Show HN: Zfs.rent
zfs.rent
zfs.rent
Also this https://radious.co/philosophy.txt
I wish such design was usable outside tech community.
You are free to ignore sites without a legal notice if you have a problem with them, but there are plenty of legitimate reasons to stay anonymous.
But then again, you are free to upload 8TB of your data to an anonymous website that was created this morning.
I heard the same thing about private WHOIS data a decade ago, and this year, despite all registrars offering WHOIS privacy by default and most for free.
I don't think your point is valid.
> But then again, you are free to upload 8TB of your data
Exactly, let the market do whatever it wants. And of course I don't mind you discussing your own predilections and why you wouldn't use it, but the broad blanket statement is just naive and too simplistic.
Even if the provider registered a limited company to their name and used that, how would it help? Who would trust such a small entity with their data?
The service is clearly targeted at people who are knowledgeable enough not to upload unencrypted sensitive data.
This is exactly what I am doing then.
but plain text is better than 99% website's aesthetics, most people just have no idea how to design anything. I think plain text belongs to "outsider art", where a lot of people practicing don't even notice how visually appealing it is (when done right).
There are more ways to communicate than a black and white wall of text. One of them is called language of design. That's why designers use visual language to communicate with their users.
Yes. When their hands are guided by PMs and marketing managers they are more often than not.
It completely fails to explain, in simple, general terms, what the hell it is or why I should care.
The first, most prominent item on the page is some numbers that mean nothing to me.
The second item says this is beta, which I don't care about since I don't know what this is.
The third items claims to be the "purpose", which is exactly what I want to know, but the one and only piece of information is a weird command line that, again, means nothing to me.
The fourth item tries to explain what this is, but just goes into a lot of detail when I am still wondering what this is for.
I still don't really know.
If you have even the slightest knowledge about zfs-snapshots, the product is self-explanatory.
They're literally trying to filter out uninformed users.
Customer discrimination is a perfectly valid tactic in certain scenarios. Do you dispute this?
But if you like Marketing-Sites..please read that:
https://www.hashicorp.com/products/consul
And now explain to me what it is please...
In case you're actually fishing for an explanation, Consul is basically a Very Nice™ DNS server for your service infrastructure.
Every piece of tech is generally worth the time investment, but you have to trust them and start learning the tool before you can realise what it actually does.
Consul is one such case...
Consul is much more than DNS of course, but at the core it's DNS.
Unless you describe it as fancy DNS with extra features, it's description page doesn't make much sense.
Oh yes Hashicorp is great please don't get me wrong, but if you just read that landing-page it's like WTF. But if you know what Hashi is doing and read the Documentation just a little bit everything becomes clear...but that page alone just sounds like pure Marketing blabla cloud stuff.
I think engineers know the reputation of hashicorp products at this point. The marketing site is for the C-suite people, documentation is for the technical people.
FWIW, their documentation pages [1] are actually a great landing page for technical readers. The comparison to other software is pretty good for "grounding" consul into your other infra.
[1]: https://www.consul.io/docs/intro [2]: https://www.consul.io/docs/intro/vs
I've used Consul, Vault, Vagrant and Terraform so far and all of them are worth the time investment (Nomad is on the todo-list but for personal reasons I cannot play with it right now).
It'd be great if Hashicorp used less marketing bs and provided a clearer description... At first I glanced the Vault description, didn't really understand what it was supposed to do and walked away thinking "I probably don't need it". Which is bad because once I started using it I really loved it.
If you don't understand how you may need their service, you do not need their service. I'd welcome being proven wrong, but I would bet money there's not a single customer they're missing by their current front page.
The only way you, as a person not knowing what ZFS is, would have any use of their service is if you, after explanation, are so motivated to pay them money that you change your own data hosting to match a service you previously didn't know you needed even if stumbling over it.
1. Near-zero conversions. If you aren't already using ZFS, you aren't going to start just to use this service. And If you do use ZFS, you likely know what `zfs send` does.
2. The few conversions that do happen are almost certain to be high-maintenance.
It's all about creating products that we would want to use. So that's exactly what we did.
https://www.servethehome.com/an-introduction-to-zfs-a-place-...
1. What OS is this whole thing running?
2. How many drives do I get? What’s the point of ZFS on a single drive? Is this just a way for me to temporarily hold a file system?
3. I don’t remember all the flags for the various zfs/zpool commands so the example command means little to me. This seems to be the main thing explaining what the service is for, yet I would have to go and look at the man pages to figure out what it does. Best I can tell: I can send my whole filesystem from my local NAS to my drive in the sky, but why would I want to do that?
4. Why do I get root access to the box? And if that’s the case, why do I need it?
5. What are the details of the data center they are in? Or the location? Or how did they arrive at their 99.999% reliability?
1. What OS is this whole thing running?.
OpenZFS + CentOS 8.2 (maintence support until 2029)
OpenZFS + Ubuntu 20.04 (maintence support until 2025)
2. How many drives do I get? What’s the point of ZFS on a single drive? Is this just a way for me to temporarily hold a file system?
We rent out KVM virtual machines with dedicated 8-TB hard drives.
Rent out multiple drives if you wish to create a RAIDZ or Mirror ZPool.
3. I don’t remember all the flags for the various zfs/zpool commands so the example command means little to me. This seems to be the main thing explaining what the service is for, yet I would have to go and look at the man pages to figure out what it does. Best I can tell: I can send my whole filesystem from my local NAS to my drive in the sky, but why would I want to do that?
4. Why do I get root access to the box? And if that’s the case, why do I need it?
5. What are the details of the data center they are in? Or the location? Or how did they arrive at their 99.999% reliability?
1. It is written on the given website: Centos 8.2 and Ubuntu 20.04
2. As many as 5U server can fit. Each drive is 8TB in passthru mode to your VM.
3. I think they are willing to work with users who know what is ZFS. At least for now.
4. Because it's your box. Do whatever you want.
5. Agree with you on that.
2. They explain you can one or more drives. They intend the product as remote snapshot storage it seems.
3. You might not be the target audience then. Not every company wants to target every possible person willing to pay.
4. As indicated on the page, you need root to provide the encryption key on a reboot.
5. They said Sacramento, but I agree, more details on which DC would be better, for their target audience anyway.
5. is indeed missing.
This service is not for you, move on.
There's a chance that this service is deliberately makes no sense to you because they don't want you as a customer.
This isn't to sound mean! It took a few reads to make sense to me as well.
> We rent out KVM virtual machines with dedicated 8-TB hard drives. There is no sharing/over-committing.
as a gate. If you don't know what to do with that description, you are probably not a good initial customer.
As a market hypothesis, there are many folks hosting their own media/storage servers for various purposes. Often these servers run zfs and they can’t. Be oushed to the cloud due to inefficient pricing for TB scale, low frequency, low concurrency, moderate latency storage solutions.
Offering raw drives in zfs configs could crack this market, if customers buy into the raw drive solution then they can work on simplifying the solution and educating their audience.
Also, another thing. I might want to use this service for storing copies of important data that I download from the net. But rather than first downloading it to my own computer and then sending it to your servers, does the account on your servers have a shell so that I can run for example wget directly on the server and is it able to connect to arbitrary remote IP or only to whitelisted IP addresses?
Also, your page lists IPv4 but not IPv6. Any word on IPv6?
This Hacker News soft launch was mainly to see if we could generate enough users to purchase a second and third server and scale-out. And by the looks of it, we can!
---
As for downloading large files directly to the instance, I'm with you 100%. That's exactly one one of the use cases. (I'm looking at you: massive linux ISOs and +30 GB Xilinx toolchains.) Personally, I like to download them first to a remote machine (takes about 5-10 minutes for 30 GB) and slowly rsync them to my local dev machine with `rsync --partial --bwlimit=10M`.
Think of this service as $5/month DigitalOcean VPS, but it happens to have a hefty native storage unit passed through.
You can SSH into it and do whatever you want with it. In fact, you don't even have to run ZFS. You could run any filesystem -- EXT4, XFS, BTRFS, etc. Compute-wise, it's not going to be anything groundbreaking -- just a 16-core Ryzen 3700 with a couple of cores passed through to you. Ryzen CPUs support ECC and are still priced as consumer-CPUs which is fantastic.
The nice thing about KVM is that even if you don't explicity allocate CPU resources, if the guest machine needs to burst compute, it will "Steal time" from the hypervisor as long as there is extra headroom. This isn't the case with RAM. (For RAM, whatever you set, that is the hard maximum limit.)
I looked recently and found the epyc and TR motherboards from the same manufacturer within $1.00. The Epyc 7302P cost $450 for the CPU, cheaper than any TR. Sure TR has some options with higher clock speeds, but generally that's less useful for fileservers.
Generally TR made sense ... before AMD brought out the single socket Epycs.
You could use TunnelBroker.net on top to get IPv6 if they don’t offer it and your use case requires it.
Anecdote: Let me introduce you to the SuperMicro 6047R-E1R36L, a 4U chassis with 36 (!) 3.5" drive bays. How, you ask? By having bays in the front and back, of course. From experience, I can tell you this baby weighs a metric fuck-ton when loaded with drives and is an absolute abomination, but it's a wonder to behold.
https://www.supermicro.com/products/system/4U/6047/SSG-6047R...
The classic Sprite Hard Drive mod comes to mind https://spritesmods.com/?art=hddhack&page=5
I'm a little skeptical to the claim that passthrough is so much faster than virtio-scsi or something similar. I'd imagine that the disk controllers and network would be the bottleneck. And as an added bonus you prevent people from doing shenanigans on disk firmware. We're talking about plate spinners here, latency isn't really what they're known for.
2. Do you intend or allow these VMs to be used as web servers or any other kind of server?
Here, we are pretty much providing a datacenter hookup for your hard-drive. If that hard-drive fails, we can ship it to you for recovery, but that's on you.
The idea for this site is that most people with ZFS pools already have a mirror/RAIDZ setup at home, and would enjoy the peace of mind if they had one-extra data-sink to send their snapshots.
Therefore, true data-loss will occur if:
* Their house burns down
* AND the remote zfs.rent hard drive fails
---
Personally, I live about 20 minutes aways from my parents' house. I have two separate ZPools (one for my place, one for theirs). I sync the two every time I visit.
I wanted an extra cloud machine to sync to once a week or so -- you know, in case Northern California fires really get out-of-hand :/
---
EDIT: I went to rsync.net just to double-check. It looks like they are running their own infrastructure now.
https://www.rsync.net/products/locations.html
Maybe my memory is going haywire, but I swear I remembered reading that they used AWS as their backend.
A few things ...
I love your website and domain name, etc. I wish more people operated like this.
I want very much not to mention, or have anyone else discuss, (other competitors) because this is your HN frontpage day.
... so I will just quickly clarify: rsync.net (which predates AWS) has always run on our own hardware that we assemble ourselves. Also, we have no cold storage option or functions - we only provide live, online, random access storage.
... and now let's return to zfs.rent and their day in the sun :)
Avoid making comments about competitors unless you are certain it is the truth.
I ended up using rsync.net with borgbackup.
I think the better comparison would be an SX-line dedicated server from Hetzner [0] and this server isn't too far off that.
This service: $10 for 8TB, or $1.25/TB.
Hetzner SX62: 64€ for 40TB, or €1.6/TB ($1.9 USD).
> rsync.net
Rsync.net is ridiculously overpriced. Even at their cheapest tier, 1TBmonth costs $15, for which you can buy 7.9TB at Hetzner. Even if you created 3 redundant copies at Hetzner, you'd be paying only half what rsync.net would charge.
[0]: https://www.hetzner.com/de/dedicated-rootserver/matrix-sx?co...
Your Hetzner numbers are impressive, but I don't think there's anything available in the US for a similar cost. Most block storage I've seen is ~$10/mo/TB. I would love to be proven wrong on that.
I know it's not what you meant but keep in mind that as-described, that's every single server provider.
And I think including what you meant (e.g. ZFS set up), it's literally a one-line command to create a new zpool on a new server.
I'm not saying that rsync.net is worthless but it's certainly not worth the price it's charging for anyone who cares about the price.
> Your Hetzner numbers are impressive, but I don't think there's anything available in the US for a similar cost.
When I was in the US I came to the same conclusion, so went with colocation.
However OVH has a datacenter just over the border in Canada and its brands Kimsufi, SoYouStart and OVH [0-2] can get you down to $5/TBmonth.
And it looks like they actually have a couple of US datacenters now, which is new. A quick look shows that you can get a 16TB server for $92/month [3], which is $5.75/TBmonth.
But overall it looks like the dedicated server market in the US is still super expensive. You can always keep your bulk data in Europe.
[0]: https://www.kimsufi.com/us/en/servers.xml
The compromises seem to be spelled out in the description in a fairly straight forward manner. There isn't flexibility beyond 8TB storage blocks, and it defaults to just one which in my opinion is fairly useless for ZFS, you lose some of the real self-healing and safety advantages (yeah you can have copies within a single drive, but that halves/thirds your space just like a mirror and it'll be slower and with less redundancy than just two half sized drives in a mirror). So basically double or more the pricing there. The data transfer limits and pricing on that is also somewhat higher, so not for high churn. I'm assuming that "1 TB included" is per month, so for offsite replication essentially there'd be a "loading fee" per drive, then bandwidth is covered going forward if churn is less than 1 TB/month. And even in the best case they don't claim you'd be able to saturate even 1 Gbps, so loading will take a while. At 100 Mbps bringing an 8 TB pool to ~3/4 would take around 6 days. There is no SLA, appears to be no autobackups or anything like that.
The pricing could use some more clarity too. Under rent-to-own for example, it's not clear what the pricing is after you have covered the drive(s). Not clear what payments they take and through what service. I wouldn't personally put keys there either, but since ZFS supports raw send/receive there is less need to trust the far side unless your metadata is sensitive as well.
So basically it's ultra bare bones, for cheap.
Which isn't a bad thing to have in the mix. I'm happy to see more space for people to experiment with very minimal focused efforts that lean more heavily on ZFS itself to try to economize vs traditional models. Personally, right now this is still a bit farther on the rough side of the line then I'd feel comfortable with (I'd want to at least have a vague idea of who they were and where exactly the data center in question is). But the basic idea is a neat one, amortizing something someone wanted to do anyway.
One question I don't see answered: If I go the rent-to-own route, then what's the monthly price (for use of the VM and to pay the colo fees) after I've purchased the drive? $5? And is there a price cut if I want to purchase the drive outright in one payment instead of split monthly?
Good luck with the beta!
Last I heard, Amazon still can't ship a hard drive properly; I would recommend a source that won't put the hard drive box in a much larger box with a garnish of air pillows, so the drive can rattle around and likely end up damaged.
Businesses and enterprise users would likely prefer an organisation with a proven track record (eg rsync.net) anyway -- even data replication aside.
But this seems to be a beta phase so I guess it's more important to iron out the issues around creating a business that works, then you can have more drives in your account.
>Rent out multiple drives if you wish to create a RAIDZ or Mirror ZPool.
Zpools with the correct level of redundancy protects against bit rot, occasional hardrive failure and the like(as you correctly point out), but not against total machine loss.
Its probably helpful to think of it as an offsite backup more than anything else.
> Rent out multiple drives if you wish to create a RAIDZ or Mirror ZPool.
"-w, -raw
For encrypted datasets, send data exactly as it exists on disk. This allows backups to be taken even if encryption keys are not currently loaded. The backup may then be received on an untrusted machine since that machine will not have the encryption keys to read the protected data or alter it without being detected. Upon being received, the dataset will have the same encryption keys as it did on the send side, although the keylocation property will be defaulted to prompt if not otherwise provided. For unencrypted datasets, this flag will be equivalent to -Lec Note that if you do not use this flag for sending encrypted datasets, data will be sent unencrypted and may be re-encrypted with a different encryption key on the receiving system, which will disable the ability to do a raw send to that system for incrementals."
Source: https://zfsonlinux.org/manpages/0.8.5/man8/zfs.8.html
However, it become a pain-in-the-ass if you have recursive snapshots where only some of the datasets are encrypted. I think... in order to accomplish this, you need to send separate batches of snapshots... one for each set of encrypted vs. decrypted.
E.g.
/zpool/tmp # not encrypted
/zpool/home/someone # not encrypted
/zpool/home/someone/thunderbird # encrypted
$ zfs snapshot -r /zpool zpool@today
# this will probably bork
$ zfs send -v -w -R -I zpool@yesterday zpool@today | ...
# but this will work
$ zfs send -v -R -I zpool/tmp@yeserday zpool/tmp@today | ...
$ zfs send -v -R -I zpool/home/someone@yeserday zpool/home/someone@today | ...
$ zfs send -v -w -R -I zpool/home/someone/thunderbird@yeserday zpool/home/someone/thunderbird@today | ...
Does the nominal bandwidth cost of $5 for 1 TB mean it's $5 each month for a TB per month? Or is it simply "every 1TB transferred adds $5", no matter how long it took to use that much?
For the rent to own plan, the only thing you're paying after payoff is the bandwidth fee? (Which for minimal usage might be under $5 or $5 depending on the above answer) Is that right?
We meant it in the sense that most users... will not exceed 1 TB/month data movement, therefore the _nominal_ or _expected_ charge would be $5. Take myself for example, once my data was in the system, I maybe move 30 GB a month and that's it.
(Although we're dealing with ZFS datahoarders here... so some assumptions may need to be discarded ...)
---
> For the rent to own plan, the only thing you're paying after payoff is the bandwidth fee? (Which for minimal usage might be under $5 or $5 depending on the above answer) Is that right?
Yup. That's correct. You're just paying for a slot in the machine + bandwidth. If you need extra bandwidth, it's $5/TB.
It looks like you just rent a VM from them that has a dedicated hard drive attached to it. How you choose to format that drive would seem to be an implementation detail that can be left to the user. If so, it seems odd to limit your potential customers to people who use ZFS?
It is not. It is useful for synchronous writes only, such as those done by a local database server, or remotely via NFS. If you’re just sshing in to send data, an SLOG device is a waste.
This zfs.rent server lets you run local programs too on it, so SLOG would help there.
So maybe some databases or some specific apps, but whatever you'd run in a 4GM RAM KVM won't be much.
Wouldn’t you be able to implement this cheaper using only 4-8GBs?
Would there be an equivalent for long term backup using a technology like tape storage? An alternative to S3 Glacier Deep Archive basically.
QNAP is now offering ZFS based systems but they are rather pricey. Is it just BYO? How can i make one that is super power efficient?
https://vermaden.wordpress.com/2019/04/03/silent-fanless-fre...
It will provide you a ZFS mirror of two 5 TB 2.5 drives.
I repaste the table from the article for convenience.
PRICE TIME SERVICE
$447 5 YEAR(S) Self Build NAS
$567 5 YEAR(S) Self Build NAS (assuming one of the drives failed)
$1500 5 YEAR(S) Amazon Drive
$1550 5 YEAR(S) Google One
$1200 5 YEAR(S) Amazon S3 Glacier Storage
$2250 5 YEAR(S) Backblaze B2 Cloud Storage"For the Backblaze B2 Cloud Storage I assumed average between upload/download price because upload is two times cheaper then download."
In fact, uploads on B2 are free, and download bandwidth is cheaper than any other supplier I know of at 1 cent per GB, with a free daily allowance of 1GB. Storage rates are 1/2 cent per GB per month, of $5/mo per TB, including redundancy.
The storage cost of 5TB for Backblaze B2 for 5 years would be $60/year or $300 for 5 years, the upload cost is free, and if you downloaded your entire 5TB each month, it would cost less than $600 over 5 years because you'd also have the 1GB/day free download allowance. HashBackup (I'm the author) uses the free B2 daily allowance to pack older backup data, which sometimes requires an download and upload cycle.
I couldn't find the upload/download bandwidth assumptions, but these numbers are way off IMO.
Overall I've been very happy with it. The only negative from a ZFS perspective is it doesn't have enough ram to support deduplication but that's not a big issue for my use case and otherwise I've been very happy with it.
You don't need more than about 4GB of RAM, however ZFS is quite good at using it for cache so the more the better.
Anyway, whatever you do, don't enable deduplication. You can't simply turn it off and it imposes sever limitations.
But you should really throw all the ram you can at zfs.
It'll just make performance better, and you'll hit the disks way less (for read operations).
[1] https://arstechnica.com/information-technology/2014/02/ars-w...
I found this setup a lot cheaper than the dedicated NAS boxes you can buy and more flexible.
zfs requires ECC[0][1], do not skimp on this.
I ran many zfs pools for >8 years without issues until silent memory corruption happened to me, and now I can't access my data anymore (zfs won't let me mount a zpool that fails scrub checks).
0: Oracle (and previously Sun) documentation recommends ECC to prevent silent corruption:
>Consider using ECC memory to protect against memory corruption. Silent memory corruption can potentially damage your data.
https://docs.oracle.com/cd/E23824_01/html/821-1448/zfspools-...
1: https://louwrentius.com/please-use-zfs-with-ecc-memory.html
All filesystems benefit from ECC memory but ZFS from all filesystems does its job best even without ECC.
Page 102 from https://is.gd/bsdstg here.
I'm getting bored repeating this over and over again.
I’d much prefer to know, but if you’re used to not knowing, knowing is scary.
It is extremely silent after cleaning the fans and replacing the aged GPU with a silent one but not power efficient, that's granted. But at least it's having actual processing power and can sustain ~120MB/s write unlike its predecessor (Lenovo ix4-300d) that could barely be bothered to do 10 MB/s...
It seems like you might have added up your costs, but not left any room to pay for your hours, unused capacity, unpaid bills, overheads, profit, etc.
"be transparent [about costs and profit margins]" isn't a necessary factor. It's useful information for the customer, and even more useful for the competitors of the business, but absolutely not the same thing.
Tell that to AWS
Hopefully the project is setup in a way so it avoids the scam AWS et al does with charging for "premium" bandwidth, meaning the cost of the bandwidth should either be a static sum each month, way cheaper or free.
Instead hosts like Hetzner and OVH uses a price closer to the true cost for bandwidth for their users. Take a look at their dedicated hosting and the pricing. I'm not sponsored by either, but I am a happy user of Hetzner.
Check out this talk by Kyle Drake that goes more into the details of bandwidth costs: https://www.youtube.com/watch?v=-i6wvix6buI
Unmetered bandwidth is reasonably real.
They offer 4GB VMs; Assuming they’re oversold and using a good backblaze-pod like design, they can fit ~32 drives per U. 32 * $5/mo is $160; Assuming they’re paying $50/U and the customer’s guaranteed bandwidth is included in their colo cost, that gives them $110 gross margin. Cost of drives is retail - If they have a better supplier that’s a few $ more, but not a lot. The hardware to run that is another real cost. I doubt it costs less than $2000/U to outfit the hardware; It doesn’t have to be beefy but SATA slots add up.
That gives a payoff time of ~18 months. Doable, but not amazing. At a scale of a few dozen racks, this could be a lifestyle business, but it’s sorta marginally so. Assuming a hardware lifetime of 3 years, you start breaking even halfway in. Lifetime margin is about the cost of the hardware - $2000/u. If this is a single man show, about 2 racks (= ~3000 customers) gives a reasonable “salary”, but the upfront costs are high.
What restrictions are those? You even say people can use a forwarding service, if that were the case, wouldn't the forwarding service be illegal too?
By saying US addresses only, they can say they had no knowledge of which country it was going to.
What's the best place to do this? I'm using google drive, but they throttle your upload quite significantly. I can see right now I'm getting about ~150kB/sec even though I have a 1GB upload speed.
Unless I'm missing something rsync.net is minimum $25/month for 1TB, and this is ~$15/month for 8TB.
rsync.net is more expensive, but has included snapshots and better fault tolerance/data retention policies. It's also been around for nearly 20 years, they own their own datacenters, and have a proven track record with thousands of users.
This is cheaper (maybe, depending on your bandwidth usage) and gives you better "ownership" of the hardware, but because it is hardware its just as likely to fail or corrupt data as a hard drive sitting under your desk. This service is also in beta with just a handful of users.
Both companies seem to have really good policies and to be unix user-friendly.
* What does the 20 mbps / 1gbps capacity indicate? How is this former number derived?
* I suspect Debian would make more sense than Ubuntu 20.04 for your target audience. At least that's my personal strong preference.
My guess is implementing with Debian would be more effort.
hd-idle and hdparm -Y alt hdparm -S 600 /dev/sdx
Honestly seems really interesting sans that tidbit, would be interesting to have unlimited 100mbps but gigabit at $5/TB or something like that. But $40 just to fill the drive is a bit much, and with being hard disk with only 4GB of RAM I'd think backups are what this is targeting, and not something with a lot of activity.
Looks like a hobby project so it's no big deal, just giving my opinion.
However, there are some product compromises that would preclude me from using this. First, I already manage drive purchases, parity, dealing with scrubs and errors, etc. for my pool. I don't want to duplicate my effort for a second backup pool. ZFS sends are, ultimately just a long binary, and I simply want them stored, without much hassle. If I'm planning a zpool expansion, the work in using this service is now doubled, having to (non-trivially) determine the best pool geometry for my own pool, and for a backup pool made up of differently sized drives.
I also don't understand why someone would want to give you their encryption passphrase. A service like this should just take encrypted sends, and send them back when requested. I don't want my plaintext anywhere other than my own hardware.
That said, your main competitor suffers from all of these problems as well.
it can, see raw sends: e.g. https://news.ycombinator.com/item?id=25148494
To securely restore you would need to be able to ZFS send back to yourself first.
You can also set copies blocks/data (copies=2) which stores two copies on one drive, allowing you to repair. Doesn't cover you for drive failures but can for bitflips. And you'd get half the storage.
Maybe I'm missing something.