Reliability Shared (all languages) active any

Overload-triggering work and resource consumption are bounded

rel-008

Intent

Prevent saturation, memory blowups, and cost explosions caused by unbounded work acceptance or resource use.

Applicability

Applies to request queues, batch fan-out, container limits, and user-controlled work sizing.

What to inspect

Queue lengths, size limits, container limits, pagination or batch size controls, and admission behavior under load.

Pass criteria

Work size and memory or queue growth are visibly bounded, and overload is rejected early.

Fail criteria

The diff allows unbounded queueing, memory consumption, or caller-controlled work sizing with no visible cap.

Do not flag

Tiny internal data structures with hard-coded small bounds.

Confidence guidance

HIGH when an unbounded queue or resource path is directly visible. MEDIUM when limits may be hidden in deployment config. LOW when runtime ownership is partial.

Remediation

Add admission limits, queue caps, body or batch size limits, and resource ceilings.

Pass example

resources:
  limits:
    memory: 512Mi
queue:
  max_depth: 1000

Fail example

queue:
  max_depth: unlimited