Workers Cache
blog.cloudflare.com
blog.cloudflare.com
I was looking forward to the "why it took us this long" explanation but it wasn't explicitly spelled out. Any Cloudflare staff here able to expand on that?
(The article does a good job of showing how many different smart design decisions went into this, but given caching is core to what a Cloudflare does I'm still a little surprised it took 9 years to get here!)
We implemented the "standard" Cache API back in the early days because it was a standard, one which was intended to be used together with the Service Workers API, which we were also building around back then.
But it was never a good fit. The get/put API was designed for a local browser cache, not a distributed cache like Cloudflare's. We probably should have realized this before implementing it, but it really became obvious over the years of actual use.
But given we had something that mostly worked for most use cases, it was hard to prioritize redoing it against all the other things on our plate. So we deferred.
More recently, architectural changes we've been making in Workers for other reasons happened to make it significantly easier to finally implement this the way we wanted, and we were able to find some engineering time to get it over the line.
Gory details for the curious: We've been improving the infrastructure around the notion of workers having multiple "entrypoints", with the ability to parameterize those entrypoints. ctx.props[0] and ctx.exports[1] are part of this. A lot of this was motivated by Dynamic Workers sandboxing, but the concept also presents a clean way to inject a cache between two parts of the same worker, by applying it to the entrypoint and having the worker call itself using ctx.exports.
Moreover, the introduction of "channel tokens" made a big difference[2][3]. Essentially I created a way to encode a token (bytes) representing an arbitrary entrypoint to a Worker, complete with its serialized parameters. I did this to enable these entrypoint stubs to be passed over RPC, which is again useful for sandboxing use cases, but it also created a convenient, encapsulated way to pass information through our cache infrastructure about what worker should run at the other end.
It's not a huge breakthrough or anything, but I think it made the architecture clearer in everyone's mind to the point that we got excited about using it to implement caching properly, finally.
[0] https://developers.cloudflare.com/workers/runtime-apis/conte...
[1] https://developers.cloudflare.com/workers/runtime-apis/conte...
[2] https://github.com/cloudflare/workerd/blob/main/src/workerd/...
[3] https://github.com/cloudflare/workerd/blob/main/src/workerd/...
Our architecture is something like:
ingress -> routing -> security -> workers -> cache -> origin
That is, workers run "in front of" cache. That's usually what you want. Workers run on the edge, so putting them in front of cache doesn't cost anything in terms of latency, and there's a lot of useful stuff you can do only if they run in front, e.g. serving pages from cache but customizing them for specific users.
Putting workers behind cache is an architectural change. All of the logic around routing to them lives in front of the cache. And it makes Workers a lot less useful.
We weren't that excited about it until we had a clearer story for how to run custom logic on both sides of the cache, but it was pretty unclear how to do that in a nice way until the recent developments around ctx.props, ctx.exports, channel tokens, etc.
This change in billing makes enabling caching not such a straightforward decision, and encourages separating cached and non-cached parts of your workers into two separate workers, which is a bit annoying.
However, I'd note that, in terms of our costs, it's often cheaper for us to run your Worker than to serve from cache. Consider that Workers run locally in whatever edge location the request landed in. We are not saving any bandwidth costs by putting the cache in front of Workers rather than behind, and we've built an architecture such that running a Worker is extremely cheap, almost free.
In that light, making cache hits free would basically be giving away money. To be honest, that is another reason why implementing cache in front of Workers has been held back so long. I think the team finally decided that the only way we can deliver it without giving away too much money is to charge full price for the requests. Workers pricing is already incredibly generous, and we do need to run a business at the end of the day.
(Of course, a cache hit still saves you the CPU pricing. CPU time pricing more directly translates to costs for us, so that makes sense.)
I admit the side effect on static assets is a little weird. But no other provider offers unlimited free static asset serving to start with -- so it's only weird relative to our own unreasonable standards. :) I think the thought here is that if you need caching in front of Workers, that implies you are doing complex work in Workers and getting a lot of value out of them. So a bit of "price segmentation" kicks in here. That said, you can work around it by serving static assets from another hostname, etc.
Worker-to-worker invocations also now participate in the cache, so despite the pricing increase, at least we’re getting some added value.
But for static assets in particular, I don’t see what value the cache is adding, so it feels like pure price segmentation.
Or, to put it another way, you were already caching the static assets and not having the workers serve them. Now you're doing the same thing, but charging for it.
1. What headers does content that will revalidate use on client. Eg can a react hook trigger if a cms json revalidates?
2. If edge worker can call database worker: if database worker sends stale while revalidate, does that revalidate whole worker chain?
btw overall very cool, esp fine grained cache for authed/etc data
On top of that the cache tags are a slick way to do invalidation. This looks like a great product.
- You still get billed per request, whether the request hits cache or not (but don't get billed for CPU time)
- You now get billed for static asset requests! This makes no sense to me. "One thing to watch: when caching is enabled, requests that are normally free — static asset requests and worker-to-worker invocations through service bindings or ctx.exports — are billed at the standard request rate, because each one now consults the cache in front of your Worker." Yeah that sounds like a bug that just happens to generate them more money.
- The cache key automatically has the worker deployment version, so even gradual deployments populate their own cache which is nice
- It seems like you can set a totally custom cache key? But that was previously Enterprise only, can't see if that's still the case here.
Until now these cache headers didn't work if you set them in a Worker, because Workers were always run in front of the cache. You could use them if you're running Cloudflare cache in front of another origin, but not if you were e.g. rendering a site in Workers. This changes that.
Yeah. It sounds like they moved the component that counts requests earlier in the pipeline so that it can count cache hits, and now it also counts these.
When Workers Cache is enabled, every cacheable request to your Worker hits Cloudflare's cache first. If there's a fresh cached response, Cloudflare returns it directly — your Worker doesn't run, and you don't pay CPU time for it. On a miss, your Worker runs, and if your response is cacheable, Cloudflare stores it for the next request. The next request from anywhere on Earth can be served straight from cache.
I know about the infamous "Enterprise plan" (especially, when your bandwidth is as high as ours in 100s of TBs) and know of at least one other tech shop that was required to pay for it ... but we haven't got that sales call, yet.
Time to migrate rarely-updating websites that still need a CMS to Cloudflare.
> One thing to watch: when caching is enabled, requests that are normally free — static asset requests and worker-to-worker invocations through service bindings or ctx.exports — are billed at the standard request rate, because each one now consults the cache in front of your Worker.
That’s a very weird limitation from a product standpoint. You might end up spending more money by enabling caching.
I wonder if static assets and worker-to-worker invocations are really more expensive to them with caching enabled, or if this is just a metering problem.
I would have made this free in the interest of making billing simpler and of making the decision to enable caching more straightforward.
If you’re min-maxing costs, this encourages awkward setups where you split workers into two, one with caching enabled and one with caching disabled.
A big worry was always "why does workers sit in front of my cache? that's a waste of an invocation if i'm returning a cached result"
But now the platform has evolved to where you’re running the entire app from Workers.
Thanks CF.
Responding to alleged slop with more slop doesn’t decrease the total amount of slop on the internet.
Why are we all of a sudden pretending like pre-LLM era blogs were these pristinely well written pieces of art or even that effort and care was put into them? In most cases they were significantly less coherent and incomplete. Don’t get me started on the mess that was the communication of this particular company or one of their competitors like AWS.
The LLM explained the core concept and features very well. But it was dull and boring to me, as I already have to read this writing style at work pretty much all day every day.
Aren't you gonna let the LLM develop for you anyway? Why bother writing and reading a post at all?
> pretending like pre-LLM era blogs were these pristinely well written pieces of art
The point is the effort and care that the writer puts which differentiates it from automatically generated text. That matters because a human can sympathize and that leads to better understanding and greater connection. That's why a post is written.
> Don’t get me started on the mess that was the communication of this particular company or one of their competitors like AWS.
And we criticize those as well. Nothing's changed. Yesterday's bad content is today's slop (plus a mind boggling amount of investment, corruption and environmental side effects).
> multi-tenant-safe cache keys
> on a server-rendered app
> byte-for-byte identical (classic)
> gets a cache-speed response
> cached-file-extensions list
Honestly, this is terrible. I had to add a "use simple words only, don't hyphenate unnecessarily" warning to my Claude config. After a full day of work, having to read these Claudisms all the time make a noticeable difference on how tired you get. It gets even worse when Claude starts to make up its own vocabulary.
The human element cannot be recreated because the human element that created the beast becomes further removed and only the beast remains.
I say beast to provide a tell that what I write is 'human'.
Even if you use em dashes and a few phrases that have become associated with AI writing, there’s still an unmistakeable sense of how much effort was put into the writing.
But I suppose there might be naive readers who don’t know how to spot this effort and would false positive on em dashes or supposed AI phrases.
> What's the competition in the gaming-capable pre-built mini-PC category
Also, most engineers will likely just be skimming this article before feeding it into their harness to implement the changes anyway, so it makes sense for it to be more heavy on context than it would be if meant for only humans to consume.
I am assuming it is a bunch of manual work.
it already uses Workers Cache for the route-level ISR cache
I just don’t understand why undermine your own announcements by delegating comms to the machine. It’s disrespectful to the reader.
I found it funny to just read through.
I think you are looking is the signal separated from the noise. I read technical documents very quickly while having very high retention and comprehension. The AI slop and AI content generation is pushing my limits.
'AI summarization' is what I naturally or learned to do growing up. I read your request as 'human readable format' being something more conversational or perhaps with known intent: inform the human.
It is not possible. Socratic method is the better path to learning. AI can make it easier.
Suggestion: AI, summarize and then ask me questions about this topic in increasingly more difficult questions, and after passing the depth of this document push further into the deeper fundamentals and first principals of the topic.