[Feature Request] ClickHouse backend support

[marswong]

Traditional RDBMS is really slow at large-scale OLAP, it's time to speed up OLAP with a better database and ClickHouse is a good choice today.

[mikecao]

We are limited to what is supported by Prisma. Adding any other database would take significant work.

[Form1ca]

We are limited to what is supported by Prisma. Adding any other database would take significant work.

i was thinking about using clickhouse.
if you are interested i can add support clickhouse

Also have to use something for the cache - for example rabbitmq

[Kdecherf]

Should ClickHouse support be added to umami, please do not make it a requirement as I turned away Plausible because of that 🙏

[Form1ca]

Should ClickHouse support be added to umami, please do not make it a requirement as I turned away Plausible because of that 🙏

what problem you get in Plausible and clickhouse?

[Kdecherf]

@Form1ca I'm using umami for a small blog on a server with restrained resources. When I tested Plausible, ClickHouse consumed too much resources for my needs, backup was not fun at all and in overall it was overkill for my workload. I don't find this stack relevant for monosite/small sites.

[Daniel15]

Did you modify the Clickhouse config to turn off logging? That's what caused a lot of the resource consumption in older versions. Plausible now ships with a default config to disable all the logging.

[entrptaher]

This is probably possible with clickhouse/postgres docker for fun projects. https://github.com/wavy/postgres-clickhouse

[Daniel15]

A column-based database like Clickhouse makes way more sense for analytics data compared to a row-based database, since practically all queries are aggregations of individual columns. You don't actually care about loading individual rows. OLAP-ish queries are much faster with a columnar database.

It doesn't have to be Clickhouse; that's just the most common one for this use case at the moment.

I agree with the other comments that it should be optional since very small deployments might not want to use it.