The Feldera Blog
An unbounded stream of technical articles from the Feldera team
Announcing our Series A and Seed
We raised $21.5M across a Series A led by Inovia Capital and a Seed led by Costanoa Ventures, to reduce database compute costs by 95% with a mathematical breakthrough.
Auth0 and Feldera: Incrementally Computing 7 Billion+ Permission Checks
See how Auth0 uses Feldera to precompute 7B+ permission checks, keeping authorization fresh in real time for AI agents and search at scale.
Agents Aren’t Coworkers, Embed Them in Your Software
Agents do not need more human-like conversation. They need software that makes conversation unnecessary. Give them the right interfaces and they stop acting like noisy copilots and start acting like infrastructure.
Making samply profiles even more useful
Feldera cut one customer's backfill time from 20 hours to 4 by getting serious about profiling. In this post, we share how we made samply profiles significantly more useful by postprocessing their output to add application-level markers — no changes to samply required.
Turns out we didn’t need that second index
Building an incremental compute engine means constantly asking whether the runtime is doing more work than it needs to. This time the answer was yes, and the fix made every pipeline that shares a join source cheaper to run.
Why incremental aggregates are difficult - part 1
Many traditional query engines may be able to handle only some kinds of queries, or only some kinds of input updates incrementally, reverting to full recomputation for unsupported operations. Feldera uniformly handles insertions and deletions and stacked views using arbitrary monotone and non-monotone queries.
Nobody ever got fired for using a struct
Rust structs are usually the obvious way to represent data. But when you serialize wide SQL tables with hundreds of nullable columns, that "obvious" layout can quietly double your storage cost. Fixing it turns out to be a surprisingly simple trick.
Can your incremental view maintenance engine do this?
Handle 217 joins, 33 materialized views, 27 aggregations, and 287 linear operators on a single 16-core machine using 15GB RAM at steady state. Here's the proof.
Introducing Feldera Health
Introducing Feldera Health: a lightweight dashboard that shows your infrastructure status without Kubernetes access. Get quick answers when pipelines fail.