← All notes

For about six years the data on the home server lived on a plain mdadm RAID1 with ext4 on top. It never lost anything, but it also never told me whether anything had quietly rotted. After reading one too many stories about silent corruption, I decided it was time to try ZFS properly instead of just in a virtual machine.

I bought two new 8 TB drives, burned them in with a full badblocks pass (which took most of two days), and created a simple mirror. The datasets are split by purpose: one for photos, one for documents, one for container volumes and one for scratch space with compression turned off. Everything else uses lz4, which costs almost nothing on this CPU.

The copy from the old array was done with rsync in a few passes, the last one with services stopped. I checked the result with a checksum comparison of the photo library, mostly for peace of mind. Then I set up a small cron job for automatic snapshots: hourly for a day, daily for a month, monthly for a year.

A monthly scrub now runs on the first Sunday of the month. The first one finished in about five hours and found nothing, which is a nice feeling. The old drives are wiped and sitting in a drawer as a cold spare.