Dgraph x YugaByte - What's the best implementation way? Event broker?

[kokoskiwi]

Hi, I got a question: what's the best design?

Example: TikTok. It's not only a social network, but with tiktok shop also an ecommerce platform now.
I want to build something similar 2-in-1.

I want to host with Hetzner in US, EU, SG.
I will host Yugabyte partitioned by geo_partition https://docs.yugabyte.com/stable/explore/multi-region-deployments/row-level-geo-partitioning/#:~:text=Geo%2Dpartitioning%20allows%20you%20to,with%20regulations%20such%20as%20GDPR

I guess I can't host dgraph in the same way by geo_partition? so i was thinking of hosting it HA&AZ in the US, and have replicas in EU SG.

https://chatgpt.com/share/6959a118-4734-8007-9570-e0a21e420683
https://www.perplexity.ai/search/3e2d3f56-ff63-46cb-a12a-de7a77ec77a1

Yugabyte is the source of truth + has the data
Dgraph is on top of it and models relationships + recommendations, ratings, calculated trust scores, algorithm scores, etc

JOINS are done via Dgraph, and then i use simple queries to get the data from yugabyte

so basically

Yugabyte User table:
userid
username
geo_partition
email
language
registration date
posts[]
etc

Yugabyte posts table:
postid
title
description
comments[]
date

Dgraph User table:
userid
username
geo_partition
trust_score
interests
friends

Dgraph posts table:
postid
likes
dislikes
rating

Postgre stores transactional, large (text posts, comments etc) data, payments, money

dgraph stores relationships + metadata ratings

I think that‘s the best way to build an app in 2026, especially a geo distributed one that shall run fast around the world (thanks to geo partitioning of yugabyte and easy replication of dgraph )

//

thanks to all the great guides online https://docs.dgraph.io/installation/ https://community.hetzner.com/tutorials/how-to-setup-a-yugabytedb-cluster it's very easy to self-host

What I try to find out now: the best way to sync dgraph from yugabyte.

E.g user makes a new blogpost. Shall I write it to Yugabyte and dgraph at the same time, and that's it?[seems like a bad idea ofc] or should i write it to yugabyte, onSuccess return to the user, and fire a write to dgraph (or an event to a message broker and then it gets added to dgraph?)

https://chatgpt.com/share/6959a4e4-69d4-8007-9cf8-251e235ffc83
ChatGPT told me: write only to Yugabyte and have an "event" table

Keep it small and practical:
	•	event_id (uuid)
	•	event_type (PostCreated, UserUpdated, LikeAdded, FollowAdded…)
	•	aggregate_type / aggregate_id (post/user id)
	•	payload (json, optional)
	•	created_at
	•	attempts, next_retry_at, status 

Is this the right thing to do? or maybe i should really use RabbitMQ? i'm willing to do this "extra step" to make it right.

How could I implement this in reality now? should i really implement RabbitMQ? https://chatgpt.com/share/6959ab54-2fcc-8007-8c18-05ef98a87361 ChatGPT gave me some ideas, but i thought i should i ask here on GitHub some real developers real humans with experience what I should do in my situation

Is RabbitMQ the right choice?

[kokoskiwi]

EDIT: I also want to pump data into typesense.org for fulltext search
So maybe RabbitMQ is the best case after all?

[matthewmcneely]

Seems like you're describing a new/prototype product. My advice: don't prematurely optimize it. Your schema is pretty simple, and the only "graph-y" looking part is the User.friends field. Why not just use YugaByte for everything in the beginning? If you find later that managing the friends field relationally becomes a bottleneck or that the full text search in YugaByte is not powerful enough, then and only then, consider bringing in new data management components such as Dgraph, etc.