Bug: Proxy plugin incorrectly parses API parameters into empty strings

[touale]

The proxy plugin incorrectly parses a valid cloud API request payload and converts structured objects into empty strings.

Expected Behavior

Given the following request body:

{
  "dimensions": {
    "dim_products": {
      "query_vector_text": "string",
      "weight": 0
    },
    "dim_partners": {
      "query_vector_text": "string",
      "weight": 0
    }
  },
  "global_filters": {
    "provinces": ["string"],
    "cities": ["string"],
    "is_listed": 0
  },
  "top_k": 0
}

The proxy should forward the payload without modifying its structure.

Actual Behavior

The proxy plugin incorrectly parses the payload as:

{
  "dimensions": "",
  "global_filters": "",
  "top_k": 0
}

Both dimensions and global_filters are reduced to empty strings, causing the downstream service to fail.

[touale]

Root Cause Analysis

This issue is caused by the use of Union[..., None] in the Pydantic model definitions.

Relevant code:

from pydantic import BaseModel


class DimensionItem(BaseModel):
    query_vector_text: str
    weight: float


class Dimensions(BaseModel):
    dim_products: DimensionItem | None = None
    dim_partners: DimensionItem | None = None


class GlobalFilters(BaseModel):
    provinces: list[str] | str | None = None
    cities: list[str] | str | None = None
    is_listed: int | None = None


class FetchCompanyByDimInfoRequest(BaseModel):
    dimensions: Dimensions
    global_filters: GlobalFilters | None = None
    top_k: int | None = None

Specifically:

• dim_products: DimensionItem | None
• dim_partners: DimensionItem | None

introduce None into a Union type.
The proxy plugin appears to mishandle these optional nested models and serializes them as empty strings instead of structured objects.