Worked on reliability hardening for the physical restore path in the percona/percona-backup-mongodb repository, focusing on backend development with Go. Addressed a critical issue where the restore process could silently hang during flush operations by implementing targeted error classification for non-recoverable shutdown scenarios. Developed a reusable helper function to centralize error handling, ensuring that fatal errors are detected and restoration halts promptly. This approach improved error visibility and reduced mean time to recovery, supporting clearer failure modes and better operator confidence. The work emphasized robust error handling and extensibility, laying groundwork for future enhancements in backup and restore reliability.