Speeding up access to resources on tools.ietf.org and www.rfc-editor.org.
I pity the tools!
A pretty simple Cloudflare Workers script to speed up access to resources on tools.ietf.org and www.rfc-editor.org. Any path on those domains should work.
In my experience, the first load time for an RFC document on tools.ietf.org is reduced from around 1.5 seconds down to 0.5 seconds). Subsequent loads are reduced further to 0.05 seconds.
For example, try measuring and comparing load time of https://tools.ietf.org/html/rfc7540 and https://tools-ietf-org.lucaspardue.com/html/rfc7540 with local caching disabled to see how much faster life can be when modern edge capabilities are harnessed.
The speed advantages are nice for humans but also work well for machines. For instance, if you're using Martin Thomson's I-D Template then you'll be familiar with turning source into IETF-formatted text, HTML or XML with
$ makeUnder the hood, citation metadata is automatically fetched and written into the output. Some metadata is fetched from the Web using HTTP and so it suffers from slow access times. You can point the document generation instead at edge-tools-ietf by passing an environment variable
$ XML_RESOURCE_ORG_PREFIX=https://xml2rfc-tools-ietf-org.lucaspardue.com/public/rfc makealternatively, you can also modify the same entry in your project's config.mk.
For now, the worker is hosted on my domain lucaspardue.com. To access these sites proxied by these workers, all you need to do is take a hostname, convert the dots to hyphens and append lucaspardue.com. The follow table lists supported substitute hosts:
| Original | Substitute |
|---|---|
| tools.ietf.org | tools-ietf-org.lucaspardue.com |
| xml2rfc.tools.ietf.org | xml2rfc-tools-ietf-org.lucaspardue.com |
| www.rfc-editor.org | www-rfc-editor-org.lucaspardue.com |
Most paths on tools are probably supported but I've mainly focused on access to HTML versions of Internet-Drafts and RFCs. Chances are other stuff could act funky, if so ping me and just fallback if needs be.
There's not really much that is clever or custom going on. When a request hits the worker, we check if the resource is stored in the edge cache. If it's not, we check a small, always warm KV store and then promote that into cache. Finally, if there are two misses, we fetch from the origin and put it in cache.
We use the Cache API, KV API and Fetch API.
The KV store is prepopulated using Worker Cron Triggers.
Since we're rewritting URLs, we need to tweak response payload a bit. We do this inline using the HTMLRewriter API.
This is still early days. The worker is hardcoded to do a lot of things and be hosted on lucaspardue.com. Future work should make the code more portable so others can experiment more easily.
Testing has been limited, so feature requests or bug reports are welcome.