fix(data-platform): promote litellm 1.100.0 to pre-production and production

[Gary Henderson (Gary-H9)]

Summary

• Promotes preproduction LiteLLM from 1.98.0 to 1.100.0 and production from 1.97.0 to 1.100.0 in environment-configuration.tf.
• Leaves development and test unchanged at 1.100.0; that first stage merged in modernisation-platform-environments#19030.
• This promotion is the required next state-machine action and therefore takes precedence over starting another upgrade. The associated Compose attempt is data-platform-app#255; it was closed without merge during the promotion run, so the Data Platform App image remains v1.98.0 and no duplicate PR is opened here.
• v1.100.1 is now the latest published non-draft, non-prerelease release, published on 2026-09-10 after this promotion PR opened. This promotion intentionally remains on 1.100.0: the rollout state machine promotes the exact version already deployed to development and test before beginning a newer upgrade.
• The post-target v1.100.0...v1.100.1 comparison changes router retry-breadcrumb retention, package metadata, tests, and the lockfile; it does not change the management API, database schema or migrations, or Helm chart files reviewed for this deployment.

Data Platform client compatibility

No breaking changes affecting the current Data Platform AI Gateway client were identified across the complete 1.97.0 to 1.100.0 range. The authoritative latest-main AIGatewayClient, tagged LiteLLM source, intervening release notes, linked upstream changes, and the current Modernisation Platform deployment configuration were reviewed.

• Access groups and organisations: GET /v1/access_group and GET /organization/list?org_alias=... retain their methods, parameters, and the consumed access_group_name, access_group_id, access_model_names, organization_alias, and organization_id fields. Upstream changes are compatible: access-group responses gained additive resolved resource names (commit c5c10bc); organisation changes concern internal typing, write validation, and v2 update handling, not the v1 list contract used here.
• Teams: POST /team/new, POST /team/delete, GET /team/info?team_id=..., POST /team/update, and GET /team/daily/activity retain their routes and the request/response data consumed by the application, including team_id, team_info, access_group_ids, results, metrics, and breakdown. Relevant changes fix partial updates preserving metadata (#36328), serialize membership/delete writes (#37969), and add per-user spend reporting (#39771); they do not remove or rename the consumed contract.
• Virtual keys: POST /key/generate, POST /key/{key}/regenerate, POST /key/delete, GET /key/info?key=..., GET /key/list with team_id, return_full_object, size, and page, and POST /key/update retain the bodies and key, keys, models, and total_pages fields consumed by the client. Changes in the range add validation for optional logging/batch-token settings not sent by this client and fix MCP grant preservation (#38463), budget-window resets (#38686), raw-key list searches (commit 9baa19c), and JWT mapping invalidation on regeneration (commit 2f7ee395).
• Models, authentication, and errors: GET /v1/model/info retains the data envelope used by list_models_v1_info. The application still authenticates with a bearer master key through pinned httpx==0.28.1 and raises its own exceptions for HTTP and transport failures; it does not import the LiteLLM Python package.
• The upstream endpoint files did change between v1.98.0 and v1.100.0; compatibility is based on direct contract and implementation review, not an unchanged-file assertion. The focused Data Platform client suite recorded in data-platform-app#255 passed all 26 tests, including auth, error handling, team operations, and /key/list pagination.

Other breaking changes

• No applicable breaking-change notice was identified in the v1.99.0, v1.99.1, or v1.100.0 release notes.
• The tagged schema.prisma contains additive nullable/defaulted columns and new tables across this range, including LiteLLM_ModelAccessGroupBudgetTable, LiteLLM_BudgetWindowSpend, LiteLLM_ProxyWorkerHeartbeat, and LiteLLM_DailyGuardrailUsageUnits. No destructive schema change applicable to this deployment was found.
• The current litellm-admin values enable migrationJob; the proxy litellm values leave it disabled. BerriAI/litellm#36975 adds optional migrationJob.activeDeadlineSeconds with a default of 1800 seconds; no values change is required here.
• The ncecere/litellm Terraform provider remains independently pinned to ~> 2.0 in versions.tf. Provider additions in BerriAI/litellm#37918 are optional and do not couple the provider version to this proxy image update.
• v1.99.1 is a Docker-only release; this deployment uses the ghcr.io/berriai/litellm image, so that packaging distinction does not block promotion.

Release notes

• LiteLLM 1.100.0 release notes
• Full upstream comparison v1.97.0...v1.100.0
• Pre-production comparison v1.98.0...v1.100.0
• Latest stable LiteLLM 1.100.1 release notes (not this promotion target)
• Post-target comparison v1.100.0...v1.100.1

[github-actions]

---

Checkov Scan Success

Show Output

*****************************

Checkov will check the following folders:
terraform/environments/data-platform/ai-gateway

*****************************

Running Checkov in terraform/environments/data-platform/ai-gateway
terraform scan results:

Passed checks: 91, Failed checks: 0, Skipped checks: 0, Parsing errors: 1

secrets scan results:

Passed checks: 0, Failed checks: 0, Skipped checks: 1


checkov_exitcode=0

---

TFLint Scan Success

Show Output

*****************************

Using default config
Installing "terraform" plugin...
The plugin was signed using a legacy PGP signing key. Please update the plugin to the latest version
Installed "terraform" (source: github.com/terraform-linters/tflint-ruleset-terraform, version: 0.15.0)
tflint checking:
terraform/environments/data-platform/ai-gateway

*****************************

Running tflint in terraform/environments/data-platform/ai-gateway
tflint_exitcode=0