02 · POSTGRESQL / SRE / CLOUD
Production Reliability & Database Architecture
Keeping infrastructure for ~140 enterprise customers available, recoverable and a bit cheaper — at AssetCues.
KubernetesPostgreSQLTerraformOpenTelemetryGrafanaAzure / AWS
- −50%
- MTTR
- −40%
- database incidents
- −70%
- detection time
- −60%
- RTO
01 Context
Production infrastructure on Azure and AWS for roughly 140 enterprise customers, with a team of 6 DevOps engineers working alongside 20 developers.
02 Engineering challenge
Recurring database incidents, slow detection, and deployments that failed more than anyone liked — with recovery procedures that existed mostly as tribal knowledge.
03 My responsibilities
- Production availability, deployment and recovery
- PostgreSQL operations and reliability
- Observability, SLIs and alerting
- Infrastructure automation and cost
04 Architecture overview
Kubernetes workloads on Azure/AWS backed by PostgreSQL with PITR, WAL archiving, logical replication and encrypted automated backups. OpenTelemetry, Prometheus, postgres-exporter, Loki, Grafana and SigNoz feed API/job/database SLIs. CI/CD with rollback strategies, environment isolation and peer-reviewed Terraform.
05 Key technical decisions
- →Recovery drills, not just backups — RTO improved by 60%.
- →SLIs defined per API, job and database so alerts are actionable.
- →Compute sizing, storage IOPS and connection pooling for a 20% cost reduction.
- →Terraform and Bash automation cut manual effort by 60%.
06 Measurable outcomes
- −50% MTTR
- −40% database incidents
- −70% detection time
- −60% RTO
07 What I learned
A backup that hasn't been restored is just optimism. The reliable path is the boring, rehearsed one.
next project
Logistics Backend Modernization