Skip to content

[Fix] writer 노드 종료시 줄바꿈 #106

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Ncs89 wants to merge 1 commit into
base: dev
Choose a base branch
from
Open

[Fix] writer 노드 종료시 줄바꿈 #106

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
32 changes: 32 additions & 0 deletions src/service/service.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -104,6 +104,12 @@
"CreditInsufficient": "system_notice",
"Fallback": "system_notice",
}
# LLM 토큰 스트리밍 완료 후 줄바꿈(\n\n)을 추가로 emit해야 하는 노드 목록
# (Writer 노드 등 feature 응답 완료 후 구분선 역할)
STREAMING_TRAILING_NEWLINE_NODES: set[str] = {
"Solve_Writer",
"Explain",
}
VISIBLE_NODES: set[str] = set(PROGRESS_MESSAGES.keys()) | {
"FinalResponse",
"Simple_response",
Expand Down Expand Up @@ -379,6 +385,8 @@ async def message_generator(

progress_messages = PROGRESS_MESSAGES
emitted_progress_nodes: set[str] = set()
# 스트리밍 노드 완료 후 줄바꿈 emit 중복 방지
streaming_newline_emitted: set[str] = set()

def emit_progress(node_name: str) -> str | None:
if node_name in emitted_progress_nodes:
Expand Down Expand Up @@ -432,6 +440,18 @@ def emit_progress(node_name: str) -> str | None:
if progress_line is not None:
yield progress_line # type: ignore[misc]

# 스트리밍 노드 완료 후 줄바꿈 토큰 emit
if (
user_input.stream_tokens
and node_name in STREAMING_TRAILING_NEWLINE_NODES
and node_name not in streaming_newline_emitted
):
streaming_newline_emitted.add(node_name)
newline_payload = json.dumps(
{"type": "token", "content": "\n\n"}, ensure_ascii=False
)
yield f"data: {newline_payload}\n\n"
Comment on lines +443 to +453
Copy link
Copy Markdown

@coderabbitai coderabbitai Bot Feb 18, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

⚠️ Potential issue | 🟡 Minor

🧩 Analysis chain

🏁 Script executed:

rg -n "skip_stream" --type py -C 3

Repository: Team-Proovy/Proovy-ai

Length of output: 10802

🏁 Script executed:

rg -n "STREAMING_TRAILING_NEWLINE_NODES" --type py -C 3

Repository: Team-Proovy/Proovy-ai

Length of output: 1515

🏁 Script executed:

rg -n "Solve_Writer\|Explain" --type py -C 2

Repository: Team-Proovy/Proovy-ai

Length of output: 47

🏁 Script executed:

# Check the context around line 548 for stream_mode == "messages" handler
sed -n '540,560p' src/service/service.py

Repository: Team-Proovy/Proovy-ai

Length of output: 1151

🏁 Script executed:

# Check lines 443-453 in detail along with surrounding context
sed -n '430,465p' src/service/service.py

Repository: Team-Proovy/Proovy-ai

Length of output: 1613

🏁 Script executed:

# Search for on_chat_model_end handler mentioned in the comment
rg -n "on_chat_model_end" --type py -C 5

Repository: Team-Proovy/Proovy-ai

Length of output: 740

🏁 Script executed:

fd -t f "\.py$" | xargs rg -l "\"Explain\".*node|add_node.*Explain" | head -10

Repository: Team-Proovy/Proovy-ai

Length of output: 127

🏁 Script executed:

rg -n "def.*[Ee]xplain|\"Explain\"" --type py -C 2 | head -50

Repository: Team-Proovy/Proovy-ai

Length of output: 2702

🏁 Script executed:

# Search for Explain node in feature graphs
rg -n "Explain" src/agents/workflows/subgraphs/step4_features/ --type py -C 2

Repository: Team-Proovy/Proovy-ai

Length of output: 2647

v1 경로의 줄바꿈 emit 로직이 skip_stream 태그를 확인하지 않아, v2 경로와 비일관적입니다.

확인 결과 STREAMING_TRAILING_NEWLINE_NODES에 포함된 Solve_Writer (solve/graph.py:428)와 Explain (explain/graph.py:100) 노드 모두 skip_stream 태그 없이 구성되어 있습니다. 따라서 현재는 실제 문제가 발생하지 않습니다.

그러나 v2 경로 (on_chat_model_end 핸들러, 1078-1082줄)는 명시적으로 if "skip_stream" in tag_list: continue를 확인하는 반면, v1 경로 (443-453줄)는 이 검사가 없습니다. 향후 이 노드들에 skip_stream을 추가할 경우 v1 경로에서만 불필요한 줄바꿈이 emit될 수 있으므로, v1 경로에도 동일한 skip_stream 확인 로직을 추가하는 것이 좋습니다.

🤖 Prompt for AI Agents 
Verify each finding against the current code and only fix it if needed.

In `@src/service/service.py` around lines 443 - 453, The v1 trailing-newline emit
block (the branch using STREAMING_TRAILING_NEWLINE_NODES,
streaming_newline_emitted, user_input.stream_tokens and node_name) lacks the
same "skip_stream" tag check present in the v2 on_chat_model_end path; update
this block to first inspect the node's tag list (e.g., the node's tags or
tag_list used elsewhere) and skip emitting the newline if "skip_stream" is
present, so the logic matches the v2 handler and avoids emitting trailing
newlines for nodes marked with skip_stream.


feature_nodes = {
"Solve",
"Explain",
Expand Down Expand Up @@ -1061,6 +1081,18 @@ async def _pump_stream(queue: asyncio.Queue[tuple[str, Any]]) -> None:
if "skip_stream" in tag_list:
continue
if active_llm_message_id is not None:
# Writer 노드 등 스트리밍 feature 응답 완료 후 줄바꿈 추가
if active_llm_node in STREAMING_TRAILING_NEWLINE_NODES:
yield emitter.emit(
"llm.token.delta",
{
"message_id": active_llm_message_id,
"node": active_llm_node,
"delta": "\n\n",
"index": active_llm_token_index,
},
)
active_llm_token_index += 1
Comment on lines +1084 to +1095
Copy link
Copy Markdown

@coderabbitai coderabbitai Bot Feb 18, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

⚠️ Potential issue | 🟠 Major

v2 경로에 v1의 streaming_newline_emitted 에 해당하는 중복 방지 로직이 없습니다.

v1 경로는 streaming_newline_emitted 집합을 통해 요청당 노드당 정확히 1회만 줄바꿈을 emit합니다. 반면 v2 경로는 on_chat_model_end 가 발생할 때마다 active_llm_node in STREAMING_TRAILING_NEWLINE_NODES 이면 줄바꿈을 emit하므로, 동일 노드 내에서 LLM 호출이 순차적으로 N회 발생하면 N개의 줄바꿈이 클라이언트에 전달됩니다.

active_llm_message_id = None 리셋은 동일 LLM 호출의 중복 emit만 방지하며, 두 번째 on_chat_model_start → on_chat_model_end 쌍이 오면 새로운 active_llm_message_id 로 다시 emit이 허용됩니다.

🐛 제안 수정: v2에 deduplication 집합 추가

message_generator_v2 상단 변수 초기화 블록에 집합 추가:

 emitted_artifacts: set[str] = set()
 last_credit_signature: tuple[Any, Any, Any] | None = None
+streaming_newline_emitted: set[str] = set()

on_chat_model_end 핸들러 내 줄바꿈 emit 조건에 중복 방지 체크 추가:

-                    if active_llm_node in STREAMING_TRAILING_NEWLINE_NODES:
+                    if (
+                        active_llm_node in STREAMING_TRAILING_NEWLINE_NODES
+                        and active_llm_node not in streaming_newline_emitted
+                    ):
+                        streaming_newline_emitted.add(active_llm_node)
                         yield emitter.emit(
                             "llm.token.delta",
🤖 Prompt for AI Agents 
Verify each finding against the current code and only fix it if needed.

In `@src/service/service.py` around lines 1084 - 1095, Add v1-style deduplication
to the v2 path by introducing a per-request set (e.g.,
streaming_newline_emitted) in message_generator_v2's top-level state and update
the on_chat_model_end handler to only emit the trailing newline for a given
(active_llm_message_id, active_llm_node) once: check (active_llm_message_id,
active_llm_node) against streaming_newline_emitted before emitting to
STREAMING_TRAILING_NEWLINE_NODES, add the tuple to the set after emitting, and
ensure the set is reset appropriately when a new request lifecycle starts so
duplicates across sequential on_chat_model_start/on_chat_model_end pairs are
prevented.

yield emitter.emit(
"llm.message.completed",
{
Expand Down