#32019 [Bug]: s3_v2 success logger does not persist all anthropic_messages
### Check for existing issues
- [x] I have searched the existing issues and checked that my issue is not a duplicate.
### What happened?
**Bug Description**
After upgrading from **v1.85.1 β v1.90.0**, the `s3_v2` success logger stopped writing the large majority of **streaming** anthropic-messages responses to S3.
We use the Anthropic Messages API endpoint (`POST /v1/messages`, i.e. litellm's `anthropic_messages` handler β `call_type='anthropic_messages'`, `request_route='/v1/messages'`, not the `llm_passthrough` route) against Bedrock Anthropic models with `success_callback: ["s3_v2"]`. After the upgrade:
- **Non-streaming** requests (`stream=false`, routed to Bedrock `.../invoke`) are mostly logged to S3, but still show a small (~3%) loss. - **Streaming** requests (`stream=true`, routed to Bedrock `.../invoke-with-response-stream`) are almost never written to S3 β no object is created β even though the request succeeds, returns 200 to the client, and a spend-log row with `status='success'` and `call_type='anthropic_messages'` is written to `LiteLLM_SpendLogs`.
Reconciling one full day of traffic (`status='success'` spend-log rows vs objects in the s3 prefix), split by whetβ¦