[pve-devel] [PATCH 00/10] Single-file-restore GUI for PBS snapshots

[Stefan Reiter]

On 22/04/2021 12:33, Dominic Jäger wrote:
> On Wed, Apr 21, 2021 at 01:15:29PM +0200, Stefan Reiter wrote:
>> Implements the necessary API for allowing single file restore via the PVE web
>> GUI.
> 
> Restoring a couple of small text files and folders with textfiles from
> containers worked really well for me.  For VMs I could also restore some text
> files & folders, but I didn't work as smooth.
> I tried both with a single .fidx/root.pxar so far.
> 
> * This
>> proxmox-file-restore failed: Error: mounting 'drive-scsi0.img.fidx/part/1' failed: all mounts failed or no supported file system (500)
> is correct but it would be nice if we could try to say something like "We found LVM stuff
> which is currently unsupported".  Currently it looks more like bug than
> "expected error" to me.

I mean, the real fix here is to support those types of storage ;)
I think "no supported file system" explains the situation reasonably 
well, but open to suggestions of course.

> 
> Maybe we could improve this (at some point in the future) so that the error
> does not block the whole panel and we can continue to select another part/number.
> Because when two out of three parts block the whole thing, then remembering the
> bad one(s), clicking X, clicking File restore->fidx->part->some number is not
> ideal IMO.
> 

argh, I already have a fix for that, but apparently I forgot to include 
my widget-toolkit patches in the series - thanks for noticing!

> 
> * Sometimes the Download button remains enabled when errors (like the previous)
>    happen and then you get a couple of 0 byte files
> 

yup, should probably disable that as well on error

> 
> * On the respective PBS sometimes this
>> 2021-04-22T11:31:01+02:00: starting new backup reader datastore 'test': "/test"
>> 2021-04-22T11:31:01+02:00: protocol upgrade done
>> 2021-04-22T11:31:01+02:00: GET /download
>> 2021-04-22T11:31:01+02:00: download "/test/vm/100/2021-04-22T09:29:17Z/index.json.blob"
>> 2021-04-22T11:31:01+02:00: GET /download
>> 2021-04-22T11:31:01+02:00: download "/test/vm/100/2021-04-22T09:29:17Z/drive-scsi0.img.fidx"
>> 2021-04-22T11:31:01+02:00: register chunks in 'drive-scsi0.img.fidx' as downloadable.
>> 2021-04-22T11:31:01+02:00: GET /chunk
>> 2021-04-22T11:31:01+02:00: download chunk "/test/.chunks/394a/394a4b4c316868a466fbece11881752a10b31a782d2740ab414c3f48b5fb4e58"
>> 2021-04-22T11:31:01+02:00: GET /chunk
>> 2021-04-22T11:31:01+02:00: download chunk "/test/.chunks/d628/d62885ad37c7a44a2fa9fdc558e85e89ecf4c9754617fab7e1f491fa3da65d38"
> appears and stays there forever and when I clicked Stop
>> 2021-04-22T11:47:00+02:00: TASK ERROR: task aborted
> it seemed to accept but the circle remained spinning forever again (OK, I only
> checked for a couple of minutes).
> 

When you clicked "stop" where? On PBS side? That would be an error in 
the server code somewhere then.

But yes, the task stays active - that is due to the nature of the beast, 
the VM used to restore keeps the channel open for disk access as long as 
it runs, once it times out the task should disappear too.

> 
> * Sometimes the PBS also logged
>> 2021-04-22T11:42:40+02:00: starting new backup reader datastore 'test': "/test"
>> 2021-04-22T11:42:40+02:00: protocol upgrade done
>> 2021-04-22T11:42:40+02:00: GET /download
>> 2021-04-22T11:42:40+02:00: download "/test/ct/102/2021-04-22T09:39:23Z/index.json.blob"
>> 2021-04-22T11:42:40+02:00: GET /download
>> 2021-04-22T11:42:40+02:00: download "/test/ct/102/2021-04-22T09:39:23Z/root.pxar.didx"
>> 2021-04-22T11:42:40+02:00: register chunks in 'root.pxar.didx' as downloadable.
>> 2021-04-22T11:42:40+02:00: GET /chunk
>> 2021-04-22T11:42:40+02:00: download chunk "/test/.chunks/ff63/ff63c49eae6b77e2933ea08af0dcf3786e74fc454ac2462ba780f0726c070023"
>> 2021-04-22T11:42:40+02:00: GET /chunk
>> 2021-04-22T11:42:40+02:00: download chunk "/test/.chunks/261e/261ec247951ab57a51df87c593ae2aaf4f76c671bb94e13245f53ebdf25cfd45"
>> 2021-04-22T11:42:40+02:00: GET /chunk
>> 2021-04-22T11:42:40+02:00: download chunk "/test/.chunks/3a60/3a60c4eb21123c01a63e981d8f550ba0f99b5e90b97340d384da9f079fa767ed"
>> 2021-04-22T11:42:40+02:00: TASK ERROR: connection error: Transport endpoint is not connected (os error 107)
> but it looked to me like it didn't really matter? I could still download my
> text files.  Note that when it started to log such messages, there were
> actually quite a lot of those entries.
> 

That is an interesting one... does it happen in the middle of the task 
log? It almost looks like the VM stopped and ended the connection. Maybe 
some intermittant network stuff?

> 
> * A few times I got in the PVE GUI when I clicked on ...img.fidx
>> malformed JSON string, neither tag, array, object, number, string or atom, at character offset 0 (before "VM 'qemu_root\\x40pa...") at /usr/share/perl5/PVE/PBSClient.pm line 220. (500)
> and I couldn't find out yet why and later it just didn't appear again.
> 

Fixed with 6ea6324bd6cc in proxmox-backup.

> 
> * This is helpful
>> executable not found '/usr/bin/proxmox-file-restore'! Proxmox backup client or file restore not installed? (500)
> but the precise package name would make it even more helpful for me.
>> proxmox-backup-file-restore

True. Do we have that as a dependency actually, or is it like with 
ifupdown2 where the user has to install it manually?

> 
> 
> Today I also tried around with existing options of restoring data with the backup client (CLI).
> And even with the previous concerns, I think that in comparison to those CLI possibilities this
> new feature makes exploring & restoring really, really convenient.
> 
> Tested-by: Dominic Jäger <d.jaeger at proxmox.com>
> 

Thanks for testing and the feedback!