Improved the robustness in the node monitoring

- Improved logging and recovery for the Record thread that dies in the middle.
- Recover from failing to monitor one node and move on to the next.