The idea: databases are pets
The industry learned to treat servers as cattle: identical, disposable, replaced rather than repaired. That works for stateless fleets. Databases are different. You provision one once, then care for it for a decade: version upgrades, restarts, failovers, hardware swaps. Provisioning should be automated and forgettable. Maintenance needs care, history and judgment. They're two problems, and they deserve two tools.
The proof
That split isn't a slogan. HASHJOIN's principal is the first-named inventor on US 11,693,748, Systems and methods for provisioning and decoupled maintenance of cloud-based database systems, from his database platform work at Auth0. The same approach to maintenance later produced a control plane that can roll an 11,000-node Vitess fleet in 35 minutes, down from more than 2 days.
The patent is assigned to Auth0 LLC. It's cited as the principal's background, not as a description of hammerdown or a license to anything.
The tool: hammerdown, made yours
hammerdown is HASHJOIN's open-source control plane for rolling restarts and upgrades: explicit workers, exact fail-fast, dry run by default, and a plain-text report you can grep. Read the story behind it. Out of the box it drives a simulated fleet. The build connects it to yours.
The build: $30,000 fixed, 3 weeks
Scope week 1
Map your inventory source (topology server, CMDB or cloud API), your restart and failover steps, and the safety checks your fleet needs. If week 1 shows your fleet doesn't fit the fixed scope, you can stop there and pay for that week only.
Build week 2
A hammerdown backend for one database engine and one inventory source, with guards for replication lag, running backups and change freezes, tested against a fake of your system.
Prove and hand over week 3
Dry runs, then a canary roll on a slice of your fleet with your team at the controls. You get the code, a runbook, and a
CLAUDE.mdso your engineers and their assistants can extend it.
50% at kickoff, 50% at handover. Stopping after week 1 costs $10,000. You own the code: the Apache-2.0 core plus the backend built for you, delivered into your repositories.
Support: keep it current
Your fleet keeps changing after handover: new database versions, new hardware, new failure modes. Support keeps hammerdown in step with it.
| Tier | Price | Response | Includes |
|---|---|---|---|
| Maintain | $2,000/mo | Next business day | About 8 hours a month. Fixes, security patches, and compatibility with new versions of your database |
| Operate | $4,000/mo | Same business day | About 16 hours a month. Everything in Maintain, plus new guards and backend changes, a monthly review of your roll logs, and on call for one scheduled upgrade window a quarter |
| Partner | $7,500/mo | Within 2 business hours | About 30 hours a month. Everything in Operate, plus on call for every scheduled upgrade window, one major-version upgrade a year run or co-piloted by HASHJOIN, and early access to the next tools in the family |
- 12-month term, billed monthly in advance; about 10% off when paid annually
- Hours don't carry over; extra hours are $250 each
- On-call covers your scheduled upgrade windows, not 24/7. Your team stays the first responder
- Fleets above 2,000 nodes are quoted individually
Already running open-source hammerdown?
Adopt: $7,500, one time. A review and hardening of your fork: the guards you're missing, the failure paths you haven't tested, and a production-readiness report. Then any support tier.
Every Adopt review tests your fork against these failure paths, with your backend's fakes:
- Two instances at once: a second roll, started while the first is still running
- A stale plan: a shard that changed (a failover, another tool) between planning and acting
- A dropped session: a hangup mid-roll, with in-flight work and locks
- A lock that asks you to delete it: any error that leaves "is the holder dead?" to the operator
Talk about your fleet
Tell us the engine, the node count, and what your last upgrade felt like.