fix(node): Improve chunk flushing #1985
Conversation
ScriptedAlchemy
force-pushed
the
feat/chunk-lush
branch
from
3f10147 to
a520756
Compare
ScriptedAlchemy
changed the title
| @@ -21,11 +22,24 @@ export const performReload = (shouldReload: any) => { | |||
| } | |||
|
|
|||
| Object.keys(req.cache).forEach((key) => { | |||
| delete req.cache[key]; | |||
There was a problem hiding this comment.
Was this intentionally left in? You can remove the if and regex below if so
There was a problem hiding this comment.
commenting it out, just to make sure i dont actually need this (if my regex covers everything)
| @@ -8,14 +8,20 @@ import { | |||
|
|
|||
| class MyDocument extends Document { | |||
| static async getInitialProps(ctx) { | |||
| await revalidate().then((shouldUpdate) => { | |||
| if (shouldUpdate) { | |||
| ctx.res.writeHead(307, { Location: ctx.req.url }); | |||
There was a problem hiding this comment.
Have you found out why this is required? We have the same - I started going deep, trying to track it down but with this workaround staring me in the face, I couldn't throw too many hours at it for now
There was a problem hiding this comment.
becuae next is already running, so if you clear require cache mid way though, its router is destroyed.
require cache clear really needs to happen before next entry points are required.
When it comes to federation, next.js is about the worst you money can buy.
Maybe when we introduce true HMR i can find a better solution.
There was a problem hiding this comment.
I worked aorund this in the past by using a proxy on get initial props and spawn a child worker to run the render and then just return the req/res via the worker and not the host.
There was a problem hiding this comment.
Ah ok, I had been exploring how I could do something like that, but inside the next.js app. Doing it outside is a rather interesting idea indeed! We haven't needed to run a custom next.js server yet but for this I think we can get one in place. Thanks I'll give it a try!
There was a problem hiding this comment.
If you are using a custom server, you can easily clear the cache by running the "clear cache" command before it hits the next route handler middleware. You can achieve this by referring to the custom server http example and placing the "require(/next)" in the middleware of HTTP server instead of at the top of the file. By doing this, you can clean and re-require Next.js whenever necessary.
It is entirely doable, but not easy to make it out of the box for users, hence the double redirect hack haha.
If you look in the server dist folder, you can see if Next.js emits some basic HTTP dev server in dev mode. If so, we can patch it via a Webpack plugin by overwriting it or something.
There was a problem hiding this comment.
@ScriptedAlchemy, we're not currently and would ideally stay away from a custom server. I had a look at your suggestion. but I don't think you can move the require(next) because it's used to prepare the server and create the route handlers.
We have some unique limitations in what we can do in our project, but I'm sure there is a solution with module federation involved. While I continue my journey into the code and try to find a solution, we've had the ok to book some time to chat with you in more detail. Hopefully, that's ok with you? We'll have a look and stick something in calendly
# Conflicts: # packages/node/global.d.ts # packages/node/src/utils/flush-chunks.ts # packages/node/src/utils/hot-reload.ts
ScriptedAlchemy
force-pushed
the
feat/chunk-lush
branch
from
a639b41 to
4c04cb5
Compare
|
@ScriptedAlchemy please could you publish a test version from main including these latest enhancements? |
Description
improve chunk flushing by using module wrappers to track usage and invocations
Better ssr printing of js assets by tracking them via wrapped module, works similar to next/dynamic hoc
Related Issue
Types of changes
Checklist