Adding a second mirror vdev to zpool

[Roman Paravicini]

I'm new to FreeNAS, so if this has been asked a lot, please forgive me, but my google-fu did not really help me here :)

In the end I want to a striped mirror zpool (the zfs pendant of a RAID10) with 4 disks (I prefer it to a 4 disk RAIDZ2), but as a concession to my wife I will start out with a two disk mirror. Am I assuming right that I can, after starting out with a two disk mirror, simply add another two disk mirror after some time, and will have nearly* the same as if starting out with 4 disk striped mirror?

I hope I am right, because this would decrease the initial cost of my setup, but will not really impact my setup on the long run.

Thanks in advance,
Roman

*nearly because I know that all my data from before adding the second mirror will be on the first mirror

 

[Ericloewe]

Am I assuming right that I can, after starting out with a two disk mirror, simply add another two disk mirror after some time, and will have nearly* the same as if starting out with 4 disk striped mirror?

Yes.

*nearly because I know that all my data from before adding the second mirror will be on the first mirror

ZFS will intelligently balance writes to keep the vdevs about even, so that will fix itself over time.

 

[Roman Paravicini]

Wow, thanks for the fast reply, and cool to know that it will even out after a while of adding new data to the pool.

 

[fricker_greg]

@Ericloewe Sorry to revive the ancient thread, but I thought it was better than making a new thread with this here.

Is it the case that over time, a new vdev will be balanced out when added to a pool over time?

Does scrubbing the data help to speed this process?

As I have a 6 disk raidz2 pool at 85% that I will be adding another 6 disk raidz2 vdev to, is there anything that I should do to make sure that the data doesnt all physically live on the first vdev and realizing none of the IO benefit of the second vdev?

Thanks - I really appreciate it.

 

[garm]

CoW is just that.. once written stays where it is. You would need to move stuff around. Create a new dataset and move everything there for instance. If it’s data being updated often then it will balance itself out by itself as new and old blocks will be spread a bit more evenly across vdevs. Scrub doesn’t move data around.

 

[fricker_greg]

Are there any best practices once one adds a new vdev to the pool aside from using as normal / creating a new dataset and moving data as is possible?