ZFS question

[xp0]

Hello all,

I was just wondering what the main difference between just doing a "mkdir" within a zpool you created compated to a 'zfs create <pool>/<dir>'

I noticed when you do the zfs create dir, it does take a few seconds longer. If I do a zfs list, it does show those directories, so It is making a snapshot? If I do a zfs create dir for everything, compared to a single folder where I just create everything in, is there much of a difference?

if I am making more snapshots, do those use much space?

Thanks in advance....
X

 

[sub.mesa]

Hey there!

ZFS is also a volume manager; your pool shares its free space across any filesystem contained in it.

A pool has one filesystem when you create the pool. Example:
zpool create tank /dev/ad1

zfs list -r tank
Now you have only one filesystem: /tank
If you can make more filesystems:

zfs create tank/music
zfs create tank/movies
zfs create tank/docs
zfs create tank/media

This is MUCH different from creating directories:

• Filesystems are self-contained and thus moving a file between two filesystems can take much longer because it needs to be physically copied instead of moving a file which takes 0.1 seconds. It also means files are placed close to eachother which have a relation.
• You can set ZFS properties per filesystem; not per directory. Thus you would need to create filesystems if you want data A to be compressed but data B not, for example.
• Sharing via NFS works per filesystem, not on directories.
• Snapshots can be made on filesystems, not on directories.

So more or less, directories behave normally but to do cool things with ZFS you need to use filesystems instead. Generally my advice is create filesystems for separate types of data; but not to the extreme.

Example: the tank/docs filesystem would store text documents and likely has compression and perhaps copies=2 for extra protection; but there is little point in making more filesystems like tank/docs/private and tank/docs/work; the private/work things should be normal directories; they do not change the type of files stored there; you still want compression and copies=2 on those files as well.

So think of your data as different types of data:
- personal photos (no compression)
- documents
- binary files
- important files
- unimportant files

Experiment and see what is better for yourself.
Snapshots are something else though:

zfs snapshot tank/docs@20101027

Now delete a document from the tank/docs directory by accident; and you can rollback to the date to get it back:

zfs rollback tank/docs@20101027

(not that it's not necessary to rollback the entire filesystem; you can just copy the snapshotted file you deleted; but this example is easier)

Snapshotting, like other ZFS properties, only works on ZFS filesystems, not on normal directories. So tank/docs works, but tank/docs/work where 'work' is a normal directory, does not.

 

[xp0]

Thanks for the great explanation!

 

[adi]

Snapshots only take up space when you change files. Meaning if you take a snapshot of tank/docs, and make zero changes to this filesystem, the space taken up is 0kb. It is only when you make changes when the snapshot starts to "use" storage. If you change one file, the snapshot keeps the version that existed when the snapshot was created, and the filesystem uses the new file. At this point the snapshot is taking up space equal to the file(s) you have changed/deleted.

 

[xp0]

Thanks ... any I need to physically create a snapshot, correct? it is not automatically creating snapshots everytime I add/delete files unless I tell it to?

 

[killagorilla187]

Yes you need to issue the zfs snapshot command, you can also make a cron job to do it at set intervals.

 

[sub.mesa]

In particular, you should realise that if you use snapshots, deleting files will no longer give you more free space back. The snapshot would still allocate the deleted file and no space is freed. Only when you delete the snapshots will you get your space back.

zfs list -t snapshot

lists all your snapshots and its size. I usually scan my snapshots and remove very big snapshots. But having something that takes snapshots every day and deletes any snapshot older than 30 days might be just what you're looking for. Then you have 30 days to recover a file in case of virus/accidental deletion/corruption. Snapshots can be very powerful and overlap with the protections backups grant you.

 

[xp0]

Right, I actually wanted to make sure it wasnt creating snapshots, because for a lot of my data I do not need snapshots

Thanks again!