Is there still a case against using ZFS in production?
I run Solaris servers and I am planning an upgrade to Solaris 10 10/09 from an earlier release.
Currently we use SVM for local disks and our vendor's technology for SAN volume management.
We are considering migrating most (if not all) of our storage to ZFS.
We want to be able to use LiveUpgrade on snapshots, easy storage management (especially growing a LUN), fast resilvering etc.
Are there any "gotchas" that we should watch out for? We have a lot of Oracle databases. Is that an issue? Has any of you had negative experiences doing what we are thinking of?
We currently have zfs in production for our app servers zones (hundreds of users) and it works great, we are currently testing it with a development oracle 10 DB and it till now (4 months) we had no problems. Please check http://www.solarisinternals.com/wiki/index.php/ZFS_for_Databases
expecially: "match the ZFS recordsize to the Oracle db_block_size"
We have a 96TB ZFS setup on an (clustered) Sun Storage 7410 unified storage appliance with multiple(8x) 10GBit/s links. It uses ZFS and OpenSolaris. This box rocks!
In our shop, we're still waiting. However, our usage is database mainly and it is very picky about filesystem and we'd rather see someone else spend hours and days testing and tuning as we don't have much resources (small shop).
At home, I'm using it, but it's a normal fileserver, not really using all the possibilities.
SmugMug.com uses the same 7410 box behind their MySql db and really loves it. The CEO gave a presentation at mysqlconf and the 7410/ZFS was one of his favorite things that they've implemented.
The only case I have against it is its licensing/patents and the fact that Oracle is being a b**ch with open source stuff it got when buying Sun. Otherwise, I wait for the day that btrfs will be on par (and in future, better than) with ZFS.
One might believe Oracle would work well with ZFS given that they are the same company now.
Lots of cool features.
However, ZFS (like other logical volume management methods) can fail and lose your data in a heartbeat due to filesystem failures that wouldn't really be an issue on more rudimentary filesystems. It's much harder to put it all back together when you don't know where everything is. RAID isn't a panacea. Failures occur that drive reporting won't solve.
I wouldn't use it to mount an OS, but it's nice for data and databases - if you have backups and you know enough not to blow away a system. I'm not talking rm -rf / level mistakes here, but things like a couple drives in a raid array failing at the same time, or not understanding ZFS import/export. Other LVM solutions have similar issues.