I’ve asked the cache team in the past and I believe they’ve said that cache logic is plan agnostic.

I’ve asked the cache team in the past and I believe they’ve said that cache logic is plan agnostic. I’m not sure where this info comes that different plan levels have different caching rules, but as far as I can establish that’s not accurate. Would love to understand if there’s some documentation or some comms from us that says otherwise so I can make sure our comms are aligned that would be helpful. If you can give me your zone ID I can take a look and see if our traces say anything. But if the snapshot of timestamps is measuring the total execution time of the worker (seems likely since Date.now only has ~second level precision and accuracy), a more likely explanation might be that your Worker isn’t getting enough traffic and is getting evicted. Or even more likely, it’s getting enough traffic to stay live, but not on every machine in the data center. There’s been a long running project to make that faster but it hasn’t finished yet. Anyway, if you’re measuring from outside the worker, there’s more things than just KV that could be in play.

ajgeiss0702•6/29/23, 4:09 AM

Ah, right. I keep forgetting that Date.now isn’t accurate.

I’m making 4 kv get requests in this route, and it regularly takes over a second for it to complete, which afaik doesn’t happen anywhere else in the worker (it’s a pages project)

As for the plan-caching misinformation, the only I’ve heard about it is from Unsmart above

zehawk•6/29/23, 6:22 AM

I believe that there are 2 latencies that affect cold read - loading time of worker, followed by loading time of KV. I've noticed a difference between 3 scenarios:

Infrequent worker + infrequent KV = ~2s
Frequent worker + infrequent KV = 600ms - 1.3s
Frequent worker + frequent KV = <100ms - 200ms

I'm on a paid plan which says "Always lowest latency" as opposed to "Lowest latency after first request", but Ive found no difference from free.

Oohqte hey is it okay if i publish my `wranger.toml` file on github if it contains my `...

Hello, I’m Allie!•6/29/23, 12:58 PM

They are unique to your account. As long as you don't store secret variables in there, you will be fine

Zzehawk I believe that there are 2 latencies that affect cold read - loading time of wor...

beelzabubOP•6/29/23, 7:59 PM

KV has enough traffic everywhere that I don’t think you’re going to see cold loads of its worker

Bbeelzabub KV has enough traffic everywhere that I don’t think you’re going to see cold loa...

Hello, I’m Allie!•6/29/23, 8:08 PM

Do you mean the user’s Worker, or the pixelcf

Worker(?) that interfaces with KV?

Bbeelzabub I’ve asked the cache team in the past and I believe they’ve said that cache logi...

Unsmart•6/29/23, 8:09 PM

I'd be really surprised if the CDN doesnt cache files longer by plan level given the testing I have done of a free plan vs enterprise... the free plan caches data for a few minutes at low volume but it will cache data for hours on enterprise

But I dont think ive ever seen it officially said from an employee fwiw so just people from the community have said it

UUnsmart I'd be really surprised if the CDN doesnt cache files longer by plan level given...

beelzabubOP•6/29/23, 8:55 PM

I just confirmed with the cache team again. That is not the behavior today (may change at some point in the future if we want) - the cache is totally agnostic and the only thing that drives eviction is relative frequency patterns. ENT customers typically have more traffic on their zones and that’s why free gets evicted more quickly.

Unsmart•6/29/23, 8:56 PM

Good to know meowthumbsup

shapeshifter•6/30/23, 2:44 AM

I see that 1M list requests are included in the paid plan. I also see that a list request has a default of 1k items, if I make a single list request and it returns the default 1k items does this count against billing as 1 list request or am I billed per key returned (1k)

Sshapeshifter I see that 1M list requests are included in the paid plan. I also see that a lis...

Hello, I’m Allie!•6/30/23, 3:19 AM

1 list request, regardless of the amount of keys returned

HHello, I’m Allie!Do you mean the user’s Worker, or the <:pixelcf:887381457201414185> Worker(?) th...

zehawk•6/30/23, 2:01 PM

Yup, I meant my worker of course.

docy•6/30/23, 4:39 PM

Let’s say I have a namespace with keys mapping to stringified objects, would it be worth it to shorten the keys of said objects to save bandwidth/storage costs?

docy•6/30/23, 4:40 PM

e.g changing a key like "releaseDate" to "r". there are going to be quite a few entries so I’m not sure if this would help at all

Unsmart•6/30/23, 4:47 PM

I mean with millions of items you will probably save a bit but it would make it really annoying dev wise so I wouldnt do it.

Original message was deleted

docy•6/30/23, 4:50 PM

Well I mean all it takes is a simple remapping function to map the stored shortened keys to readable keys

docy•6/30/23, 4:51 PM

I just want to know if I’m overoptimizing here, I mean it’s going to contain every movie/tv show on IMDb so

docy•6/30/23, 4:54 PM

Ye alright, thanks

vassbence•7/1/23, 6:58 PM

Does KV perform good when used via the HTTP API?

vassbence•7/1/23, 6:59 PM

If I'm working in an environment where I can call the HTTP API or call a Worker which then uses service bindings which will be faster?

Vvassbence If I'm working in an environment where I can call the HTTP API or call a Worker ...

Hello, I’m Allie!•7/1/23, 7:48 PM

For production apps, I would generally recommend using a Worker, since the Worker KV API can be cached locally, while(iirc) the HTTP API has to hit core

Erisa•7/1/23, 7:50 PM

The API is also rate limited

reavatar•7/2/23, 11:21 AM

I have next.js site on cloudflare pages. I want lowest latency. Each request is essentially a read + write both. Wondering what to use cloudflare kv or upstash redis.

Each key is associated to a user, so I only need read/write from closed point to user. Users are situated globally.

Unsmart•7/2/23, 1:39 PM

Durable objects would be the fastest if the user is only reading and writing their own key and no one else is

Phatso•7/3/23, 12:14 AM

Could someone help me understand some stuff in this part of the docs?
https://developers.cloudflare.com/workers/runtime-apis/kv/#parameters

Question 1:

Due to the eventually consistent nature of Workers KV, concurrent writes can end up overwriting one another. It is a common pattern to write data from a single process via Wrangler or the API. This avoids competing, concurrent writes because of the single stream. All data is still readily available within all Workers bound to the namespace.

Does this mean "concurrent writes to the same key"? Or just concurrent writes to any keys in general?
Can I store two values with different keys at the same time? One immediately after the other?

Question 2:

Writes are immediately visible to other requests in the same global network location, but can take up to 60 seconds to be visible in other parts of the world. Refer to How KV works for more information on this topic.

This section is a bit vague at best, and at worst, conflicts with the "How KV Works" page.

On the "How KV Works" page it says:

Initial reads from a location do not have a cached value. The data must be read from the nearest central data store, resulting in a slower response.

Implying that, so long as its the first Read in the location, it will reach out to the central data store and get + cache the value.
However, based on how the API docs are written, it implies that it could take up to 60 seconds for a "Write" to be visible globally.

Which is it?

Question 3:

Will a .put.put ever "fail"? If so, what does it look like?
i.e. what if the data or the key were too big? Are there other reasons it could fail not listed in the docs?
How do I catch and handle these errors?

On the .put.put docs it says:

This method returns a Promise that you should await on in order to verify a successful update.

Does this mean I should expect an error if it fails? Or set it up with a

.catch

.catch

or something?

PPhatso Could someone help me understand some stuff in this part of the docs? <https://d...

Chaika•7/3/23, 12:39 AM

Does this mean "concurrent writes to the same key"? Or just concurrent writes to any keys in general?

Same key, you can write to different keys, not affected by that

Which is it?

It's both, I don't see how those conflict, the first statement is a bit vague though. Writes update local colo (location) cache. If you write to ORD (Chicago), any KV Requests for that same key in Chicago will see the updated key instantly. If you had requested that key in EWR (Newark) recently enough that it is still in cache, it can take a minute or so to update to that new value, depending on when you fetched it. If you had never fetched that key from LHR, then a request frrom lhr would hit one of the central data stores and see the updated value instantly.

Note recently there was an update to decouple the background refresh rate from the cache ttl: https://blog.cloudflare.com/faster-workers-kv-architecture/

Question 3:

I believe yes on network exceptions and such edge cases where it just fails to write/confirm entirely, as well as usual validation of too long key/value. The method returns a Promise<void>, it would just throw an error like Type: Error, Message: KV PUT failed: 414 UTF-8 encoded length of 1064 exceeds key length limit of 512.KV PUT failed: 414 UTF-8 encoded length of 1064 exceeds key length limit of 512.

Phatso•7/3/23, 12:56 AM

Thank you, that was very informative.

My clients are getting 404404s when trying to access data that was just created.. but only sometimes.
Trying to track down what I've done wrong.

Phatso•7/3/23, 12:59 AM

And if they retry the same request enough times, eventually they find it

Kkian Do you have code to reproduce it?

Phatso•7/3/23, 6:06 AM

Yes, now I do. I submitted a support ticket about it - would it be better to share my test project/info here?

Erisa•7/3/23, 9:20 AM

Hey @phatso are you able to confirm the ticket number you raised? Wondering if it's the one I was on just now, or another

matt•7/3/23, 9:22 AM

is there any way to push a keyvalue out to all edge locations and not have them expire? trying to check api keys for incoming requests

EErisa Hey @phatso are you able to confirm the ticket number you raised? Wondering if i...

Phatso•7/3/23, 9:27 AM

Yep! That's me

PPhatso Yep! That's me

Erisa•7/3/23, 9:30 AM

Hey

do you have a link to the message you were referring to, so I can understand the context?

docy•7/3/23, 3:12 PM

is there any way to get keys in bulk? googled but couldn't find anything

something like this:

env.NAMESPACE.get(['foo', 'bar', 'bang'])

env.NAMESPACE.get(['foo', 'bar', 'bang'])

kian•7/3/23, 3:13 PM

Not with the Worker bindings, no

docy•7/3/23, 3:13 PM

why is it available over http api but not worker bindings?

kian•7/3/23, 3:13 PM

I don't see it in the API either

kian•7/3/23, 3:14 PM

There's bulk writes and deletes, but no gets.

docy•7/3/23, 3:14 PM

oh yeah i misread

docy•7/3/23, 3:14 PM

that's weird

docy•7/3/23, 3:14 PM

wonder why

eg_scl•7/3/23, 5:26 PM

I'm struggling with using kv, are there any good whole project examples. I see in the docs there is a tutorial for rust, but not ES. I'm just getting null when I call env.NS.get('whatever') and I'm sure it's got to be a setting so I'm looking for a whole working project to compare too. Thanks.

kian•7/3/23, 5:51 PM

https://developers.cloudflare.com/workers/runtime-apis/kv/#reading-key-value-pairs

getget will be
```
null
```
```
null
```
if that key doesn't exist.

eg_scl•7/3/23, 5:58 PM

Thanks @kian - thing is, I can get the correct value from the command line "wrangler kv:key get --namespace-id=### "{key}" but it just doesn't in the worker.js. My binding has the same preview_id as I used in the commandline call. so I'll keep trying.

kian•7/3/23, 6:04 PM

Is it a published Worker?

kian•7/3/23, 6:04 PM

Wrangler dev defaults to local development, so it has no access to your real KV namespace.

eg_scl•7/3/23, 6:40 PM

@Kian oh. I did not see that mentioned in the docs. I'm using local. So is there a local kv to setup and use?

I’ve asked the cache team in the past and I believe they’ve said that cache logic is plan agnostic.

Question 1:

Question 2:

Question 3:

Similar Threads

Question 1:

Question 2:

Question 3: