Agreed. This is probably the trickiest part of building managed Postgres on local NVMe. Postgres already gives you a lot of the building blocks: physical streaming replication, synchronous replication, WAL archiving/PITR with backup toold, etc. The hard part is making failovers, recovery, and node replacements seamless.
We’ve spent a fair bit of time on this at ClickHouse through our operational expertise huilding Postgres managed services multiple times. We’re fortunate to have deep internal and partner Postgres expertise, folks who originally built WAL-G back in the Citus Data days and are active contributors now and solid partnership with Ubicloud.
It’s definitely more engineering work, but the performance/scalability gains make it worth it. Relaying this based on 100s of customers who migrated to ClickHouse Managed Postgres - new, powerful world surpassing previous expectations.
We’ve spent a fair bit of time on this at ClickHouse through our operational expertise huilding Postgres managed services multiple times. We’re fortunate to have deep internal and partner Postgres expertise, folks who originally built WAL-G back in the Citus Data days and are active contributors now and solid partnership with Ubicloud.
It’s definitely more engineering work, but the performance/scalability gains make it worth it. Relaying this based on 100s of customers who migrated to ClickHouse Managed Postgres - new, powerful world surpassing previous expectations.