August Product Update 2026

August moved fast around here, and we've got a full roundup for you.

Cut Database Infra with Self-Tuning Caching

On August 18, our team hosted a live session on what changes when caching moves into your infrastructure and continuously tunes itself. They covered why database costs keep climbing, why more hardware isn't the answer, and how to get more throughput from the hardware you already own without spending engineering hours on cache maintenance.

Product Updates

• Faster Shallow Cache Serving: Reduced adapter CPU on shallow-cached queries by computing each query's ID once per request instead of on every status-cache access, and removed a per-query status-cache lock acquisition that was limiting throughput under high concurrency. Also switched per-query view and reader lookups to a faster hash function.
• Stronger Shallow Cache Coalescing Under Load: Closed a small race window in shallow cache coalescing that let misses in rapid succession bypass the mechanism and query the upstream without waiting. A sudden burst of queries that would previously all miss now properly coalesce into a single upstream query, protecting the upstream during traffic spikes.
• New Runtime Sharding for High-Concurrency Workloads: Added --runtime-shards to the adapter, which spreads client connections across additional runtimes so connection readiness no longer funnels through a single I/O thread. Size the added runtimes with --shard-worker-threads. The default of 0 keeps the previous single-runtime behavior.
• Authorization on Shallow Cache Reads: With authentication enabled, shallow cache reads now enforce the connecting user's upstream authorization, so a user who lacks access to a table is no longer served its cached rows.
• Manage Readyset When Upstream Is Down: Users can now connect to the readyset schema (USE readyset on MySQL or SET search_path = readyset on PostgreSQL) to inspect and manage Readyset even when the upstream database is down.
• Better MySQL Compatibility: Proxied queries now follow the collation the client advertises in its connection handshake, matching a direct MySQL connection. Cached results for sessions using utf8mb3 correctly convert supplementary characters to ?. Case-insensitive collations now match on LIKE, and utf8mb4_general_ci, utf8mb4_unicode_ci, and utf8mb3_unicode_ci correctly match accented and unaccented equivalents in point lookups and GROUP BY (a resnapshot is necessary for deep caching).
• Sharper Postgres Type Handling: Text casting and coercion under PostgreSQL now follow Postgres's own input functions, so a value parses or is rejected the way upstream does. Casts to boolean and to the elements of an array are affected. PostgreSQL's ->>, #>>, and json_extract_path_text now return the extracted value with its JSON quoting intact.
• And more!

Check out the full release here.

New on the Blog & YouTube

Your Database Bill Is Climbing Faster Than Your Traffic: A look at why database spend keeps rising even when traffic is flat, and where that hidden cost usually lives: memory-heavy instances that grow with the load pattern, not the traffic count.

How Readyset Rewrites Your SQL: Subqueries in HAVING, ORDER BY, JOIN ON: Readyset now caches queries with correlated subqueries in four new positions: HAVING, ORDER BY, INNER JOIN ON, and LEFT OUTER JOIN ON. Many previously unsupported queries can now be rewritten by Readyset, allowing them to be cached and maintained incrementally.

Introducing Adaptive Cache for Dynamically Managing Freshness: A walkthrough of how Readyset's adaptive cache decides on its own what queries to keep hot, how often to refresh them, and when to let them drop, all without hand-tuned TTLs.

Upcoming Events

Percona Live Amsterdam 2026

Readyset is sponsoring Percona Live in Amsterdam September 9-11. On Friday, September 11 from 3:10-3:40 PM in Matterhorn 1, Marcelo Altmann will present "What Happened to the MySQL Query Cache?" His talk looks at why the original MySQL Query Cache failed, why key/value caches became the dominant workaround, and how Readyset combines ideas from both to get past their limitations.

MySQL BR Conf 2026

Our very own Vinicius Grippa and Marcelo Altmann are co-organizing another year of MySQL Brazil Conference, and Readyset is excited to be sponsoring. Tanmay Sinha will present "Building the Invisible Cache" from 10:10-10:45 in the Library. The talk explores why MySQL's built-in query cache failed under real concurrency, why Redis and Valkey still leave platform teams owning the hard parts, and how Readyset progressed from shallow and deep caching to Adaptive Cache, which decides what to cache and when without application changes or manual invalidation.

Get Involved

Questions, bugs, ideas, or just want to say hi? Find us on Slack. Looking to contribute? Our newcomer issues are a good place to start.