mirror of
https://github.com/n8n-io/n8n.git
synced 2024-12-24 20:24:05 -08:00
7ce7285f7a
* Changes to types so that credentials can be always loaded from DB This first commit changes all return types from the execute functions and calls to get credentials to be async so we can use await. This is a first step as previously credentials were loaded in memory and always available. We will now be loading them from the DB which requires turning the whole call chain async. * Fix updated files * Removed unnecessary credential loading to improve performance * Fix typo * ⚡ Fix issue * Updated new nodes to load credentials async * ⚡ Remove not needed comment Co-authored-by: Jan Oberhauser <jan.oberhauser@gmail.com>
808 lines
26 KiB
TypeScript
808 lines
26 KiB
TypeScript
import * as fs from 'fs';
|
|
import {
|
|
Command,
|
|
flags,
|
|
} from '@oclif/command';
|
|
|
|
import {
|
|
UserSettings,
|
|
} from 'n8n-core';
|
|
|
|
import {
|
|
INode,
|
|
INodeExecutionData,
|
|
ITaskData,
|
|
} from 'n8n-workflow';
|
|
|
|
import {
|
|
ActiveExecutions,
|
|
CredentialsOverwrites,
|
|
CredentialTypes,
|
|
Db,
|
|
ExternalHooks,
|
|
IExecutionsCurrentSummary,
|
|
IWorkflowDb,
|
|
IWorkflowExecutionDataProcess,
|
|
LoadNodesAndCredentials,
|
|
NodeTypes,
|
|
WorkflowCredentials,
|
|
WorkflowRunner,
|
|
} from '../src';
|
|
|
|
import {
|
|
sep,
|
|
} from 'path';
|
|
|
|
import {
|
|
diff,
|
|
} from 'json-diff';
|
|
|
|
import {
|
|
getLogger,
|
|
} from '../src/Logger';
|
|
|
|
import {
|
|
LoggerProxy,
|
|
} from 'n8n-workflow';
|
|
|
|
import {
|
|
pick,
|
|
} from 'lodash';
|
|
|
|
export class ExecuteBatch extends Command {
|
|
static description = '\nExecutes multiple workflows once';
|
|
|
|
static cancelled = false;
|
|
|
|
static workflowExecutionsProgress: IWorkflowExecutionProgress[][];
|
|
|
|
static shallow = false;
|
|
|
|
static compare: string;
|
|
|
|
static snapshot: string;
|
|
|
|
static concurrency = 1;
|
|
|
|
static debug = false;
|
|
|
|
static executionTimeout = 3 * 60 * 1000;
|
|
|
|
static examples = [
|
|
`$ n8n executeAll`,
|
|
`$ n8n executeAll --concurrency=10 --skipList=/data/skipList.txt`,
|
|
`$ n8n executeAll --debug --output=/data/output.json`,
|
|
`$ n8n executeAll --ids=10,13,15 --shortOutput`,
|
|
`$ n8n executeAll --snapshot=/data/snapshots --shallow`,
|
|
`$ n8n executeAll --compare=/data/previousExecutionData --retries=2`,
|
|
];
|
|
|
|
static flags = {
|
|
help: flags.help({ char: 'h' }),
|
|
debug: flags.boolean({
|
|
description: 'Toggles on displaying all errors and debug messages.',
|
|
}),
|
|
ids: flags.string({
|
|
description: 'Specifies workflow IDs to get executed, separated by a comma.',
|
|
}),
|
|
concurrency: flags.integer({
|
|
default: 1,
|
|
description: 'How many workflows can run in parallel. Defaults to 1 which means no concurrency.',
|
|
}),
|
|
output: flags.string({
|
|
description: 'Enable execution saving, You must inform an existing folder to save execution via this param',
|
|
}),
|
|
snapshot: flags.string({
|
|
description: 'Enables snapshot saving. You must inform an existing folder to save snapshots via this param.',
|
|
}),
|
|
compare: flags.string({
|
|
description: 'Compares current execution with an existing snapshot. You must inform an existing folder where the snapshots are saved.',
|
|
}),
|
|
shallow: flags.boolean({
|
|
description: 'Compares only if attributes output from node are the same, with no regards to neste JSON objects.',
|
|
}),
|
|
skipList: flags.string({
|
|
description: 'File containing a comma separated list of workflow IDs to skip.',
|
|
}),
|
|
retries: flags.integer({
|
|
description: 'Retries failed workflows up to N tries. Default is 1. Set 0 to disable.',
|
|
default: 1,
|
|
}),
|
|
shortOutput: flags.boolean({
|
|
description: 'Omits the full execution information from output, displaying only summary.',
|
|
}),
|
|
};
|
|
|
|
/**
|
|
* Gracefully handles exit.
|
|
* @param {boolean} skipExit Whether to skip exit or number according to received signal
|
|
*/
|
|
static async stopProcess(skipExit: boolean | number = false) {
|
|
|
|
if (ExecuteBatch.cancelled === true) {
|
|
process.exit(0);
|
|
}
|
|
|
|
ExecuteBatch.cancelled = true;
|
|
const activeExecutionsInstance = ActiveExecutions.getInstance();
|
|
const stopPromises = activeExecutionsInstance.getActiveExecutions().map(async execution => {
|
|
activeExecutionsInstance.stopExecution(execution.id);
|
|
});
|
|
|
|
await Promise.allSettled(stopPromises);
|
|
|
|
setTimeout(() => {
|
|
process.exit(0);
|
|
}, 30000);
|
|
|
|
let executingWorkflows = activeExecutionsInstance.getActiveExecutions() as IExecutionsCurrentSummary[];
|
|
|
|
let count = 0;
|
|
while (executingWorkflows.length !== 0) {
|
|
if (count++ % 4 === 0) {
|
|
console.log(`Waiting for ${executingWorkflows.length} active executions to finish...`);
|
|
executingWorkflows.map(execution => {
|
|
console.log(` - Execution ID ${execution.id}, workflow ID: ${execution.workflowId}`);
|
|
});
|
|
}
|
|
await new Promise((resolve) => {
|
|
setTimeout(resolve, 500);
|
|
});
|
|
executingWorkflows = activeExecutionsInstance.getActiveExecutions();
|
|
}
|
|
// We may receive true but when called from `process.on`
|
|
// we get the signal (SIGNIT, etc.)
|
|
if (skipExit !== true) {
|
|
process.exit(0);
|
|
}
|
|
}
|
|
|
|
formatJsonOutput(data: object) {
|
|
return JSON.stringify(data, null, 2);
|
|
}
|
|
|
|
shouldBeConsideredAsWarning(errorMessage: string) {
|
|
|
|
const warningStrings = [
|
|
'refresh token is invalid',
|
|
'unable to connect to',
|
|
'econnreset',
|
|
'429',
|
|
'econnrefused',
|
|
'missing a required parameter',
|
|
'insufficient credit balance',
|
|
'request timed out',
|
|
];
|
|
|
|
errorMessage = errorMessage.toLowerCase();
|
|
|
|
for (let i = 0; i < warningStrings.length; i++) {
|
|
if (errorMessage.includes(warningStrings[i])) {
|
|
return true;
|
|
}
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
|
|
async run() {
|
|
|
|
process.on('SIGTERM', ExecuteBatch.stopProcess);
|
|
process.on('SIGINT', ExecuteBatch.stopProcess);
|
|
|
|
const logger = getLogger();
|
|
LoggerProxy.init(logger);
|
|
|
|
const { flags } = this.parse(ExecuteBatch);
|
|
|
|
ExecuteBatch.debug = flags.debug === true;
|
|
ExecuteBatch.concurrency = flags.concurrency || 1;
|
|
|
|
const ids: number[] = [];
|
|
const skipIds: number[] = [];
|
|
|
|
if (flags.snapshot !== undefined) {
|
|
if (fs.existsSync(flags.snapshot)) {
|
|
if (!fs.lstatSync(flags.snapshot).isDirectory()) {
|
|
console.log(`The parameter --snapshot must be an existing directory`);
|
|
return;
|
|
}
|
|
} else {
|
|
console.log(`The parameter --snapshot must be an existing directory`);
|
|
return;
|
|
}
|
|
|
|
ExecuteBatch.snapshot = flags.snapshot;
|
|
}
|
|
if (flags.compare !== undefined) {
|
|
if (fs.existsSync(flags.compare)) {
|
|
if (!fs.lstatSync(flags.compare).isDirectory()) {
|
|
console.log(`The parameter --compare must be an existing directory`);
|
|
return;
|
|
}
|
|
} else {
|
|
console.log(`The parameter --compare must be an existing directory`);
|
|
return;
|
|
}
|
|
|
|
ExecuteBatch.compare = flags.compare;
|
|
}
|
|
|
|
if (flags.output !== undefined) {
|
|
if (fs.existsSync(flags.output)) {
|
|
if (fs.lstatSync(flags.output).isDirectory()) {
|
|
console.log(`The parameter --output must be a writable file`);
|
|
return;
|
|
}
|
|
}
|
|
}
|
|
|
|
if (flags.ids !== undefined) {
|
|
const paramIds = flags.ids.split(',');
|
|
const re = /\d+/;
|
|
const matchedIds = paramIds.filter(id => id.match(re)).map(id => parseInt(id.trim(), 10));
|
|
|
|
if (matchedIds.length === 0) {
|
|
console.log(`The parameter --ids must be a list of numeric IDs separated by a comma.`);
|
|
return;
|
|
}
|
|
|
|
ids.push(...matchedIds);
|
|
}
|
|
|
|
if (flags.skipList !== undefined) {
|
|
if (fs.existsSync(flags.skipList)) {
|
|
const contents = fs.readFileSync(flags.skipList, { encoding: 'utf-8' });
|
|
skipIds.push(...contents.split(',').map(id => parseInt(id.trim(), 10)));
|
|
} else {
|
|
console.log('Skip list file not found. Exiting.');
|
|
return;
|
|
}
|
|
}
|
|
|
|
if (flags.shallow === true) {
|
|
ExecuteBatch.shallow = true;
|
|
}
|
|
|
|
|
|
// Start directly with the init of the database to improve startup time
|
|
const startDbInitPromise = Db.init();
|
|
|
|
// Load all node and credential types
|
|
const loadNodesAndCredentials = LoadNodesAndCredentials();
|
|
const loadNodesAndCredentialsPromise = loadNodesAndCredentials.init();
|
|
|
|
// Make sure the settings exist
|
|
await UserSettings.prepareUserSettings();
|
|
|
|
// Wait till the database is ready
|
|
await startDbInitPromise;
|
|
|
|
let allWorkflows;
|
|
|
|
const query = Db.collections!.Workflow!.createQueryBuilder('workflows');
|
|
|
|
if (ids.length > 0) {
|
|
query.andWhere(`workflows.id in (:...ids)`, { ids });
|
|
}
|
|
|
|
if (skipIds.length > 0) {
|
|
query.andWhere(`workflows.id not in (:...skipIds)`, { skipIds });
|
|
}
|
|
|
|
allWorkflows = await query.getMany() as IWorkflowDb[];
|
|
|
|
if (ExecuteBatch.debug === true) {
|
|
process.stdout.write(`Found ${allWorkflows.length} workflows to execute.\n`);
|
|
}
|
|
|
|
// Wait till the n8n-packages have been read
|
|
await loadNodesAndCredentialsPromise;
|
|
|
|
// Load the credentials overwrites if any exist
|
|
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);
|
|
|
|
// Send a shallow copy of allWorkflows so we still have all workflow data.
|
|
const results = await this.runTests([...allWorkflows]);
|
|
|
|
let { retries } = flags;
|
|
|
|
while (retries > 0 && (results.summary.warningExecutions + results.summary.failedExecutions > 0) && ExecuteBatch.cancelled === false) {
|
|
const failedWorkflowIds = results.summary.errors.map(execution => execution.workflowId);
|
|
failedWorkflowIds.push(...results.summary.warnings.map(execution => execution.workflowId));
|
|
|
|
const newWorkflowList = allWorkflows.filter(workflow => failedWorkflowIds.includes(workflow.id));
|
|
|
|
const retryResults = await this.runTests(newWorkflowList);
|
|
|
|
this.mergeResults(results, retryResults);
|
|
// By now, `results` has been updated with the new successful executions.
|
|
retries--;
|
|
}
|
|
|
|
if (flags.output !== undefined) {
|
|
fs.writeFileSync(flags.output, this.formatJsonOutput(results));
|
|
console.log('\nExecution finished.');
|
|
console.log('Summary:');
|
|
console.log(`\tSuccess: ${results.summary.successfulExecutions}`);
|
|
console.log(`\tFailures: ${results.summary.failedExecutions}`);
|
|
console.log(`\tWarnings: ${results.summary.warningExecutions}`);
|
|
console.log('\nNodes successfully tested:');
|
|
Object.entries(results.coveredNodes).forEach(([nodeName, nodeCount]) => {
|
|
console.log(`\t${nodeName}: ${nodeCount}`);
|
|
});
|
|
console.log('\nCheck the JSON file for more details.');
|
|
} else {
|
|
if (flags.shortOutput === true) {
|
|
console.log(this.formatJsonOutput({ ...results, executions: results.executions.filter(execution => execution.executionStatus !== 'success') }));
|
|
} else {
|
|
console.log(this.formatJsonOutput(results));
|
|
}
|
|
}
|
|
|
|
await ExecuteBatch.stopProcess(true);
|
|
|
|
if (results.summary.failedExecutions > 0) {
|
|
this.exit(1);
|
|
}
|
|
this.exit(0);
|
|
|
|
}
|
|
|
|
mergeResults(results: IResult, retryResults: IResult) {
|
|
|
|
if (retryResults.summary.successfulExecutions === 0) {
|
|
// Nothing to replace.
|
|
return;
|
|
}
|
|
|
|
// Find successful executions and replace them on previous result.
|
|
retryResults.executions.forEach(newExecution => {
|
|
if (newExecution.executionStatus === 'success') {
|
|
// Remove previous execution from list.
|
|
results.executions = results.executions.filter(previousExecutions => previousExecutions.workflowId !== newExecution.workflowId);
|
|
|
|
const errorIndex = results.summary.errors.findIndex(summaryInformation => summaryInformation.workflowId === newExecution.workflowId);
|
|
if (errorIndex !== -1) {
|
|
// This workflow errored previously. Decrement error count.
|
|
results.summary.failedExecutions--;
|
|
// Remove from the list of errors.
|
|
results.summary.errors.splice(errorIndex, 1);
|
|
}
|
|
|
|
const warningIndex = results.summary.warnings.findIndex(summaryInformation => summaryInformation.workflowId === newExecution.workflowId);
|
|
if (warningIndex !== -1) {
|
|
// This workflow errored previously. Decrement error count.
|
|
results.summary.warningExecutions--;
|
|
// Remove from the list of errors.
|
|
results.summary.warnings.splice(warningIndex, 1);
|
|
}
|
|
// Increment successful executions count and push it to all executions array.
|
|
results.summary.successfulExecutions++;
|
|
results.executions.push(newExecution);
|
|
}
|
|
});
|
|
}
|
|
|
|
async runTests(allWorkflows: IWorkflowDb[]): Promise<IResult> {
|
|
const result: IResult = {
|
|
totalWorkflows: allWorkflows.length,
|
|
summary: {
|
|
failedExecutions: 0,
|
|
warningExecutions: 0,
|
|
successfulExecutions: 0,
|
|
errors: [],
|
|
warnings: [],
|
|
},
|
|
coveredNodes: {},
|
|
executions: [],
|
|
};
|
|
|
|
if (ExecuteBatch.debug) {
|
|
this.initializeLogs();
|
|
}
|
|
|
|
return new Promise(async (res) => {
|
|
const promisesArray = [];
|
|
for (let i = 0; i < ExecuteBatch.concurrency; i++) {
|
|
const promise = new Promise(async (resolve) => {
|
|
let workflow: IWorkflowDb | undefined;
|
|
while (allWorkflows.length > 0) {
|
|
workflow = allWorkflows.shift();
|
|
if (ExecuteBatch.cancelled === true) {
|
|
process.stdout.write(`Thread ${i + 1} resolving and quitting.`);
|
|
resolve(true);
|
|
break;
|
|
}
|
|
// This if shouldn't be really needed
|
|
// but it's a concurrency precaution.
|
|
if (workflow === undefined) {
|
|
resolve(true);
|
|
return;
|
|
}
|
|
|
|
if (ExecuteBatch.debug) {
|
|
ExecuteBatch.workflowExecutionsProgress[i].push({
|
|
workflowId: workflow.id,
|
|
status: 'running',
|
|
});
|
|
this.updateStatus();
|
|
}
|
|
|
|
await this.startThread(workflow).then((executionResult) => {
|
|
if (ExecuteBatch.debug) {
|
|
ExecuteBatch.workflowExecutionsProgress[i].pop();
|
|
}
|
|
result.executions.push(executionResult);
|
|
if (executionResult.executionStatus === 'success') {
|
|
if (ExecuteBatch.debug) {
|
|
ExecuteBatch.workflowExecutionsProgress[i].push({
|
|
workflowId: workflow!.id,
|
|
status: 'success',
|
|
});
|
|
this.updateStatus();
|
|
}
|
|
result.summary.successfulExecutions++;
|
|
const nodeNames = Object.keys(executionResult.coveredNodes);
|
|
|
|
nodeNames.map(nodeName => {
|
|
if (result.coveredNodes[nodeName] === undefined) {
|
|
result.coveredNodes[nodeName] = 0;
|
|
}
|
|
result.coveredNodes[nodeName] += executionResult.coveredNodes[nodeName];
|
|
});
|
|
} else if (executionResult.executionStatus === 'warning') {
|
|
result.summary.warningExecutions++;
|
|
result.summary.warnings.push({
|
|
workflowId: executionResult.workflowId,
|
|
error: executionResult.error!,
|
|
});
|
|
if (ExecuteBatch.debug) {
|
|
ExecuteBatch.workflowExecutionsProgress[i].push({
|
|
workflowId: workflow!.id,
|
|
status: 'warning',
|
|
});
|
|
this.updateStatus();
|
|
}
|
|
} else if (executionResult.executionStatus === 'error') {
|
|
result.summary.failedExecutions++;
|
|
result.summary.errors.push({
|
|
workflowId: executionResult.workflowId,
|
|
error: executionResult.error!,
|
|
});
|
|
if (ExecuteBatch.debug) {
|
|
ExecuteBatch.workflowExecutionsProgress[i].push({
|
|
workflowId: workflow!.id,
|
|
status: 'error',
|
|
});
|
|
this.updateStatus();
|
|
}
|
|
} else {
|
|
throw new Error('Wrong execution status - cannot proceed');
|
|
}
|
|
});
|
|
}
|
|
|
|
resolve(true);
|
|
});
|
|
|
|
promisesArray.push(promise);
|
|
}
|
|
|
|
await Promise.allSettled(promisesArray);
|
|
|
|
res(result);
|
|
});
|
|
}
|
|
|
|
updateStatus() {
|
|
|
|
if (ExecuteBatch.cancelled === true) {
|
|
return;
|
|
}
|
|
|
|
if (process.stdout.isTTY === true) {
|
|
process.stdout.moveCursor(0, - (ExecuteBatch.concurrency));
|
|
process.stdout.cursorTo(0);
|
|
process.stdout.clearLine(0);
|
|
}
|
|
|
|
|
|
ExecuteBatch.workflowExecutionsProgress.map((concurrentThread, index) => {
|
|
let message = `${index + 1}: `;
|
|
concurrentThread.map((executionItem, workflowIndex) => {
|
|
let openColor = '\x1b[0m';
|
|
const closeColor = '\x1b[0m';
|
|
switch (executionItem.status) {
|
|
case 'success':
|
|
openColor = '\x1b[32m';
|
|
break;
|
|
case 'error':
|
|
openColor = '\x1b[31m';
|
|
break;
|
|
case 'warning':
|
|
openColor = '\x1b[33m';
|
|
break;
|
|
default:
|
|
break;
|
|
}
|
|
message += (workflowIndex > 0 ? ', ' : '') + `${openColor}${executionItem.workflowId}${closeColor}`;
|
|
});
|
|
if (process.stdout.isTTY === true) {
|
|
process.stdout.cursorTo(0);
|
|
process.stdout.clearLine(0);
|
|
}
|
|
process.stdout.write(message + '\n');
|
|
});
|
|
}
|
|
|
|
initializeLogs() {
|
|
process.stdout.write('**********************************************\n');
|
|
process.stdout.write(' n8n test workflows\n');
|
|
process.stdout.write('**********************************************\n');
|
|
process.stdout.write('\n');
|
|
process.stdout.write('Batch number:\n');
|
|
ExecuteBatch.workflowExecutionsProgress = [];
|
|
for (let i = 0; i < ExecuteBatch.concurrency; i++) {
|
|
ExecuteBatch.workflowExecutionsProgress.push([]);
|
|
process.stdout.write(`${i + 1}: \n`);
|
|
}
|
|
}
|
|
|
|
startThread(workflowData: IWorkflowDb): Promise<IExecutionResult> {
|
|
// This will be the object returned by the promise.
|
|
// It will be updated according to execution progress below.
|
|
const executionResult: IExecutionResult = {
|
|
workflowId: workflowData.id,
|
|
workflowName: workflowData.name,
|
|
executionTime: 0,
|
|
finished: false,
|
|
executionStatus: 'running',
|
|
coveredNodes: {},
|
|
};
|
|
|
|
|
|
|
|
const requiredNodeTypes = ['n8n-nodes-base.start'];
|
|
let startNode: INode | undefined = undefined;
|
|
for (const node of workflowData.nodes) {
|
|
if (requiredNodeTypes.includes(node.type)) {
|
|
startNode = node;
|
|
break;
|
|
}
|
|
}
|
|
|
|
// We have a cool feature here.
|
|
// On each node, on the Settings tab in the node editor you can change
|
|
// the `Notes` field to add special cases for comparison and snapshots.
|
|
// You need to set one configuration per line with the following possible keys:
|
|
// CAP_RESULTS_LENGTH=x where x is a number. Cap the number of rows from this node to x.
|
|
// This means if you set CAP_RESULTS_LENGTH=1 we will have only 1 row in the output
|
|
// IGNORED_PROPERTIES=x,y,z where x, y and z are JSON property names. Removes these
|
|
// properties from the JSON object (useful for optional properties that can
|
|
// cause the comparison to detect changes when not true).
|
|
const nodeEdgeCases = {} as INodeSpecialCases;
|
|
workflowData.nodes.forEach(node => {
|
|
executionResult.coveredNodes[node.type] = (executionResult.coveredNodes[node.type] || 0) + 1;
|
|
if (node.notes !== undefined && node.notes !== '') {
|
|
node.notes.split('\n').forEach(note => {
|
|
const parts = note.split('=');
|
|
if (parts.length === 2) {
|
|
if (nodeEdgeCases[node.name] === undefined) {
|
|
nodeEdgeCases[node.name] = {} as INodeSpecialCase;
|
|
}
|
|
if (parts[0] === 'CAP_RESULTS_LENGTH') {
|
|
nodeEdgeCases[node.name].capResults = parseInt(parts[1], 10);
|
|
} else if (parts[0] === 'IGNORED_PROPERTIES') {
|
|
nodeEdgeCases[node.name].ignoredProperties = parts[1].split(',').map(property => property.trim());
|
|
} else if (parts[0] === 'KEEP_ONLY_PROPERTIES') {
|
|
nodeEdgeCases[node.name].keepOnlyProperties = parts[1].split(',').map(property => property.trim());
|
|
}
|
|
}
|
|
});
|
|
}
|
|
});
|
|
|
|
return new Promise(async (resolve) => {
|
|
if (startNode === undefined) {
|
|
// If the workflow does not contain a start-node we can not know what
|
|
// should be executed and with which data to start.
|
|
executionResult.error = 'Workflow cannot be started as it does not contain a "Start" node.';
|
|
executionResult.executionStatus = 'warning';
|
|
resolve(executionResult);
|
|
}
|
|
|
|
let gotCancel = false;
|
|
|
|
// Timeouts execution after 5 minutes.
|
|
const timeoutTimer = setTimeout(() => {
|
|
gotCancel = true;
|
|
executionResult.error = 'Workflow execution timed out.';
|
|
executionResult.executionStatus = 'warning';
|
|
resolve(executionResult);
|
|
}, ExecuteBatch.executionTimeout);
|
|
|
|
|
|
try {
|
|
|
|
const runData: IWorkflowExecutionDataProcess = {
|
|
executionMode: 'cli',
|
|
startNodes: [startNode!.name],
|
|
workflowData: workflowData!,
|
|
};
|
|
|
|
const workflowRunner = new WorkflowRunner();
|
|
const executionId = await workflowRunner.run(runData);
|
|
|
|
const activeExecutions = ActiveExecutions.getInstance();
|
|
const data = await activeExecutions.getPostExecutePromise(executionId);
|
|
if (gotCancel || ExecuteBatch.cancelled === true) {
|
|
clearTimeout(timeoutTimer);
|
|
// The promise was settled already so we simply ignore.
|
|
return;
|
|
}
|
|
|
|
if (data === undefined) {
|
|
executionResult.error = 'Workflow did not return any data.';
|
|
executionResult.executionStatus = 'error';
|
|
} else {
|
|
executionResult.executionTime = (Date.parse(data.stoppedAt as unknown as string) - Date.parse(data.startedAt as unknown as string)) / 1000;
|
|
executionResult.finished = (data?.finished !== undefined) as boolean;
|
|
|
|
if (data.data.resultData.error) {
|
|
executionResult.error =
|
|
data.data.resultData.error.hasOwnProperty('description') ?
|
|
// @ts-ignore
|
|
data.data.resultData.error.description : data.data.resultData.error.message;
|
|
if (data.data.resultData.lastNodeExecuted !== undefined) {
|
|
executionResult.error += ` on node ${data.data.resultData.lastNodeExecuted}`;
|
|
}
|
|
executionResult.executionStatus = 'error';
|
|
|
|
if (this.shouldBeConsideredAsWarning(executionResult.error || '')) {
|
|
executionResult.executionStatus = 'warning';
|
|
}
|
|
} else {
|
|
if (ExecuteBatch.shallow === true) {
|
|
// What this does is guarantee that top-level attributes
|
|
// from the JSON are kept and the are the same type.
|
|
|
|
// We convert nested JSON objects to a simple {object:true}
|
|
// and we convert nested arrays to ['json array']
|
|
|
|
// This reduces the chance of false positives but may
|
|
// result in not detecting deeper changes.
|
|
Object.keys(data.data.resultData.runData).map((nodeName: string) => {
|
|
data.data.resultData.runData[nodeName].map((taskData: ITaskData) => {
|
|
if (taskData.data === undefined) {
|
|
return;
|
|
}
|
|
Object.keys(taskData.data).map(connectionName => {
|
|
const connection = taskData.data![connectionName] as Array<INodeExecutionData[] | null>;
|
|
connection.map(executionDataArray => {
|
|
if (executionDataArray === null) {
|
|
return;
|
|
}
|
|
|
|
if (nodeEdgeCases[nodeName] !== undefined && nodeEdgeCases[nodeName].capResults !== undefined) {
|
|
executionDataArray.splice(nodeEdgeCases[nodeName].capResults!);
|
|
}
|
|
|
|
executionDataArray.map(executionData => {
|
|
if (executionData.json === undefined) {
|
|
return;
|
|
}
|
|
if (nodeEdgeCases[nodeName] !== undefined && nodeEdgeCases[nodeName].ignoredProperties !== undefined) {
|
|
nodeEdgeCases[nodeName].ignoredProperties!.forEach(ignoredProperty => delete executionData.json[ignoredProperty]);
|
|
}
|
|
|
|
let keepOnlyFields = [] as string[];
|
|
if (nodeEdgeCases[nodeName] !== undefined && nodeEdgeCases[nodeName].keepOnlyProperties !== undefined) {
|
|
keepOnlyFields = nodeEdgeCases[nodeName].keepOnlyProperties!;
|
|
}
|
|
executionData.json = keepOnlyFields.length > 0 ? pick(executionData.json, keepOnlyFields) : executionData.json;
|
|
const jsonProperties = executionData.json;
|
|
|
|
const nodeOutputAttributes = Object.keys(jsonProperties);
|
|
nodeOutputAttributes.map(attributeName => {
|
|
if (Array.isArray(jsonProperties[attributeName])) {
|
|
jsonProperties[attributeName] = ['json array'];
|
|
} else if (typeof jsonProperties[attributeName] === 'object') {
|
|
jsonProperties[attributeName] = { object: true };
|
|
}
|
|
});
|
|
});
|
|
});
|
|
|
|
});
|
|
});
|
|
});
|
|
} else {
|
|
// If not using shallow comparison then we only treat nodeEdgeCases.
|
|
const specialCases = Object.keys(nodeEdgeCases);
|
|
|
|
specialCases.forEach(nodeName => {
|
|
data.data.resultData.runData[nodeName].map((taskData: ITaskData) => {
|
|
if (taskData.data === undefined) {
|
|
return;
|
|
}
|
|
Object.keys(taskData.data).map(connectionName => {
|
|
const connection = taskData.data![connectionName] as Array<INodeExecutionData[] | null>;
|
|
connection.map(executionDataArray => {
|
|
if (executionDataArray === null) {
|
|
return;
|
|
}
|
|
|
|
if (nodeEdgeCases[nodeName].capResults !== undefined) {
|
|
executionDataArray.splice(nodeEdgeCases[nodeName].capResults!);
|
|
}
|
|
|
|
if (nodeEdgeCases[nodeName].ignoredProperties !== undefined) {
|
|
executionDataArray.map(executionData => {
|
|
if (executionData.json === undefined) {
|
|
return;
|
|
}
|
|
nodeEdgeCases[nodeName].ignoredProperties!.forEach(ignoredProperty => delete executionData.json[ignoredProperty]);
|
|
});
|
|
}
|
|
});
|
|
|
|
});
|
|
});
|
|
});
|
|
}
|
|
|
|
const serializedData = this.formatJsonOutput(data);
|
|
if (ExecuteBatch.compare === undefined) {
|
|
executionResult.executionStatus = 'success';
|
|
} else {
|
|
const fileName = (ExecuteBatch.compare.endsWith(sep) ? ExecuteBatch.compare : ExecuteBatch.compare + sep) + `${workflowData.id}-snapshot.json`;
|
|
if (fs.existsSync(fileName) === true) {
|
|
|
|
const contents = fs.readFileSync(fileName, { encoding: 'utf-8' });
|
|
|
|
const changes = diff(JSON.parse(contents), data, { keysOnly: true });
|
|
|
|
if (changes !== undefined) {
|
|
// we have structural changes. Report them.
|
|
executionResult.error = `Workflow may contain breaking changes`;
|
|
executionResult.changes = changes;
|
|
executionResult.executionStatus = 'error';
|
|
} else {
|
|
executionResult.executionStatus = 'success';
|
|
}
|
|
} else {
|
|
executionResult.error = 'Snapshot for not found.';
|
|
executionResult.executionStatus = 'warning';
|
|
}
|
|
}
|
|
// Save snapshots only after comparing - this is to make sure we're updating
|
|
// After comparing to existing verion.
|
|
if (ExecuteBatch.snapshot !== undefined) {
|
|
const fileName = (ExecuteBatch.snapshot.endsWith(sep) ? ExecuteBatch.snapshot : ExecuteBatch.snapshot + sep) + `${workflowData.id}-snapshot.json`;
|
|
fs.writeFileSync(fileName, serializedData);
|
|
}
|
|
}
|
|
}
|
|
} catch (e) {
|
|
executionResult.error = 'Workflow failed to execute.';
|
|
executionResult.executionStatus = 'error';
|
|
}
|
|
clearTimeout(timeoutTimer);
|
|
resolve(executionResult);
|
|
});
|
|
}
|
|
|
|
}
|