How to transition from mirrors to raidz?

[Robert Rehnmark]

So I have decided to get an LSI SAS/SATA 8 port raid card for my hackintosh and change my pool from…
2 mirrors of 2x3TB Seagate Barracuda (4 drives) —TO—> 1 RaidZ2 of 6x3TB Seagate Barracuda
My intention is to set up a complete backup of the whole pool also but since that is not done yet I can’t juggle the data that way.

Are there ANY options other than creating the raidz2 in a new pool, transferring all the data and then scrapping the old pool?
Can I clone the pool with settings, filesystems, mountpoints and all onto a new pool?
Can I mirror the old pool with the raidz and then scrap the old part of the mirror? This sounds dangerous though, making a vdev containing other vdevs.. (if even possible)

Thanks in advance for any advice.
Robert

[Jason Belec]

Well a few things, if you don't have a backup already don't go any further. Your just asking for tears of anguish.

This procedure to do what your asking has been around almost since the beginning of ZFS, a little Googling can find the steps. I suggest anything with the name Oracle at the top of the page.

You say 2 mirrors, do you mean they are stripped together? How did your original creation command look? To receive aid, you need to be detailed in asking as you could get advice based on assumption that will have you back to the tears thing....

Are you looking for speed, security, redundancy? Because 2 drives mirrored and striped across 2 more drives mirrored is pretty safe and can be grown by another 2 drives and repeated. In fact I usually encourage this for most people rather than RaidZ2 (something you cannot alter after creation).

--
Jason Belec
Sent from my iPad

> --
>
> ---
> You received this message because you are subscribed to the Google Groups "zfs-macos" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to zfs-macos+...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[Robert Rehnmark]

The tears are something I really want to avoid.
Therefore I have already taken the financial hit and ordered drives and an external enclosure for backup purposes.
I will not do anything, not even resilvering, until tomorrow when I have been able to make a full backup.
(As of now, only the most critical data is backed up.)
I would like to avoid two-way mirrors in the future since I read that resilvering often triggers failure of the remaining drive if it’s in bad shape.

Yes, the pool is now two mirrors that are striped.
As you say, it’s easy to grow and it’s fast but I’d like the extra redundancy that a raidz2 gives.
I will see how it works out in terms of performance but the CPU and Memory should not be the problem.
I’ve got a hexa core xeon @ 4,1 GHz and 24 GB RAM.

> This procedure to do what your asking has been around almost since the beginning of ZFS, a little Googling can find the steps. I suggest anything with the name Oracle at the top of the page.

What procedure exactly are you referring to here?

Thank you for offering your tips and help.
I’ll let you know how it goes. :)

/Robert

[BelecMartin]

Not sure what you have read about mirrors, I have yet to see any such issue across 15 such setups.

Now Raidz2 also has its caveats, and for the average individual I really do recommend mirrors for actual redundancy. If you build Raidz2 then you need to do that as a mirror of another raidz2 or have a very good backup strategy and that almost always goes to mirrors. ;)

I really don't see many people running into the memory issues unless you turn dedup on, ouch! I can only speak from experience but I have a good size client base all on ZFS running daily for years. All issues can be tracked back to hardware failure and that is to be expected. Also almost always easy to recover from with ZFS.

However everyone makes their own decisions on data at hand and you may be ahead of the curve.

Jason Belec
Sent from my "It's an iPod, a Phone, and an Internet Device..."

[Robert Rehnmark]

The problem would not be the mirror..
I use consumer standard drives and they fail quite often.
With two-way mirrors I would risk a lot more trouble if the other drive dies too while resilvering.
(Transferring as fast as possible from one disk to another with no other bottlenecks or ”breaks” and therefore stressing the disk to a maximum.)
Of course this might not be a great risk in real life but I read that of all the disks that fail, a quite big part do so while stressing them like this.

Anyways, if the speed of one raidz2 is enough for me I will probably stick with it since it feels a little more safe than two-way mirrors.
It’s mounted in the computer cabinet so there isn’t all that much room for growing it except by migrating to larger drives.
In any case, I’m investing in more and better hardware and if the speeds are to slow with the raidz2 I can go for 3 two-way mirrors in the pool.
Of course the full backup will save me quite a bit of worry and that’s why I have now spent cash on something that will hopefully work out well for me in that department.

Another thing.
I’m feeling good about moving away from the built-in sata controllers, and getting a controller card that I hope will be a bit more dependable.
Is it advisable to distribute the disks for a certain vdev across more than one controller?
..in case the controller introduces errors?
Or is that not recommended for some reason?

Thank you very much for taking the time.

/Robert

[Jason Belec]

I wish you nothing but the best. I still totally disagree. ;)

The controllers usually aren't the fail point. But distributing is always better. I do it so one side of a mirror is on card A and the other on B as an example. I still run quite a few legacy controllers as sata2 is more than fast enough for many things but they are all connected to some form of Thunderbolt device now. My overall structure is similar to big storage companies in that I lay everything out with each component as a modular part so I can swap something new in easily. I stay away from anything with a backplane though!

Currently on a 8 disk 4 mirrored pairs striped I get 2G/s on sata2 controllers and sata2 drives attached to Thunderbolt. Very nice. I've done some SSD tests where I swap the rotating rust and that gets awesome numbers but not really practical space wise.

Note: everything is consumer grade even the over priced stuff claiming otherwise.

Do those backups. ;)

--
Jason Belec
Sent from my iPad

[Bjoern Kahl]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


Hi Robert!

Jason, as usual, said all the important stuff, so little to add, but ...


Am 03.11.14 um 18:34 schrieb Robert Rehnmark:

> The problem would not be the mirror.. I use consumer standard
> drives and they fail quite often. With two-way mirrors I would
> risk a lot more trouble if the other drive dies too while
> resilvering. (Transferring as fast as possible from one disk to
> another with no other bottlenecks or ”breaks” and therefore
> stressing the disk to a maximum.) Of course this might not be a
> great risk in real life but I read that of all the disks that fail,
> a quite big part do so while stressing them like this.

This effect of failing the one remaining "redundant" disk, while the
other is resilvering, is indeed a problem and I have seen this twice.
But the same thing can happen in any configuration, not only with
mirrors, but also in RaidZ. Of course, the more dead disks a
configuration can tolerate, the higher the chance to survive such a
double-failure.

Another thing I always recommend, no matter if setting up a mirror or
a RaidZ:

Buy disk from different manufacturers (not just different brands,
really from different makers), and have them ship in two batches.

The idea here is, that the two, three, ... disks making up one mirror
or one RaidZ set do not have the exact same physical structure and do
not have experienced the exact same shocks when falling of the truck at
your gentle parcel delivery service.

In the past 5+ years where I followed this rule, I have not seen a
single double-failure (but three disk failure in total).


Another thing:

Do regular scrubs, and monitor your SMART values. Although some disks
seem to lie about their SMART health reports, it sometimes gives an
early warning that a disk is getting dementia.

(And some disks don't like to be monitored: I had two Seagate disks
that reproducibly dropped of the bus when a host write request and
SMART status inquiry arrived almost simultaneously.)


Best regards

Björn

| Bjoern Kahl +++ Siegburg +++ Germany |
| "googlelogin@-my-domain-" +++ www.bjoern-kahl.de |
| Languages: German, English, Ancient Latin (a bit :-)) |
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1

iQCVAgUBVFfaNFsDv2ib9OLFAQKbeAP/eYoi0VYTWZOYQkSMCGmbGE0HIznbv3Pi
RnARwjENWT+d93d5waM0qjLldVuovzednaj/63bGCBVp+SzFU+o9Or7l+q4f+xO7
MtmYmY1ouUOJmGYnjpsCpnFbjjJaIdnxtSV02zPEzDP/fabBSrDVLV2H357Xri/w
2VrTpwBJ2r0=
=vnaG
-----END PGP SIGNATURE-----

[Raoul Callaghan]

Here’s how I’ve dealt with personal-use/home-use disk management and ZFS in the past, which in terms of a timeline would date back to disk-size when a 250GB IDE disk was considered “decent”.

I would always buy my disks in sets of “three”.

I used to have two pools for years:

Pool A - a very important pool – being 3-way mirror.  Not the fastest, but that wasn’t my goal with this pool.

Pool B - a not so important pool for everything else – being clusters of three disks in RAIDz1. This pool tended to grow ;))

Pool B eventually grew to be 15 disks (5 x 3 sets of raidz1) which ranged over the years from disk sets of 250GB, 500GB, 1TB and 2TB in size.

I rarely used onboard connectors so I used aCard 6885M for IDE and Silicon Image 3132 for SATA.

All this was stuffed inside an old Apple ANS500 case (the PCI backplane eventually failed so I sadly guttered it and retrofitted whatever mobo was spare over the years)

Every year, I would “retire” disks that reached 4 years of age (4YoA) as they were running 24x7 at home.  I also felt it was time to reduce the number of spindles each time I upgraded, as I knew I was carrying a fair bit of risk having so many spindles!

Being able to add another raidz1 set to Pool B though is a great feature! run out of space? no big deal, just add more if you have spare ports…  (I dealt with a lot of PAL-DV video footage)

Today, I’m down to just the one pool of  5 x 4TB Seagate disks hanging off a RocketRaid 2744 which can drive 16 disks should I ever need more space quickly.

Over the years, I've lost a few disks here and there and so I would just bring forward my yearly replace > 4YoA program and retire the whole set of 3 disks. (I would use the remaining 2 disks in other projects)

All this is backed up onto old Xraids (7 disk raidz1 clusters) that are also running ZFS. I perform a scrub first, then push the snapshot to them.

These things are so damn loud that I only backup every fortnight and then they're all off again! 

About every 90 days I update another ZFS striped backup of my data and keep that in a briefcase at my parents house.

The above might sound a bit extreme for most, but the beauty of ZFS is that it gives you flexibility and scaleability to suit your own needs.

Whatever “design” is chosen for data management, make sure you have discipline to maintain it.

I find human neglect is just as often the root-cause of people’s misery when things don’t go to plan. 

Cheers,