ZFS dataset sizes changed after copying data to new setup

[dionv]

I've been reading through the forums a fair bit before making the leap from 0.7 to 8.3.0, so I understand the cautions about ZFS. I have also seen some folks using less than the recommended specs with apparent success, so please bear with me.

I have a P4 3GHz system with 3GB of RAM. It was running 0.6 and then 0.7 with a WD 1TB drive formatted UFS just fine for the past couple years. For XMas I received 2x Seagate 1TB drives. I decided to upgrade to 8.3.0 and try a ZFS Mirror since it's essentially just me accessing the server at this point, and only for basic file storage and the occasional iTunes streaming.

Initial setup went lickety-split and the ZFS mirror, which I named "storage" showed up just fine with 908GB available. I then started creating ZFS Datasets to replicate my previous layout, each one under "storage": "home", "iTunes", "iso" and some others. I configured CIFS/Samba sharing and shared out those datasets and mapped to them from my Win 7 x64 box. Everything showed up just fine, and each "drive" showed a total of 908GB available (which I am used to seeing).

I then connected my old 1TB drive, imported it into the server as "oldstorage", dropped to the shell and copied by files from it to the new mirror. Everything seemed to be ok. Remembering that copying can fail, and re-copying is a pain, I refreshed myself with rsync, and rsync'd a folder over. Still good. So, I went whole-hog and rsync'd the contents of the old drive to the new mirror.

After almost 24 hours (sigh), it was done. Yay!

But now FreeNAS is reporting that each dataset has a different total size, but they all have 55GB free. Windows Explorer shows the same. For example, my home drive/dataset shows that it is a 469GB drive, with 55GB free. My iso drive/dataset says it is a 64GB drive with 55GB free. I tried creating a new dataset, but it, too, reported the odd sizes. I did not specify any quotas on any of the datasets.

A few file tests seem to show that the data is ok, but I am wondering if this is a normal behaviour? (My initial testing in a VirtualBox did not exhibit this, but I was testing on a 64-bit system) Not seeing any errors in the logs. Only error was something from django that flashed by during boot-up, complaining about some field being NULL.

Since I still have the old drive, I can wipe the new setup and try again. Worst-case scenario, I'll revert back to UFS. Just wondering if anyone has seen this happen before?

Thanks!

Dion

 

[JaimieV]

Yep, that's normal. It's due to the way ZFS handles datasets, which is to say that the total *possible* size of your Home dataset is now 469gig, which should be the current file-load of 414gig plus the 55gig that's free space on the mirror set.

Each of the other drives will be similar. Note that due to various odd things (search for forum for "ZFS size" for examples) the sums might not quite add up...

If you wanted things to be simpler, you could have ignored the whole dataset thing entirely and just made folders and shared them individually. Then each share would show as a 1Tb drive.

 

[cyberjock]

If you wanted things to be simpler, you could have ignored the whole dataset thing entirely and just made folders and shared them individually. Then each share would show as a 1Tb drive.

+++++1

For home use I'm not really sure why a dataset may be useful at all. The only thing I can think of would be if you created personal shares for your children and you don't want them to do something weird and fill the zpool to 100%. With a dataset you could prevent them from filling more than their allocated size.

 

[JaimieV]

I've got a domestic use case for that: I run Macs, and the builtin backup system Time Machine will happily fill any drive eventually. So I've got 1Tb split off as a dataset with 1Tb quota for the four Macs to use as backup space. Works perfectly, the Macs autotrim their backups when they hit what they are told is the drive size. I suspect there are backup engines for other platforms that'll do something similar.

There's also still software out there that craps itself if it sees its data drive as larger than it can count - bigger than 2gig/128gig/whatever. Older versions of Quicken spring to mind.

 

[Stephens]

I'm definitely datasets advocate here too. I'd rather have each of my various types of content in their own "collections" with their own rules and security separate from sharing protocols.

 

[bollar]

I'm with the multiple dataset people. On top of the reasons already mentioned, snapshots and replication are dataset-driven, so this gives you flexibility on handling the frequency of each.

 

[dionv]

JaimieV: Thanks for the initial response and tips.

noobsauce80, Stephens, and bollar: Thanks for weighing in with your points about datasets.

Not too worried at this point about limiting anything, but the point about snapshots and replication is good. There's a couple of areas that I think I want handled differently in that regard (iTunes library, versus my photography backups, for example). I'll have to think on that a bit more.

Dion

 

[JaimieV]

No probs.

Since you mentioned photos... Remember that RAID isn't a backup, it's for availability. Nor are snapshots, because they're on the same disks as the originals - although they do allow you to replicate out to a real backup location. Any important data needs to exist in at least two physical places, preferably three or more, otherwise it hardly counts.

 

[dionv]

JaimieV: Yup, Currently have photos on two different systems (main PC and FreeNAS). Next steps are to get off-site backup worked out (plus get a bigger 64-bit system going for FreeNAS)

I've reworked my layout a bit and see the sizes changing accordingly. Gonna play around a bit, but I'm feeling better overall, now. Thanks again!

Dion

 

[JaimieV]

Excellent!

I've been using www.crashplan.com for offsite stuff for a year or three now, using the free version where you backup to a friend's computer. It works well for backup and restore.

 

[pirateghost]

Excellent!

I've been using www.crashplan.com for offsite stuff for a year or three now, using the free version where you backup to a friend's computer. It works well for backup and restore.

i use the paid version with a slightly convoluted setup.
pics from mine, my wife's, and my son's computers all sync up with synkron on a regular basis to a VM attached to an iscsi volume. the vm handles backing that up to crashplan cloud so I only have to pay for one copy, and it handles all my family's pics

 

[dionv]

Excellent!

I've been using www.crashplan.com for offsite stuff for a year or three now, using the free version where you backup to a friend's computer. It works well for backup and restore.

I've looked at CrashPlan in the past, but didn't have someone offsite to backup to. With the way things worked out I may have a spare drive or two, so I might look at setting up a copy/replication to an external drive which would be stored offsite.

As it all turned out, the P4 has now been retired, and my old Athlon 64 x2 4600+ desktop system with 4GB of RAM is now my FreeNAS box. Making the switch was a breeze. Powered down the P4, moved the drives and USB key to the Athlon box, powered up and boom, done. A little hiccup with the change in Ethernet cards, but everything seems to be working fine.