ZFS Troubleshooting Boilerplate

Operating Systems & Open Source
1

Couple of ZFS troubleshooting things:

  • make sure ashift = 12. if it isn’t later model drives won’t be matching logical and physical cluster size
  • unless you’ve tested to confirm, de-dupe probably isn’t worth it
  • learn what VDEVs are and how they related to IO throughput before building your pool
  • learn what atime (access time tracking) is and how it can impact performance on your pool under certain workloads.
4 Likes
2

Record size as well.

2 Likes
3

Also, check for SMR drives. Last I checked they’re a firm “bad for raid” / no buy for ZFS.

3 Likes
4

Stop using RAIDZ it’s not the sexy silver bullet you think it is, unless you know what you’re doing.

6 Likes
5

Be generous with the ram, ZFS will love you for it

1 Like
6

Don’t add L2ARC until you’ve maxed out your RAM. Understand ZIL/SLOG and if it will helpful in your use case

5 Likes
7

&
Dont forget that adding l2arc causes arc/ram usage also

3 Likes
8

If one asks for a paste of pool + dataset properties, it’d show if compression and or duplication are on, and ashift/recordsize. Even atime would show up.

I guess suggesting it be wrapped in the spoiler tags would help reduce visible post length

9

Oh I wanna do it too.

  • If you are running sync workloads like serving NFS sharing or virtual machines consider the usage of a SLOG dedicated SSD. A ZIL dedicated device greatly decreases the possibility of fragmentation.
  • Having a SLOG device will also increase NFS sync write performance. So if you are having performance issues with NFS writes into a ZFS volume you might want to do that or use async in case you don’t care about the data.
2 Likes
10

Can you stop attacking me publicly like this?

4 Likes