Clera, an AI talent agent, says it moved more than 500 database tables from Europe to the US overnight while production traffic continued, then saw typical CPU use fall from 100% to 10–20%. The account offers a detailed, if vendor-hosted, look at how one company tested and migrated to ClickHouse Managed Postgres.
ClickHouse Watch analysis
What happened
Clera says its previous database struggled when analytics queries and bursts of scraped data competed with routine lookups. The company reports that some candidate searches took about three seconds and that CPU use once remained at 100% for six days.
In a guest post on ClickHouse’s site, Clera’s co-founders describe comparing Supabase, Neon, PlanetScale and ClickHouse Managed Postgres. They tested around 500GB of production data under simulated workloads, then ran sysbench TPC-C tests. Clera says ClickHouse Managed Postgres led most of its production-workload comparisons and delivered 16–42% more throughput than PlanetScale in the benchmark. The test was not fully TPC-C-compliant, and the teams did not use identical default PostgreSQL settings.
The team says it moved more than 500 tables overnight using ClickPipes, with production still running, and had the migration mostly complete by morning. It reports typical CPU use of 10–20% afterwards, without changing settings.
Why it matters
This is a useful case study in a very specific problem: keeping routine database lookups responsive while analytics and data-ingestion work pile on. Clera says the move left it with more headroom, while letting it keep using managed Postgres rather than switching its operational database to ClickHouse’s analytics database.
The numbers are the customer’s account, not an independent comparison of database providers. The testing setup also matters: the production workload was replayed using direct connections, and the standardised benchmark had stated limitations. Still, Clera provides enough detail for technical readers to see what it tested and what it says changed.
Our read
The most interesting result is not a tidy benchmark victory; it is the reported overnight migration and the reduction in CPU use on Clera’s own workload. Teams weighing managed Postgres options should treat the post as a reason to run their own representative tests, not as a substitute for them. Database decisions rarely improve when a chart is asked to do all the thinking.
What to watch
- Whether Clera publishes follow-up results as its data grows beyond the roughly 500GB used in testing.
- Whether other customers report similar outcomes on their own workloads.
- Whether the open-source benchmark materials make the comparisons straightforward to reproduce.
Discussion spark: When choosing a managed database, should a detailed customer workload test carry more weight than standardised benchmarks, or is that too easy to tailor to one vendor?
Sources and evidence
- Source update (8 October 2026, 20:51 UTC)
not affiliated with or endorsed by ClickHouse