Linux repository with S3 filesystem?

[danswartz]

So I've been playing with backblaze and they just rolled out support for AWS S3 buckets. I installed fuse-s3 on my linux repository, and mounted the bucket/filesystem. I can copy files to it manually (at least in limited testing), but if I try a Veaam B&R job to a linux-style repository on that server (with the S3 mountpoint as the target), it fails almost instantly. It seems to be able to create the vbm file okay (which is very small), but then immediately aborts with '5/4/2020 7:45:30 PM :: Error: Input/output error POSIX: Failed to flush file buffers. File: [/mnt/backblaze-s3/Backblaze Backup Job/Backblaze Backup JobD2020-05-04T161144_8742.vbk].
Failed to open storage for read/write access. Storage: [/mnt/backblaze-s3/Backblaze Backup Job/Backblaze Backup JobD2020-05-04T161144_8742.vbk]."

It's odd, because being a vanilla linux repository, I assume B&R is sending the data over to a process which writes to files on the S3 mountpoint, so I'm not understanding what is going wrong. This isn't a big deal, as I am just experimenting, but if anyone has tried something like this, I'd be curious what s3fs mount options they used, or some clue as to why this is imploding like this.

'

[HannesK]

Hello,
The software expects "block storage behavior". That means that it can access a file on whatever position it likes to do it. Object storage works completely different than block storage.

So the failed job is expected behavior on any object storage.

As you mention that Backblaze has a S3 compatible API now... you could use it as capacity tier for copy & move. But object storage today is not possible as backup target.

Best regards,
Hannes

[Andreas Neufert]

See the 4th point here: https://helpcenter.veeam.com/docs/backu ... ml?ver=100
Officially not supported as we know it causes issues. Even in some smaller labs it owkrs sometimes but will fail over time.
As Hannes said, use the official supported methods.

[danswartz]

Thanks, I figured it was something like this. Unfortunately, my use case was for a very small number of critical VMs as a belt&suspenders backup for just those 4 VMs, so capacity tier isn't really helpful to me. What I think will work: do the B&R backup to a folder on the linux server. When done, from Linux CLI, do an 'mv' of all contents to the S3 filesystem. Always good to learn something new

[dalbertson]

How often are you wanting to take these backups? While backing up to a linux repo and then copying to s3 may work it is also adding the potential for issues as there are alot of moving parts here that would be needed if you wanted to restore.

With that said, one option i would recommend would be to schedule VeeamZip backups to the linux machine and then copy those to s3. This would be the "safer" option as you only need that file back and not a chain to perform a restore. But this would depend on your backup plan for these machines.

[danswartz]

Ah, good point. The repository in question has max restore points = 1, so the intent was to just be able to copy the entire bucket contents back. But a veeamzip does seem cleaner. I'll give that a shot, thanks. Almost forgot: this is part of my paranoid monthly backup. I backup all VMs to a hot-plug SATA disk, and also the 4 VMs mentioned to the cloud.

[dalbertson]

Then yeah i would for sure go the VeeamZip way. You could script it out to make it more automated and send you alerts.

https://helpcenter.veeam.com/docs/backu ... ml?ver=100

https://helpcenter.veeam.com/docs/backu ... ml?ver=100

[danswartz]

The one annoying thing (is there a workaround?) I tried shift-selecting 2 small VMs to test this out. Unfortunately, they end up in the same backup blob, so if I wanted to restore 1 of them from the cloud, I'd have to DL the entire backup blob, and then restore? Bummer if so... Yeah, I know I could do 4 separate backup runs, manually selecting each VM, but it would be nice to avoid that

[dalbertson]

Yeah its all in one file. However, you could have one script that kicks off 4 individual jobs and that way you have them in individual files but only one "task"

[danswartz]

Good point. Thanks for the ideas!