Sizing self-managed GitLab with the published reference architectures and choosing how Gitaly, PostgreSQL, Redis and Sidekiq scale as users and CI load grow.
Matching an organization to the 1K to 50K reference architectures by requests per second rather than headcount alone, adjusting for monorepos and CI load, and when to pick Cloud Native over VMs.
The smallest standard HA layout: node roles for Rails, Sidekiq, Gitaly Cluster, Patroni PostgreSQL, Redis Sentinel and Consul, the scaled-down variant and what zero-downtime upgrades require of it.
Running Webservice and Sidekiq in Kubernetes while PostgreSQL, Redis, Gitaly and object storage stay outside the cluster, and the managed cloud services that fit each stateful role.
Deciding whether repository-level HA justifies Gitaly Cluster: synchronous replication and failover against its known limitations, no snapshot backups, Kubernetes beta status and the sharded alternative.
The database tier of GitLab HA: Patroni-managed PostgreSQL, PgBouncer pooling, version and extension requirements, and when a managed cloud database is the better option.
Choosing Redis 7.2 or Valkey 7.2 for GitLab, Sentinel-based HA, why Redis Cluster and serverless offerings are unsupported, and eviction policy settings.
How large monorepos and heavy CI concurrency change GitLab sizing: Gitaly resources, starting large and scaling down, and the workloads GitLab lists as performance risks.
Validating an environment against reference architecture targets with the k6-based GitLab Performance Tool before go-live or after changes, and reading results against RPS targets.
Using GitLab's Terraform and Ansible toolkit to build reference architectures on AWS, GCP or Azure, its documented limitations and the manual steps still needed for production.
When one Linux package server is enough: CPU, memory and storage baselines, automated backups as the recovery plan and the signals that it is time for the 3K HA design.
Planning Sidekiq processes and memory as job volume grows, separating Sidekiq nodes from Rails and the per-process memory guidance above 10,000 users.
Tell us what you're working on, whether it's a deployment, an audit, a security test or a cyber range. You'll speak with an engineer who can help you scope it.