Clickhouse: How to add Graph Computing Engine on the top of ClickHouse?

[theseusyang]

Hi team,
I have some requirements about Graph Computing, and want to use ClickHouse as an underlying Graph DataStore. ClickHouse stored many vertex and edge information, and execute Computing using Graph Engine.

the problem is that how to integrate with a Graph Computing Engine and ClickHouse?

[alexey-milovidov]

It's dependent on what kind of graph computations do you want to execute in ClickHouse.
Let's start with any example of real task and we will look how it will be possible to adapt ClickHouse for it.

[theseusyang]

@alexey-milovidov Yes, we want to do some graph query and graph analysis like Titan or Neo4j, and also especially like enterprise product GraphSQL , a table with 2 paradigms , graph and relational database.

[alexey-milovidov]

It's difficult to integrate complex graph processing directly into ClickHouse,
because ClickHouse gains its performance advantages from flat, uniform, well structured data.

I see only few possibilities:

• using ClickHouse as a storage layer for large graphs, as you can quickly fetch data from ClickHouse based on some simple condition, and perform further analysis in other engine;
• using ClickHouse for analytics on "shingles" (subpaths of N nodes in a graph) extracted from graphs: some tasks can be solved from this data model instead of real graph.

[theseusyang]

@alexey-milovidov Yes, We are researching how to use ClickHouse as an graph storage layer, maybe We need extend the underlying Storage API to complete it.

[cesarmarinhorj]

@theseusyang is possible doing some simple experiments with graph with clickhouse? It helps in situations where there are too many edges in same node? Thanks.

[theseusyang]

yes, It is feasible that clickhouse is begarded as a backend storage engine for distributed graph database.