PR test in mdn/translated-content is slower than before

[yin1999]

Summary

I just notice that after we migrating to ESM. The PR test workflow in translated-content is slower that before.

See the following workflow run:

• now: https://github.com/mdn/translated-content/actions/runs/4261747120/jobs/7416424977#step:7:63
• before: https://github.com/mdn/translated-content/actions/runs/4048314445/jobs/6963395796#step:7:30

As @queengooborg mentioned in translated-content, maybe we need synchronous file reading for this step.

URL

None

Reproduction steps

See workflow run.

Expected behavior

The workflow could run faster (maybe we could start from the gather translation step).

Actual behavior

The workflow is a bit slow.

Device

Desktop

Browser

Chrome

Browser version

Stable

Operating system

Linux

Screenshot

No response

Anything else?

No response

Validations

• I have read the Community Participation Guidelines.
• I have verified that there isn't already an issue that reports the same bug to avoid creating a duplicate.
• I have checked that this is a concrete bug. For Q&A open a GitHub Discussion.

[caugner]

Please note that individual workflow runs are not necessarily comparable, as workflow runners use a variety of CPUs.

[yin1999]

Please note that individual workflow runs are not necessarily comparable, as workflow runners use a variety of CPUs.

Sorry about ignoring that, I just made a local test, and found that:

run node node_modules/@mdn/yari/build/cli.js files/zh-tw/learn/common_questions/web_mechanics/what_is_a_web_server/index.md

use @mdn/[email protected]:

13.639s is used to gather translations.

use @mdn/[email protected]:

1.831s is used to gather translations.

Test platform:

OS: Linux debian 6.1.0-3-amd64 #1 SMP PREEMPT_DYNAMIC Debian 6.1.8-1 (2023-01-29) x86_64 GNU/Linux
Node.js: v18.13.0
Yarn: 1.22.19 (Debian package: [email protected]+~cs24.27.18-2)
CPU: 13th Gen Intel(R) Core(TM) i5-13600K

With 1TB SSD and 32GB memory

[caugner]

I compared @mdn/[email protected] and @mdn/[email protected] and indeed the build performance has degraded significantly:

• 1.36.4: 19.0 seconds
• 2.0.0: 49.0 seconds

[queengooborg]

On bf23810, I ran yarn build files/zh-tw/learn/common_questions/web_mechanics/what_is_a_web_server/index.md and the command completed in ~19 seconds:

❯ yarn build files/zh-tw/learn/common_questions/web_mechanics/what_is_a_web_server/index.md
yarn run v1.22.19
$ cross-env NODE_ENV=production ts-node build/cli.ts files/zh-tw/learn/common_questions/web_mechanics/what_is_a_web_server/index.md
(node:36300) ExperimentalWarning: Importing JSON modules is an experimental feature and might change at any time
(Use `node --trace-warnings ...` to show where the warning was created)
CONTENT_ROOT:            /Users/queengooborg/Developer/Gooborg/mdn-content/files
CONTENT_TRANSLATED_ROOT: /Users/queengooborg/Developer/Gooborg/mdn-translated-content/files
Time to gather all translations: 16.302s
Language does not exist: svelte
/Users/queengooborg/Developer/Gooborg/yari/client/build/zh-tw/docs/learn/common_questions/web_mechanics/what_is_a_web_server
Built 1 pages in 17.5 seconds, at a rate of 0.1 documents per second.
Peak heap memory usage: 300 MB


Total_Flaws_Count
broken_links  8
macros        1


✨  Done in 19.45s.

The time spent gathering all the translations is definitely quite high, I wonder what's causing the sudden change in build times, hmm...

[queengooborg]

In v2.5.5 (the latest as of this comment) on a brand new package and folder made specifically for testing Yari, I found that the time spent went down significantly to ~4.3s:

❯ node node_modules/@mdn/yari/build/cli.js files/zh-tw/learn/common_questions/web_mechanics/what_is_a_web_server/index.md
(node:46939) ExperimentalWarning: Importing JSON modules is an experimental feature and might change at any time
(Use `node --trace-warnings ...` to show where the warning was created)
CONTENT_ROOT:            /Users/queengooborg/Developer/Gooborg/mdn-content/files
CONTENT_TRANSLATED_ROOT: /Users/queengooborg/Developer/Gooborg/mdn-translated-content/files
Time to gather all translations: 4.300s
Language does not exist: svelte
/Users/queengooborg/Developer/tmp-yari-test/node_modules/@mdn/yari/client/build/zh-tw/docs/learn/common_questions/web_mechanics/what_is_a_web_server
Built 1 pages in 5.2 seconds, at a rate of 0.2 documents per second.
Peak heap memory usage: 252 MB


Total_Flaws_Count
broken_links  8
macros        1

And in v1.36.4, I get around the same time.

---

When I was trying to dig into profiling and performance, I ran a profiler through Chrome's debug panel (within Yari, using the latest branch), and I found something interesting:

I don't see much of a reason that the front-matter package would be causing such an issue... Maybe it's something to do with NodeJS wrapping CJS to ESM?

[yin1999]

After the fix (#8806) is applied, the build time is reduced to 3.5s. Let's see if there is side-effect made by the fix for a few days. If nothing happend, I'll close this PR :)