Service levels

No SLA today

Headband is in beta. We will publish an SLA once we have enough measured history to stand behind a number. Until then, we do not quote an uptime percentage. If you need a contractual commitment, talk to us before you rely on Headband for it.

What we aim for

• The API, the dashboard and hosted engines answering around the clock.
• Knowing about an outage before you tell us: our checks run every minute and alert us by email.
• Posting incidents that affect customers on the status page.

These are goals, not promises, and there are no credits when we miss them.

Each hosted engine has one home server

Hosted engines run in one region, ewr1 in New York, on two servers: ewr1-a and ewr1-b. Each engine lives on one of them, which takes all of its writes.

Each always-on Meilisearch or Elasticsearch engine on bucket storage and a paid plan normally has a live read-only copy on our other server. If its server fails, searches move to the copy within seconds, a little slower and up to a few minutes behind. Writes return in about 2 to 3 minutes, when the copy's server takes over. If the server failed in the moment it was saving, writes return after about 12 minutes. A move off a server that died can lose recent writes: up to about a minute of them. They are not in the engine after it moves. If that server comes back, they stay on its disk for 30 days. If its saves to storage were already failing before it died, writes wait until that server returns or a person steps in.

Typesense and Visor engines, and engines on the free plan, have no copy yet.

• If an engine's server goes down, an engine that keeps its data in bucket storage moves to the other server by itself, in about 2 to 3 minutes. It starts there from the bucket.
• Engines not on bucket storage still move by hand. Such an engine is down until the server comes back or we move it to the other server. A person first confirms the old server is off, so two servers never write the same engine. These engines are restored from their latest backup.
• A move off a server that died can lose recent writes: up to about a minute of them for an engine on bucket storage, and everything since the last backup for other engines. If the server comes back before an engine moves, that engine starts again with its data.
• Once our checks mark a server unreachable, within about 90 seconds, requests for its engines get a 503 with a Retry-After header at once, instead of waiting for a timeout.

If an engine going down for a while would hurt you, plan for it: keep your source data, retry with backoff, and consider a second index on another engine or an engine you bring.

Other single points

• The control plane (the API, the dashboard and their database) runs on Vercel and BataData. If either has an outage, Headband's API and dashboard are affected.
• There is one region. A New York area outage affects every hosted engine.
• Engines you bring are as available as you or their host make them. We do not monitor them for you.

Support

Email support@headband.dev. How soon a person first answers depends on your plan:

• Community support (Free): no stated response time.
• Email support (Starter and Pro): first response within 1 business day.
• Priority support (Scale): first response within 4 business hours.

Business days and business hours are Monday to Friday, 9am to 6pm US Eastern time, excluding US federal holidays. A first response is a person replying to you, not an automatic receipt. It is not a promise to fix the problem in that time. There is no phone line.