We're also considering adding a way so that you can request a write have more immediate consistency

We're also considering adding a way so that you can request a write have more immediate consistency which in practice will cause all global reads to see the new value within a second.

beelzabubOP•12/6/22, 8:42 PM

Because of the use of caching, in practice availability will be very high. In practice it's generally quite difficult to guarantee a lot of 9s of availability for a strongly consistent object store. Spanner has 7 9s of availability if IIRC but that's a traditional database, not a KV object store.

beelzabubOP•12/6/22, 8:43 PM

so if there is a way to force a kv namespace to be within one datacenter or region that should be enough for us

KV exists in all Cloudflare data centers.

beelzabubOP•12/6/22, 8:43 PM

I'm also not sure what the distinction is between TTFB and network latency.

beelzabubOP•12/6/22, 8:43 PM

To me TTFB = network latency + cost of application code to retrieve asset

beelzabubOP•12/6/22, 8:47 PM

FWIW "cost of application code to retrieve asset" internally is always itself dominated by network latencies (unless you're caching & the data is available locally). CPU time makes up almost 0% of the time for data storage but that's probably semantics..

beelzabubOP•12/6/22, 8:47 PM

I'm going to move this thread to the KV room.

zegevlier•12/7/22, 9:53 AM

Has it been wrong for a while? I believe it doesn't update live, but only ever hour(?)

zegevlier•12/7/22, 9:57 AM

And it's under-reporting the size?

zegevlier•12/7/22, 9:59 AM

That might be a bug then (over-reporting could be due to partial multi-part uploads, they don't show up in the dash)... I haven't heard that one before.

andrew•12/7/22, 12:36 PM

I actually thought it was just 2 locations... what is the 3rd one, other than US and EU?

Kkian & https://twitter.com/Aly_Cabral/status/1571917591228325888

andrew•12/7/22, 12:45 PM

I'm still a little confused... it's now kosher to use the worker cache API to cache non-HTML content that comes from an R2 bucket?

kian•12/7/22, 12:46 PM

Yes

andrew•12/7/22, 12:50 PM

Oh, wow

andrew•12/7/22, 12:50 PM

cool

Karew•12/7/22, 12:53 PM

It might be a good idea to pin something about the R2+Cache rules

kian•12/7/22, 12:54 PM

The authoritative wording (self-serve agreement & supplemental terms) doesn’t say it’s OK so I’d always recommend taking it with a grain of salt

kian•12/7/22, 12:54 PM

If there’s no legalese backing then I’d be careful relying on it

Karew•12/7/22, 12:57 PM

I guess then I should change my suggestion to: Someone should put the semi-official R2+Cache policy into the supplemental product terms to make it super clear

Karew•12/7/22, 12:57 PM

Also R2+Stream too, I think

kian•12/7/22, 12:58 PM

What’s the note about Stream?

Karew•12/7/22, 12:58 PM

That's the correct way to host videos 100% on Cloudflare, right?

kian•12/7/22, 1:00 PM

Stream is the fully managed pipeline for encoding/subtitles/their own player/etc - you can also just throw videos in R2

andrew•12/7/22, 1:06 PM

good to know, thanks.

kian•12/7/22, 3:46 PM

R2 is strongly consistent so as soon as the write is finished, reads will show the new file.

kian•12/7/22, 3:46 PM

If you're using caching on custom domains or within Workers, that's just dependent on when the cache evicts the old asset

BBrammy After this was resolved I added back all my files and it only seems to add the n...

Isaac McFadyen•12/7/22, 4:11 PM

Probably related to unfinished multipart uploads?

Isaac McFadyen•12/7/22, 4:11 PM

In which case they'll be cleared automatically after a few days.

Isaac McFadyen•12/7/22, 4:11 PM

Oh yeah, you're right

We're also considering adding a way so that you can request a write have more immediate consistency

Similar Threads

Similar Threads

Similar Threads