Build time & bundle size testing
- 1 Build Time
- 1.1 Development Build Time
- 1.1.1 Results
- 1.2 Production Build Time
- 1.2.1 Results
- 1.1 Development Build Time
- 2 Bundle Sizes
- 2.1 Production
- 2.1.1 Traditional Build
- 2.1.2 Guest/Module Build
- 2.1.3 Results
- 2.1.3.1 Worst Case Scenario
- 2.1.3.2 Best Case Scenario
- 2.1 Production
For this test, we’re going to measure the time it takes to build dev and production bundles of the frontend-app-mf-guest repository. We’ll measure the build time in two different app configurations:
A traditional MFE which bundles all of its dependencies and includes the header, footer, and all of the shared dependencies.
A new “guest” MFE with a ModuleFederationPlugin configuration that shares dependencies, and with only the shared dependencies it now needs to be loaded as a runtime module into a “host” MFE. That means the header, footer, core-js, regenerator-runtime, and brand are no longer included at all.
Build Time
Development Build Time
Test Run | Traditional (ms) | Guest/Module (ms) |
---|---|---|
1 | 9133 | 6558 |
2 | 5582 | 3308 |
3 | 5510 | 3300 |
4 | 5610 | 3354 |
5 | 5634 | 3342 |
6 | 5671 | 3284 |
7 | 5693 | 3322 |
8 | 5640 | 3297 |
9 | 5623 | 3347 |
10 | 5580 | 3376 |
Average | 5967.6 | 3648.8 |
This improvement is because we’re bundling less stuff in the guest/module architecture. If the MFE retains its index page which loads core-js, regenerator-runtime, and the header/footer, the build time is actually slightly higher on average.
Results
Development build time for a guest/module MFE is 61.1% of the time of a traditional MFE.
Production Build Time
Test Run | Traditional (ms) | Guest/Module (ms) |
---|---|---|
1 | 14896 | 10935 |
2 | 14982 | 10415 |
3 | 14844 | 10362 |
4 | 14840 | 10741 |
5 | 14776 | 10320 |
6 | 15003 | 10449 |
7 | 15161 | 10308 |
8 | 15170 | 10182 |
9 | 14982 | 10456 |
10 | 15309 | 10343 |
Average | 14996.3 | 10451.1 |
Similar to the development build times, we’re bundling less actual code, so can expect there’s less work to do.
Results
Production build time for a guest/module MFE is 69.6% of the time of a traditional MFE.
Bundle Sizes
Production
Traditional Build
File | Size | Notes |
---|---|---|
440b1fd39e91e51e7522b7475338e338.png | 526 | An image I can’t find the source of - maybe a header or footer? |
588.ccafd157cd20348f5d12.js | 1179475 | Dependencies (with Treeshaking) |
95ec738c0b7faac5b5c9126794446bbd.svg | 342 | User menu icon |
app.2733c7b2743ced7571ec.css | 510993 | Our massive stylesheet |
app.2733c7b2743ced7571ec.js | 1279 | Our application code |
index.html | 31299 | Big because it includes New Relic script. |
runtime.e844be7d2206edc6dff5.js | 2070 | Webpack overhead |
Total: 1,725,984 bytes
Guest/Module Build
File | Size | Notes |
---|---|---|
272.9f559385b83edb423687.js | 66629 | Module federation runtime |
440b1fd39e91e51e7522b7475338e338.png | 526 | Same blank image… unsure why this is here. |
446.88baa1ec0097e5a731bb.css | 1 | Presumably a stub for paragon’s stylesheet somehow, since we’re not importing it. |
446.88baa1ec0097e5a731bb.js | 929360 | Paragon |
483.f8bdb2bc9f890dc5a6d3.js | 123479 | react-dom - singleton, so impossible to use |
810.39346add3075df3c2e86.js | 7123 | react - singleton, so impossible to use |
826.48c8a2cae4305c5e8390.js | 914 | prop-types |
95ec738c0b7faac5b5c9126794446bbd.svg | 342 | A “user” icon. Unclear why this has been bundled. |
__federation_expose_GuestPage.74bf4d7962b592080a4d.css | 115 | Our module CSS! |
__federation_expose_GuestPage.74bf4d7962b592080a4d.js | 2240 | Our module JS! |
app.c2789d2465bb34d2a8e5.js | 168 | Unused |
guest_partial.1a68511e4abda6f1aea1.js | 666 | Overhead |
index.html | 31525 | Unused as a module |
remoteEntry.js | 75855 | Manifest |
runtime.3e9d4ef61501a655f753.js | 7147 | Overhead |
Results
Worst Case Scenario
Worst case bundle size: 1,167,317 bytes
The worst case scenario is how many bytes we’d have to load if our host application didn’t have a compatible shared dependency version for any of our dependencies (like Paragon).
The above bundle size number won’t make sense with the sizes in the above table. While running this test, I realized that because my MFE didn’t use our other shared dependencies, they weren’t included in the bundle. Rather than start over, I went back to the Shared Dependencies page, counted up the total size of all the dependencies that we would have to load if our host didn’t have a compatible copy of them, and added that to the base size of our best case scenario. That number was 1,081,960, and includes @openedx/paragon, classnames, prop-types, react-redux, react-router, react-router-dom, and redux. React, react-dom, and @edx/frontend-platform don't count because they are singletons and can’t be loaded more than once.
Best Case Scenario
Best case bundle size: 78,210 bytes
In the best case scenario, we load our GuestPage JS and CSS and the remoteEntry.js file. Remote entry is the size it is because of how many shared dependencies we have - it needs to act as a manifest for all the other files we might need to load.
In our worst case scenario, our MFE is now 67.6% of the size of a traditional MFE.
In our best case scenario, our MFE is now 4.5% of the size of a traditional MFE.
As the application code of an MFE increases the percentage attributable to the shared dependencies will decrease - application code of 1MB (a very large MFE) will be roughly 50% as large as a traditional MFE in the best case scenario. We can leverage lazy loading and bundle splitting to keep the incremental download for guest MFEs small.