May was a busy month! Between Percona Live, Web Summit Vancouver, and a packed release, there is a lot to catch up on.

New Case Study: Lemit Reaches 100K QPS on MySQL

Lemit, a Brazilian data enrichment and credit intelligence company, was hitting the ceiling on MySQL performance; vertical scaling, read replicas, and MEMORY tables weren't enough. After deploying Readyset, they reached over 109,000 queries per second with under 1ms cached query latency, 8x read throughput increase, and zero lines of code changed.

Read the full story.

Product Updates

Readyset Core

• MCP Support: Readyset now includes built-in MCP support, allowing  AI assistants like Claude Code and Cursor to introspect deployments and manage caches directly. Access is gated by Bearer tokens with configurable scopes (read_only, cache_admin, full), managed via CREATE / SHOW / ALTER / DROP MCP TOKENfor PostgreSQL.
• PostgreSQL Group Commit: Group commit is now available for PostgreSQL replication, batching committed transactions before flushing to reduce replication overhead and improve throughput under high write rates.  Uses the same configuration parameters as MySQL: group_commit_max_trx, group_commit_wait_us, and replication_batch_size.
• UNTIL WRITE Caching Policy: New CREATE CACHE ... UNTIL WRITE keyword serves queries from cache until the current transaction observes a write, then proxies upstream. Auto-created shallow caches in --cache-mode=shallow adopt this policy automatically, which means Supabase Python drivers, ORMs with autocommit off, and JDBC defaults get query acceleration without any configuration changes.
• PostgreSQL Query Hints: Readyset now supports Postgres query hints, including /*rs+ CREATE SHALLOW CACHE */ and /*rs+ SKIP CACHE */ across both simple and extended protocols, bringing feature parity with MySQL.
• CREATE CACHE Now Returns Metadata: Returns a row with the query id, name, query text, and cache type for the just-created cache, making scripting and automation easier. 
• New Virtual Relations: readyset.deep_caches can now be queries via SQL to inspect deep cache state while readyset.shallow_caches now exposes per cache hits, misses, and refreshes 
• Parallelized RocksDB Lookups: Batched upqueries now use parallel RocksDB lookups, reducing cold cache-miss latency for large key batches.
• And more! 

Check out the full release here. 

New on the Blog & YouTube

Replication Internals: Decoding the MySQL Binary Log (Parts 10 and 11)

Marcelo's 11-part series is now complete. The final two parts cover the rotate_event that closes out a binary log file and the new serialization framework MySQL 8.4 introduced for tagged GTIDs.

• Part 10: ROTATE_EVENT — Closing the File and Pointing at the Next One
• Part 11: GTID_TAGGED_LOG_EVENT — Tagged GTIDs and MySQL's New Serialization Framework

Caching Real GitLab Queries on Azure Postgres with Readyset:  Vinicius ran real GitLab CE traffic through Readyset on Azure Postgres and found 83% of query shapes cacheable, the slowest query dropped from 85ms to 3ms, and the instance scaled down from $577/mo to $140/mo at the same throughput.

From Parser Combinators to Hand-Written TDOP: On Adopting sqlparser-rs at Readyset: Why Readyset replaced its homegrown SQL parser, how the switch delivered 4x throughput improvement and 44% reduction in tail latency, and what Pratt parsing has to do with it.

April Event Recap:

rdst Fleet Audit: Live Demo: We ran rdst live against a live RDS fleet, surfacing the queries responsible for the majority of cost and latency and showing how to address them directly from the terminal. If you missed it, the recording is now live.

Vancouver Web Summit: Great conversations with engineering teams thinking through database performance and infrastructure costs. If you missed us there, book time with our team here.

Percona Live 2026: We sponsored Percona Live this past week. Vinicius's talk "Beyond Manual Caching: How Readyset Offloads MySQL and Postgres Workloads" took place on May 28.

Upcoming Events

Web Summit Rio | June 8-11, Riocentro, Rio de Janeiro

We will be at Web Summit Rio next month at booth E318.

And on June 10, Gautam Gopinadhan is giving a talk on Stage 3 at 4:15pm: "AI Is Querying Your Database. Are You In Control?". Catch him on stage for a look at what actually changes for database infrastructure when AI becomes the primary interface to your data. 

More event details here.

🧑‍💻Request Features

Keen to use Readyset but your database or cloud environment is not supported? Reach out to [email protected] and let us know.

Get Involved!

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.