Find what breaks before it does

Draw your racks, servers, containers, networks and power. Fail a zone, a rack or a feed and see what your users lose.

Zone A availability zone
Zone B availability zone
Visitorssource
Load balancerload-balancer
Appweb-01 · :8080
Appweb-02 · :8080
Postgres primarydb-01 · :5432
Postgres standbydb-02 · :5432
infrWeb stack across two zonesFailure test
641
processors
1,593
servers, instances and VPS plans
1,051
network devices and services
237
storage systems
936
UPSs, PDUs and generators
44
multi-service platforms
Failure domains

Put things where they really live

A container runs on a server, in a rack, in a zone. Draw it that way and the analysis follows, because when the outer thing fails, everything in it goes with it.

  • Single points of failure

    Every component and every pair of components is taken down in turn, and each loss is described in words.

  • Nesting that matches the real world

    Racks hold rack hardware only, so a drawing cannot claim something that could not be built.

  • Quorum and redundancy

    A cluster that loses its majority stops serving, even the members that are still running.

Zone
Rack
Server
Container

A node is down when anything around it is down.

Capacity

See the survivor, not just the failure

Losing one half of a pair is rarely the end. The question is whether the other half can carry the load, and how much headroom it has left.

  • Traffic follows your links

    Set requests per second at the source and the call ratio on each link; the load on every tier is worked out for you.

  • Failover load

    Take out a zone, a rack or a server and see which tier goes over what it can serve.

App in zone Azone lost
App in zone Btakes the traffic

The line marks what the app can serve. Past it, the cluster is not N+1.

Power

Power is a failure domain too

Utility feeds, UPSs, PDUs, transfer switches, generators and batteries feed your hardware over cables you draw. Cut the power and see what survives.

  • Runtime that scales with load

    A UPS carries its load for the runtime you set, and trips if its load goes over its rating.

  • Outages with a length

    Test a ten minute cut or an open-ended one; a battery that outlasts the first may not outlast the second.

Utility feed
UPS
Server

up on battery down

Keep your diagrams in the cloud

Save up to 50 diagrams, encrypted, and open them wherever you sign in. Your browser copy keeps working either way.

infr plus is £3 a month

Walk a team through it, live

Host a session and share a code. Up to 50 people watch your canvas change, with no account needed.

infr plus is £3 a month

Start with a blank canvas

No account needed. Everything stays in your browser until you decide otherwise.

Open the editor