ZFS: replace disks with smaller one

[Alessandro 123]

Let's assume a RAIDZ2 with SSDs from different vendors. When creating the vdev, I can safely mix a 400GB and a 480GB device with no issue (obviously, vdev would be created based on the smaller disk)

But what If I have to replace a disk with another one, same nominal size, but different vendor?
Not all vendor agree on nominal size, in example, a 480GB SSD from Samsung could have a bounch of sector less than a 480GB from Intel.

In this case, ZFS will refuse to replace because the real size doesn't match?
How to prevent this ?

With SSD, using different brands is safer than using a single brand, but how can I ensure that both disks would be 'swappable'? In case of failure, I have to replace the failed disk with the first available one.

 

[alexskysilk]

Generally speaking, dont mix and match your vdev members. but if you MUST:
1. define what your absolute smallest physical device would be.
2. create a partition on each of your disks with the size defined above
3. create your vdevs using the PARTITIONS, not the DEVICES.

Again, this is not a recommendation.

 

[Alessandro 123]

So, are you suggesting to use only 1 vendor?
With SSD this is not a good practise.

 

[Nemesiz]

Make partition size 479G or similar to avoid disk replacement problem.

 

[alexskysilk]

So, are you suggesting to use only 1 vendor?
With SSD this is not a good practise.

yes, that is common practice. I'm curious as to why you would assume this is not a good idea. When the intention is to run a system in production the primary concern is to make sure everything works correctly and in a supportable manner; when you mix drives any potential incompatibilities (firmware, etc- including the concern you voiced about differing sizes) complicates resolution tremendously. the cost of the drives is insignificant vs the cost of qualified engineers.

 

[Q-wulf]

I think there is some confusion as to SSD and same series/vendor usage, which somehow got adapted from the HDD "best practice", where people suggest you use same capacity drives of different charges / lines / series / vendors, as to mitigate a complete charge being bad and you loosing all your data.
IMHO these people do not realize that Raid is not a replacement for a proper backup strategy. It is just for performance gains and redudancy (read convenience and low downtime).

I have always found that sticking to drives of the same series, doing a proper burn in test and then employing a proper backup strategy is the way to go. Makes these setups more performant and maintenance free, which translates to less hassle.


As with SSD's you definitely wanna stick with same vendor and series, if not same firmware for your NAND, as alexskysilk is pointing out. just looking at SSD benchmarks should make this very obvious. Not to mention vendor specs like TBW (Tera Byte Written), MTBF, etc ...

 

[LnxBil]

His problem is maybe that they can fail if they hit the DWPD limit all at once ... but then again, you have other problems. I saw a similar approach on the advertisement on the (maybe new) Synology 19'' rack NAS with inhomogeneous writing to SSDs. So maybe he is referring to that.

We have a (enterprise) SSD-only RAID in production for over 1.5 years and the "health" of the disks is still 100% (claimed by the SAN manager). So, the wearout is somehow negligible. Often you do not write that often to the disks. At least for Fujitsu, you can order 'read-intensive', 'write-intensive' or 'mixed-use' SSDs that fit your need perfectly.