Deprecating Custom Tasks Runner
The Custom Task Runners API was created many years ago and has not been supported for several years. It was developed before we introduced a formal method for extending Nx through plugins. The API potentially allows modifications to the lifecycle of the Nx command execution, which breaks important invariants that Nx depends on. The new API leverages the plugins API for better performance and functionality.
This page guides you on how to migrate off the now deprecated Custom Task Runner API for both:
- choosing the best remote cache option for your organization
- implementing pre/post processing logic when running tasks
Remote Cache
There are several ways to set up remote caching with Nx. Read more about remote cache options.
The preTasksExecution and postTasksExecution hooks
Starting with Nx 20.4, a dedicated plugin-based API has been introduced that allows you to safely hook into the task running lifecycle. This new approach provides pre and post execution hooks without compromising Nx's internal operations. For comprehensive documentation on this feature, see Hook into the Task Running Lifecycle.
Let's look into some concrete examples of how you can migrate from the previous custom task runners to this new API.
Before: Custom Tasks Runner Version
Let's imagine that you have implemented a custom task runner as follows:
1function serializeTasksResults(taskResults: { [taskId: string]: TaskResult }) {
2 // ...
3}
4
5function validateEnv() {
6 // ...
7}
8
9export default async function customTasksRunner(tasks, options, context) {
10 if (process.env.QA_ENV) {
11 process.env.NX_SKIP_NX_CACHE = 'true';
12 }
13 if (!validateEnv()) {
14 throw new Error('Env is not set up correctly');
15 }
16 const allTaskResults = {};
17 const lifeCycle = {
18 endTasks(taskResults) {
19 taskResults.forEach((tr) => {
20 allTaskResults[tr.task.id] = tr;
21 });
22 },
23 };
24 const ret = await defaultTasksRunner(
25 tasks,
26 { ...options, lifeCycle },
27 context
28 );
29 if (options.reportAnalytics) {
30 await fetch(process.env.DATACAT_API, {
31 method: 'POST',
32 headers: {
33 'Content-Type': 'application/json',
34 },
35 body: serializeTasksResults(allTaskResults),
36 });
37 }
38 return ret;
39}
40
And let's imagine you configured it in nx.json
as follows:
1{
2 "tasksRunnerOptions": {
3 "default": {
4 "runner": "npm-package-with-custom-tasks-runner",
5 "options": {
6 "reportAnalytics": true
7 }
8 }
9 }
10}
11
After: Using the New API
The new API includes preTasksExecution
and postTasksExecution
hooks that plugins can register. These hooks do not affect task execution and cannot violate any invariants.
The custom task runner above can be implemented as follows:
1function serializeTasksResults(taskResults: { [taskId: string]: TaskResult }) {
2 // task results contain timings and status information useful for gathering analytics
3}
4
5function validateEnv() {
6 // ...
7}
8
9// context contains workspaceRoot and nx.json configuration
10export async function preTasksExecution(options: any, context) {
11 if (process.env.QA_ENV) {
12 process.env.NX_SKIP_NX_CACHE = 'true';
13 }
14 if (!validateEnv()) {
15 throw new Error('Env is not set up correctly');
16 }
17}
18
19// context contains workspaceRoot, nx.json configuration, and task results
20export async function postTasksExecution(options: any, context) {
21 if (options.reportAnalytics) {
22 await fetch(process.env.DATACAT_API, {
23 method: 'POST',
24 headers: {
25 'Content-Type': 'application/json',
26 },
27 body: serializeTasksResults(context.taskResults),
28 });
29 }
30}
31
Because it's a regular plugin, it can be configured as follows in nx.json
:
1{
2 "plugins": [
3 {
4 "plugin": "npm-package-with-the-plugin",
5 "options": {
6 "reportAnalytics": true
7 }
8 }
9 ]
10}
11
As with all plugin hooks, the preTasksExecution
and postTasksExecution
hooks must be exported so that they load properly when the "npm-package-with-the-plugin" is initialized.
Multiple Tasks Runners
You can implement multiple tasks runners using the new hooks.
Imagine you have the following nx.json
:
1{
2 "tasksRunnerOptions": {
3 "a": {
4 "runner": "package-a"
5 },
6 "b": {
7 "runner": "package-b"
8 }
9 }
10}
11
You can replace it with two plugins:
1{
2 "plugins": [
3 {
4 "plugin": "package-a"
5 },
6 {
7 "plugin": "package-b"
8 }
9 ]
10}
11
Simply add a condition to your hooks as follows:
1export async function preTasksExecution() {
2 if (process.env.RUNNER != 'a') return;
3}
4
5export async function postTasksExecution(options, tasksResults) {
6 if (process.env.RUNNER != 'a') return;
7}
8
You can then choose which hooks to use by setting the RUNNER env variable.
Passing Options
You can no longer augment options passed to the default tasks runner, so instead you need to set env variables in the preTasksExecution
hook.
Composing Plugins
Implementing hooks in plugins offers several advantages. It allows multiple plugins to be loaded simultaneously, enabling a clean separation of concerns.
Keeping State Across Command Invocations
By default, every plugin initiates a long-running process, allowing you to maintain state across command invocations, which can be very useful for advanced analytics.
Unhandled Custom Tasks Runner Use Cases?
If you have a use case that the new API doesn't handle, please open an issue.