What kind of traffic can workers handle if they're being used to serve up static pages from R2? Is t
What kind of traffic can workers handle if they're being used to serve up static pages from R2? Is there any caching involved?
UMillions per second for workers. You can use the cache api if you want. You can also use public buckets to add your domain instead of using a worker.
UR2 can handle a few thousand rps (like 15-20k iirc). But cache can push that further since it won’t hit R2
CYea R2 would be your bottleneck there, not workers. There's no cache by default with Workers, but you could use the cache api (or use an existing library like kotx/render which does it for you) and save on some Class Bs and faster responses (at the cost of consistency).
If you can use R2 Custom Domains even better pricing wise since you're not paying for the Worker, and then you can just use a Cache Rule to force caching
If you can use R2 Custom Domains even better pricing wise since you're not paying for the Worker, and then you can just use a Cache Rule to force caching
DDo cron triggers support smart placement?
KNope
OCache was super easy to get running -- thanks for the tip!
OOkay, one last question -- is it possible to direct traffic from a client's domain using a cname/similar so example.client.com/endpoint could route to the same worker as nativedomainexample.com/endpoint?
CSounds like a use case for CF for SaaS https://developers.cloudflare.com/cloudflare-for-platforms/cloudflare-for-saas/, special setup for Workers: https://developers.cloudflare.com/cloudflare-for-platforms/cloudflare-for-saas/start/advanced-settings/worker-as-origin/
It's what Pages/R2 use internally for Custom Domains
You'd have to take over the entire hostname/subdomain though, ex: example.client.com, can't just route specific paths. You could of course 404 on all but /endpoint
It's what Pages/R2 use internally for Custom Domains
You'd have to take over the entire hostname/subdomain though, ex: example.client.com, can't just route specific paths. You could of course 404 on all but /endpoint
Learn how to use a Worker as the fallback origin for your SaaS zone.
OThat sounds perfect! Thank you so much
Jexcuse me ... waht is Workers Constellation
UThat doesn’t exist anymore it’s just #workers-ai-beta now
Joh I saw it in the navigation then dissapear
Pjust had that, clearing cache and refreshing fixed it
Pthe broken update may still be cached in your region, nothing to do except wait
CSeeing some reports of it. It has been raised and being looked into
DIs
cacheKey still an enterprise-only feature?DThought it was, but don't see it mentioned in the docs anymore
DIs this still the case?
IAs far as I know, yes.
Doh.
II
Wwhat happened if I need more than 10ms for my operation?
Wlet's say 12ms
WI see that my median worker cpu time is 12ms
Wbut not sure why it's still running, cause I have exceeded the limit
AIt’s not exact in my experience, I had a worker that would sometimes exceed CPU usage (on the old plans) by 20-30 ms and still “succeed” and some that were 5ms over and fail - the paid plan ($5/month) has a much higher CPU limit per invocation though if you do hit it
EThe limit is applied quite loosely and doesnt stop execution unless you go over it by a lot or several times. But as a general rule if your median cpu time is above the limit then even if it didn't trigger this time it might in future so its better to get usage down or switch to a paid plan to avoid interruption
EPaid plans allow for up to 30 seconds of CPU usage per request, where you pay more if you use more (above the amount you get for $5/mo)
Sa bit confusing? it's utterly confusing 
SThank you for the tip. I think i will opt for 
documentation URL in this case 
documentation URL in this case 
Vdoes cloudflare have a nice solution for OAuth CSRF tokens? I don’t want to use kv, as it’s not really designed for that… can i keep the worker waiting somehow?
UKV is perfectly fine for CSRF tokens 
can also use durable objects probably cheaper and faster also
can also use durable objects probably cheaper and faster also
Vthe main problem is the write pricing
VThough i was clearly misremembering the pricing, as i thought the free tier was 1000 a month and its 1000 a day
UDurable objects are a lot cheaper. $1/million writes + 0.15/million for the invokes. D1 can also be a choice (for non production critical apps currently)
Dwhy not D1?
D@kian Is there any downside to storing S3 objects with gzip compression?
Vnah, for my volumes 1000 writes/day is plenty
DIs the Cache API shared with
fetchif you use aws4fetch, you can get free caching without needing to fiddle with the Cache API manually.
DWhy?
DDo I pay for the idle time on the setTimeout?
JIn Unbound, you would pay for that wall time, yes. With Bundled, or the new Standard, no you will not
Dthanks!
