mirror of
https://github.com/BerriAI/litellm.git
synced 2026-09-30 01:52:18 +00:00
The route resolved every declared provider secret before preparing the call, so a caller that passed api_key and api_base still hit the secret manager for all four names, paying its latency and failing the call if a read errored. Python reads each secret lazily through `api_key or get_secret_str(...)`. resolve_on_demand in litellm-secrets now runs the pure prepare step against the secrets fetched so far and fetches only the first name it asks for that is not yet known, repeating until nothing is missing. Reads happen in the same order and with the same short circuits as Python, and the per-config secret_names list is gone since nothing needs to be declared up front |
||
|---|---|---|
| .. | ||
| crates | ||
| .gitignore | ||
| Cargo.lock | ||
| Cargo.toml | ||
| clippy.toml | ||