ZFS, Actually Storage you can finally reason about.

Your First Pool

procedure
depth 2/3 − deeper + surface

Guides present zpool create—the command that builds a pool—as mere setup: type it, get storage, move on. Actually, that one command makes the only decisions in ZFS you can never revise: how the disks are teamed (mirrors or RAIDZ), and the disk cell-size setting (ashift). Everything else—compression, locations, space limits, datasets—can be changed over morning coffee. The discipline is knowing which decisions are the foundation pour and giving them the ten deliberate minutes they deserve.

⚠ DESTRUCTIVE—zpool create wipes and takes over the disks you name. One mistyped disk name destroys that disk's contents. Verify with ls -l /dev/disk/by-id/ before running.

The shape of the command, judgments annotated:

zpool create -o ashift=12 \
  -O compression=lz4 -O atime=off \
  tank mirror ata-DISK1-SERIAL ata-DISK2-SERIAL
  1. Vdev shape (mirror ... / raidz2 ...)—permanent. Random-I/O workloads → mirrors; bulk sequential → RAIDZ2 (the whole argument lives in Mirrors and RAIDZ).
  2. ashift=12—permanent; don't let ZFS trust a lying drive ("set at vdev creation and it is immutable"—OpenZFS docs).
  3. By-id device names—not permanent, but the cheapest gift to the future you who'll someday map a dead disk to a physical serial (see Replacing a Disk).
  4. -O compression=lz4 (and atime=off—write amplification for a timestamp nobody reads)—changeable later, but properties only touch new writes, so setting them at birth is the one time they cover everything.

Then immediately make datasets—the pool root is a policy anchor, not a place to put files: zfs create tank/services, and policy boundaries per thing you'll snapshot or tune separately.