Skip to main contentSkip to user menuSkip to navigation
Interactive design lab

Load Balancer Visualizer

What is load balancing?

Load balancing routes each request to an eligible backend. A safe design must distribute work, eject unhealthy instances, preserve spare capacity, and prevent retries from multiplying an incident.

Step 1

Model

Change traffic, routing policy, affinity, backend capacity and weight, health ejection, and caller retry behavior.

Step 2

Observe

Trace request distribution, saturation, queueing, health ejection, availability, tail latency, and the consequence users experience.

Step 3

Challenge

Compare a node failure, slow instance, affinity hotspot, zone loss, retry storm, and gradual recovery. Find the policy that keeps failures isolated instead of moving pressure elsewhere.

Live routing workbench

Route traffic, then break the pool

Compare routing policy, backend pressure, health ejection, and the user impact of retries without hiding the failure path.

Availability

100.00%

0 user rps lost

Estimated p95

54.9 ms

routing plus queue delay

Queued

0 rps

attempts above capacity

Ejected

0 / 4

not eligible for new work

Retry load

+2%

1,275 total attempts

Request topology

Where every attempt goes

Line width represents routing share. Select a backend to tune its capacity and weight.

1,275 attempts / second
Load balancerCapacity aware

Failure laboratory

Inject a production condition

Active condition: Healthy

Two control loops

Separate routing intent from failure behavior

1. Traffic and routing

Arrival rate and algorithm change distribution before any health policy is applied.

Routing algorithm

2. Health and recovery

Ejection and retry policy determine whether a fault stays isolated or multiplies.

Health ejection
Caller retry policy

Selected backend

API A1Zone A

Healthy

73% utilized at 48 ms observed latency

0 queued

Pool pressure

Distribution, saturation, and queueing

The marker at 100% is the tested capacity boundary.

API A1

Healthy

26% distribution73% saturation
335 in0 queued

API A2

Healthy

21% distribution73% saturation
262 in0 queued

API B1

Healthy

30% distribution73% saturation
379 in0 queued

API B2

Healthy

23% distribution73% saturation
299 in0 queued

Observed outcome

The pool is serving the declared traffic.

Users see stable response times and the fleet keeps usable headroom.

Raise traffic or change the algorithm to find the first pressure point.

Capacity is a tested per-backend throughput, not CPU percentage.Queueing begins when attempts exceed effective capacity.This is a teaching model, not a production sizing guarantee.
No quiz questions available
Could not load questions file
Spotted an issue or have a better explanation? This page is open source.Edit on GitHub·Suggest an improvement