Using the paid Typesense (cloud) database, is ther...
# community-help
s
Using the paid Typesense (cloud) database, is there some sort of "duplicate database" feature? What i need is a fast way to create a "staging" env, currently our staging env is connected to nothing or live vector, which is just bad
f
You can clone your cluster by going to your cluster's Cluster Configuration page and clicking the Clone button:
s
Thanks. Does that create a new, separate instance(s)? Are those added to billing? Thanks!
Oh, and how long does that take usually? Is it as expensive as upserting? Or is it something like a copy operation behind the scenes done on raw data?
j
It does create a new cluster, and will be billed separately / in-addition to the source cluster
Time it takes to clone depends on the size of your dataset. We still have to re-index the data into memory on the new cluster after copying it. So it won't be faster than doing upserts on a new cluster - it just saves you the hassle of doing it manually yourself.
s
After the clone, I get `Unhandled Promise Rejection: Error: 404 - Could not find a field named
embedding
in the schema.` even thou it is there (I mean, on the "source" from which I cloned, clearly we have a embedding)
j
Could you share a curl request of the search that throws this error (minus your hostname and api key)?
s
This happens on simple load of the search page, so even before I perform any filtering. It is the 1:1 same code as on "live", just that I changed the nearest node and API keys, and I have no alternative nodes
Copy code
{
  "results": [
    {
      "code": 404,
      "error": "Could not find a field named `embedding` in the schema."
    }
  ]
}
j
multi_search requests should also have a POST body
s
Copy code
{
  "searches": [
    {
      "query_by": "Content,Transcript,Episode,Title,embedding",
      "exclude_fields": "embedding,Transcript,Summary,Topic,Guest,Channel,Author,Category,Tag,Access,Episode,Date",
      "prefix": false,
      "stopwords": "stopword_set1",
      "sort_by": "_text_match(buckets: 10):desc,Date:desc",
      "highlight_full_fields": "Content,Transcript,Episode,Title,embedding",
      "collection": "pp_posts",
      "q": "*",
      "facet_by": "Date,Access,Topic,Guest,Category,Author",
      "max_facet_values": 1000,
      "page": 1,
      "per_page": 15
    }
  ]
}
This is 1:1 the same as is used on live site, the code did not change. The payload is the same on live, just that there it gets a response since embedding is found, while on clone, it is not.
j
When the cloned dataset was being re-indexed, looks like the embedding model was not accessible (we host them on hugging face), so the embedding field wasn't loaded into memory because of that
Just fixed it on your cluster.
Could you try the same query again
s
It works now
j
(I've also deleted your hostname and API key above, so it isn't indexed publicly in the archive)
I vaguely remember us improving model download in recent versions
s
Oh thanks It is search api key so anyone can grab it anyway, but thanks!
j
So I'd recommend upgrading to a more recent version
s
Perhaps you can delete my comments? So the website is not there etc (sorry, did not think before I posted)
So I'd recommend upgrading to a more recent version
You mean typesense itself?
j
Yup
s
Yes will do by next chance
j
You can do that by going to Cluster Configuration > Modify
🙏 1
Also deleted your message with the site
🙏 1