Reviews database migrations in pull requests and flags locks, data loss and slow backfills before deploy
Many of the worst outages engineering teams describe to us are not application bugs but database migrations: an index built without CONCURRENTLY that locks a 200M-row table, a column rename that breaks the old app version mid-deploy, a backfill that saturates the replica. Reviewers rarely know table sizes or traffic, so these slip through code review.
Quebracho reads each migration in a pull request alongside production schema statistics, table sizes and query traffic, and simulates the lock and runtime impact on a shadow copy. It comments on the PR with the specific risk, the estimated lock time and a safer rewrite, such as splitting a change into expand and contract steps, for Postgres and MySQL.
Engineering teams running Postgres or MySQL in production with frequent deploys; over 150,000 companies worldwide, starting with SaaS and fintech teams of 20 to 300 engineers.
Per-repository subscription: US$49 a month for small teams and US$399 a month for the Team plan with production statistics; enterprise contracts add self-hosting.
Valentina RuizCEO and Co-founder
Martín IglesiasCTO and Co-founder
Lucía ParedesFounding EngineerOur plant systems team runs Postgres for production tracking, and the generic AI reviewer we tried only commented on style. Quebracho knew our station-events table had 180M rows and that writes peak at shift change, and its shadow-copy simulation put a number on the writes a migration would block. Our two sites on a managed cloud database are not covered yet.
A junior engineer added an index on a 200M-row ledger table without CONCURRENTLY, and two reviewers approved it. Quebracho commented with an estimated six-minute write lock and the exact rewrite, which we merged. We have since made the missing lock timeout check blocking across all our repos.
Reply from Valentina Ruiz, CEO
Glad the ledger table survived, Lucas. Making the lock timeout check blocking was your suggestion in the first place.
Our order database grows fast in November, and Quebracho warned that a backfill on the orders table would saturate the read replica during the sale. We batched it as suggested and nobody noticed the change. The policy settings took some tuning before it stopped warning about small tables.
Quebracho reached 74 paying teams with almost no sales staff, mostly through engineers sharing PR comments that saved them from an outage. Valentina and Martín understand database internals deeply. The next step is selling to larger organisations that need self-hosting, which is a different skill set they are now hiring for.
Coming soon. Testimonials are invited by the founder and shown separately from open reviews.
Coming soon. Follow Quebracho to hear about product and funding updates when this launches.
