Burst

Engines

Burst

When your search traffic spikes, burst gives your engine more CPU within a minute and takes it back when the spike ends. Your plan's searches a second rise with it.

At a glance
PlansOn by default on Pro and Scale
StartsAfter an engine is short of CPU for 30 seconds
How muchDoubles the CPU, then can double it again: up to 4 times, at most 16 CPUs
EndsWhen the load stays under half the lower level for 10 minutes. Each lift lasts at least 10 minutes
Price$0.05 (Pro) or $0.04 (Scale) per burst GB-hour
Included20 GB-hours (Pro) or 100 (Scale) a month
Cap$25.00 (Pro) or $100.00 (Scale) a month by default. You never pay more

How it works

  • Each engine size comes with 2 CPUs per GB. When an engine has been short of CPU for 30 seconds, burst doubles its CPU, and can double it again, up to 4 times its size's CPU and at most 16 CPUs.
  • While an engine is lifted, your organization's searches a second and in-flight requests rise by the same factor.
  • When the load stays under half of the lower level for 10 minutes, burst steps back down. Each lift lasts at least 10 minutes, unless you turn burst off, reach your monthly cap or move off Pro and Scale, which end it at once.
  • The engine keeps running the whole time. Nothing restarts and your data does not move.
  • Burst uses CPU the server has free. When the server is busy, burst waits, and your engine runs at its size's CPU.

Pricing

Burst is billed in burst GB-hours. One burst GB is 2 extra CPUs, the CPU of one GB of size. Lifting a 1 GB engine from 2 to 8 CPUs for an hour is 3 burst GB-hours.

PlanPer burst GB-hourIncluded each monthDefault cap
Pro$0.0520 GB-hours$25.00 a month
Scale$0.04100 GB-hours$100.00 a month

On Pro, searches a second go from 100 to up to 400 while lifted. On Scale, from 500 to up to 2000.

Turning it off, and the cap

Burst is on for every Pro and Scale organization, with the plan's default cap. In Settings, Billing you can set your own monthly cap in dollars, or turn burst off. The cap counts every engine in your organization together. The included GB-hours cost nothing and count toward no cap. We email the owners and admins at 80% of the cap. At the cap, burst pauses until the next billing period and nothing more is billed for it. Raise the cap to turn it back on.

Settings, Billing also shows how many burst GB-hours you have used this period, and whether an engine is lifted right now.

Which engines

Meilisearch, Elasticsearch and always-on Typesense, at every size. Each size's own CPU is 2 per GB, at most 6, and burst lifts it to at most 16. Visor and sleeping Typesense are not lifted. Burst is part of the Pro and Scale plans.

Your standby takes the overflow

When your engine is at its limit of searches in flight, or at 80% of it for 10 seconds, new searches can go to a warm standby on a second server while it has room, instead of answering 429. It costs nothing extra.

  • Who gets it. Every always-on Meilisearch, Elasticsearch and Visor (0.1.3 or later) engine on a paid plan, Starter included, on bucket storage (the default).
  • How fresh it is. The standby answers from your engine's last published copy. Your engine publishes about once a minute, and the standby takes it up about 10 seconds later for Meilisearch and Visor, or about 90 seconds for Elasticsearch, as measured. Overflow never goes to a standby whose copy is more than 5 minutes old.
  • When you need your latest write. Send the header x-hb-read: primary: while your engine runs and answers, the search stays on it, and gets a 429 at the limit as before.
  • An Elasticsearch _msearch never spills on its own.