Piotr Oleszczyk
092fd87606
When Gemini stops generation early (e.g. due to safety filters or thinking-model quirks), finish_reason != STOP but no exception is raised, causing the caller to receive truncated JSON and a confusing 502 "invalid JSON" error. Now: - finish_reason is extracted from candidates[0] and stored in ai_call_logs - any non-STOP finish_reason raises HTTP 502 with a clear message - Alembic migration adds the finish_reason column to ai_call_logs Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
102 lines
3.4 KiB
Python
102 lines
3.4 KiB
Python
"""Shared helpers for Gemini API access."""
|
|
|
|
import os
|
|
import time
|
|
from contextlib import suppress
|
|
|
|
from fastapi import HTTPException
|
|
from google import genai
|
|
from google.genai import types as genai_types
|
|
|
|
_DEFAULT_MODEL = "gemini-flash-latest"
|
|
|
|
|
|
def get_gemini_client() -> tuple[genai.Client, str]:
|
|
"""Return an authenticated Gemini client and the configured model name.
|
|
|
|
Raises HTTP 503 if GEMINI_API_KEY is not set.
|
|
"""
|
|
api_key = os.environ.get("GEMINI_API_KEY")
|
|
if not api_key:
|
|
raise HTTPException(status_code=503, detail="GEMINI_API_KEY not configured")
|
|
model = os.environ.get("GEMINI_MODEL", _DEFAULT_MODEL)
|
|
return genai.Client(api_key=api_key), model
|
|
|
|
|
|
def call_gemini(
|
|
*,
|
|
endpoint: str,
|
|
contents,
|
|
config: genai_types.GenerateContentConfig,
|
|
user_input: str | None = None,
|
|
):
|
|
"""Call Gemini, log full request + response to DB, return response unchanged."""
|
|
from sqlmodel import Session
|
|
|
|
from db import engine # deferred to avoid circular import at module load
|
|
from innercontext.models.ai_log import AICallLog
|
|
|
|
client, model = get_gemini_client()
|
|
|
|
sys_prompt = None
|
|
if config.system_instruction:
|
|
raw = config.system_instruction
|
|
sys_prompt = raw if isinstance(raw, str) else str(raw)
|
|
if user_input is None:
|
|
with suppress(Exception):
|
|
user_input = str(contents)
|
|
|
|
start = time.monotonic()
|
|
success, error_detail, response, finish_reason = True, None, None, None
|
|
try:
|
|
response = client.models.generate_content(
|
|
model=model, contents=contents, config=config
|
|
)
|
|
with suppress(Exception):
|
|
finish_reason = response.candidates[0].finish_reason.name
|
|
if finish_reason and finish_reason != "STOP":
|
|
success = False
|
|
error_detail = f"finish_reason: {finish_reason}"
|
|
raise HTTPException(
|
|
status_code=502,
|
|
detail=f"Gemini stopped early (finish_reason={finish_reason})",
|
|
)
|
|
except HTTPException:
|
|
raise
|
|
except Exception as exc:
|
|
success = False
|
|
error_detail = str(exc)
|
|
raise HTTPException(status_code=502, detail=f"Gemini API error: {exc}") from exc
|
|
finally:
|
|
duration_ms = int((time.monotonic() - start) * 1000)
|
|
with suppress(Exception):
|
|
log = AICallLog(
|
|
endpoint=endpoint,
|
|
model=model,
|
|
system_prompt=sys_prompt,
|
|
user_input=user_input,
|
|
response_text=response.text if response else None,
|
|
prompt_tokens=(
|
|
response.usage_metadata.prompt_token_count
|
|
if response and response.usage_metadata
|
|
else None
|
|
),
|
|
completion_tokens=(
|
|
response.usage_metadata.candidates_token_count
|
|
if response and response.usage_metadata
|
|
else None
|
|
),
|
|
total_tokens=(
|
|
response.usage_metadata.total_token_count
|
|
if response and response.usage_metadata
|
|
else None
|
|
),
|
|
duration_ms=duration_ms,
|
|
finish_reason=finish_reason,
|
|
success=success,
|
|
error_detail=error_detail,
|
|
)
|
|
with Session(engine) as s:
|
|
s.add(log)
|
|
s.commit()
|
|
return response
|