fix(proxy): backport team member spend jsonb flush to rc/1.103.0 (#43029)

[yuneng-berri]

TLDR

Problem this solves:

• After one all-$0 member spend flush, later member spend never lands
• Team, key and user spend keep rising while member rows freeze

How it solves it:

• Cherry-picks fix(proxy): pass team member spend rows as jsonb so a $0 flush cannot poison the pool connection #43029 onto rc/1.103.0 unchanged
• Member rows travel as one jsonb document typed in SQL

User Flow

Before: a team that mixes a free model with a paid one sees every member's spend frozen at $0 while the team total keeps growing

1. The admin creates a team with a $0 custom-priced model and a paid model, adds two users and generates a key for each
2. The free-model user sends POST http://localhost:4000/v1/chat/completions with the free model and gets 200
3. The paid-model user sends POST http://localhost:4000/v1/chat/completions with the paid model and gets 200
4. GET http://localhost:4000/team/info?team_id=... shows team spend above 0 but that member still at 0.0

After: the same requests credit the member row alongside the team total

1. The admin sets up the same team, models, users and keys
2. The free-model user sends the same request and gets 200
3. The paid-model user sends the same request and gets 200
4. GET http://localhost:4000/team/info?team_id=... shows the paid member's spend matching the team's increase

Relevant issues

Backport of #43029

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have added meaningful tests
• The handful of test files covering my change pass locally, e.g. uv run pytest tests/test_litellm/<your_test_file>.py -v. Leave the suites (make test-unit-*, make test-unit) to CI: it finishes in ~15 minutes where a laptop takes an hour or more
• My PR passes all required CI/CD checks (e.g., lint, schema.d.ts sync check, etc.)
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review (Greptile reviews automatically once the PR is opened; only comment @greptileai to re-request a review after pushing changes)

The cherry-pick applied cleanly with no conflicts, and test_db_spend_update_writer.py passes on this branch (134 tests). The tests are the ones #43029 added on main

Type

🐛 Bug Fix

Caveats (if any)

Medium

• Live proxy proof on rc/1.103.0 is not captured yet
  • The new integration test needs a real Postgres and was not run locally

[greptile-apps]

Confidence Score: 5/5

[Critical risk] Changes how team member spending data is written to the database.

The PR appears safe to merge based on the reviewed changes.

Summary

The PR changes team-member spend flushes to send one JSONB document instead of parallel arrays, addressing fractional spend writes after a zero-cost flush.

• Updates unit assertions for the new payload.
• Adds a single-connection database integration test for the reported sequence.

_{Reviews (1) · Last reviewed commit: "fix(proxy): pass team member spend rows ..."}