Realistic query performance

[ahctangU (ahctang)]

I am trying to evaluate dgraph for use in a new project for my company.

However, I ran into a major issue with regards to query performance.

Right now I am trying doing a text search using the term and text indexers on almost 5M nodes.

At the moment, the term search latency varies wildly between sub 50ms to 10s depending on the length of the string and uniqueness of the terms. The text search seems to be faster (usually under 1s) but produces fewer useful results.

What kind of performance is realistic so I know if I’m doing anything wrong?

[mrjn (Manish R Jain)]

Generally expect results in milliseconds.

[ahctangU (ahctang)]

Thank you for your response.

In that case, I don’t understand why my queries are so slow.

I am running 3 nodes on GKE with n1-standard-4 nodes. I am using the HA deployment with 3 Zeroes and 3 Alphas.

This is my query.
query Me($name: string){
me(func: allofterms(name@ja, $name), first: 10) {
uid
name@ja:en:.
nameKana
}
}

Name is indexed with term, trigram, and fulltext. There are approximately 5M nodes with name@ja.

However, depending on the name the query time can vary from 50ms to 10 seconds. Also under load (around 50rps) this index query time quickly rises to over 30s.

Am I missing something?

[ashishgoswami (Ashish)]

Hey @ahctangU, thanks for reaching to us with more information related to query and deployment setup. We would like to look more into the issue. Can you share your testing data with us? We think it might be because of the lang tag specified in name predicate.

[ahctangU (ahctang)]

Hello,

I don’t mind sharing, but how do I share it with you?

[ashishgoswami (Ashish)]

Hey @ahctangU, you can share it via email. My mail is ashish@dgraph.io