Deployment of the AI Gateway using Runway

Deployment of the AI Gateway (for Self-Hosted models) using Runway.

Exit Criteria

• Develop a deployment process for the AI Gateway using Runway.

Contraints

Runway is being considered as a deployment method for self-managed instances. However, the current underlying runtime of Runway is Cloud Run, which may not work for air-gapped customers. We're is looking into expanding the underlying runtime of Runway to support more complex services in the future. So, while it's not currently fully suited for self-managed instances, especially those that are air-gapped, it's a possibility being explored for the future.

Links

• Multi-region support for AI Gateway (gitlab-com/gl-infra&1206 - closed) • Unassigned
• https://runway-docs-4jdf82.runway.gitlab.net/guides/onboarding/

See also https://gitlab.com/gitlab-com/gl-security/security-assurance/fedramp/fedramp-certification/-/issues/452#note_1832261170

Context

The following discussion from !148599 (merged) should be addressed:

• @andrewn started a discussion: (+5 comments)

  I am strongly in favour of self-hosted Runway being the delivery mechanism here, as we'll be able to reuse what we build for future components too, and building a consistent self-managed platform will ensure best supportability, observability, compliance adherence, and consistency across deployments. @rnienaber @kwanyangu @cfeick @fforster is there an issue in the Runway tracker for this, that we can point this item at? I would rather have this pointing at a well defined issue than a discussion in the Fedramp tracker.

  Also, I feel Omnibus is the least favourable option. Would it be possible to reverse the order of this list, as a hint to that.