In this article, I will present two concrete cases of implicit dependencies and show you how to configure your Nx workspace and projects to deal with them.
You can find the repository that illustrates those cases on GitHub.
Before we dive into the cases, let’s take a moment to understand how Nx represents dependencies and how it can help us work with implicit dependencies.
In the context of a workspace, Nx maintains a graph of all projects it contains, including internal and external dependencies. It does its best to deduct relationships by continuously analyzing the workspace components, such as the source code, the projects' configuration, the installed dependencies, etc. All being explicit dependencies. For the rest - the implicit dependencies that only a human could be aware of - Nx offers a way to declare them so they can also be part of the graph.
We can use the following graph extract to understand its composition:
{
"graph": {
"nodes": {
"ts-interfaces": {
"name": "ts-interfaces",
"type": "lib",
"data": {
"root": "libs/shared/ts-interfaces",
"name": "ts-interfaces",
"targets": {
// ...
},
"$schema": "../../../node_modules/nx/schemas/project-schema.json",
"sourceRoot": "libs/shared/ts-interfaces/src",
"projectType": "library",
"implicitDependencies": ["schemas"],
"tags": ["platform:js", "type:core"]
}
},
"schemas": {
"name": "schemas",
"type": "lib",
"data": {
"root": "libs/shared/schemas",
"name": "schemas",
"targets": {
// ...
},
"$schema": "../../../node_modules/nx/schemas/project-schema.json",
"sourceRoot": "libs/shared/schemas/src",
"projectType": "library",
"tags": ["platform:shared", "type:core"],
"implicitDependencies": []
}
},
"coffee-dealer": {
"name": "coffee-dealer",
"type": "app",
"data": {
"root": "apps/coffee-dealer",
"name": "coffee-dealer",
"targets": {
// ...
},
"$schema": "../../node_modules/nx/schemas/project-schema.json",
"sourceRoot": "apps/coffee-dealer/src",
"projectType": "application",
"tags": ["type:app", "platform:node"],
"implicitDependencies": []
}
}
},
"dependencies": {
"ts-interfaces": [
{
"source": "ts-interfaces",
"target": "schemas",
"type": "implicit"
}
],
"schemas": [],
"coffee-dealer": [
{
"source": "coffee-dealer",
"target": "ts-interfaces",
"type": "static"
}
]
}
}
}
Note
- This graph is generated by running
nx graph --file=output.json && cat output.json
in the repository root folder.- If you are unfamiliar with Nx, you might want to head to Nx’s documentation to understand the benefits of the project graph and the mental model.
Under graph.nodes
, each node represents a project in the workspace. The graph.dependencies
include the edges defining the relationship between nodes/projects. The type
property can be static
or implicit
, depending on whether the dependency is explicit or implicit.
Case 1: The Libraries Hidden Bridge
This case occurred in an integrated repository composed of multiple apps and libraries. I reproduced a similar situation with the following setup:
-
coffee-dealer
, is a NestJS application importing TS interfaces fromts-interfaces
-
ts-interfaces
, is composed of auto-generated Typescript interfaces fromschemas
-
schemas
, is composed of JSON schemas exported in a TS file
You can see the current relation between the projects in the project graph:
And the tasks graph for the three projects:
What is wrong with these graphs?
The schemas
library is not linked to the ts-interfaces
library, meaning that when building the application with nx run coffee-dealer:build
:
- The
schemas:bundle
task will not be included in the task pipeline, inducing the risk of using outdated schemas while generating the Typescript interfaces. - The
schemas:bundle
andts-interfaces:bundle
tasks runner is not caching inputs, so Nx cannot check whether the source definitions have changed at the nextbundle
run.
In the following sections, I will further present the two libraries and show you how to solve this issue.
schemas
These JSON schemas represent the domain entities and value objects; in other words, they are a source of truth and can be derived into types for different programming languages (Go struct, Typescript interfaces … ).
The JSON schema definitions organization follows this good advice so that each definition represents a single domain entity or value object, making work easier during development. Before being shared, the definitions are bundled and exported into larger schemas representing business operations.
The logic to bundle the JSON schemas is located outside the src
folder, in the internals
folder, to avoid including those scripts and schemas in the build output. To still benefit from Typescript in our IDE, a separate config is created under tsconfig.editor.json
and referenced in tsconfig.json
to include the internals
folder.
// libs/schemas/tsconfig.editor.json
{
"extends": "./tsconfig.json",
"compilerOptions": {
"outDir": "../../dist/out-tsc",
"declaration": true,
"resolveJsonModule": true,
"esModuleInterop": true,
"types": ["node"]
},
"include": ["src/**/*.ts", "internals/**/*.ts"],
"exclude": ["jest.config.ts", "test/**/*.spec.ts", "test/**/*.ts"]
}
Now let’s zoom in on the project.json
- It contains an extra
namedInputs
, which references the content ofinternals
, allowing Nx to use this as a cache input for a task - In the
build
task, I added a dependency on thebundle
task, meaning that before compiling the Typescript code, Nx will execute thebundle
task - The
bundle
task executes the code containing the logic to bundle schemas into thesrc/lib
folder. Thecache
property is set to true to enable task runner caching. Theinputs
use theinternals
named input, and theoutputs
reference the glob{projectRoot}/src/lib/*.json
, which matches all bundled schemas.
Note
If the cache is hit because the input has not changed, Nx will restore the existing outputs.
// libs/schemas/project.json
{
"name": "schemas",
"$schema": "../../../node_modules/nx/schemas/project-schema.json",
"sourceRoot": "libs/shared/schemas/src",
"projectType": "library",
"namedInputs": {
"internals": ["{projectRoot}/internals/*.{json,ts}"]
},
"targets": {
"build": {
"executor": "@nx/js:tsc",
"outputs": ["{options.outputPath}"],
"options": {
"outputPath": "dist/libs/shared/schemas",
"main": "libs/shared/schemas/src/index.ts",
"tsConfig": "libs/shared/schemas/tsconfig.lib.json",
"assets": [
"libs/shared/schemas/*.md",
"libs/shared/schemas/src/lib/*.{json,js}"
]
},
"dependsOn": ["bundle"]
},
"bundle": {
"executor": "nx:run-commands",
"cache": true,
"inputs": ["internals"],
"outputs": ["{projectRoot}/src/lib/*.json"],
"options": {
"command": "npx ts-node libs/shared/schemas/internals/bundle.ts",
"cwd": "."
}
}
},
"tags": ["platform:shared", "type:core"]
}
ts-interfaces
The ts-interfaces
library is composed of auto-generated Typescript interfaces. A custom script, located in internals
for the same reason as the schemas
library, loads the JSON schemas using the File system API instead of imports to avoid including schemas
dependencies to generate the interfaces.
The drawback is that Nx can't detect the dependency between schemas
and ts-interfaces
; let’s see how to declare the relationship between those two libraries by examining the project.json
. It contains a configuration similar to the schemas
project with the following noticeable additions:
- The
implicitDependencies
declare a dependency on theschemas
library, which creates the missing link in the project graph - In the
bundle
target,dependsOn
is now set to^bundle
, meaning allbundle
tasks from the dependencies will run first when callingnx run ts-interfaces:bundle
- The
bundle.inputs
now reference theinternals
named input from the current library - The
bundle.outputs
now reference the path (glob) of the generated Typescript interfaces
// libs/ts-interfaces/project.json
{
"name": "ts-interfaces",
"$schema": "../../../node_modules/nx/schemas/project-schema.json",
"sourceRoot": "libs/shared/ts-interfaces/src",
"projectType": "library",
"namedInputs": {
"internals": ["{projectRoot}/internals/*.ts"]
},
"implicitDependencies": ["schemas"],
"targets": {
"build": {
"executor": "@nx/js:tsc",
"outputs": ["{options.outputPath}"],
"options": {
"outputPath": "dist/libs/shared/ts-interfaces",
"main": "libs/shared/ts-interfaces/src/index.ts",
"tsConfig": "libs/shared/ts-interfaces/tsconfig.lib.json",
"assets": ["libs/shared/ts-interfaces/*.md"]
},
"dependsOn": ["bundle"]
},
"bundle": {
"executor": "nx:run-commands",
"cache": true,
"dependsOn": ["^bundle"],
"inputs": ["^internals"],
"outputs": ["{projectRoot}/src/lib/*.ts"],
"options": {
"command": "npx ts-node --project libs/shared/ts-interfaces/tsconfig.editor.json libs/shared/ts-interfaces/internals/bundle.ts",
"cwd": "."
}
}
// ...
},
"tags": ["platform:js", "type:core"]
}
After applying these changes, the project graph contains the expected relationship between the schemas
and ts-interfaces
libraries.
And the tasks graph shows that the schemas:bundle
task is now part of the coffee-dealer:build
pipeline.
Enable cache for the bundle tasks
There are two ways to enable caching of tasks :
- on a project basis by setting the
cache
property totrue
and theinputs
to cache in the target, as shown previously for theschemas
andts-interfaces
libraries - globally, by setting the
bundle
default values intargetDefaults
and declaringinternals
innamedInputs
innx.json
// nx.json
{
"$schema": "./node_modules/nx/schemas/nx-schema.json",
"targetDefaults": {
// ...
"bundle": {
"cache": true,
"dependsOn": ["^bundle"],
"inputs": ["^internals"]
}
// ...
},
"namedInputs": {
// ...
"internals": [
"{projectRoot}/internals/**/*.[jt]s",
"{projectRoot}/internals/**/*.json"
]
// ...
}
}
Tips
The technique above allows you to declare a global rule for the cache and the named inputs. It is useful when you have many projects sharing the same configuration. This can also be applied to executors, such as
@nx/jest:jest
or@nx/eslint:lint
.
Have a look at the Nx docs, to learn how to further tweak the cache rules
Our case is now closed, and we can move on to the next one.
Case 2: The software developer's optimism
This story is another case of implicit dependencies impacting developers' experience. I reproduced a similar situation with the following setup:
-
coffee-dealer
, is a NestJS application serving an HTTP API on port 3000 -
coffee-dealer-cli
, is a CLI tool to interact withcoffee-dealer
through its HTTP API -
coffee-dealer-e2e
, is an end-to-end test suite sending HTTP requests tocoffee-dealer
In this case, the developers would like to ensure that when they work locally on the coffee-dealer
application with the watch mode enabled, the coffee-dealer-cli
and coffee-dealer-e2e
are always running against the latest build of the coffee-dealer
application.
You can see the current relation between the projects in the project graph:
What can we deduct from this graph?
- We can see an implicit dependency between
coffee-dealer
andcoffee-dealer-e2e
, created by default by the NestJS generator plugin, and no dependency betweencoffee-dealer
andcoffee-dealer-cli
. There is no task dependency between thecoffee-dealer-e2e:e2e
task and thecoffee-dealer:serve
task, meaning that the end-to-end tests could run before the application is ready. It is the same for thecoffee-dealer-cli:run
task, which could run before the application's API is available. With Nx, it is currently impossible to declare a task dependency for a long-running task, such asserve
, so we must find an alternative solution. See related issues: nx issue #5570 and nx issue #3748. - When running the application with
nx run coffee-dealer:serve
, it starts with the watch mode enabled. The watcher does not instantly detect changes; there is a delay (which depends on the compiler or task manager implementation and/or configuration) between the file(s) change detection and the build completion. It means there is a small window where the CLI or the end-to-end tests could run against a previous build version.
coffee-dealer-e2e
To dramatically decrease the risk of encountering the abovementioned issues, we can :
- Create a new task
check-port
and add it todependsOn
of the tasks that should check whether thecoffee-dealer
port is open before running the end-to-end tests - Declare an extra
namedInput
to use the output of thecoffee-dealer:build
task as a cache input for thecoffee-dealer-e2e:e2e
task, using thedependentTasksOutputFiles
property - Add a dependency on the
build
tasks to create a race condition between thecoffee-dealer:build
tasks thatcoffee-dealer:serve
andcoffee-dealer-e2e:e2e
depend on.
// apps/coffee-dealer-e2e/project.json
{
"name": "coffee-dealer-e2e",
"$schema": "../../node_modules/nx/schemas/project-schema.json",
"implicitDependencies": ["coffee-dealer"],
"namedInputs": {
"coffee-dealer": [
{
"dependentTasksOutputFiles": "{workspaceRoot}/dist/apps/coffee-dealer/**/*.js",
"transitive": false
}
]
},
"projectType": "application",
"targets": {
"e2e": {
"executor": "@nx/jest:jest",
"dependsOn": ["build", "check-port"],
"inputs": ["^production", "coffee-dealer"],
"outputs": ["{workspaceRoot}/coverage/{e2eProjectRoot}"],
"options": {
"jestConfig": "apps/coffee-dealer-e2e/jest.config.ts",
"passWithNoTests": true
},
"configurations": {
"ci": {
"ci": true
},
"development": {}
}
},
"check-port": {
"executor": "nx:run-commands",
"cache": false,
"options": {
"command": "lsof -i:${PORT:-3000} -t"
}
},
"lint": {
"executor": "@nx/eslint:lint",
"outputs": ["{options.outputFile}"]
}
}
}
Finally, ensure that coffee-dealer-e2e:e2e
and coffee-dealer:serve
will use the development
configuration to use the same build output by setting the defaultConfiguration
to development
in the serve
and build
targets of the coffee-dealer
project.
// apps/coffee-dealer/project.json
{
"name": "coffee-dealer",
"$schema": "../../node_modules/nx/schemas/project-schema.json",
"sourceRoot": "apps/coffee-dealer/src",
"projectType": "application",
"targets": {
"build": {
"executor": "@nx/webpack:webpack",
"outputs": ["{options.outputPath}"],
"defaultConfiguration": "development",
"options": {
"target": "node",
"compiler": "tsc",
"transformers": [
"typia/lib/transform",
{
"name": "@nestia/core/lib/transform",
"options": {
"validate": "assert",
"stringify": "assert"
}
}
],
"outputPath": "dist/apps/coffee-dealer",
"main": "apps/coffee-dealer/src/main.ts",
"tsConfig": "apps/coffee-dealer/tsconfig.app.json",
"assets": ["apps/coffee-dealer/src/assets"],
"webpackConfig": "apps/coffee-dealer/webpack.config.js"
},
"configurations": {
"development": {},
"production": {}
}
},
"serve": {
"executor": "@nx/js:node",
"defaultConfiguration": "development",
"options": {
"buildTarget": "coffee-dealer:build"
},
"configurations": {
"development": {
"buildTarget": "coffee-dealer:build:development"
},
"production": {
"buildTarget": "coffee-dealer:build:production"
}
}
}
// ...
},
"tags": ["type:app", "platform:node"]
}
We can now check that the task pipeline is correct :
Note
In this tasks graph, even though the
coffee-dealer:serve
task is not directly added to thecoffee-dealer-e2e:e2e
pipeline, they will both produce and consume the same build artifacts. Which is how we implictly tie thecoffee-dealer-e2e:e2e
task to thecoffee-dealer:serve
task.
We can test the solution with the script test-race-watch.js
, which emulates a user's behavior. The goal is to demonstrate that it is possible to ensure that the e2e tests always run against the latest application version.
The script will change an API response from the coffee-dealer
app and revert it after a few seconds, triggering a rebuild of the application. The e2e tests will run before and after each rebuild, and we should see the following output:
- e2e tests will fail when the API is changed, and the task won't use the cache
- e2e tests will pass when the API change is reverted, and the tasks will use the cache (when it is populated) for all tasks
// test-race-watch.js
const { exec } = require('node:child_process');
const { assert } = require('node:console');
const { readFile, writeFile } = require('node:fs').promises;
const { join } = require('node:path');
const { PerformanceObserver, performance } = require('node:perf_hooks');
const { once } = require('node:stream');
const { format } = require('node:util');
const coffeeDealerAppServiceTsPath = join(
__dirname,
'apps',
'coffee-dealer',
'src',
'app',
'app.service.ts'
);
const port = 3010;
let counter = 0;
let isOriginalApi = true;
function runE2ETests(suite = '#1') {
const shouldFail = !isOriginalApi;
console.warn(
'➡️ running E2E tests suite %s %s',
suite,
shouldFail ? 'should fail' : 'should pass'
);
const coffeeDealerE2EProcess = exec('npx nx run coffee-dealer-e2e:e2e', {
env: { ...process.env, PORT: port },
});
coffeeDealerE2EProcess
.once('spawn', () => {
performance.mark(format('e2e-start:%s', suite));
performance.mark(format('e2e-deps-start:%s', suite));
})
.once('exit', (code, signal) => {
console.warn('🏁 test finished with code %s and signal %s', code, signal);
performance.mark(format('e2e-end:%s', suite));
performance.measure(
format('e2e-run:%s', suite),
format('e2e-start:%s', suite),
format('e2e-end:%s', suite)
);
assert(
shouldFail ? code !== 0 : code === 0,
e2e tests </span><span class="p">${</span><span class="nx">suite</span><span class="p">}</span><span class="s2"> should </span><span class="p">${</span><span class="nx">shouldFail</span> <span class="p">?</span> <span class="dl">'</span><span class="s1">fail</span><span class="dl">'</span> <span class="p">:</span> <span class="dl">'</span><span class="s1">pass</span><span class="dl">'</span><span class="p">}</span><span class="s2">
);
});
coffeeDealerE2EProcess.stdout.setEncoding('utf8').on('data', (data) => {
// uncomment to see the output of the e2e tests task
// console.log(data.trim());
if (data.includes('Setting up')) {
performance.mark(format('e2e-deps-end:%s', suite));
performance.measure(
format('e2e dependencies tasks:%s', suite),
format('e2e-deps-start:%s', suite),
format('e2e-deps-end:%s', suite)
);
}
});
coffeeDealerE2EProcess.stderr.pipe(process.stderr);
}
async function updateAppService() {
const data = await readFile(coffeeDealerAppServiceTsPath, 'utf8');
/**
- Hello API is the value that should be returned by the API and what the e2e tests expect
*/
const index = data.search('Hello API');
let updatedData = '';
if (index > 0) {
// in that case e2e tests will fail
updatedData = data.replaceAll('Hello API', 'Hello World!');
isOriginalApi = false;
} else {
// in that case e2e tests will pass
updatedData = data.replaceAll('Hello World!', 'Hello API');
isOriginalApi = true;
}
console.warn(
'🎬 updating API to %s version',
isOriginalApi ? 'original' : 'new'
);
runE2ETests('#1');
/**
- should trigger build and reload of cofee-dealer app
**/
writeFile(coffeeDealerAppServiceTsPath, updatedData).then(() => {
counter++;
console.warn('➡️ updated API #%d', counter);
runE2ETests('#3');
});
runE2ETests('#2');
}
function main() {
const obs = new PerformanceObserver((items) => {
console.log(
'task %s duration : %d ms',
items.getEntries()[0].name,
items.getEntries()[0].duration
);
performance.clearMarks(items.getEntries()[0].name);
});
obs.observe({ type: 'measure' });
const coffeeDealerServerProcess = exec(
'npx nx run coffee-dealer:serve --inspect=false',
{
env: {
...process.env,
PORT: port,
},
}
);
// uncomment to see the output of the server
// coffeeDealerServerProcess.stdout.pipe(process.stdout);
coffeeDealerServerProcess.stderr.pipe(process.stderr);
once(coffeeDealerServerProcess, 'exit').then(([code, signal]) =>
console.log('coffeeDealerServerProcess exit', code, signal)
);
/**
- update API every 3-7 seconds
/
setInterval(() => {
updateAppService();
}, (Math.floor(Math.random() 4) + 3) * 1000);
}
main();
coffee-dealer-cli
We can apply the same techniques to the coffee-dealer-cli
application by applying similar changes to the project.json
file.
Key Takeaways
To conclude our cases, Nx offers many ways to manage implicit internal and external dependencies. We can declare them in the project.json
file, use named inputs to cache the output of a task, and use the dependsOn
property to create a task pipeline. We can also use the targetDefaults
property in the nx.json
file to set global rules for the cache and the named inputs.
I hope this article has helped you understand the importance of managing implicit dependencies in your workspace or inspired you to improve it. If you have any questions or feedback, feel free to reach out to me on Twitter or Linkedin.
In the next episode, I plan to explore the usage of nx watch
to observe the workspace changes and create a custom tasks pipeline triggered based on the changes.
Top comments (0)