Prometheus 2.20: Trustworthy Metrics at Scale

Introduction

On June 15, 2020, the Prometheus community released version 2.20, tackling the scaling and reliability challenges of cloud native observability. Write-ahead log compression, remote-write improvements, and smarter rule evaluation keep metrics pipelines trustworthy.

WAL Compression & Durability

• New Snappy-based WAL compression cuts disk usage by up to 70% while improving restart times.
• Checkpoint logic reduces duplicate samples after crashes or restarts.
• promtool tsdb create-blocks-from utilities simplify offline debug and disaster recovery.

Remote Write Sharding & Backpressure

1. Queue-based sharding balances load across remote-storage endpoints, preventing hot shards.
2. Configurable backpressure and on-disk buffering smooth over short outages downstream.
3. Label limits and relabeling guard against unbounded cardinality before shipment.

Rule Evaluation & UI Polish

• Scheduler improvements prevent thundering herds when multiple rules fire simultaneously.
• promtool gains linting for recording/alerting rules, reducing production errors.
• Web UI adds query history, autocomplete, and exemplars groundwork for tracing correlations.

Getting Started

kubectl set image deployment/prometheus-server prometheus=quay.io/prometheus/prometheus:v2.20.0 -n monitoring
kubectl rollout status deployment/prometheus-server -n monitoring

Tune remote write:

remote_write:
- url: https://metrics.example.com/api/v1/write
  queue_config:
    max_shards: 10
    min_shards: 4
    max_samples_per_send: 2000

Summary

Prometheus 2.20 gives SRE teams confidence that their dashboards and alerts remain accurate even as clusters scale out.