the reason I want to do it with workers is because I don't want any site data/load to be sent before
the reason I want to do it with workers is because I don't want any site data/load to be sent before the user is verified
IWhat do you mean by sent?
IOh, as in, you want it to immediately block them without serving an HTML page?
Pyes
IYou should use SvelteKit middleware in this case, not _worker.js. SvelteKit already uses a _worker.js so you can't add your own.
ISorry, SvelteKit calls it Hooks not Middleware
IPOh. But I mostly wanted to make it using a worker since I want to use this in other projects as well, guess this will work though
IBasically, check their auth (cookie or header or whatever) and if it's invalid just don't call
resolve, return an error instead.
AWould love to get some insights here: https://canary.discord.com/channels/595317990191398933/1174021841023664288
I'm currently trying to fix
I'm currently trying to fix
_routes.json
DI have question need your help: How to run local for pages project (example run react app and have function in it)
DWhen I deploy by command npm run pages: deploy, it work well,
DBut still not run app in locally
HNo
ZI would be very surprised if it isn't planned. This is an often-requested feature. It seems like there are just other features that are prioritised over it
ZYou can always deploy a worker with a cron trigger or use DO alarms.
HI’m guessing this is a why bother, it will work in convergence situation
DSo, is there a good way to prevent CSRF attacks on pages with forms? I figured that I could have javascript request a form from a worker after the webpage has been loaded, but I'm uncertain that would work.
J
K
JI would have to put the build step in the dashboard then? Can it do a cargo build?
JI looked into this, but I do not think this will work. The wasm-pack wrapper is hard-coded to always put the worker in the
/build/worker/shim.mjs file, instead of the expected /functions/_worker.js.KI’d probably just
&& mv it in my build commandJUnlucky me, I thought I had tried everything. I need to know whether this'll do it now.
JAh, I found the issue. This doesn't work with
wrangler2 pages dev
HYou could import shim from _worker.js
JThese are all great ideas, I am experimenting
JI am getting the wildest behaviour. I right now have the following build command:
When doing
However, this error occurs while yarn is still building the frontend, as in, it'll be in the middle of the build logs. Not sure what is going on there. I get a similar error from the worker:
As you can see, despite CloudFlare alleging the worker doesn't exist, it compiles successfully and actually runs. All of this output is produced while the build command is still running.
When doing
wrangler2 pages dev frontend/dist, I get an error:✘ [ERROR] Could not validate _routes.json at /home/jeroen/project/frontend/dist: Error: ENOENT: no such file or directory, open '/home/jeroen/project/frontend/dist/_routes.json'
However, this error occurs while yarn is still building the frontend, as in, it'll be in the middle of the build logs. Not sure what is going on there. I get a similar error from the worker:
✘ [ERROR] 1 error(s) and 0 warning(s) when compiling Worker.
▲ [WARNING] Failed to bundle _worker.js. Error: Build failed with 1 error:
../../.wrangler/tmp/bundle-sGxLQ0/middleware-insertion-facade.js:1:43: ERROR: Could not resolve
"/home/jeroen/project/frontend/dist/_worker.js"
at failureErrorWithLog
(/usr/lib/node_modules/wrangler/node_modules/esbuild/lib/main.js:1636:15)
at /usr/lib/node_modules/wrangler/node_modules/esbuild/lib/main.js:1048:25
at /usr/lib/node_modules/wrangler/node_modules/esbuild/lib/main.js:1512:9
at process.processTicksAndRejections (node:internal/process/task_queues:95:5) {
errors: [
{
detail: undefined,
id: '',
location: [Object],
notes: [],
pluginName: '',
text: 'Could not resolve
"/home/jeroen/project/frontend/dist/_worker.js"'
}
],
warnings: []
}Compiled Worker successfully
As you can see, despite CloudFlare alleging the worker doesn't exist, it compiles successfully and actually runs. All of this output is produced while the build command is still running.
JEven more wildly, if I delete the
frontend/dist directory, my worker will not work. It won't even be detected. So CloudFlare seems to insist my worker already exists before the build command is ran, and then compiles it while the build command runs.JI believe this is caused by some form of live-reloading that cannot be disabled. If I replace my build command with a
sleep 60, at no point does CloudFlare try to build the worker until the sleep is over. However, if I add a && sleep 60 to my build command, cloudflare will try to build my worker before the sleep has even started.JBasically, cloudflare is trying to bundle my worker while it is being built, which causes errors
HIs this in your
wrangler.tomlyep
Jbasically the only way I can get it to work is:
- Comment out command = in my wrangler.toml.
- Run the command manually
- Run wrangler pages dev frontend/dist
JAny other way wrangler will try to bundle my worker and routes while the files are being produced still
HThat might be the issue. It is a bug that
wrangler reads your config when using PagesJI see, I was under the impression it was intended behaviour so tried to get it to work nicely. That also explains why some values in the wrangler.toml are seemingly ignored such as the compatibility date
HIIRC it is because it shares a good bit of its code with
wrangler dev
JWell, I assume some of it should be read, such as key-value namespaces?
HNo, bindings are configured via CLI flags
JI see, got it
JAlright, my only final hurdle is that I now need to deploy it, which requires building it. I assume building it on the cloud is not the way to go here.
JI got it to work this way! I did have to disable Cloudflare's built-in build process though, and upload manually with wrangler.
One side-effect is that all my backend code is accessible through the web now, as people can navigate to ./shim.mjs etc. Is there a way to stop that? I am thinking of making a redirect rule, but perhaps something more elegant works?
One side-effect is that all my backend code is accessible through the web now, as people can navigate to ./shim.mjs etc. Is there a way to stop that? I am thinking of making a redirect rule, but perhaps something more elegant works?
HThe
shim.mjs shouldn't be in the output directory at all. Maybe try moving it somewhere else?J@HardlyHolidayin’ If I upload manually, I can only upload the output directory.
JAnd cloudflare's built-in build process doesn't work as it doesn't support cargo compiling.
