Server migration – convert .htaccess to Nginx conf

[adamziel]

playground.wordpress.net is shared hosting and sometimes becomes unusably slow. There's a new server @brandonpayton set up, but it runs Nginx and not Apache.

We need to express the below rules using Nginx. These could be useful:

• Latest NGINX equivalent for .htaccess to host a playground #1096 (comment)
• https://wordpress.github.io/wordpress-playground/architecture/host-your-own-playground/#nginx-configuration

AddEncoding x-gzip .gz

<FilesMatch "index\.html">
Header unset ETag
Header set Cache-Control "max-age=0, no-cache, no-store, must-revalidate"
</FilesMatch>
<FilesMatch "index\.js|blueprint-schema\.json|logger.php|wp-cli.phar|wordpress-importer.zip">
Header set Access-Control-Allow-Origin "*"
Header unset ETag
Header set Cache-Control "max-age=0, no-cache, no-store, must-revalidate"
</FilesMatch>

SetEnv ENV_VARIABLE ****

AddType application/wasm .wasm
AddType application/octet-stream .data

<FilesMatch "iframe-worker.html$">
  Header set Origin-Agent-Cluster: ?1
</FilesMatch>
<FilesMatch "store.zip$">
  SetEnv no-gzip 1
  SetEnv no-brotli 1
  Header set Access-Control-Allow-Origin: *
</FilesMatch>

RewriteEngine on
RewriteRule ^scope:.*?/(.*)$ $1 [NC]
RewriteRule plugin-proxy$ /plugin-proxy.php [NC]
RedirectMatch 301 /wordpress-browser.html /

RewriteCond %{HTTP_REFERER} ^https://developer/\.wordpress\.org/
RewriteRule wordpress.html /index.html [R=302,L]

RewriteCond %{HTTP_REFERER} ^https://wordpress/\.org/
RewriteRule wordpress.html /index.html [R=302,L]

[brandonpayton]

With the new site, we don't have direct access to nginx config, but we may be able to get there with a combination of

• PHP-handled redirects
• Requesting support of additional default MIME types

The site is currently set up as a non-WP site, but we could create a different WP-based site if it turns out we'd like to use other features like a page cache based on memcached.

[bgrgicak]

There is also one .htaccess in the root folder on the server which would need to be migrated.

[brandonpayton]

So far, it looks like we will need to use PHP to handle requests so to dynamically handle different URI and to add custom HTTP headers. Using .htaccess is certainly more convenient, but it is not supported. PHP will be slower than directly serving static files, but we have edge caching available that should speed things up.

It's easy to turn on edge caching for a site, but I need to look into how we can automate cache invalidation when deploying Playground updates.

For now, I'm working on handling the different requests with PHP. Once that is working, we can worry about edge caching.

[brandonpayton]

Serving Playground from the test site seems to be working OK:
https://https://playground-dot-wordpress-dot-net.atomicsites.blog/

Because we want to customize HTTP headers and perform redirects and do not have the ability to customize nginx config, all files including static files are served via PHP using a platform feature custom-redirects.php.
https://gist.github.com/brandonpayton/9e3da0845791f6e5c833013d83cf86d6#file-custom-redirects-php

It's super verbose compared to htaccess, but it works and is surprisingly fast IMO given that PHP is also serving every static asset -- at least when it isn't under heavy load.

Before we move this to production, we can enable edge caching so those responses can be served from the cache.

What is left:

• Create a separate deployment workflow that deploys to the test site.
• Turn on edge caching and verify it is working
• Update new deployment workflow to purge the edge cache when it completes

[bgrgicak]

// TODO: Set these
/*
SetEnv GITHUB_TOKEN --secret--

Does this need to be set in PHP or is there a more secure way? I know that VIP allows admins to set env variables.

[brandonpayton]

Does this need to be set in PHP or is there a more secure way? I know that VIP allows admins to set env variables.

@bgrgicak, on WP Cloud, there is are APIs for setting and accessing secrets, so I was thinking we'd use those.

Btw, the comments at the end of custom-redirects.php are just parts of the htaccess file that haven't been addressed yet. Maybe that was clear, but I feel like explaining since they just look like cruft at the end :)

[adamziel]

@brandonpayton let's start a new issue to create and document a Playground serving flow that uses just PHP dev server or... Playground itself :-) it's not a high priority work, but let's still track it. It's cool how we can handle all the rewrite rules headers etc. using just PHP. I'm already thinking through scenarios where it makes everything self-contained and removes even a webserver dependency.

[brandonpayton]

Some updates:

• There is a new deployment workflow under Add initial workflow for deploying the website to WP Cloud #1293 which is ready for review.
• We are seeing some 429 statuses due to rate-limiting on the new host, likely to the large number of static file requests served by PHP.

For the 429s, we might need to request a greater allowance for these requests, which will hopefully be OK once edge caching is enabled. Also, we might consider carefully diffing files between website builds and requesting cache invalidation only for changed files.

[brandonpayton]

@adamziel, sure! Created #1294. Please feel free to add more context if I missed anything. 🙇

[brandonpayton]

I am currently looking into how to avoid the 429s and to get the edge cache to work for requests for Chrome. It is priming the cache for 2+ request for the same resource via curl, but when requesting in Chrome, the dev tools Network tab just shows cache misses. It's later in the day, so perhaps what is going on will be obvious to fresh eyes in the morning.

[brandonpayton]

It is priming the cache for 2+ request for the same resource via curl, but when requesting in Chrome, the dev tools Network tab just shows cache misses.

I think the hideExperimentalNotice cookie is probably causing edge cache misses. Will look at what we can do for this.

[brandonpayton]

I think the hideExperimentalNotice cookie is probably causing edge cache misses. Will look at what we can do for this.

Safari does not allow using WebStorage in private windows, so if we want Playground to run in that context (and I would guess we want it to run in as many contexts as reasonable), we cannot rely upon localStorage.

One possibility is to augment the fetch request for static files to omit credentials.

[adamziel]

One possibility is to augment the fetch request for static files to omit credentials.

Sounds like a great idea

[brandonpayton]

One possibility is to augment the fetch request for static files to omit credentials.

Sounds like a great idea

One downside is that this may cause issues for anyone wanting to host a Playground behind some kind of auth requirement. If we need to do it, it seems fine for now, and we could later add configurability so creds can still be relayed by Service Worker fetch.

That said, I've been thinking, and we may be able to have nginx serve most of our static files directly. We don't really have special rules for most files, so I think we may be able to deploy most static files in a way that nginx can find them directly without involving PHP. And when we do want to add custom headers we can omit those files from their usual location and have PHP serve them with custom headers instead.

@adamziel, is there a reason we cannot rewrite "scope:xyz/" paths in the Service Worker so that doesn't need handled on the web server? If we can do that, I think we can cut PHP out of the loop on the web server for most static file requests, and that should address the 429 errors we are seeing due to rate limiting (which can be due to using too many PHP workers at a time).

[adamziel]

One downside is that this may cause issues for anyone wanting to host a Playground behind some kind of auth requirement. If we need to do it, it seems fine for now, and we could later add configurability so creds can still be relayed by Service Worker fetch.

Good point, let's open a new issue to keep track of this. Someone will report it sooner or later, let's prioritize then.

we may be able to deploy most static files in a way that nginx can find them directly without involving PHP

That sounds good, too!

@adamziel, is there a reason we cannot rewrite "scope:xyz/" paths in the Service Worker so that doesn't need handled on the web server?

I thought we did that already? Do requests with scope:xyz in them make it to the server?

[brandonpayton]

Good point, let's open a new issue to keep track of this. Someone will report it sooner or later, let's prioritize then.

I created a PR to omit credentials from the Service Worker requests for static files here. When that is merged, I will create an issue to acknowledge credentials are omitted in case anyone has a problem with that.

@adamziel, is there a reason we cannot rewrite "scope:xyz/" paths in the Service Worker so that doesn't need handled on the web server?

I thought we did that already? Do requests with scope:xyz in them make it to the server?

Interesting! The Chrome dev tools show that as the effective URL, and even in the request's "Initiator" sub-tab, it looked like the final URL included the scope:xyz. But when I log requests on the server, all I see for scope mentions are HTTP Referer headers like:

'HTTP_REFERER' => 'https://playground-dot-wordpress-dot-net.atomicsites.blog/scope:0.8919731437847068/',

we may be able to deploy most static files in a way that nginx can find them directly without involving PHP

That sounds good, too!

Great! My plan is to:

• Rework the code that performs redirects and adds custom headers so we can ask whether a given path needs special treatment
• Create a playground-files directory on the web server to contain all Playground files
• Mirror directory structure of playground-files in /srv/htdocs/ dir which is the web root
• Symlink all files that do not need special treatment into their corresponding dir under /srv/htdocs so nginx locates and serves those directly
• Rely on nginx to defer to PHP when files are not found on disk so we can continue customizing headers for specific files.

We can do all of this in a working dir outside of the web root and then rsync everything into place under /srv/htdocs/.

🤞 I think this will take care of the 429 errors, and it should be faster regardless.

[brandonpayton]

I have scripts written to do the above mentioned setup on the host. It is mostly working, but there is a confusing bug where PHP claims that str_ends_with() does not exist even though PHP_VERSION reflects 8.1.

Planning to continue this work in the morning.

[brandonpayton]

After merging #1331 and the fix #1333, Nginx is now serving most Playground files directly without involving PHP. The 429s have disappeared, and I am seeing more edge cache HITs.

The last thing AFAIK is to get the logger working. This should be straightforward. The platform has an API for setting and retrieving secrets, and we can use the same custom-redirects-lib.php file to notice requests to logger.php and set secret-related environment vars ahead of time (so logger.php doesn't need to change). I am planning to finish that in the morning.

To see the current state of the site, check out:
https://playground-dot-wordpress-dot-net.atomicsites.blog/

[brandonpayton]

#1337 wired up the error logger on the new WP Cloud site.

A few things left:

• Redirects to / for /wordpress.html requests with wordpress.org "Referer" are not working properly. This was derived from our existing htaccess file. @adamziel is this still needed?
• Add a brief README to website-deployment/ to explain our setup on WP Cloud
• Move mime-types.php into a shared JSON file and generate mime-types.php during deployment process

The only possible MUST before switching the playground.wordpress.net domain over to WP Cloud is the wordpress.html redirect. We probably also want to do more thorough testing before making the switch.

[adamziel]

Redirects to / for /wordpress.html requests with wordpress.org "Referer" are not working properly. This was derived from our existing htaccess file. @adamziel is this still needed?

@brandonpayton we’d have to patch and re-deploy this line (and potentially another one in the same repo):

https://github.com/WordPress/wporg-wasm/blob/ce5c76146eb46578a8a80239cfb2f05cd7ac7dfe/source/wp-content/themes/wporg-wasm/src/wasm-demo/src/components/playground.js#L14

[brandonpayton]

@brandonpayton we’d have to patch and re-deploy this line (and potentially another one in the same repo):

@adamziel, I think we should be able to fix the redirect on our side as well. It's just broken at the moment. I can take a quick look at fixing the redirect, and if that doesn't work, we can adjust the behavior under wporg-wasm.

[adamziel]

@brandonpayton cool! FYI that code powers the Playground demo embedded on https://w.org/playground.

[brandonpayton]

@adamziel, thank you for the tip about how the redirect is used by https://w.org/playground.

I fixed the redirect to work. There is a secondary bug where the redirect may not happen when edge-caching has already cached a non-redirecting /wordpress.html response, and that is being fixed under #1351.

It looks like https://w.org/playground is embedding an iframe referring to wasm.wordpress.net. Is there any difference between wasm.wordpress.net and playground.wordpress.net?

[adamziel]

Is there any difference between wasm.wordpress.net and playground.wordpress.net?

both point to the same server, that redirect is the only difference

[brandonpayton]

Here is what I believe remains before switching to the new site:

• Learn what kind of GitHub token is used for plugin-proxy.php, and obtain one for use with the WP Cloud site (thinking that it is poor practice to reuse secrets in different places, though if this token is just allowed access to public repos, reusing should be totally fine)
• Update deployment workflow to create mime-types.php from the shared mime-types.json file (low priority but might as well be done before the WP Cloud site becomes production)
• Thoroughly test new site -- test it personally from various angles and then post a call for testing

@WordPress/playground-maintainers any other thoughts on this?

[brandonpayton]

Deduplication of MIME type mappings is done. I also discovered oauth.php used secrets and made an update for that.

What is left is:

• Work out whether to reuse secrets or create new ones in the case of auth tokens
• Thoroughly test new site -- test it personally from various angles and then post a call for testing

[brandonpayton]

@adamziel, I updated the secrets but am running into an issue testing GitHub Export with https://playground-dot-wordpress-dot-net.atomicsites.blog.

Trying to Connect to GitHub redirects to the following URL:
https://playground.wordpress.net/?error=redirect_uri_mismatch&error_description=The+redirect_uri+MUST+match+the+registered+callback+URL+for+this+application.&error_uri=https%3A%2F%2Fdocs.github.com%2Fapps%2Fmanaging-oauth-apps%2Ftroubleshooting-authorization-request-errors%2F%23redirect-uri-mismatch

Is it possible to update the app settings to allow redirecting to the test site domain?

[brandonpayton]

We migrated the website to WP Cloud today and this can be closed.