Skip to main content

ZFS Gotchas

Operational knowledge from running ZFS in production.

Verification Status​

FieldValue
Last verified—
ZFS version—
OS—

Critical Knowledge​

Pools Cannot Shrink​

ZFS pools grow only. Vdevs cannot be removed (with limited mirror exceptions).

Implication: Plan topology before creation. Mirrors offer flexibility over raidz.

RAIDZ Expansion​

OpenZFS 2.3+ supports raidz expansion. Earlier versions do not.

Implication: Verify ZFS version before assuming expansion capability.

Dedup Requires Significant RAM​

Deduplication requires ~5GB RAM per TB of deduplicated data.

Recommendation: Use compression (lz4) instead. Nearly free and effective.

Scrubs Are Required​

ZFS detects silent corruption only on read or scrub.

Requirement: Monthly scrubs. Non-negotiable.

zpool scrub poolname
zpool status  # Check scrub state

ECC RAM​

ZFS checksums data. Memory corruption can corrupt data before checksum.

Recommendation: ECC RAM for data integrity. Not required, but strongly advised.

Configuration​

ARC Size​

ZFS uses available RAM for ARC. Can starve applications.

# /etc/modprobe.d/zfs.conf
options zfs zfs_arc_max=8589934592  # 8GB limit

Ashift​

Sector size must be correct at pool creation.

# Always specify ashift
zpool create -o ashift=12 poolname ...

Recordsize​

Default 128K inappropriate for databases.

# Database datasets
zfs set recordsize=16K poolname/database

Operations​

Send/Receive​

Large transfers can fail. Use resume tokens (ZFS 0.7+).

zfs send -t <token> | zfs receive ...

Snapshots vs Backups​

Snapshots are point-in-time recovery on same pool. Not backup.

Backup: Different physical media. Different location.

Current Configuration​

zpool status
zfs get compression,recordsize,atime poolname
DatasetRecordsizeCompressionPurpose
————

Revision History​

DateChange
—Initial document