We released Neki in platform preview yesterday. To celebrate, we tested running 1 million queries per second on Neki. We hit this goal quickly on 5 shards and pushed further. The next run ended with 512 shards running 118 million queries per second with 1.22 PiB of data.

The benchmark was simple: a single-shard point select, one row per query by primary key. No writes, joins, or cross-shard queries. Each shard's workload was isolated. Our target was 200k QPS per shard, then grow the cluster. Five shards, then fifty, then 512. Ten times the shards, ten times the throughput. At 512, shards still had headroom, so each settled at 231k QPS.

We sustained 118,538,803 QPS for 16 minutes across 512 shards and 1.22 PiB of data. Largest recording: 118,747,267. 512 shards, each with one Postgres primary on an r8g.16xlarge; 480 Neki routers on 8xlarge instances; p99 latency 6.06ms at router, 13.95ms at client; 67 errors/sec (~1 in 1.8 million); 15.8M read IOPS; over 2 Tb/s network.

Worth noting: shards were primary-only, no replicas; read-only workload; no failover during the window. We'll detail the engineering in a future article.