woongmin commited on
Commit
37c1245
·
1 Parent(s): 62bd0bb

[Delta-02] Drill C retry/backoff 대응 코드 추가

Browse files
Files changed (2) hide show
  1. app/embedding_utils.py +7 -5
  2. app/main.py +25 -5
app/embedding_utils.py CHANGED
@@ -1,7 +1,7 @@
1
  from typing import List
2
  import nltk
3
  import numpy as np
4
- from openai import OpenAI, RateLimitError
5
  from sentence_transformers import CrossEncoder
6
  from tenacity import retry, wait_exponential, stop_after_attempt, retry_if_exception
7
  from app.config import OPENAI_API_KEY, OPENAI_EMBEDDING_MODEL
@@ -41,17 +41,19 @@ def _simple_split_by_chars(
41
  return chunks
42
 
43
 
44
- def _is_rate_limit_retryable(e: Exception) -> bool:
45
- """insufficient_quota는 재시도 불필요, rate_limit_exceeded만 재시도"""
 
 
46
  if isinstance(e, RateLimitError):
47
  return e.code != "insufficient_quota"
48
  return False
49
 
50
 
51
  @retry(
52
- retry=retry_if_exception(_is_rate_limit_retryable),
53
  wait=wait_exponential(multiplier=1, min=2, max=30),
54
- stop=stop_after_attempt(2),
55
  )
56
  def _openai_embed(text: str) -> List[float]:
57
  response = client.embeddings.create(
 
1
  from typing import List
2
  import nltk
3
  import numpy as np
4
+ from openai import OpenAI, RateLimitError, APITimeoutError, APIConnectionError
5
  from sentence_transformers import CrossEncoder
6
  from tenacity import retry, wait_exponential, stop_after_attempt, retry_if_exception
7
  from app.config import OPENAI_API_KEY, OPENAI_EMBEDDING_MODEL
 
41
  return chunks
42
 
43
 
44
+ def _is_transient_error(e: Exception) -> bool:
45
+ """타임아웃·연결 오류·rate limit(quota 부족 제외)을 재시도 대상으로 판단"""
46
+ if isinstance(e, (APITimeoutError, APIConnectionError)):
47
+ return True
48
  if isinstance(e, RateLimitError):
49
  return e.code != "insufficient_quota"
50
  return False
51
 
52
 
53
  @retry(
54
+ retry=retry_if_exception(_is_transient_error),
55
  wait=wait_exponential(multiplier=1, min=2, max=30),
56
+ stop=stop_after_attempt(3),
57
  )
58
  def _openai_embed(text: str) -> List[float]:
59
  response = client.embeddings.create(
app/main.py CHANGED
@@ -25,7 +25,13 @@ from .config import (
25
  )
26
 
27
  import pdfplumber
28
- from openai import OpenAI, RateLimitError, AuthenticationError
 
 
 
 
 
 
29
  from tenacity import retry, wait_exponential, stop_after_attempt, retry_if_exception
30
  import hashlib
31
  import time
@@ -42,17 +48,19 @@ async def health_check():
42
  return {"status": "healthy"}
43
 
44
 
45
- def _is_rate_limit_retryable(e: Exception) -> bool:
46
- """insufficient_quota는 재시도 불필요, rate_limit_exceeded만 재시도"""
 
 
47
  if isinstance(e, RateLimitError):
48
  return e.code != "insufficient_quota"
49
  return False
50
 
51
 
52
  @retry(
53
- retry=retry_if_exception(_is_rate_limit_retryable),
54
  wait=wait_exponential(multiplier=1, min=2, max=30),
55
- stop=stop_after_attempt(2),
56
  )
57
  def _call_llm(messages: list, model: str) -> str:
58
  response = client.chat.completions.create(model=model, messages=messages)
@@ -239,6 +247,12 @@ async def upload_document(file: UploadFile = File(...)):
239
  status_code=500,
240
  detail="서버 설정 오류가 발생했습니다. 관리자에게 문의해주세요.",
241
  )
 
 
 
 
 
 
242
  except Exception as e:
243
  print(f"오류 발생: {str(e)}")
244
  raise HTTPException(status_code=500, detail=str(e))
@@ -374,6 +388,12 @@ async def question_answering(question: Dict[str, str] = Body(...)):
374
  status_code=500,
375
  detail="서버 설정 오류가 발생했습니다. 관리자에게 문의해주세요.",
376
  )
 
 
 
 
 
 
377
  except RateLimitError as e:
378
  if e.code == "insufficient_quota":
379
  raise HTTPException(
 
25
  )
26
 
27
  import pdfplumber
28
+ from openai import (
29
+ OpenAI,
30
+ RateLimitError,
31
+ AuthenticationError,
32
+ APITimeoutError,
33
+ APIConnectionError,
34
+ )
35
  from tenacity import retry, wait_exponential, stop_after_attempt, retry_if_exception
36
  import hashlib
37
  import time
 
48
  return {"status": "healthy"}
49
 
50
 
51
+ def _is_transient_error(e: Exception) -> bool:
52
+ """타임아웃·연결 오류·rate limit(quota 부족 제외)을 재시도 대상으로 판단"""
53
+ if isinstance(e, (APITimeoutError, APIConnectionError)):
54
+ return True
55
  if isinstance(e, RateLimitError):
56
  return e.code != "insufficient_quota"
57
  return False
58
 
59
 
60
  @retry(
61
+ retry=retry_if_exception(_is_transient_error),
62
  wait=wait_exponential(multiplier=1, min=2, max=30),
63
+ stop=stop_after_attempt(3),
64
  )
65
  def _call_llm(messages: list, model: str) -> str:
66
  response = client.chat.completions.create(model=model, messages=messages)
 
247
  status_code=500,
248
  detail="서버 설정 오류가 발생했습니다. 관리자에게 문의해주세요.",
249
  )
250
+ except APITimeoutError:
251
+ print("오류 발생: OpenAI API 요청 시간 초과")
252
+ raise HTTPException(
253
+ status_code=504,
254
+ detail="요청 처리 시간이 초과되었습니다. 잠시 후 다시 시도해주세요.",
255
+ )
256
  except Exception as e:
257
  print(f"오류 발생: {str(e)}")
258
  raise HTTPException(status_code=500, detail=str(e))
 
388
  status_code=500,
389
  detail="서버 설정 오류가 발생했습니다. 관리자에게 문의해주세요.",
390
  )
391
+ except APITimeoutError:
392
+ print("오류 발생: OpenAI API 요청 시간 초과")
393
+ raise HTTPException(
394
+ status_code=504,
395
+ detail="요청 처리 시간이 초과되었습니다. 잠시 후 다시 시도해주세요.",
396
+ )
397
  except RateLimitError as e:
398
  if e.code == "insufficient_quota":
399
  raise HTTPException(