June kept the Readyset team moving. Between Web Summit Rio, new technical content, and a round of improvements to caching performance and reliability under load, here's what happened last month.
We spent June 8-11 at Web Summit Rio, and on June 10 Gautam took the stage with a talk on a question more teams are running into: AI Is Querying Your Database. Are You In Control? The talk explored how AI is changing the way applications interact with data, why AI-generated queries create a fundamentally different database workload, and why scaling databases is becoming as much a control problem as a performance problem.
We had great conversations with engineering teams thinking through AI, database performance, and infrastructure costs. If we missed you there, you can book time with our team here.
- Top K Queries On by Default: Support for Top K queries (
ORDER BY with LIMIT) is now enabled by default, so eligible queries can be accelerated out of the box. You can opt out per process with --feature-topk=false.
- More Reliable Shallow Cache Coalescence: When many requests wait on the same shallow-cache fill, a failed upstream query no longer makes waiting requests sit out the full timeout, and slower upstream queries can still benefit from request coalescing. The result is lower latency and fewer stalls during cache-miss bursts.
- Stronger PostgreSQL Replication Under Load: Fixed a rare dropped write that could occur in PostgreSQL replication when group commit coalesced transactions under heavy write load, tightening correctness for high-throughput Postgres deployments.
- Stricter Empty Array Handling: Bare empty arrays (
ARRAY[]) are now correctly rejected during query lowering instead of being silently accepted, matching Postgres behavior. Empty arrays with an explicit cast (ARRAY[]::INT[]) remain valid. Worth noting on upgrade: cached queries that previously succeeded with a bare empty array will now return a type error, so check for that pattern.
- New JSON Array Builders: Added support for PostgreSQL's
json_build_array and jsonb_build_array built-in functions, allowing cached queries to build JSON and JSONB arrays from variadic arguments.
- And more!
Check out the full release here.
Your AI Now Speaks Readyset, Announcing the MCP Server: Readyset now talks MCP, so agents like Claude and Cursor can inspect proxied queries, see what's cacheable, and create or drop caches directly, no more copying SQL into a chat window and pasting the results back.
Caching Supabase with Readyset over IPv6: If you're running Supabase and want lower read latency without changing application code, this post is for you. Vinicius walks through getting Readyset running in front of an IPv6-only Supabase project end-to-end, including the AWS and Docker networking that has to line up.
PGDay Porto Alegre | July 9, Caldeira Institute, Porto Alegre, Brazil
Marcelo Altmann is speaking at PGDay Porto Alegre on "Scaling PostgreSQL Reads with Readyset." If you'll be there and want to talk Postgres internals, replication, or caching, come find him. Register here.
Keen to use Readyset but your database or cloud environment is not supported? Reach out to support@readyset.io and let us know.
Want deeper dives on database performance, MySQL and Postgres internals, and what the industry is talking about? Readyset Weekly lands in inboxes every Tuesday.
Questions, bugs, ideas, or just want to say hi? You can find us on GitHub, Slack, and LinkedIn. Looking to contribute? Check out our newcomer issues; it's a great place to start.
Readyset