Cloudflare's 150ms global cache purge | Deep Dive

Agreed that the old version of purge was perplexing. Why did it take 1500ms to send a message to all the datacenters that a certain URL should be on the purge list? And was every single request checked against that purge list instead of being removed from the cache? That seems extremely wasteful. Also doesn't make sense that the simplest case of purge, the URL purge is 240ms, while a prefix or tag based one is 150?

Nice video! I recently became interested in cloudflare’s offering.

I came across their Hyperdrive product which aims to make SQL queries faster by distributing a single region database globally using cache, similar to prisma accelerate.

Would it be possible for a deep dive video and benchmarking the improvement in terms of latency? Some of us want to see if this is a viable solution as an alternative to read replicas.

Thank you for your work!

NSA backed company

I was watching your old video.
Why not to shard database

What a surprise.
I am just exploring exactly cloudflare cdn

Do we know what queue system do they use? how do you look through the contents of a queue fast enough that you can afford to do it before every request?

We have CDN77 here and those guys have some pretty insane tech, they do craazy stuff with recoding video. It's nice seeing good technology used where it's actually supposed to for once. No it's not related to the video sorry.

Listening to this, i found out that i have to pronounce my s as sh.
Not sure where that bit of knowledge came from but no, we dont pronounce Australia as 'Aushtralia'. Not saying we dont do weird things, but this is new to me.

Great video, but you've been mislead about how to pronounce S in Australia. It's not Sh. You were saying it properly already. Likewise for strawberry.

Yet there's lots of accents in the world, so some would say it that way.

My suggestion (that s definitely doesn't have an Sh) is just speak how you speak. We understand you perfectly fine.

What if they use separate database for each domain instead of single db. In that case they don't need to remove millions of row for a purge.