improve session resilience

[joshadambell]

Summary

Adds session resilience: automatic session recovery on backend and stream reconnection on frontend for seamless experience when sessions are deleted or pages reload during active tasks.

• Backend session recovery: Automatically recreates sessions on 404 errors during event append, with in-flight task detection
• Frontend stream resumption: Reconnects to in-flight tasks (working/submitted state) when page is reloaded
• Bug fix: Session events now properly populated on load (events=events instead of events=[])

What Changed

Backend (python/packages/kagent-adk)

• _session_service.py: New _recreate_session() method handles 404 during append_event() — recreates the session preserving ID/user/agent, handles 409 conflicts from concurrent recreation, fetches and logs in-flight tasks after recovery. Single retry prevents infinite recursion.
• _session_service.py: Bug fix — get_session() now returns events=events instead of empty list
• test_session_service.py: 9 new tests covering append 404 recovery, session recreation, conflict handling, and in-flight task detection

Frontend (ui/src)

• a2aClient.ts: New resubscribeTask() method sends tasks/resubscribe JSON-RPC to resume SSE streams
• messageHandlers.ts: extractMessagesFromTasks() now returns TaskExtractionResult with optional pendingTask for in-flight task detection
• ChatInterface.tsx: On page load, detects pending tasks and auto-resubscribes to their streams. Properly cleans up abort controller on unmount. HITL approval takes priority over resubscribe when both apply.
• AgentCallDisplay.tsx: Updated for new extractMessagesFromTasks return type

Tests

• a2aClient.test.ts: New test suite for resubscribeTask() and existing client methods
• messageHandlers.test.ts: New tests for pending task detection (working, submitted, completed states)

How It Works

1. Session deleted during active task → Backend 404 handler recreates session, logs in-flight tasks
2. Page reload during active task → Frontend detects pending task via extractMessagesFromTasks
3. Frontend calls resubscribeTask() → Reconnects to SSE stream for the in-flight task
4. Stream resumes → User sees continued updates without interruption
5. If HITL approval is pending → Approval takes priority over resubscribe (gated by !hasPendingApproval)

[Copilot]

Pull request overview

This pull request implements session resilience features to automatically recover from deleted sessions and reconnect to in-flight tasks after page reloads.

Changes:

• Backend session recovery: Added _recreate_session() method in _session_service.py that recreates deleted sessions and detects in-flight tasks, with 404 handling in append_event() to trigger recreation before retrying
• Frontend stream reconnection: Implemented resubscribeTask() method and enhanced extractMessagesFromTasks() to detect pending tasks (working/submitted states), with auto-resubscription on page load in ChatInterface
• Bug fix: Corrected get_session() to populate events (changed events=[] to events=events)

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
python/packages/kagent-adk/src/kagent/adk/_session_service.py	Added session recreation logic and 404 handling; fixed events population bug
python/packages/kagent-adk/tests/unittests/test_session_service.py	Complete test rewrite with comprehensive coverage for append_event recovery and session recreation
ui/src/lib/messageHandlers.ts	Enhanced to return TaskExtractionResult with pending task detection
ui/src/lib/a2aClient.ts	Added resubscribeTask() method and updated request type parameters
ui/src/lib/__tests__/messageHandlers.test.ts	Updated tests for new return type and added pending task detection tests
ui/src/lib/__tests__/a2aClient.test.ts	New comprehensive test suite for A2A client methods
ui/src/components/chat/ChatInterface.tsx	Added pending task detection and auto-resubscription logic with proper cleanup
ui/src/components/chat/AgentCallDisplay.tsx	Updated to use new extractMessagesFromTasks return type

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The bug fix in get_session() changing events=[] to events=events lacks test coverage. While new tests were added for the append_event recovery logic, the tests that previously validated event loading in get_session were removed entirely. There should be at least one test ensuring that events returned from the API are properly populated in the session object.

[Copilot]

The pendingTask return type uses state: string which is less precise than needed. Since the state is guaranteed to be either 'working' or 'submitted' (from the conditions on lines 22-23), consider typing this as { taskId: string; state: 'working' | 'submitted' } or { taskId: string; state: TaskState } for better type safety. This would eliminate the need for the as TaskState cast on line 162 of ChatInterface.tsx.