[FF] use_vertex_codestral_code_completion -- Vertex Codestral as the code completion default

Summary

Roll out the feature currently behind the use_vertex_codestral_code_completion feature flag.

• DRI: @mikewronski
• Team Slack channel: #g_ai_coding

Part of epic https://gitlab.com/groups/gitlab-org/modelops/ai-model-validation-and-research/ai-evaluation/-/work_items/60. When enabled, the SaaS code completion default model changes from Fireworks-hosted Codestral 25.08 to Vertex-hosted Codestral 25.08 on the direct-access and v2 proxy paths.

What could go wrong?

• Vertex quota exhaustion (429s at peak). Do not start the rollout before the quota raise in https://gitlab.com/gitlab-org/modelops/applied-ml/code-suggestions/ai-assist/-/work_items/2736 is approved and verified. Peak need: ~7M input TPM in europe-west4, ~4M in us-central1.
• Asia-region HTTP 500. The AI gateway v2 guard _allow_vertex_codestral sends asia-pod requests for this model to a dead Fireworks route. Do not start the rollout before the AI gateway fix ships.
• Revert latency on direct connections. The v2 proxy reverts instantly on disable; direct-access IDE clients converge within the token lifetime (~1 hour).
• Watch the AI gateway Runway dashboards on https://dashboards.gitlab.net (Mimir - Runway): model_inferences_total error rates by error_type, and Vertex 429s, split by region.

Rollout

Run all production /chatops in #production and cross-post the results to #g_ai_coding. Background: incremental rollout process, feature actors.

Non-production

/chatops gitlab run feature set use_vertex_codestral_code_completion 50 --actors --dev --pre --staging --staging-ref
/chatops gitlab run feature set use_vertex_codestral_code_completion true --dev --pre --staging --staging-ref

Production — percentage rollout (wait ≥15 min between steps, watch dashboards):

/chatops gitlab run feature set use_vertex_codestral_code_completion <percentage> --actors

Before global rollout

• Quota raise approved and verified in both regions (https://gitlab.com/gitlab-org/modelops/applied-ml/code-suggestions/ai-assist/-/work_items/2736)
• AI gateway asia-guard fix deployed
• Docs follow-up MR merged (default-model table in model_selection.md, data-retention wording in data_usage.md, prompt-caching note)
• Data team informed: reported model changes from fireworks_ai/codestral-2508 to gitlab/codestral_2508_vertex

Cleanup

Remove the flag once deemed stable. Self-managed instances get the Vertex default when the flag is removed (or default_enabled: true).

/chatops gitlab run release check <merge-request-url> <milestone>
/chatops gitlab run feature delete use_vertex_codestral_code_completion --dev --pre --staging --staging-ref --production

Rollback

/chatops gitlab run feature set use_vertex_codestral_code_completion false