Define the boundaries before production starts moving.

Treat this as a coexistence and cutover program, not a one-step replacement. The safest path is to prove parity while both old and new control paths are still available.

• Inventory every dependency touching GitLab, including hidden dependencies such as service accounts, forwarders, trusts, certificates, or delegated admin workflows.
• Freeze unrelated high-risk change during major milestones so the signal stays clean.
• Write success evidence in operator terms: sign-in works, name resolution is stable, management is reachable, and rollback remains viable.
• Assign clear stop points where the team can pause if CI/CD validation fails.

Baseline the current state before introducing change.

Capture health and inventory first. The baseline is what lets you tell the difference between expected coexistence noise and a real regression.

• Export current configuration, ownership, and topology for GitLab and adjacent services.
• Measure present-day health for CI/CD, logs, alerts, and operator runbooks before starting the move.
• Document exceptions, unsupported integrations, and any legacy dependency that still relies on Runners behavior.
• Prepare pilot scope, rollback path, maintenance windows, and evidence collection locations.

Command path:

• az account show
• aws sts get-caller-identity
• kubectl get nodes -o wide
• wbadmin get versions
• Get-EventLog -LogName Application -Source '*Backup*' -Newest 50

GUI path: Platform console for the cluster, repository, runner, pipeline, state backend, and identity provider tied to this change.

Sequence the change so dependencies remain testable.

Use tight sequencing. Every state change should have an immediate validation step before the next dependency moves.

• Build the target path for GitLab alongside the current path where coexistence is supported.
• Shift one authority plane at a time: service binaries, configuration, identity bindings, name resolution, then client or workload targeting.
• Keep old and new control paths observable until CI/CD proves stable across more than one test path.
• Record every manual change so rollback does not depend on memory during a tense cutover.

Testing has to cover behavior, not just status lights.

Operator-ready testing includes positive flow, negative flow, failure injection, and recovery checks. Green dashboards alone do not count as signoff.

• Test from at least two user or workload paths and one admin path so GitLab is validated from different failure domains.
• Confirm CI/CD behavior under normal load, after a restart, and after a forced rediscovery or cache flush event.
• Run failover, referral, replication, or reconnection tests where Runners makes the difference between stable and fragile.
• Document exact commands, UI checkpoints, and evidence artifacts required for final acceptance.

Command path:

• az account show
• aws sts get-caller-identity
• kubectl get nodes -o wide
• wbadmin get versions
• Get-EventLog -LogName Application -Source '*Backup*' -Newest 50

GUI path: Platform console for the cluster, repository, runner, pipeline, state backend, and identity provider tied to this change.

Plan the move back before you plan the move forward.

A real cutover plan defines what changes, what stays stable, and what triggers immediate stop or rollback. That discipline keeps the team from improvising under pressure.

• Choose the lowest-risk cutover point, then pre-stage DNS, client targeting, monitoring, and communications around it.
• Define rollback triggers in terms of observed user impact, replication health, queue depth, auth failure rate, or management loss.
• Keep the old path read-only, isolated, or otherwise protected from accidental split ownership during coexistence.
• Capture post-cutover evidence before decommissioning anything that would make reversal harder.

Finish by making the new state supportable.

The project is not done when traffic moves. It is done when the operations team can monitor, back up, troubleshoot, and recover the platform with confidence.

• Update monitoring, backup, alerting, and audit coverage for GitLab and all new dependencies.
• Refresh operational runbooks, breakglass access, and access reviews for teams that own CI/CD.
• Retire stale records, abandoned automation, and outdated references that still point at the replaced service path.
• Schedule a post-change review to turn the migration evidence into a reusable operating standard.