I really want to use kv, but the cold read latency is kinda killing me. For context, I am starting a

I really want to use kv, but the cold read latency is kinda killing me. For context, I am starting an API gateway service for API provides (similar to something like an API marketplace), and I use KV to store config details about the API and the end user. KV is god tier when I load test with tons of reqs / sec, but for more realistic scenarios of bursted requests at a smaller scale, I am getting P90 of ~300ms which is just not feasible. Would love to stay on Cloudflare services such as D1, but Upstash Redis seems like a good fix. Any thoughts?

KKing Kazu I really want to use kv, but the cold read latency is kinda killing me. For cont...

Elliot Hesp•1/21/25, 6:30 PM

How often does that data change? The ttl property on get requests may help.

EElliot Hesp How often does that data change? The ttl property on get requests may help.

King KazuOP•1/21/25, 6:40 PM

Not often but only thing is data changes should be reflected quickly (e.x certain api key gets banned, don’t want to wait super long)

Original message was deleted

se•1/23/25, 10:39 AM

Hello, only based on object TTL; i.e. after you update a value, the safest is to purge after max TTL defined for your objects + some small buffer.

iLearn•1/23/25, 6:57 PM

Following is my Question and Answer(given by ChatGPT)
Please confirm / update the Answer given by ChatGPT. Thanks.

QUESTION :
In the following Cloudflare Worker Code to List by Prefix .. how will the billing happen ? .. billing only for the Keys which are read and correspond to prefix or billing for complete scan of the KV namespace ?

export default {
async fetch(request, env, ctx) {
const value = await env.NAMESPACE.list({ prefix: "user:1:" });
return new Response(value.keys);
},
};

ANSWER by CHATGPT :
In Cloudflare Workers, billing for KV operations depends on the specific API calls and the data retrieved. Here's how the billing works for your provided code:

KV List API Billing

The KV.list()KV.list() operation is billed based on the number of keys returned that match the prefixprefix filter, not the total number of keys in the namespace.
The filtering by prefixprefix is handled internally by Cloudflare, so the operation scans only the relevant part of the namespace that matches the specified prefixprefix.

### 2. Breakdown of Costs

List Operation: Each call to list()list() counts as a single read operation, regardless of the number of keys returned.
Key Retrieval: You are billed for each key returned in the value.keysvalue.keys array. If the list()list() operation retrieves 10 keys, it will count as 10 read operations.

### 3. Example Billing Scenario

Assume NAMESPACE.list({ prefix: "user:1:" })NAMESPACE.list({ prefix: "user:1:" }) returns 50 keys.
You are billed for:
- 1 list operation (base cost for calling the list()list() method).
- 50 read operations (one for each key returned in the result).

### 4. What Is NOT Charged

You are not billed for scanning or evaluating keys that do not match the prefixprefix. The filtering happens on Cloudflare's side, and only matching keys are returned and billed.

This means you are billed based on the keys matching your prefix and retrieved by the list()list() operation, not for scanning the entire namespace.

Mathis @ blaxel.ai•1/23/25, 10:12 PM

Hello, does the expiration set at the key creation persists if I put on the key without specifying expiration?

Mathis @ blaxel.ai•1/23/25, 10:14 PM

E.g in this code:

await env.NAMESPACE.put(key, value, {
  expiration: secondsSinceEpoch,
});

await env.NAMESPACE.put(key, value);

await env.NAMESPACE.put(key, value, {
  expiration: secondsSinceEpoch,
});

await env.NAMESPACE.put(key, value);

MMathis @ blaxel.ai Hello, does the expiration set at the key creation persists if I put on the key ...

se•1/24/25, 2:42 PM

No, it does not inherit the expiration settings of the previous overwritten object.

�

😈 Donkey 💫•1/28/25, 1:00 PM

Requests from around the world, pretty higher than expected

iLearn•1/28/25, 10:25 PM

QQ : Can "Storage" for "Workers KV per Namespace" store 200GB of data ? .. Limits section says "Unlimited", but still want this to be confirmed .

thomasgauvin•1/31/25, 3:56 PM

Hey folks. We're designing a new feature for Workers KV to make it easier to retrieve multiple kv pairs at once. I'd love to get some feedback on some early designs. If you're up for it, snag 15 minutes on my calendar and I'll share with you some of the things we're thinking (or DM me) https://calendar.app.google/ABB6TtNyg7jPg82j8

Appointments

Tthomasgauvin Hey folks. We're designing a new feature for Workers KV to make it easier to ret...

森

森@VerseLink (RW)•1/31/25, 5:44 PM

Scheduled! First time doing this, would it be on discord?

jd•2/3/25, 7:40 PM

hi qq -- im trying to convert a largely kv-based lookup metadata of ~7GB in size of json and gets 10+ GB in size later with parallel reads, updates. Metadata needs persistence storage for consistency and availability too.

My metadata has 10M+ entries where each entry is a small dict of mostly nested strings of a loose schema, mixed with floats.

Is this size a good use case for cloudflare kv? thanks!

alotabits•2/4/25, 2:35 PM

Sorry for the cross-post. I asked this in workers before I knew about this channel.

Are values written to KV automatically compressed for storage?

iLearn•2/4/25, 9:25 PM

how much is "expirationTtl (option in Workers-KV)" reliable ? .. i mean, if ttl is set for 10-minutes, can i expect the key to deleted after 10-minutes ? or what error margin need to be taken in consideration ?

RRob We will not serve an object after expiration/expirationTtl has passed. This is a...

iLearn•2/5/25, 4:57 PM

Brilliant. Thanks for confirmation.

RRob We will not serve an object after expiration/expirationTtl has passed. This is a...

iLearn•2/5/25, 5:00 PM

QQ : if i use get(my_key) after expiration/expirationTTL .. i shall be getting "null" / "undefined" as returned value ?

IiLearn QQ : if i use get(my_key) after expiration/expirationTTL .. i shall be getting "...

1984 Ford Laser•2/6/25, 4:21 AM

It will be

null

null

cosbgn•2/7/25, 2:58 PM

What's the difference between expirationTtl vs cachettl ?

cosbgn•2/7/25, 2:59 PM

If I use for example: expirationTtl: 21600 expirationTtl: 21600 do I also need to use cacheTtl?

cosbgn•2/7/25, 2:59 PM

or it will be cached by default?

Ccosbgn What's the difference between expirationTtl vs cachettl ?

Chaika•2/7/25, 3:09 PM

expirationTtl: after this time, the key (and thus its value) will be gone forever
cacheTtl: Cache it x long, to reduce cold requests to KV's Central Stores, but of course also impacts how quickly you'll see updates if you do delete/update it

Ccosbgn or it will be cached by default?

Chaika•2/7/25, 3:09 PM

cacheTtl is 60 seconds minimum. If you don't set it, it'll be 60s.

cosbgn•2/7/25, 3:14 PM

If I have a key which won't change till I manually delete it, I normally use expirationTtl: 21600expirationTtl: 21600 so I know that even if something goes wrong it will expire in 6 hours.

I guess it would help a lot to also add cacheTtl: 21600, right?

cache will be delete when I delete the key manually?

cosbgn•2/7/25, 3:19 PM

For example I do:

const [key, value] = ["something", "whatever"]

binding.put(key, value, { expirationTtl: 21600 })

Then after 5 minutes I do: 

binding.get(key, {cacheTtl:21600})

Then after 5 minutes I do:

binding.delete(key)

Then after 5 minutes I do:

binding.get(key, {cacheTtl:21600}) 
Will this last one return null or the cached value?

const [key, value] = ["something", "whatever"]

binding.put(key, value, { expirationTtl: 21600 })

Then after 5 minutes I do: 

binding.get(key, {cacheTtl:21600})

Then after 5 minutes I do:

binding.delete(key)

Then after 5 minutes I do:

binding.get(key, {cacheTtl:21600}) 
Will this last one return null or the cached value?

pandalambda2•2/9/25, 12:31 AM

it isn't clear to me, that if cacheTtl is absent what happens - is the default min of 60 seconds applied, or is it no cache what so ever?

Walshy•2/9/25, 12:38 AM

default 60s yeah

FFlikTeoh Was the KV tutorial page always this empty? https://developers.cloudflare.com/kv...

James•2/15/25, 3:14 PM

That looks like a bug. You may want to file an issue at https://github.com/cloudflare/cloudflare-docs/issues

Constantin•2/16/25, 10:08 PM

hey there, any wrapper around KV to easly query thinks?

James•2/16/25, 10:15 PM

What kind of wrapper are you looking for? The KV bindings in Workers work really well, for example to read values: https://developers.cloudflare.com/kv/api/read-key-value-pairs/

James•2/16/25, 10:15 PM

Was there an issue you hit or some use-case or workflow where you think a wrapper for something would be helpful?

Constantin•2/16/25, 10:30 PM

well, i'm looking for a more orm like .. thing; i'm using KV to store objects, that have some relations stored via metadata

Constantin•2/16/25, 10:31 PM

i was just wondered if there's something out there that simplifies this, by adding some syntax sugar

Constantin•2/16/25, 10:33 PM

but probably using prefix, would make more sense

matt•2/18/25, 1:10 AM

would KV be a good replacement for upstash redis?

matt•2/18/25, 1:11 AM

low-latency batch reads (eg. HMGET) would be a requirement

matt•2/18/25, 1:11 AM

fast global replication also a must, i remember a couple of years ago reading that KV was up to 30 seconds, but there was a performance release a few months ago right?

matt•2/18/25, 1:39 AM

oh boyo, that 30 seconds is now 60 seconds :\

matt•2/18/25, 1:40 AM

what is cloudflare kv used for? I thought it was intended to be a redis-like db, but the performance profile is too detached from what redis offers

Mmatt fast global replication also a must, i remember a couple of years ago reading th...

Isaac McFadyen•2/18/25, 2:33 AM

It's always been 60 seconds except for a short time during a rollout that ended up being rolled back.

Isaac McFadyen•2/18/25, 2:33 AM

KV is primarily designed for read-heavy workloads that rarely write.

Isaac McFadyen•2/18/25, 2:34 AM

For example, KV backs Cloudflare Pages and Workers Assets, where you deploy a site infrequently but people access often and quickly.

Isaac McFadyen•2/18/25, 2:35 AM

KV caches at your nearest colo (that's why the 60 second limit), and future reads during the time it's cached will be very low-latency.

Mmatt low-latency batch reads (eg. HMGET) would be a requirement

Chaika•2/18/25, 3:12 AM

not a thing currently but they talked about doing stuff for it recently: kv

Mmatt fast global replication also a must, i remember a couple of years ago reading th...

Chaika•2/18/25, 3:14 AM

worth keeping in mind along with what Isaac said, KV doesn't replicate as much as it does cache: https://developers.cloudflare.com/kv/concepts/how-kv-works/
Two Central stores, US and EU, all the colos/data reads are behind cache and cacheTtl. It's not actively pushing, they're pulling

Cloudflare Docs

How KV works · Cloudflare Workers KV

KV is a global, low-latency, key-value data store. It stores data in a small number of centralized data centers, then caches that data in Cloudflare's data centers after access.

nukely•2/24/25, 6:03 PM

Which is faster and has lower latency for random reads/writes: KV or D1?

Nnukely Which is faster and has lower latency for random reads/writes: KV or D1?

Isaac McFadyen•2/24/25, 6:20 PM

Likely KV because it's cached much more heavily than d1 (D1 is single-location), but updates may take up to 60s to be visible globally.

o7•2/28/25, 7:08 AM

I racked up 700K (free plan) delete operations testing worst case cleanup operations, then 1.27M with a paid plan.
The paid plan includes 1M deletes, so will I be charged fractionally for 270k deletes? Or rounded up to $5?

I really want to use kv, but the cold read latency is kinda killing me. For context, I am starting a

KV List API Billing

Similar Threads

Similar Threads

KV List API Billing

Similar Threads