Mid-market SaaS (pattern case) · 2019 · cost index 4/5

Service mesh before the traffic arrived

Platform team of six; product incidents from mesh misconfig outpaced actual load issues

What happened

A fifty-engineer product company installed a service mesh because conference talks said Kubernetes was incomplete without one. Latency budgets were fine; certificate rotation and sidecar resource burn were not. The mesh became the outage surface area while the real bottleneck remained an unindexed query.

Failure mode

Cargo-cult hyperscale: adopting FAANG topology without FAANG traffic or ops culture. Success was measured in mesh coverage percentage, not customer outcomes. Debugging hopped through sidecar logs nobody owned.

Lesson

If your scale problem fits in a spreadsheet and one Postgres, you need ownership and indexes — not mTLS theater. Mesh when traffic and team topology demand it; until then it is cosplay with an on-call rotation.

Sources

  • Industry pattern — premature service mesh adoption
  • CNCF surveys — mesh complexity vs benefit
  • Platform engineering retrospectives (anonymized)

Related tech

← All cases

© 2026 Fadstack · Shane Code

Opinionated history · not a ranking