";
import * as config from '../config';
import * as Bull from 'bull';
:sparkles: Separate webhooks from core (#1408) * Unify execution ID across executions * Fix indentation and improved comments * WIP: saving data after each node execution * Added on/off to save data after each step, saving initial data and retries working * Fixing lint issues * Fixing more lint issues * :sparkles: Add bull to execute workflows * :shirt: Fix lint issue * :zap: Add graceful shutdown to worker * :zap: Add loading staticData to worker * :shirt: Fix lint issue * :zap: Fix import * Changed tables metadata to add nullable to stoppedAt * Reload database on migration run * Fixed reloading database schema for sqlite by reconnecting and fixing postgres migration * Added checks to Redis and exiting process if connection is unavailable * Fixing error with new installations * Fix issue with data not being sent back to browser on manual executions with defined destination * Merging bull and unify execution id branch fixes * Main process will now get execution success from database instead of redis * Omit execution duration if execution did not stop * Fix issue with execution list displaying inconsistant information information while a workflow is running * Remove unused hooks to clarify for developers that these wont run in queue mode * Added active pooling to help recover from Redis crashes * Lint issues * Changing default polling interval to 60 seconds * Removed unnecessary attributes from bull job * Added webhooks service and setting to disable webhooks from main process * Fixed executions list when running with queues. Now we get the list of actively running workflows from bull. * Add option to disable deregistration of webhooks on shutdown * Rename WEBHOOK_TUNNEL_URL to WEBHOOK_URL keeping backwards compat. * Added auto refresh to executions list * Improvements to workflow stop process when running with queues * Refactor queue system to use a singleton and avoid code duplication * Improve comments and remove unnecessary commits * Remove console.log from vue file * Blocking webhook process to run without queues * Handling execution stop graciously when possible * Removing initialization of all workflows from webhook process * Refactoring code to remove code duplication for job stop * Improved execution list to be more fluid and less intrusive * Fixing workflow name for current executions when auto updating * :zap: Right align autorefresh checkbox Co-authored-by: Jan Oberhauser <jan.oberhauser@gmail.com>
2021-02-09 14:32:40 -08:00
import * as Queue from '../src/Queue';
:sparkles: Unify execution id + Queue system (#1340) * Unify execution ID across executions * Fix indentation and improved comments * WIP: saving data after each node execution * Added on/off to save data after each step, saving initial data and retries working * Fixing lint issues * Fixing more lint issues * :sparkles: Add bull to execute workflows * :shirt: Fix lint issue * :zap: Add graceful shutdown to worker * :zap: Add loading staticData to worker * :shirt: Fix lint issue * :zap: Fix import * Changed tables metadata to add nullable to stoppedAt * Reload database on migration run * Fixed reloading database schema for sqlite by reconnecting and fixing postgres migration * Added checks to Redis and exiting process if connection is unavailable * Fixing error with new installations * Fix issue with data not being sent back to browser on manual executions with defined destination * Merging bull and unify execution id branch fixes * Main process will now get execution success from database instead of redis * Omit execution duration if execution did not stop * Fix issue with execution list displaying inconsistant information information while a workflow is running * Remove unused hooks to clarify for developers that these wont run in queue mode * Added active pooling to help recover from Redis crashes * Lint issues * Changing default polling interval to 60 seconds * Removed unnecessary attributes from bull job * :zap: Improved output on worker job start Co-authored-by: Jan Oberhauser <jan.oberhauser@gmail.com>
2021-02-08 23:59:32 -08:00
export class Worker extends Command {
static description = '\nStarts a n8n worker';
static examples = [
`$ n8n worker --concurrency=5`,
];
static flags = {
help: flags.help({ char: 'h' }),
concurrency: flags.integer({
default: 10,
description: 'How many jobs can run in parallel.',
}),
};
static runningJobs: {
[key: string]: PCancelable<IRun>;
} = {};
static jobQueue: Bull.Queue;
static processExistCode = 0;
// static activeExecutions = ActiveExecutions.getInstance();
/**
* Stoppes the n8n in a graceful way.
* Make for example sure that all the webhooks from third party services
* get removed.
*/
static async stopProcess() {
console.log(`\nStopping n8n...`);
// Stop accepting new jobs
Worker.jobQueue.pause(true);
try {
const externalHooks = ExternalHooks();
await externalHooks.run('n8n.stop', []);
const maxStopTime = 30000;
const stopTime = new Date().getTime() + maxStopTime;
setTimeout(() => {
// In case that something goes wrong with shutdown we
// kill after max. 30 seconds no matter what
process.exit(Worker.processExistCode);
}, maxStopTime);
// Wait for active workflow executions to finish
let count = 0;
while (Object.keys(Worker.runningJobs).length !== 0) {
if (count++ % 4 === 0) {
const waitLeft = Math.ceil((stopTime - new Date().getTime()) / 1000);
console.log(`Waiting for ${Object.keys(Worker.runningJobs).length} active executions to finish... (wait ${waitLeft} more seconds)`);
}
await new Promise((resolve) => {
setTimeout(resolve, 500);
});
}
} catch (error) {
console.error('There was an error shutting down n8n.', error);
}
process.exit(Worker.processExistCode);
}
async runJob(job: Bull.Job, nodeTypes: INodeTypes): Promise<IBullJobResponse> {
const jobData = job.data as IBullJobData;
const executionDb = await Db.collections.Execution!.findOne(jobData.executionId) as IExecutionFlattedDb;
const currentExecutionDb = ResponseHelper.unflattenExecutionData(executionDb) as IExecutionResponse;
console.log(`Start job: ${job.id} (Workflow ID: ${currentExecutionDb.workflowData.id} | Execution: ${jobData.executionId})`);
let staticData = currentExecutionDb.workflowData!.staticData;
if (jobData.loadStaticData === true) {
const findOptions = {
select: ['id', 'staticData'],
} as FindOneOptions;
const workflowData = await Db.collections!.Workflow!.findOne(currentExecutionDb.workflowData.id, findOptions);
if (workflowData === undefined) {
throw new Error(`The workflow with the ID "${currentExecutionDb.workflowData.id}" could not be found`);
}
staticData = workflowData.staticData;
}
const workflow = new Workflow({ id: currentExecutionDb.workflowData.id as string, name: currentExecutionDb.workflowData.name, nodes: currentExecutionDb.workflowData!.nodes, connections: currentExecutionDb.workflowData!.connections, active: currentExecutionDb.workflowData!.active, nodeTypes, staticData, settings: currentExecutionDb.workflowData!.settings });
const credentials = await WorkflowCredentials(currentExecutionDb.workflowData.nodes);
const additionalData = await WorkflowExecuteAdditionalData.getBase(credentials);
additionalData.hooks = WorkflowExecuteAdditionalData.getWorkflowHooksWorkerExecuter(currentExecutionDb.mode, job.data.executionId, currentExecutionDb.workflowData, { retryOf: currentExecutionDb.retryOf as string });
:sparkles: Unify execution id + Queue system (#1340) * Unify execution ID across executions * Fix indentation and improved comments * WIP: saving data after each node execution * Added on/off to save data after each step, saving initial data and retries working * Fixing lint issues * Fixing more lint issues * :sparkles: Add bull to execute workflows * :shirt: Fix lint issue * :zap: Add graceful shutdown to worker * :zap: Add loading staticData to worker * :shirt: Fix lint issue * :zap: Fix import * Changed tables metadata to add nullable to stoppedAt * Reload database on migration run * Fixed reloading database schema for sqlite by reconnecting and fixing postgres migration * Added checks to Redis and exiting process if connection is unavailable * Fixing error with new installations * Fix issue with data not being sent back to browser on manual executions with defined destination * Merging bull and unify execution id branch fixes * Main process will now get execution success from database instead of redis * Omit execution duration if execution did not stop * Fix issue with execution list displaying inconsistant information information while a workflow is running * Remove unused hooks to clarify for developers that these wont run in queue mode * Added active pooling to help recover from Redis crashes * Lint issues * Changing default polling interval to 60 seconds * Removed unnecessary attributes from bull job * :zap: Improved output on worker job start Co-authored-by: Jan Oberhauser <jan.oberhauser@gmail.com>
2021-02-08 23:59:32 -08:00
let workflowExecute: WorkflowExecute;
let workflowRun: PCancelable<IRun>;
if (currentExecutionDb.data !== undefined) {
workflowExecute = new WorkflowExecute(additionalData, currentExecutionDb.mode, currentExecutionDb.data);
workflowRun = workflowExecute.processRunExecutionData(workflow);
} else {
// Execute all nodes
// Can execute without webhook so go on
workflowExecute = new WorkflowExecute(additionalData, currentExecutionDb.mode);
workflowRun = workflowExecute.run(workflow);
}
Worker.runningJobs[job.id] = workflowRun;
// Wait till the execution is finished
const runData = await workflowRun;
delete Worker.runningJobs[job.id];
return {
success: true,
};
}
async run() {
console.log('Starting n8n worker...');
// Make sure that n8n shuts down gracefully if possible
process.on('SIGTERM', Worker.stopProcess);
process.on('SIGINT', Worker.stopProcess);
// Wrap that the process does not close but we can still use async
await (async () => {
try {
const { flags } = this.parse(Worker);
// Start directly with the init of the database to improve startup time
const startDbInitPromise = Db.init().catch(error => {
console.error(`There was an error initializing DB: ${error.message}`);
Worker.processExistCode = 1;
// @ts-ignore
process.emit('SIGINT');
});
// Make sure the settings exist
await UserSettings.prepareUserSettings();
// Load all node and credential types
const loadNodesAndCredentials = LoadNodesAndCredentials();
await loadNodesAndCredentials.init();
// Load the credentials overwrites if any exist
const credentialsOverwrites = CredentialsOverwrites();
await credentialsOverwrites.init();
// Load all external hooks
const externalHooks = ExternalHooks();
await externalHooks.init();
// Add the found types to an instance other parts of the application can use
const nodeTypes = NodeTypes();
await nodeTypes.init(loadNodesAndCredentials.nodeTypes);
const credentialTypes = CredentialTypes();
await credentialTypes.init(loadNodesAndCredentials.credentialTypes);
// Wait till the database is ready
await startDbInitPromise;
const redisConnectionTimeoutLimit = config.get('queue.bull.redis.timeoutThreshold');
:sparkles: Separate webhooks from core (#1408) * Unify execution ID across executions * Fix indentation and improved comments * WIP: saving data after each node execution * Added on/off to save data after each step, saving initial data and retries working * Fixing lint issues * Fixing more lint issues * :sparkles: Add bull to execute workflows * :shirt: Fix lint issue * :zap: Add graceful shutdown to worker * :zap: Add loading staticData to worker * :shirt: Fix lint issue * :zap: Fix import * Changed tables metadata to add nullable to stoppedAt * Reload database on migration run * Fixed reloading database schema for sqlite by reconnecting and fixing postgres migration * Added checks to Redis and exiting process if connection is unavailable * Fixing error with new installations * Fix issue with data not being sent back to browser on manual executions with defined destination * Merging bull and unify execution id branch fixes * Main process will now get execution success from database instead of redis * Omit execution duration if execution did not stop * Fix issue with execution list displaying inconsistant information information while a workflow is running * Remove unused hooks to clarify for developers that these wont run in queue mode * Added active pooling to help recover from Redis crashes * Lint issues * Changing default polling interval to 60 seconds * Removed unnecessary attributes from bull job * Added webhooks service and setting to disable webhooks from main process * Fixed executions list when running with queues. Now we get the list of actively running workflows from bull. * Add option to disable deregistration of webhooks on shutdown * Rename WEBHOOK_TUNNEL_URL to WEBHOOK_URL keeping backwards compat. * Added auto refresh to executions list * Improvements to workflow stop process when running with queues * Refactor queue system to use a singleton and avoid code duplication * Improve comments and remove unnecessary commits * Remove console.log from vue file * Blocking webhook process to run without queues * Handling execution stop graciously when possible * Removing initialization of all workflows from webhook process * Refactoring code to remove code duplication for job stop * Improved execution list to be more fluid and less intrusive * Fixing workflow name for current executions when auto updating * :zap: Right align autorefresh checkbox Co-authored-by: Jan Oberhauser <jan.oberhauser@gmail.com>
2021-02-09 14:32:40 -08:00
Worker.jobQueue = Queue.getInstance().getBullObjectInstance();
:sparkles: Unify execution id + Queue system (#1340) * Unify execution ID across executions * Fix indentation and improved comments * WIP: saving data after each node execution * Added on/off to save data after each step, saving initial data and retries working * Fixing lint issues * Fixing more lint issues * :sparkles: Add bull to execute workflows * :shirt: Fix lint issue * :zap: Add graceful shutdown to worker * :zap: Add loading staticData to worker * :shirt: Fix lint issue * :zap: Fix import * Changed tables metadata to add nullable to stoppedAt * Reload database on migration run * Fixed reloading database schema for sqlite by reconnecting and fixing postgres migration * Added checks to Redis and exiting process if connection is unavailable * Fixing error with new installations * Fix issue with data not being sent back to browser on manual executions with defined destination * Merging bull and unify execution id branch fixes * Main process will now get execution success from database instead of redis * Omit execution duration if execution did not stop * Fix issue with execution list displaying inconsistant information information while a workflow is running * Remove unused hooks to clarify for developers that these wont run in queue mode * Added active pooling to help recover from Redis crashes * Lint issues * Changing default polling interval to 60 seconds * Removed unnecessary attributes from bull job * :zap: Improved output on worker job start Co-authored-by: Jan Oberhauser <jan.oberhauser@gmail.com>
2021-02-08 23:59:32 -08:00
Worker.jobQueue.process(flags.concurrency, (job) => this.runJob(job, nodeTypes));
const versions = await GenericHelpers.getVersions();
console.log('\nn8n worker is now ready');
console.log(` * Version: ${versions.cli}`);
console.log(` * Concurrency: ${flags.concurrency}`);
console.log('');
Worker.jobQueue.on('global:progress', (jobId, progress) => {
// Progress of a job got updated which does get used
// to communicate that a job got canceled.
if (progress === -1) {
// Job has to get canceled
if (Worker.runningJobs[jobId] !== undefined) {
// Job is processed by current worker so cancel
Worker.runningJobs[jobId].cancel();
delete Worker.runningJobs[jobId];
}
}
});
let lastTimer = 0, cumulativeTimeout = 0;
Worker.jobQueue.on('error', (error: Error) => {
if (error.toString().includes('ECONNREFUSED') === true) {
const now = Date.now();
if (now - lastTimer > 30000) {
// Means we had no timeout at all or last timeout was temporary and we recovered
lastTimer = now;
cumulativeTimeout = 0;
} else {
cumulativeTimeout += now - lastTimer;
lastTimer = now;
if (cumulativeTimeout > redisConnectionTimeoutLimit) {
console.error('Unable to connect to Redis after ' + redisConnectionTimeoutLimit + ". Exiting process.");
process.exit(1);
}
}
console.warn('Redis unavailable - trying to reconnect...');
} else if (error.toString().includes('Error initializing Lua scripts') === true) {
// This is a non-recoverable error
// Happens when worker starts and Redis is unavailable
// Even if Redis comes back online, worker will be zombie
console.error('Error initializing worker.');
process.exit(2);
} else {
console.error('Error from queue: ', error);
}
});
} catch (error) {
this.error(`There was an error: ${error.message}`);
Worker.processExistCode = 1;
// @ts-ignore
process.emit('SIGINT');
}
})();
}
}