Solaris 10: now to put zones on the damn thing.

[reddragdiva]

hauntedunix's suggestion worked: mount the /var zpool as legacy. All hail!

The trick now is that the system won't install zones properly. The zone creates okay, verifies as installable, when I go to install it installs 3 packages and of course won't boot. This is because, desperate for space on Friday, I cleared all that faff out of /var/sadm/pkg ... which is where zone packages are installed from. D'oh!

So today I tried jumpstarting it. And IT appear to have set up the network so stuff won't tftpboot. ARGH. Tomorrow we make an appointment to go down to the server room and ... put the 10u8 SPARC DVD in the drive.

It's been a learning experience.

Oh, and the Solaris 10u8 i386 VirtualBox image that Sun heartily encourages everyone to download and try is a dysfunctional piece of complete dogshit that seems to have been put together with the express aim of warning people off Solaris.

Comments

[hirez.livejournal.com]

We went through a bit of a random boostrap process to get the first two Suns up: install OS on box1, realise we needed to have Jumpstarted it so as to config the h/w RAID controller, so install enough j/s swerver to bring up box2 as we really wanted, install j/s on box2 so we could splat & j/s box1.

The ZFS learning curve has been much flattened by already having several Netapps lying about, so the concepts of optimal raidgroup sizing and how to utterly hose a CoW f/s are already common and/or wikified.

There are regular exchanges of the form 'What d'you mean ZFS doesn't do (x)? I thought it was a Netapp killer. Everyone says so. Does this mean they've not actually tried to do (x)?'

[reddragdiva]

I'm interested in your approaches to utterly hosing a CoW fs!

[hirez.livejournal.com]

a. Fill several 1Tb volumes.
b. Delete it all at once.
c. Profit!

Actually, (c) expands to 'watch the Filer have a complete head-fit and peg the CPUs for several minutes. Since throughput has just fallen through the floor, watch all the businessy-businessy apps expire messily and then deal with the throng of angry suits between you and the coffee machine.'

The Netapp tech-note boils down to 'Don't do that. Performance can be sub-optimal.'

[reddragdiva]

Yeah, most of the approaches to trashing ZFS are along similar lines: get it to tie itself in knots and use 75% system CPU trying to get itself sorted out. "With you in a moment ... a moment ... a moment ..." The famed killer ZFS block allocator bug (lots of small fast writes to a large zpool that's over 80% full, block allocator shits itself trying to allocate perfectly rather than well enough, box turns to molasses, requires reboot, workaround is to leave 25% of your huge expensive zpool empty) is one of these.

[reddragdiva]

And ZFS is mostly meant to be better than Leengux (which it oh so is); it's a "Netapp killer" in that cheaper and good-enough is vicious competition. An X4500 full of disks does mostly enough well enough as a NAS cheaper. Of course, learning the boundaries of "mostly well enough" is invaluable.

[hirez.livejournal.com]

Oh God Yes. The weenies are shitting themselves with hatred, I understand.

I know damn fine that I'd worry more about the data if it were on things that weren't Filers. (He said, being non-committal and leaving out a bunch of stuff)

[nisaba.livejournal.com]

Very true. Two words are key to this: maintenance, and resiliency.

Netapp's support is pretty damn good and it means I don't have to phaff about detecting the inevitable failed disks or even reporting them. And Netapp's clustering is good and has only been getting better in the two yers I've worked with them. I wouldn't want to run anything really key on an x4500 (and that's without going into the problems we had during a PoC running an oracle DB on same with zfs on top as it dealt with random reads/writes really really badly).

We still have a truckload of netapps at our place, and I still love them. Where OracleSun end up when their zfs-based product (Unified? Formerly Amber Rd) is more mature however, that remains to be seem (we trialled their beta when it was the only one they had in europe, it was in serious need of getting the basics right let alone the clever stuff).

[reddragdiva]

Well, yeah. You pays yer money and yer data is safe. We have great big NetApps at the hosting centre which all the actual data lives on.

[xlerb.livejournal.com]

There are regular exchanges of the form 'What d'you mean ZFS doesn't do (x)? I thought it was a Netapp killer. Everyone says so

Also, RAID-Z is full of goodness and light and bunnies and is like RAID except better in every way and will solve every problem in the world…

…except the ones involving lots of small files.