Monitoring and Maintenance

Running bcachefs day-to-day involves the usual monitoring you'd do for any filesystem. Check dmesg or syslog for bcachefs errors because the kernel will log if it encounters checksum mismatches or I/O errors on a device. Use the bcachefs fsck in read-only mode to scrub the filesystem periodically. Especially if you have replication similar to a ZFS scrub or Btrfs scrub, this will read all data and verify checksums – bcachefs can then repair any mismatches by using the redundant copy. The bcachefs user tools also provide commands such as bcachefs device stats to check device health within the filesystem and bcachefs attach/detach to manage devices.

On Ubuntu systems, ensure you have the latest bcachefs-tools package if you're using a newer kernel than the default. Sometimes the userspace tools need to closely match the kernel module version. If you compiled your own kernel, you may also need to compile the tools from source to get the absolute latest features or bug fixes. It's also wise to run with CONFIG_BCACHEFS_DEBUG disabled in production kernels because the upstream code has debug options that, when off, give better performance; distribution kernels will have these appropriately set for release builds.

Backup Strategy

Until bcachefs proves itself over more time, a conservative admin will treat it like any "RAID is not backup" scenario – do regular backups and/or use it initially for data that is reproducible or not mission-critical. Bcachefs does support send/receive in a rudimentary form to export sub-volumes or snapshots indirectly, though it's not as polished as ZFS send or Btrfs send yet. You might instead rely on higher-level backup tools (rsync, Borg, zfs-send via a different interim filesystem, etc.) to keep copies of data stored on bcachefs. The presence of snapshots, however, means you can make consistent backups even of Live data: Take a snapshot and then back that snapshot up without worrying about the data changing underneath.

Conclusion

Bcachefs represents the most ambitious new Linux filesystem in years, and it comes at a time when storage needs are as demanding as ever. It's pragmatic and administrator-friendly in design, borrowing the best ideas from predecessors and adding its own innovations (such as configuration-less tiering).

The Author

Marcin Gastol is a Microsoft MVP, Microsoft Certified Trainer, and conference speaker. He works as a senior DevOps engineer and has extensive experience in Azure technologies. Visit his blog at https://marcingastol.com/.