Cloudflare Developers•2y ago

i dont get the idea of revalidating assets when they are going to be different on each deployment an

i dont get the idea of revalidating assets when they are going to be different on each deployment anwyay ?

OOleg Isonen - document is potentially dynamic - check out for example this css resource from...

bunOP•2/18/24, 6:54 PM

this is weird ? my browser doesnt cache the asset even though Cache-ControlCache-Control headers are present

OOleg Isonen - document is potentially dynamic - check out for example this css resource from...

Erisa•2/18/24, 6:55 PM

What is the location for that bucket? And as I mentioned it will depend whether the asset is cached or not. When I access it from the UK I get a HIT and it loads in between 40 and 50 ms.
Is there a reason you are deploying these assets to R2 instead of a product designed for website hosting like Pages?

Bbun i dont get the idea of revalidating assets when they are going to be different o...

Erisa•2/18/24, 6:55 PM

What setup specifically are you referring to?

bunOP•2/18/24, 6:55 PM

his provided css url

EErisa What is the location for that bucket? And as I mentioned it will depend whether ...

Oleg Isonen•2/18/24, 6:55 PM

Western Europe (WEUR)

Bbun his provided css url

Erisa•2/18/24, 6:56 PM

What does revalidating have to do with that?

Original message was deleted

bunOP•2/18/24, 6:56 PM

yeah

EErisa What is the location for that bucket? And as I mentioned it will depend whether ...

Oleg Isonen•2/18/24, 6:56 PM

yes there is a reason, we are a SaaS, designed to host milions of sites. consider us ourselves a cloudflare pages.

OOleg Isonen Western Europe (WEUR)

Erisa•2/18/24, 6:57 PM

Testing it from the other side of the world is going to be slow unless cached, this is expected behaviour due to the distance and R2 only storing in a single region currently + not supporting Tiered Cache. The more an asset is requested the more likely it is to be in cache and serve quick.

Or are you seeing slow requests for cf-cache-status: HITcf-cache-status: HIT? Because that's a whole other question.

EErisa What is the location for that bucket? And as I mentioned it will depend whether ...

Oleg Isonen•2/18/24, 6:57 PM

Western Europe (WEUR) explains your fast response, its fast from germany too, but globally it's slow, massively, up to 500ms

Unsmart•2/18/24, 6:58 PM

63ms in Texas for me

EErisa Testing it from the other side of the world is going to be slow unless cached, t...

Oleg Isonen•2/18/24, 6:59 PM

the working assumption here is that we simply don't get the cache hit and fetching from origin across half planet, and I am genially at a loss how this is not something that has been long figured out at cloudflare

EErisa What does revalidating have to do with that? 🤔

bunOP•2/18/24, 7:00 PM

oh i thought you asked about something else.

i was refering to what you said:

Increasing the cache TTL will keep those caches around for longer but increase time to revalidate when you push new updates.

why would it increase time to revalidate if asset (JS/CSS) paths are going to be changed anyway as the name (paths) of those files changes everytime the project is bundled. assuming the index.html containing those asset paths remains dynamic (not cached)

Original message was deleted

Oleg Isonen•2/18/24, 7:01 PM

that's not what I am trying to figure out

Bbun oh i thought you asked about something else. i was refering to what you said: >...

Erisa•2/18/24, 7:01 PM

It won't. If you change the names of the assets then that's perfect. The TTL consideration is one to make mainly when trying to cache HTML pages or assets that don't change filename.

Original message was deleted

bunOP•2/18/24, 7:03 PM

its super weird. only firefox seems to cache out of chrome and safari

Oleg Isonen•2/18/24, 7:04 PM

@Erisa | Support Engineer if that helps here is what the worker logic looks like that serves that css file, see the /build/ branch

https://gist.github.com/kof/914c593049325eda72b25ad392737647

Gist

dynamic dispatcher

dynamic dispatcher. GitHub Gist: instantly share code, notes, and snippets.

Erisa•2/18/24, 7:05 PM

Is env.STATIC_ASSET_CDN_ORIGINenv.STATIC_ASSET_CDN_ORIGIN an R2 public bucket with Custom Domain?

OOleg Isonen @Erisa | Support Engineer if that helps here is what the worker logic looks like...

bunOP•2/18/24, 7:08 PM

i dont see u using caching here. i believe you can just add cfcf properties to the

fetch

fetch

request

EErisa Is `env.STATIC_ASSET_CDN_ORIGIN` an R2 public bucket with Custom Domain?

Oleg Isonen•2/18/24, 7:08 PM

correct

Ivan Starkov•2/18/24, 7:09 PM

It's cached by default fetch behavior it's seen with Cf-Cache-Status: HITCf-Cache-Status: HIT header

Original message was deleted

bunOP•2/18/24, 7:11 PM

yeah but does it retrieve the response from disk cache or still make a request?

bunOP•2/18/24, 7:15 PM

wdym both? it should display (disk cache) if it actually retrieves the asset from chrome cache

bunOP•2/18/24, 7:15 PM

why doesnt it do the same for faviconfavicon ? maybe its because content-type is not provided for this css asset? its still odd. i dont see this behaviour for traditional websites network request flows

Original message was deleted

Oleg Isonen•2/18/24, 7:17 PM

that's interesting and I am just

how it can be an unsolved problem out of the box in a straight forward way, because its not like I am trying to shave off 20ms, I am talking real shit-show slow, 500ms in many edges means we are not passing core web vitals "because we are serving static assets too slow" and that's a crazy problem to have in 2024

SSoulHarsh007 👋 I am using rclone to upload files to my s3 bucket, and today I found this ver...

dave•2/18/24, 7:17 PM

If possible, I really recommend everyone avoid multipart uploads.

OOleg Isonen that's interesting and I am just 🤯 how it can be an unsolved problem out of the...

bunOP•2/18/24, 7:21 PM

i get 10 ms for that css

Ddave If possible, I really recommend everyone avoid multipart uploads.

SoulHarsh007•2/18/24, 7:21 PM

Yep, I reuploaded the file without multipart, because one of our tooling depends on the etag being md5sum ^^

Bbun i get 10 ms for that css

Oleg Isonen•2/18/24, 7:21 PM

that's nice, must be a cache hit, though even with cache hit its usually 50ms, you must be living in the datacenter or you are testing it wrong

Oleg Isonen•2/18/24, 7:22 PM

150 is more like it when its not too far away from european R2

Oleg Isonen•2/18/24, 7:23 PM

150ms with cache hit is a s..t show as well in my opinion

Erisa•2/18/24, 7:24 PM

That will also heavily depend on the internet connection and routing quality of the individual requesting it. For example I would probably get 150ms if I used my phones mobile data or had an ISP with poor quality peering

Erisa•2/18/24, 7:25 PM

There are two ends of every connection

EErisa That will also heavily depend on the internet connection and routing quality of ...

Oleg Isonen•2/18/24, 7:27 PM

you can compare with this static file to see clear difference https://eu.posthog.com/static/array.js

Unsmart•2/18/24, 7:28 PM

thats exactly the same speed for me on cache miss and cache hit from texas

(compared to yours)

Oleg Isonen•2/18/24, 7:29 PM

actually this one seems doing well https://cdn.jsdelivr.net/npm/psl/dist/psl.min.js

Unsmart•2/18/24, 7:29 PM

cloudflare seems to do much better when cached though the posthog aws one seems to have higher end latencies when refreshing multiple times

Original message was deleted

Oleg Isonen•2/18/24, 7:29 PM

that would be an acceptable performance to me if its globally in that range, I am talking really 500ms bad globally, not trying to shave off 20ms

bunOP•2/18/24, 7:31 PM

bunOP•2/18/24, 7:31 PM

its like 200 ms

bunOP•2/18/24, 7:31 PM

globally

Bbun Click to see attachment

Oleg Isonen•2/18/24, 7:31 PM

what resource are you going for with this test, the document? we are discussing static asset, e.g. this one https://webstudio.is/build/_assets/index-6JJ5GLFW.css