ZFS, Actually Storage you can finally reason about.

Reading zpool status

procedure
depth 3/3 − deeper + surface
A disk just died 1/4 ← prev next →

zpool status is the command that shows the pool's health report. Scary capitalized words in a report mean crisis in most of computing, so DEGRADED reads as "half-broken—touching it might tip it over." Actually, DEGRADED means the design is working: a disk failed, the spare copies absorbed it, and the pool is running normally with less safety margin than before. The report itself is a calm document—what state things are in, what caused it, and almost always an explicit recommended next step, written right there in an action: line. It tells you what to do; the skill is believing it.

The reading order:

  1. state:—ONLINE (healthy), DEGRADED (redundancy absorbed something; act deliberately), FAULTED/UNAVAIL (pool-level trouble—now it's serious).
  2. status: and action:—plain-English diagnosis and the recommended move. Start there, not the forums.
  3. The device tree—indentation is the vdev structure (this is also your layout audit—a bare disk at vdev level is a single point of failure hiding in plain sight). Each line: state plus READ / WRITE / CKSUM error counts.
  4. The scan: line—last scrub/resilver and its result.

Reading the error columns: READ/WRITE are I/O failures (dying disk, or the perennially guilty cable); CKSUM means the device returned wrong data—silent corruption, caught. A handful, once, cleared with zpool clear and watched: reasonable. A count that climbs after clearing: that disk (or cable) is telling you something—swap the cheap cable before condemning the expensive disk.

What the panic-model gets wrong, structurally: DEGRADED is not an unstable state sliding toward loss—it's a stable state with reduced margin. The correct response is neither speed nor paralysis but deliberateness: the real risks in a degraded pool are precision risks (replacing the wrong disk—see Replacing a Disk) and window risks (a second failure during the resilver—see Resilvering). Rushing serves the first risk; freezing serves the second.

One habit upgrade: don't wait for a bad day to read this output. zpool status on a healthy pool, regularly—or wired into monitoring (the -x flag exists exactly for scripts: silence when all pools are healthy)—is how DEGRADED becomes a Tuesday instead of an event.