[A-F1V3]

(Mux co-founder here) The Postgres system described as the predecessor to Clickhouse here actually was cstore_fdw with a pretty heavily customized (and outdated ) CitusDB. CitusDB is a great system for sharding and distributing postgres, but we found that the compression and query performance for this specific analytics use case was much better served by Clickhouse.

[caust1c]

Cloudflare also moved from Citus to Clickhouse

[tomnipotent]

The only place Citus is mentioned in the post is in the title of a chart. Might be worth updating to include this, would also clarify note #3:

"ClickHouse is still faster than the sharded Postgres setup at retrieving a single row, despite being column-oriented and using sparse indices. I'll describe how we optimized this query in a moment."

[mmcclure]

Not A-F1V3, but also a Mux co-founder.

To be transparent, we included Citus in original versions of the post, but decided to take it out because it didn't feel like it was a fair representation of Citus™. As Adam mentioned, what we were using was heavily customized and based on a pretty outdated fork by the time we transitioned.

[tomnipotent]

This should be clarified in the post, otherwise you lead the reader to believe you're representing stock Postgres performance instead of the performance of a forked Citus. This is important information for an informed reader.