Let’s dive into our first product update of 2026!
Readyset Core
- MySQL Replication Server UUID Support: Added the --replication-server-uuid option to set a server UUID in MySQL. This UUID is now reported in SHOW REPLICAS.
- OOM Prevention for Large Parameterized IN Predicates: Fixed potential out-of-memory errors by rejecting queries with parameterized IN predicates that would expand to too many lookup keys.
- Configurable Sampler Query Timeouts: Added support for configuring the timeout for sampler queries.
- Duplicate Index Prevention: Added support for preventing the creation of duplicate indexes.
- Faster RocksDB Index Creation: Improved index creation performance on RocksDB column families by up to 20%.
- Top-Level Function Predicates in WHERE: Allowed functions to be used as top-level predicates in WHERE clauses (e.g., WHERE COALESCE(col, TRUE) or WHERE JSON_OVERLAPS(col, '[42]')).
- Correct Schema Resolution in Multi-Schema Joins: Fixed incorrect schema resolution for unqualified tables in JOIN ON clauses when multiple schemas contain tables with the same name.
- Improved Grafana Dashboard Imports: Updated Grafana dashboards to use a parameterized Prometheus data source, allowing imports to map existing data sources without hardcoded UIDs.
Check out the full release here.
Predictions About the Effect of AI on Enterprise Software and Infrastructure
We’ve made some predictions about how AI will dramatically accelerate enterprise software development, while making systems harder to control as complexity and deployment velocity rise. Agentic workloads will create volatile, expensive database access patterns that push infrastructure to its limits. The core takeaway: caching, guardrails, and cost-aware controls will become essential as databases become the primary bottleneck for agentic AI at scale.
The Database Is About to Lose Its Last Line of Defense
AI-generated SQL is eliminating the last human checkpoint before queries hit production. And that breaks a lot of assumptions. Agentic systems can produce unpredictable, high-risk queries at runtime, creating governance failures that existing database controls were never built to handle. This is already happening. Organizations will need new layers of oversight and guardrails to keep machine-generated data access safe, compliant, and under control.
The Hidden Complexity of Postgres Scaling Architectures
We enjoyed reading the recent OpenAI Postgres scaling write-up and wanted to share a quick follow-up on what it really takes to run Postgres at extreme scale, from caching layers to query routing to the ongoing operational effort most teams don’t talk about. What often gets overlooked is the hidden long-term complexity and developer tax that comes with application-level caching. The real question isn’t whether Postgres can scale, it’s where you want the complexity of scaling to live and how much ongoing cost you’re willing to absorb.
The Readyset team was in Belgium last week for two standout gatherings in the database community: MySQL Belgian Days and FOSDEM.
A major highlight for us: Marcelo presented “Fast on Hits, Faster on Misses: Join Execution for MySQL Workloads in Readyset” at MySQL Belgian Days, and was recognized with the MySQL Rockstar Award 2025.
FOSDEM also marked the return of one of our favorite traditions: the annual Readyset Database & Drinks happy hour. Huge thanks to everyone who joined us to talk database performance, swap stories, and spend an evening with the community.
Keen to use Readyset but your database or cloud environment is not supported? Reach out to support@readyset.io and let us know.
We love hearing from you! If you want to get involved with Readyset, you can:
Readyset Team
