Posts in category 'selfhosting'
Revisiting IPv6
Many years ago I experimented with running IPv6 in my home network (dual-stacked, not IPv6-only… I’m not that crazy!). At the time this was mainly an intellectual exercise. While a lot of major services already offered IPv6 (including Google, Facebook, and Netflix), the big draw of v6 is the ability to completely do away with NAT and simplify access to services and P2P applications running out of my home. But without broad v6 support, even if my home network was available via v6, the rest of the world wouldn’t be able to access it, which pretty severely curtailed the utility of the whole thing.
But, it was still an interesting exercise!
Until, that is, Netflix started cracking down on VPNs.
The way v6 was deployed in my network was via a tunnel supplied by Hurricane Electric. That tunnel terminated in California, and, while not intentional, it allowed me to watch US Netflix in Canada.
That is until Netflix realized people were abusing those tunnels and started blocking inbound traffic via HE.
I considered potential workarounds, but I could never figure out a satisfying solution (in large part thanks to closed devices like Chromecasts).
And so I shut down v6 in my network. While, previously, v6 didn’t provide a lot of value, it also didn’t cause me any problems. Once this issue surfaced, it was no longer worth the effort.
Recently I decided to take another look at the situation to see if anything had changed.
Well, unfortunately Netflix still blocks traffic coming from Hurricane Electric traffic originating in the US.
However, it turns out, back in 2013, HE added new Points of Presence (POPs) in both Calgary and Manitoba. That meant I could set up a tunnel with an exit point inside the country.
Would Netflix block that?
It turns out, the answer is: No!
So I now have IPv6 back up in my home network.
But has the connectivity story changed? Yes!
Much to my astonishment, I discovered that in the last couple of years, AT&T, Rogers, and Telus have all deployed native IPv6 inside their networks. That means that, when I’m out and about in both Canada and the US, I have direct v6 connectivity back to my home network! Even my mother-in-law’s house has access thanks to her Telus internet package.
That’s a huge expansion in coverage!
In fact, ironically enough, of the places I frequent, the only location that lacks v6 connectivity is my workplace. Go figure. But, in that case, I can always just tunnel through my linode VPS, which has had v6 connectivity for many many years.
IPv6 adoption may be taking a while, but it is happening!
Homegrown Backups Redux
Over the last couple of years I’ve written extensively about backup solutions. The whole thing started as I tried to find a use for my NUC, which I initially turned into a Hackintosh, a solution that was, frankly, in search of a problem.
macOS ran fairly nicely on the thing, but eventually I ran into issues which ultimately lead me to just converting the thing over to an Ubuntu 18.04 installation. In the end, Linux is just, at least in my experience, a much better home server OS for mixed-OS environments (taking the SMB issues on the Mac as a perfect example).
Anyway, I still needed a backup solution, and I originally settled on a combination of a few things:
- For Windows machines ** A Samba file share on the server ** Windows 10 built in file copy backup capabilities
- For Linux machines ** Syncthing for real-time storage redundacy ** rclone for transferring backups to Google Drive for off-site replication.
The whole thing stalled out when I screwed up the rclone mechanism and inadvertently deleted a bunch of items in my broader Google Drive account.
Oops.
And so I became gun shy and paused the whole thing.
The other big change is I switched over to Ubuntu on my X1 Carbon, which meant that I now needed to sort out the backup solution for a Linux client as well. Syncthing is great for redundancy, but it’s not itself a backup solution.
So a couple of things changed, recently, that allowed me to close those gaps and resolve those issues.
First off, when it comes to rclone and Google Drive, I enabled two features:
- Set the authentication scope to “drive.file”
- Set the root_folder_id to the location on Drive where I want the backups stored
The first setting authenticates rclone to only be able to manipulate files it creates. So Google Drive should prevent rclone from accidentally touching anything else but the backups it’s transferring.
The second setting is belt-and-suspenders. By setting the root_folder_id, even if Google Drive somehow screwed up, rclone would never look outside of the target folder I selected.
So, the accidental deletion problem should be well behind me.
The issue of backups with Linux was to expand my use of Syncthing to include additional folders on my laptop I want stored on my backup server. This ensures that my laptop is always maintaining a real-time replica of critical data in another location.
Finally, I adopted Restic for producing snapshot backups of content that I replicate to my backup server.
Basically, I create a local replica of data on the server (either with Syncthing, rclone, lftp, or other mechanisms) and then use Restic to produce a backup repository from those local copies. Restic then takes care of de-duplication, snapshotting, restoration, and other mechanisms. The Restic repositories then get pushed out to Google Drive via rclone.
I’ve also extended this backup strategy to the contents of my linode instance (where this blog is hosted), and to Lenore’s blog. Specifically, I use rclone (or lftp) to create/update a local copy of the data on those respective servers, and then use Restic to produce a backup repository from those copies. And, again, those repositories are then pushed out to Drive.
Overall, I think this stack should work nicely! And I like that it neatly separates the various stages of the process (data transfer, backup, off-siting) into a set of discrete stages that I can independently monitor and control.
Update:
Just a quick handy tidbit: When using rclone for backup purposes like this, it’s a good idea to create a custom OAuth API key for use with Google Drive. By default rclone uses a default API key shared by all other rclone users, which means you’re sharing the API quota as well. As a result, you get much better performance with your own key (though, unless you’re willing to jump through a lot of hoops, you’re stuck with “drive.file” scope… which, again, for this purpose isn’t just fine, it’s desirable).
Hackintosh as NAS
So in my previous post I mentioned some challenges I encountered using macOS on my Hackintosh as a NAS, and my ultimate success in getting it working with Windows as a backup server… after moving the actual NAS’ing to a Linux VM.
What I didn’t realize then, but I know now, is that at least on my NUC, for some reason, the IntelMausiEthernet is not actually stable! I don’t know if it’s tied to high/sustained load, but for whatever reason, over time the NIC would lose connectivity with the network. Re-plugging the network cable resolved the issue, but it would quickly recur.
This rapidly became a dealbreaker, as not only did it render the machine useless for backups, it also made it useless as a Transmission server.
Now, before you ask, no, I haven’t spent any time debugging the issues and don’t plan to. So I haven’t a clue what was actually wrong.
My solution was a lot simpler: I just bought a USB Ethernet dongle and moved on with my life. That, fortunately, has worked like an absolute charm and solved all of my network stability issues!
macOS as NAS
So, as I mentioned previously, one of my ideas for my hackintosh server was to turn it into a backup server/NAS for my home. As a server, the NUC is an excellent option, being low power, quiet, and incredibly compact. And while I can do some amount of backing up to cloud storage (i.e. Drive), for regular day-to-day backups a proper local solution is preferable.
Now, Lenore and I both have Windows 10 equipped laptops, which means we can take advantage of the File History feature to actually perform backups to a designated network drive. So, it would seem that simply setting up a drive share on the Mac, and pointing our laptops at it, would do the job nicely!
Au contraire.
A few releases back macOS moved away from Samba to their own implementation of SMB (the Windows file sharing protocol). Well, apparently that implementation of SMB does not work with File History. And I have no idea why. The errors you get make no sense, and there’s basically no solutions out there on the internets.
You’d be amazed how long I spent pulling my hair out over this one.
Ironically, the solution I arrived at was as silly as it was obvious: I deployed an Ubuntu Server VM running headless on the Mac via VirtualBox. The VM mounts the macOS filesystem and shares it using Samba.
But it works! We now have backups!
And while I was at it, I also finally set up Transmission and Flexget so I could move my bittorrent activity to the Mac as well. The downloaded content is shared using the built-in macOS drive sharing features… for basic reads it seems to work just fine. For now, anyway.