so whats the latency like with KV or R2 with workers in the same region?
so whats the latency like with KV or R2 with workers in the same region?
SHow does CF cname flattening work with CF proxying and workers?
If I flatten a cname will my worker never trigger because the response the DNS gives will cause the request to just skip cloudflares network?
If I flatten a cname will my worker never trigger because the response the DNS gives will cause the request to just skip cloudflares network?
CR2 can be pretty quick, it's not the most consistent performance though. I think ~100ms or so uncached to R2, like EWR/ORD to ENAM bucket on average is around that. Just using workers binding and not cache.
If you use cache, either with r2 custom domain or cache binding, of course gets super quick, like under 40-50ms from all locations as long as you hit cache
KV is more complex. You don't have region control, the two central stores are always EU/US. Cached it's great, espec after recent perf changes, maybe knock off 10ms from what r2 cached would be. Uncached is more complex since no region control.
They both serve very different purposes though, KV is eventually consistent, infrequent updates, lots of writes, R2 is strongly consistent, more mixed usage
If you use cache, either with r2 custom domain or cache binding, of course gets super quick, like under 40-50ms from all locations as long as you hit cache
KV is more complex. You don't have region control, the two central stores are always EU/US. Cached it's great, espec after recent perf changes, maybe knock off 10ms from what r2 cached would be. Uncached is more complex since no region control.
They both serve very different purposes though, KV is eventually consistent, infrequent updates, lots of writes, R2 is strongly consistent, more mixed usage
CWorker Custom Domains or Routes?
Cwell -- you can't modify custom domains regardless afaik, and enabling flattening on proxied records wouldn't do anything
E40-50 :pepethink:
Sah ya.. I guess it would not work because a the cname would just get replaced by a cname for the worker 
CI'm counting end to end from my perf monitoring, from the worker it's only going to be ~10ms or so I'd imagine
Cerr actually I think I have that data lol
E10 ms :pepethink:
FAs far as I understand CNAME flattening only remove intermediary CNAME pointer records and the CF nameservers respond directly with A/AAAA records?
Cyes which happens anyway when proxied
Sroutes don't cause a cname to be applied, only custom domains ya?
Eu would be an absolute life saver if u do
Cr2 cached, timing in worker
Ehow big is your data?
Ethe file you requested
C10kb
E:pepethink:
CI don't think size is too much of a factor though, once you get it streaming -- it well streams back at pretty decent speeds
Cr2 uncached from enam region to enam bucket
CKV's pretty solid
S@Chaika sorry I confused myself but to be clear, a CNAME that is proxied and flattened can or can not be intercepted by a worker with a route setup to catch something on that cname/subdomain?
Cflattening shouldn't matter at all, not even sure why it shows up on proxied records currently
CProxy inherently needs to unflatten it to connect, as the client sure can't
CSo the answer is yes* as long as you're not CNAMEing to a CF for SaaS partner on the legacy integration like Shopify
Ebeautiful
Emight have to resort to KV
Ehow long is it presistent for in memory
Cas long as it's getting frequent requests but it's complex in reality
CWorkers run on the same edge machine which handled your original request. There is no affinity to routing to a specific machine which already has your worker running. There is keepalive. Chance of two random connections hitting the same machine and thus getting the same worker is pretty low, espec when it can be any machine at any cf pop. However, keepalive can make it confusing since your browser would keep hitting the same
CIf you're worried about cold starts, CF does a good job of that, taking advantage of the tls handshake to warm up the worker during initial connection, elimating cold starts unless you've got a big worker/a lot of code in global scope
Cbut def don't rely on global scope for anything
Eits not about coldstarts
Eits more about i have templates that i need to serve
Estatic html, css
IWhere did you get these stats from?
Eits his own
E.
CIf you're using KV and you want to measure it's real easy and nothing fancy, just performance.now() and writing to Cloudflare's Analytics Engine, ex:
Then you can use Grafana w/ AE and build nice dashboards relatively easily
Then you can use Grafana w/ AE and build nice dashboards relatively easily
Equick question since you seem like you know the ins and outs. How would you go about serving static content? with wildcard domains
would you just store everything in a r2 bucket, then have a worker in the front responding with the proper files?
would you just store everything in a r2 bucket, then have a worker in the front responding with the proper files?
Cwildcard domains being one of your domains, all subdomains, or as in you'd want to hook up customer domains?
Eboth
EI thought i might get away with doing cd of CF pages
EBut doesnt seem like i can
Cwith the same set of static assets for all, or would customers have different ones or something?
EDifferent
CFor customer domains you need CF For SaaS on top, and you can't do wildcard with CF For SaaS without Enterprise
Cyou could go Workers Static Assets, pay nothing for requests like pages, and use wildcard routes for your own domain
Cthen you can use CF For SaaS easily with plain Workers
Cif you need lots of customer assets though, well you can only have 500 workers even on paid, and there's limits on assets/would be messy to intermingle, you'd probably want to just throw their custom stuff into KV
