Cloudflare Developers•3y ago

perfect, re performance is the same for namespace filtering and metadata filtering ?

Sscotto perfect, re performance is the same for namespace filtering and metadata filteri...

Jerome•12/7/23, 12:46 PM

no, namespace filtering is generally faster than metadata filtering

Old Man Maker•12/7/23, 4:02 PM

refreshing my memory here - we're still looking at 1000 namespaces max per account?

OOld Man Maker refreshing my memory here - we're still looking at 1000 namespaces max per accou...

Jerome•12/7/23, 4:24 PM

yep; https://developers.cloudflare.com/vectorize/platform/limits/

OOld Man Maker refreshing my memory here - we're still looking at 1000 namespaces max per accou...

Matt Silverlock•12/7/23, 7:37 PM

Per index

Frank•12/7/23, 8:01 PM

whats the expected error when we exceed # of vec, namespace per index

Frank•12/7/23, 10:59 PM

hopfully the hard limit on namespace will be lifted post beta

채

채찬•12/8/23, 2:55 AM

After connecting to Vectorize, there is no results data when requesting a D1 query.
Previously, it worked fine.
In the dashboard, the number of query requests is increasing.
Probably a bug.

MMatt Silverlock Per _index_

Old Man Maker•12/8/23, 4:58 AM

Yep, the per index part is important

BBhanu Teja P Awesome. How to request an increase in limits. Right now, it shows 200k vectors ...

Vy•12/8/23, 3:29 PM

With open beta product today, we can't support that large of an index. However, it's something we're working on! Are you often (1) searching across all 5M vectors or (2) search space is smaller but used to operating a single index?

huw•12/11/23, 11:51 PM

Hello! Is there a limit on the maximum number of IDs for getByIdsgetByIds and deleteByIdsdeleteByIds? I am scheduling a bulk delete and want to know if I should chunk it, or if that limit is well above the memory size of a Worker. (EDIT for searchers: Limit is 20)

Workers API · Vectorize

This page covers the Vectorize API available within Cloudflare Workers, including usage examples.

BBhanu Teja P We have a single index and the vectors of all my customers would go to that sing...

yevgen•12/12/23, 11:08 AM

Hi Bhanu, Eng manager here. Based on the info you provided I suggest splitting your 5M vectors index into smaller indexes per customer id. Let me know if 100 indexes per account is sufficient for you (we can bump this limit if necessary). You will get much better performance searching in smaller indexes. We are working hard to bump vectors per index limit, we don't have ETA for this to be ready though.

BBhanu Teja P I will rather wait for sometime until Vectorize becomes stable enough to handle ...

yevgen•12/12/23, 11:36 AM

fair enough

Yyevgen Hi Bhanu, Eng manager here. Based on the info you provided I suggest splitting y...

huw•12/12/23, 11:15 PM

This would also be more reasonable if there was a way to programatically create Workers bindings for indices, but you have to use

wrangler.toml

wrangler.toml

right now which hard-codes your limit

huw•12/12/23, 11:16 PM

I’m not sure about the specifics—you can use API calls to create and query indices dynamically, sure, but aren’t those API calls slower?

Hhuw Hello! Is there a limit on the maximum number of IDs for [`getByIds` and `delete...

huw•12/13/23, 9:41 AM

(I have found out that this limit is 20)

divby0•12/13/23, 12:19 PM

Are pages bindings here yet? The API doesn't support delete so I can't really use it. Gotta wait for pages bindings :/

scottoOP•12/13/23, 8:32 PM

hey - we started to get a strange 502 error from search API -

Ccfnathan checking into this for you 👀

scottoOP•12/13/23, 8:45 PM

https://api.cloudflare.com/client/v4/accounts/920b1a6e159cf77dab28969103a4765b/vectorize/indexes/customers-650a02c2a5be79517a7f79e2/query

https://api.cloudflare.com/client/v4/accounts/920b1a6e159cf77dab28969103a4765b/vectorize/indexes/customers-650a02c2a5be79517a7f79e2/query

xeon06•12/14/23, 12:48 AM

Hey folks, love the work on the AI stuff! Just wanted to drop in to chime in that I would love it for the limits on Vectorize index to be way higher, to the tune of a couple of indices per user of my app. That's all!

Xxeon06 Hey folks, love the work on the AI stuff! Just wanted to drop in to chime in tha...

Jerome•12/14/23, 9:46 AM

for survey purposes, what limit would fit your use case?

JJerome for survey purposes, what limit would fit your use case?

xeon06•12/14/23, 8:56 PM

I don't want to give a number because it would scale with the number of users my app has, e.g. I need to be able to make vector searches against documents belonging only to a single user.

Even if it was 10,000, then what do I do at 10k users!

xeon06•12/14/23, 9:00 PM

There's a world where I can filter post search, but that doesn't seem inefficient and probably severely limited by the topKtopK limit

Xxeon06 There's a world where I can filter post search, but that doesn't seem inefficien...

huw•12/14/23, 11:59 PM

The in between here is namespaces, right now if you maximally used indices & namespaces you could support 100k users

huw•12/15/23, 12:00 AM

tbh it’s not really clear whether a user-segmented app is supposed to use separate indices or separate namespaces right now—only except that namespaces have the advantage of using proper service bindings but if you go for nn indices you’d have to use the API

xeon06•12/15/23, 1:13 AM

Oh wow somehow I completely missed namespaces

xeon06•12/15/23, 1:14 AM

Yeah I suppose without creating new indices with the API in order to "bin" the namespaces you're limited at 1000

xeon06•12/15/23, 1:14 AM

But I guess my request now is more along the lines of increasing the namespace limit

divby0•12/15/23, 2:06 PM

I am getting "error code: 502" errors for the last few minutes on two different vectorize indices

divby0•12/15/23, 2:07 PM

The body literally only says "error code: 502" if I see this right. I am doing a normal request (that used to work a few days ago) on /query. It uses topK: 2 with a namespace filter and returnMetadata: true

divby0•12/15/23, 2:08 PM

This happens on both of my indices, if a cloudflare staff can DM me, I can provide the specific indices maybe there's more info on the cause on your isde

JJerome @cosbgn vector deletion on the REST api is on the backlog. Upsert will ship soon...

dbk•12/19/23, 2:15 PM

Greetings! Are there any hints on when we can expect vector DELETE and GET to be available through the REST api?

Ddbk Greetings! Are there any hints on when we can expect vector DELETE and GET to be...

Vy•12/19/23, 3:46 PM

we're working on vector DELETE. For GET, is that get vector by id?

dbk•12/19/23, 3:47 PM

Hi @Vy, yes, that'd be what I'm interested in

Ddbk Hi @Vy, yes, that'd be what I'm interested in

Vy•12/19/23, 11:56 PM

thanks, we should be able to tackle GET same time as DELETE

natanparmigiano•12/20/23, 11:59 PM

Hey all!
I have a question regarding billing, if using metadata filters, would total queried vector dimensions be calculated towards the entire index or only the vectors matching the metadata filter?

natanparmigiano•12/21/23, 12:00 AM

I am currently studying the use case of partitioning user documents using filters, but don't want to be billed on a search through all vectors for a question regarding a single document

natanparmigiano•12/21/23, 1:23 AM

Sorry everyone, dumb question, it seems I misunderstood the pricing, a query is billed as 1 queried vector regardless of how many vectors are in the index, right? Now it makes sense why my estimations were coming out so absurd

Nnatanparmigiano Sorry everyone, dumb question, it seems I misunderstood the pricing, a query is ...

Vy•12/21/23, 4:21 PM

no, number of vectors in your index does impact billing since that's the total search space for a query. See https://developers.cloudflare.com/vectorize/platform/pricing/#calculating-vector-dimensions for some examples.

Pricing · Vectorize

Vectorize bills based on:

natanparmigiano•12/21/23, 4:25 PM

But that would affect the "stored vectors" cost, correct?
My understanding is that the "per query" cost is not affected by the stored vectors count (as in, if I have 100000 vectors stored, the cost per query is the same as if I have 200000, but the stored vectors bill would increase)

Is my understanding correct in that sense?

Nnatanparmigiano But that would affect the "stored vectors" cost, correct? My understanding is th...

Matt Silverlock•12/21/23, 4:47 PM

Nnatanparmigiano But that would affect the "stored vectors" cost, correct? My understanding is th...

Matt Silverlock•12/21/23, 4:48 PM

If this doesn't make sense, help us understand where you're stuck. There's also some examples further down:

Matt Silverlock•12/21/23, 4:49 PM

Put another way:
- If you store 10k vectors of 1536 dimension each and make ZERO queries, it's (10000)*1536*(0.040/1000000)(10000)*1536*(0.040/1000000) = $0.6144 for the month
- If you query that index 50k times that month, it's (10000+50000)*1536*(0.040/1000000)(10000+50000)*1536*(0.040/1000000) = $3.68. The query component is $3.072. You just add them up.

natanparmigiano•12/21/23, 4:56 PM

In the end it does make sense!
The docs are quite clear, I just got it wrong at first, since the provided formula includes both the query and storage costs, I at first assumed that index size would affect per query costs

I think it would be clearer to newcomers if the pricing examples had the cost components (stored and queried vectors) shown (and calculated) separately.

perfect, re performance is the same for namespace filtering and metadata filtering ?

Similar Threads