Benchmarks
Independent Supabase ETL performance results, tested versions, and plans for future benchmarks.
Independent benchmark
Hardbyte's independent CDC benchmarks compare six configurations across steady-state delivery, mixed-speed consumers, and consumer outages.
The table shows selected results from separate benchmark scenarios, with all six configurations included. Each metric heading links to its source table. Bold numbers mark the lowest reported value in each column, including ties at the published precision.
| CDC configuration | Fast-consumer tail latency | Fast-consumer memory | Mixed-speed tail latency | Source slot WAL during outage |
|---|---|---|---|---|
| Supabase ETL | 26 ms | 13.0 MB | 273 ms | 57.2 MB |
| Debezium + Kafka | 961 ms | 1,385.4 MB | 1,074 ms | 11.0 MB |
| Debezium Server | 848 ms | 1,629.3 MB | 1,181 ms | 57.3 MB |
| Raw pgoutput (SQL polling) | 70 ms | 59.7 MB | 267 ms | 56.6 MB |
| Sequin | 33 ms | 595.2 MB | 33 ms | 642.0 MB |
| Sequin with message grouping | 33 ms | 624.3 MB | 33 ms | 679.9 MB |
For methodology, tested versions, configurations, and complete results, see Hardbyte's benchmark repository.
Performance differences may also reflect configuration, workload, and deployment choices, rather than the CDC tools alone.
Our own benchmarks
We plan to run and publish our own benchmarks as Supabase ETL matures and product stability improves.