Lesson 19 · Active health probing - Build a Load Balancer

Active health probing

Passive checks only learn from real traffic - so a down backend, getting no traffic, could never recover. Active probing fixes that by running an out-of-band health check against every backend, up or down.

The goal

Run one probe round over all backends, recording each result as a success or failure streak.

Start here - the target

TO DO

Scenario: One probe round records an outcome for every backend

Givena pool of A (up), B (up), and C (down), and a probe function that reports A healthy, B unhealthy, C healthy

Whenthe prober runs one round

ThenA.SuccessStreak() is 1, B.FailStreak() is 1, and C.SuccessStreak() is 1

AndC is probed even though it is down - that is how a down backend can later be brought back

Background

Active health checking runs a dedicated probe against each backend - a health endpoint request, a TCP connect - independent of user traffic. It exists to solve a problem passive checking cannot: a backend marked Down receives no requests, so without an out-of-band probe it would have no way to prove it has recovered and would stay ejected forever. That is why the prober iterates All() backends, not just the Available() ones.

Today the prober only records each probe’s outcome into the same streak counters your passive checks feed; it deliberately does not yet transition a down backend back to up. Reusing RecordSuccess and RecordFailure means active and passive signals accumulate in one place, so a backend recovering under probes and a backend serving real traffic are treated identically. Keep the probe function injected, exactly like the transport, so this stays a deterministic offline test rather than a real network call. Turning a run of successful probes back into Up is the next lesson.

Make it work

// Pool gains All() returning every backend, up or down
type Probe func(b *Backend) bool
type Prober struct { pool *Pool }
func (pr *Prober) RunOnce(probe Probe) {
  for _, b := range pr.pool.All() {   // ALL backends, not just Available()
    if probe(b) { b.RecordSuccess() } else { b.RecordFailure() }
  }
}

CheckpointDONE

An active probe round records a fresh outcome for every backend, including down ones. Commit and stop here.