Skip to main content
Seed tiers from anonymized production-scale measurements. Infrastructure teams block on this page. Owner: SRE lead.

Sizing tiers

Per-component sizing

  • Kafka — broker count, partition strategy, retention vs disk.
  • Flink — task managers; RocksDB state sizing (dedup-state size drives both Flink memory and Redis).
  • Neo4j — heap/pagecache vs graph size.
  • ClickHouse — storage/retention math for the incident store.
  • Redis Cluster — memory vs dedup window per tier.

Kubernetes resources

Requests/limits table per workload · storage classes · anti-affinity rules.

Capacity assumptions

Ingest bandwidth, inter-AZ traffic, growth headroom guidance.