Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
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
4 changes: 4 additions & 0 deletions .jules/bolt.md
Original file line number Diff line number Diff line change
@@ -1,3 +1,7 @@
## 2025-05-15 - [Sequential session destruction in SDKs]
**Learning:** All Copilot SDKs (Node.js, Python, Go, .NET) were initially implementing session destruction sequentially during client shutdown. This leads to a linear increase in shutdown time as the number of active sessions grows, especially when individual destructions involve retries and backoff.
**Action:** Parallelize session cleanup using language-specific concurrency primitives (e.g., `Promise.all` in Node.js, `asyncio.gather` in Python, `Task.WhenAll` in .NET, or WaitGroups/Channels in Go) to ensure shutdown time remains constant and minimal.

## 2025-05-16 - [Sequential session destruction in Go and .NET SDKs]
**Learning:** While Node.js and Python SDKs have been optimized for parallel session destruction, Go and .NET SDKs still implement this sequentially. This leads to a linear increase in shutdown time as the number of active sessions grows in those languages.
**Action:** Parallelize session cleanup in Go using goroutines/WaitGroups and in .NET using Task.WhenAll to ensure consistent O(T) shutdown time across all SDKs.
34 changes: 27 additions & 7 deletions python/copilot/client.py
Original file line number Diff line number Diff line change
Expand Up @@ -313,13 +313,33 @@ async def stop(self) -> list["StopError"]:
sessions_to_destroy = list(self._sessions.values())
self._sessions.clear()

for session in sessions_to_destroy:
try:
await session.destroy()
except Exception as e:
errors.append(
StopError(message=f"Failed to destroy session {session.session_id}: {e}")
)
async def destroy_with_retry(session: CopilotSession) -> Optional[StopError]:
last_error = None
# Try up to 3 times with exponential backoff
for attempt in range(1, 4):
try:
await session.destroy()
return None
except Exception as e:
last_error = e
if attempt < 3:
# Exponential backoff: 100ms, 200ms
delay = 0.1 * (2 ** (attempt - 1))
await asyncio.sleep(delay)

msg = f"Failed to destroy session {session.session_id}"
msg += f" after 3 attempts: {last_error}"
return StopError(message=msg)

if sessions_to_destroy:
# Parallelize session destruction to ensure O(T) shutdown time
Copy link

Copilot AI Feb 10, 2026

Choose a reason for hiding this comment

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

The comment claiming this “ensures O(T) shutdown time” is misleading: there’s still O(N) overhead to schedule tasks and send N JSON-RPC requests, and JsonRpcClient serializes writes via _write_lock. Consider rewording to something like “reduces wall-clock shutdown time by running destroys concurrently (bounded by the slowest destroy)”.

Suggested change
# Parallelize session destruction to ensure O(T) shutdown time
# Parallelize session destruction to reduce wall-clock shutdown time,
# with overall duration largely bounded by the slowest destroy.

Copilot uses AI. Check for mistakes.
results = await asyncio.gather(
*[destroy_with_retry(s) for s in sessions_to_destroy],
Copy link

Copilot AI Feb 10, 2026

Choose a reason for hiding this comment

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

Using a list comprehension here forces building an intermediate list before argument expansion. For large session counts, it’s slightly more memory/overhead than necessary; consider passing a generator (or using map) directly into asyncio.gather.

Suggested change
*[destroy_with_retry(s) for s in sessions_to_destroy],
*(destroy_with_retry(s) for s in sessions_to_destroy),

Copilot uses AI. Check for mistakes.
return_exceptions=False,
)
for result in results:
if result:
errors.append(result)

# Close client
if self._client:
Expand Down
Loading