Category Archives: Tips

Bootstrap SSH on Ubuntu Core with out Ubuntu SSO credentials

I was playing around with Snaps and I wanted to try out Ubuntu Core as well. I still had some of those  Orange Pi Zero boards laying around and when you go to the Core Download page – there’s an option for Orange Pi right there – sweet! I downloaded the .img file, wrote it to my microsd card with dd, slapped it in my Orange Pi Zero, found the new IP in my DHCP server and off I went to SSH in.

Then I saw this step on install docs:

you will be asked to enter your Ubuntu SSO credentials

– Core Install Docs

Whhhaaat? Oh, I see, Core’s whole shtick is that it’s secure by default. They say, “Secure by default – Automatic updates ensure that critical security issues are addressed in the field, even if a device is unattended.”.  Cool, I can get behind that. IoT needs some thought leaders in IoT security. However…I still just want to SSH in and poke around a bit – I don’t want to have to set up an account at Ubuntu. 

Then I thought, “What if I just create a .ssh directory in /root/ and put my public key in the authorized keys file?” Assuming you’re on an Ubuntu system, logged in as mrjones and just stuck in your microSD card with the core image on it, that’d look like this:

cd /media/mrjones/writable/system-data/root/
mkdir .ssh
chmod 700 .ssh/
vim .ssh/authorized_keys 
chmod 700 .ssh/authorized_keys

After unmounting the card, inserting it into and rebooting my Pi, I SSHed as root and it just worked – that’s awesome! Now you know how to do it as well!  In fact, this likely will work with the Raspbian, Armbian and Ubuntu images for all kinds of Pi boards as well.

Stay tuned for my next post where I’ll massively simplify my stubby and pi-hole how to!


Teaching Kids About Busy Signals with a PBX on LXD for $15

Linksys RTP300 Circa 2005

The other day the kids and I were at our local, awesome Goodwill store. I guess they’re all awesome, eh? I was looking over the pile of small electronics while the kids picked out a shirt (“any shirt you want!!”), and stumbled across a device with Ethernet jacks (RJ45) and phone jacks (RJ11). It’s a router with two ATA ports – cool! It was only $5 bucks!! A quick internet search suggests that it may be locked to a provider like Vonage, but that there’s lots of folks unlocking it. Given this was Goodwill, there were was a plethora of POTs devices handy – two more phones at just $5/ea – no problem!

Now that I had the whole kit home, it was time to see what the ATA was up to. It’s a Linksys RTP300 – and the voice section was definitely locked down :( But that means it’s time to start hacking! I dug up the site I initially found before buying it – luckily all the downloads still work! The post is 8 years old, forever in ‘net time, so I was impressed.

I’m pretty pleased with myself because I fully understood everything and was able to improve the process because of this. Mainly:

Using inspector tool to make the ping input field longer
  • Being a web developer I already had a web server running. This is much easier to use instead of setting up a new TFTP server as they suggested.
  • Being a web developer (still) I was kinda in awe that I’d already forgotten how amazeballs Firebug was, and how I take for granted that modern browsers have such good developer tools built in.
  • Knowing all the linux commands used (wget, chmod, cd & dd) I was able to explore the device a bit
  • I figured out that the guide wanted you to download into /var/tmp which didn’t exist on my device, but /var worked
  • They have you download some *.img, but running it through strings and checking the size, it appeared to match the 3.1.24 version. Completing the install confirmed this. However, md5sum outputs don’t match.
That, sir, is an odd response to a ping ;)

After all this I had a generic, if not quite dated, ATA that was ready to talk to a VoIP provider. Time to install Asterisk! This is where LXD comes in. I already had my home server running Stubby and Pi Hole – time to add another. After spinning up an Ubuntu 16.04 box, hitting a snag, then spinning up an 18.04 box, a quick apt-get install asterisk and I was ready to go!

I created 4 extensions, one for each of us, by following some simple guides I’d found (edit /etc/asterisk/sip.conf, /etc/asterisk/extensions.conf and a reload). I plunked in the LXD Asterisk IP, extension and password and BAM! both lines registered easy peasy on the ATA.

With two hard phones set up (just plug ’em in to the ATA), I set up me and the wife using the built in client on our phones. Now we all had a phone and could call each other. See the gallery below for the visual story!

On a personal note, this was pleasing for a two reasons:

  1. I saved 3 small electronic devices from the landfill.  More and more I’m trying to be conscientious about buying less or buying used.
  2. My kids got to look up at me and say, “it’s making a weird sound!?” I explained to them that it was a “busy signal”.  See, cell phones, office phones and 99% of home phones don’t do this any more. They have either call waiting or voice mail, or both ; )

Poon Hill Trek

Be sure to read my prior post about attending the APRICOT 2018 conference in Nepal before the trek!

Intro & Highlights

This was my first time trekking and my first time trekking in Nepal. While we had considered a more rigorous route of the Annapurna Base Camp (ABC), the max elevation of 4130m (13,400ft) had me concerned. I opted for a much more tame Poon Hill route, which tops out at 3210m (10,000ft) which was more my speed. In the end the elevation ended up being a non-issue. In fact, just about all my concerns were non issues ;)

We started on day 1 with celebrating Holi. At first we were reticent, but then joined the festivities by having the local kids do us up. It was lovely.

My favorite day for views was day 3. Poon Hill was clear for our pre-dawn hike and the sunrise and Annapurna mountains were breath taking. Bring money to pay for the entrance fee and hot drinks at the top of the hike! The rest of the day the mountain views kept us company and they were epic. Thanks Nepal!!

The rest of the trip was amazing. While you can feel modern amenities creeping in every town, the both benefit the towns folk and pull the “rural-ness” away from what you can tell was once a remote village. You know, the old man in me wants to have no tech when I hike/trek. I’m there to experience the place and nature, not check me email ;)

Farm animals where everywhere and it was a delight to see them. All the locals were nice, though some drove a harder bargain for their tea house prices. Check around a bit and don’t settle for the first one.

Click any of the photos in this post to see the gallery! Note, you’ll see pictures of my work conference too ;)

Finally, there was a really…oddly comforting I guess, moment when the cabi on the way to Pokhora put on “Like a Virgin“. I loved it (it’s the second clip in this video). Sorry I was too lazy to edit this video down, it’s 3 min long:


Based on this write up, among others, we ended up doing this route (all prices in US Dollars and are per person):

  • Day 1: Kathmandu -> Pokhora – 30 Minute Flight $250
  • Day 1: Pokhora -> Nayapul – 1.5 hour Taxi $30
  • Day 1: Nayapul -> Hile – 1 hour Jeep $30
  • Day 1: Hile -> Uleri – trek
  • Day 2: Uleri -> Ghorepani -> Poon Hill -> Ghorepani (day hike to Poon Hill after dropping packs ;)- trek
  • Day 3: Ghorepani -> Poon Hill -> Ghorepani -> Tadapani – trek
  • Day 4: Tadapani -> Taulung -> Chomrong -> Taulung -> Chinu/Jhinudanda – trek
  • Day 5: Chinu/Jhinudanda -> Tolka – trek
  • Day 5: Tolka -> Pokhora – 1 hour Jeep $15
  • Day 5: Pokhora -> Kathmandu – 30 Minute Flight paid for on first item

I strongly encourage you pick up a map in Kathmandu or Pokhora. They’re cheap and really handy to have. We got this one which was readily available in map stores in Thamel.

Packing list

My pack was 9.5kg. I think if I didn’t carry spare running sneakers, USB battery, as fully stocked first aid kit, and micro 4 3rds camera, I’d have been around 7.5 or 8. I have no regrets though! My list is taken (and culled) from The “two shirts, one for hiking, one for not” was really great. During the warmer months you could totally skip the sleeping bag to save some real weight. The tea houses all had blankets. While I took photos with my nice camera I mentioned above, I actually was fine with cell phone’s pictures, given it’s a newer phone. Walking poles are optional, but I loved them. My gaiters and rain pants went unused the entire trip.

Hiking Clothes

  • Hiking Pants
  • 2 short sleeve shirt (preferably a “technical” shirt with synthetic fibers to help keep you dry)
  • 2 pairs of socks (in case one pair gets wet during the day, you can change socks)
  • upper and lower base layer
  • Down jacket
  • Sun hat
  • warm hat
  • Gloves (look for ones with Gore-Tex)
  • Raincoat
  • Rainpants
  • gaiters
  • Hiking Boots Water proof
  • Sunglasses
  • running shoes/flip flops


  • Backpack
  • Phone & Cord (doubles as camera)
  • Map
  • Sleeping bag
  • Water purification tabs
  • Water bladder (2 liter size)
  • Headlamp
  • Pack rain cover
  • Passport & Permits (you’ll be asked for these at every checkpoint)
  • Cash –
  • Ziplock bags for organizing/waterproof


Toothpaste, soap and TP can be bought along the way if you want.

  • Toilet Paper
  • Soap
  • Toothbrush
  • Toothpaste
  • Sunscreen
  • Lip balm with SPF
  • nail trimmer
  • ear plugs (tea houses can be LOUD)
  • First aid kit with at least: band aids, moleskin, ibuprofen, antidiarrheal, tweezers


  • Power Plug adapter (just about every tea house had a universal outlet that fit every wall charger)
  • Pillowcase
  • Trekking poles
  • Real Camera & Extra battery

You’ll want a small pack – no bigger than 40L likely – to fit this in. That big-ass “I fit it all on my back” pack that you carry your tent, stove, pots and pans and food for a week is too big! Here’s my pack (foot for scale) and a day pack I checked at my hotel in Kathmandu with a change of clothes and other stuffs I didn’t want to bring with:


From zero to LXD: Installing a private compute cloud on a Cisco C220 M4SFF


Having some nice server grade hardware on hand, I knew I needed to put it to good use.  I’m talking about the Cisco C220 M4SFF:

This lil’ guy comes with some nice specs. The one I have is configured thusly:

  • 1U
  • 64GB RAM
  • 8 x 240GB SSD
  • 2 x 10 core 2.4Ghz E5-2640 v4 Xeon  (40 threads)
  • 2 x PSU
  • Cisco 12G SAS RAID controller (more on this later)
  • 2 x 1Gbit on-board NIC

While having only 2 of his 24 RAM slots full, we can make do with this – besides, filling it up to 1.5TB of RAM will cost $14,400 – ouch! Specifically, I think we can make this into a nice host of lots and lots of virtual machines (VMs) via a modern day hypervisor.

Just 4 steps!!

Looking about at what might be a good solution for this, I cast my eyes upon LXD. I’m already a huge fan of Ubuntu, the OS LXD works best with. As well, with 2.0, Canonical is really pumping up LXD as a viable, production worthy solution for a running your own cloud compute. Given that 2.0 now ships with Ubuntu 16.04, I figured I’d give it for a spin.

I’m guessing a techie at Canonical got a hold of a marketing person at Canonical; they realized that with LXD baked into Ubuntu, anyone comfy on the command could line run 4 steps to get an instant container.  This meant that while there’s a lot more complexities to getting a LXD running in a production environment (LAN bridge, ZFS Pools, ZFS Caches, JBOD on your RAID card, static IPs dynamically assigned, all covered below), it was indeed intensely satisfying to have these 4 steps work exactly as advertised.  Sign me up!

But what about the other options?  Did I look at more holistic approaches like OpenStack?  Or even the old, free as in beer vSphere? What about the up and coming SmartOS (not to be confused with the unrelated TrueOS or PureOS)? Given the timing of the demise of SmartOS’s roots in Solaris, maybe it’s time to pay my respects? Finally, there is of course KVM.

While I will still will take up my friend’s offer to run through an OpenStack install at a later date, I think I’m good for now.

This write up covers my learning how to provision my particular bare metal, discovering the nuances of ZFS and finally deploying VMs inside LXD.  Be sure to see my conclusion for next steps after getting LXD set up.

Terms and prerequisites

Since this post will cover a lot of ground, let’s get some of the terms out of the way. For experts, maybe skip this part:

  • Ubuntu – the host OS we’ll install directly onto the C220, specifically 16.04 LTS
  • LXD – say “Lex-dee” (and not this one) – the daemon  that controls LXC. LXD  ships with 16.04 LTS. All commands are lxc
  • LXC –  Linux Containers – the actual hypervisor under LXD. Yes, it’s confusing – even the man says so ;)
  • container – interchangeably used with “VM”, but it’s a secure place to run a whole instance of an OS
  • ZFS – the file system that allows for snapshots, disk limits, self healing arrays, speed and more.
  • JBOD/RAID – Two different disk configurations we’ll be using.
  • CIMC – Cisco Integrated Management Interface which is a out of band management that runs on Cisco server hardware. Allows remote console over SSL/HTML.
  • Stéphane Graber – Project leader of LXC, LXD and LXCFS at Canonical as well as an Ubuntu core developer and technical board member. I call him out because he’s prolific in his documentation and you should note if you’re reading his work or someone else’s.
  • KVM – “Keyboard/Video/Mouse” – The remote console you can access via the CIMC via a browser. Great for doing whole OS installs or for when you bungle your network config and can no longer SSH in.
  • KVM – The other KVM ;)  This is the Kernel Virtual Machine hypervisor, a peer of LXC. I only mention it here for clarification.  Hence forth I’ll always be referencing the Keyboard/Video/Mouse KVM, not the hypervisor.

Now that we have the terms out of the way, let’s give you some home work.  If you had any questions, specifically about ZFS and LXD, you should spend a couple hours on these two sites before going further:

  • Stéphane Graber’s The LXD 2.0: Blog post series – These are the docs that I dream of when I find a new technology: concise, from the horses mouth, easy to follow and, most of all, written for the n00b but helpful for even the most expert
  • Aaron Toponce’s ZFS on Debian GNU/Linux – while originally authored back on 2012, this series of posts on ZFS is canon and totally stands the test of time.  Every other blog post or write up I found on ZFS that was worth it’s salt, referenced Aaron’s posts as evidence that they knew they were right.  Be right too, read his stuff.

Beyond that, here’s a bunch of other links that I collected while doing my research. I will call out that Jason Bayton’s LXD, ZFS and bridged networking on Ubuntu 16.04 LTS+ gave me an initial burst of inspiration that I was on the right track:

LXD Info:

ZFS Info

Prep your hardware

Before starting out with this project, I’d heard a lot of bad things about getting Cisco hardware up and running.  While there may indeed be hard-to-use proprietary jank in there somewhere, I actually found the C220M4 quite easy to use.  The worst part was finding the BIOS and CIMC updates, as you need a Cisco account to download them.  Fortunately I know people who have accounts.

After you download the .iso, scrounge up a CD-R drive, some blank media and  burn the .iso. Then, plug in a keyboard, mouse and CD drive to your server, boot from it with the freshly burned disk, and upgrade your C220. Reboot.

Then you should plug the first NIC into a LAN that your workstation is on that has DHCP.  What will happen is that the CIMC will grab an address and show it to you on the very first boot screen.

You see that IP address in there (click for large image)?  That means you can now control the BIOS via a KVM over the network.  This is super handy. Once you know this IP, you can point your browser to the c220 and never have to use a monitor, keyboard or mouse directly connected to it.  Handy times!  To log into the CIMC the first time, default password is admin/password (of course ;).

The upgraded CIMC looks like this:

I’ve highlighted two items in red: the first, in the upper left, is the secret, harder-to-find-than-it-should-be menu of all the items in the CIMC.  The second, is how to to get to the KVM.

For now, I keep the Ubuntu 16.04 install USB drive plugged into the C220.  Coupled with the remote access to the CIMC and KVM, this allows to me to easily re-install on the bare metal, should anything go really bad.  So handy!

While you’re in here, you should change the password and set the CIMC IP to be static so it doesn’t change under DHCP.

Prep your Disks

Now it’s time to set up the RAID card to have two disks be RAID1 for my Ubuntu boot drive, and the rest show up as JBOD for ZFS use.  When accessing the boot process, wait until you see the RAID card prompt you and hit ctrl+v.  Then configure 2 of your 6 drives as a RAID1 boot drive:

And then expose the rest of your disks as JBOD.  The final result should look like this:

Really, this is too bad that this server has a RAID card.  ZFS really wants to talk to the devices directly and manage them as closely as possible.  Having the RAID card expose them as JBOD isn’t ideal.  The ideal set up would be to have a host bus adapter (HBA) instead of the raid adapter. That’d be a Cisco 9300-8i 12G SAS HBA for my particular hardware. So, while I can get it to work OK, and you often see folks set up their RAID cards as JBOD just like I did, it’s sub-optimal.

Install Ubuntu 16.04 LTS + Software

F6 to select boot drive

As there’s plenty of guides on how to install Ubuntu server at the command line, I won’t cover this in too much detail.  You’ll need to:

  1. Download 16.04 and write it to a USB drive
  2. At boot of the C220, select F6 to select your USB drive as your boot device (see right)
  3. When prompted in the install processtt, select the RAID1 partition you created above.  For me this was /dev/sdg, but will likely be different for you. I used LVM with otherwise default partitions.
  4. Set your system to “Install security updates automatically” when prompted with “Configuring tasksel
  5. For ease of setup, I suggest selecting “OpenSSH Server” when prompted for “Software select”. This way we won’t have to install and enable it later.
  6. Finally, when you system has rebooted, login as your new user and you can install the core software we’ll need for the next steps:
sudo apt-get install lxd zfsutils-linux bridge-utils
sudo apt install -t xenial-backports lxd lxd-client


ZFS is not for the faint of heart.  While the LXD can indeed be run on just about any Ubuntu 16.04 box and the default settings will just work, and I do now run it on my laptop regularly, getting a tuned ZFS install was the hardest part for me.  This may be because I have more of a WebDev background and less of a SysAdmin background.  However, if you’ve read up on Aaron Toponce’s ZFS on Debian GNU/Linux’s posts, and you read my guide here, you’ll be fine ;)

After reading and hemming and hawing about which config was best, I decided that getting the most space possible was most important.  This means I’ll allocate the remaining 6 disks to the ZFS pool and not have a hot spare.  Why no hot spare?  Three reasons:

  1. This isn’t a true production system and I can get to it easily (as opposed to an arduous trip to a colo)
  2. The chance of more than 2 disks failing at the same time seems very low (though best practice says I should have different manufacturer’s batches of drives – which I haven’t checked)
  3. If I do indeed have a failure where I’m nervous about either the RAID1 or RIADZ pool, I can always yank a drive from one pool to another.

Now that I have my 6 disks chosen (and 2 already running Ubuntu), I reviewed the RAID levels ZFS offers and choose RAIDZ-2 which is, according to Aaron, “similar to RAID-6 in that there is a dual parity bit distributed across all the disks in the array. The stripe width is variable, and could cover the exact width of disks in the array, fewer disks, or more disks, as evident in the image above. This still allows for two disk failures to maintain data. Three disk failures would result in data loss. A minimum of 4 disks should be used in a RAIDZ-2. The capacity of your storage will be the number of disks in your array times the storage of the smallest disk, minus two disks for parity storage.”

In order to have a log and cache drive (which can be the same physical disk), I’ll split it so 5 drives store data and 1 drive with two partitions store my log and cache. Read up those log and cache links to see how these greatly improve ZFS performance, especially if your data drives are slow and you have a fast cache drive (SSD or better).

To find the devices which we’ll use in our pool, let’s first see a list of all devices from dmesg:

sudo dmesg|grep \\[sd|grep disk
[ 17.490156] sd 0:0:0:0: [sdh] Attached SCSI removable disk
[ 17.497528] sd 1:0:9:0: [sdb] Attached SCSI disk
[ 17.497645] sd 1:0:8:0: [sda] Attached SCSI disk
[ 17.499144] sd 1:0:10:0: [sdc] Attached SCSI disk
[ 17.499421] sd 1:0:12:0: [sde] Attached SCSI disk
[ 17.499487] sd 1:0:11:0: [sdd] Attached SCSI disk
[ 17.499575] sd 1:0:13:0: [sdf] Attached SCSI disk
[ 17.503667] sd 1:2:0:0: [sdg] Attached SCSI disk

This shows us that sdh is our USB drive (“removable”) and that the rest are the physical drives.  But which is our boot drive comprised of a RAID1 volume?  It’s the one mounted with an ext4 file system:

sudo dmesg|grep mount|grep ext4
[ 21.450146] EXT4-fs (sdg1): mounting ext2 file system using the ext4 subsystem

So now we know that sdb, sda, sdc, sde, sdd and sdf are up for grabs. The hard part with ZFS is just understanding the full scope and impact of how to set up your disks.  Once you decide that, things become pleasantly simple.  Thanks ZFS!  To set up our RAIDZ-2 pool of 5 drives and enable compression to it is just these two command:

sudo zpool create -f -o ashift=12 lxd-data raidz1 sda sdb sdc sdd sde
sudo zfs set compression=lz4 lxd-data

Note that there’s no need to edit the fstab file, ZFS does all this for you. Now we need to create our log and cache partitions on the remaining sdf disk.  First find out how many blocks there are on your drive, 468862128 in my case, using parted:

parted /dev/sdf 'unit s print'|grep Disk|grep sdf
 Disk /dev/sdf: 468862128s

Aaron’s guide suggests we can make both partitions in one go with this command, which will make a 4GB log partition and use the remainder to make a ~230GB cache partition:

parted /dev/sdf unit s mklabel gpt mkpart log zfs 2048 4G mkpart cache zfs 4G 468862120

However, this didn’t work for me. I had to go through the parted using the interactive mode. Run this twice starting both times with the sdf device:

parted /dev/sdf unit s

I forgot to record this session when I formatted my partitions, but you should end up with this when you’re done:

sudo parted /dev/sdf print
Model: ATA SAMSUNG MZ7LM240 (scsi)
Disk /dev/sdf: 240GB
Sector size (logical/physical): 512B/512B
Partition Table: gpt
Disk Flags:

Number Start  End    Size   File system Name Flags
 2     1049kB 3999MB 3998MB zfs         log
 1     4000MB 240GB  236GB  zfs         cache

OK – almost there!  Now, because linux can mount these partitions with different device letters (eg sda vs sdb), we need to use IDs instead in ZFS.  First, however, we need to find the label to device map:

sudo ls -l /dev/disk/by-partlabel/
 total 0
 lrwxrwxrwx 1 root root 10 Sep 15 11:20 cache -> ../../sdf2
 lrwxrwxrwx 1 root root 10 Sep 15 11:20 log -> ../../sdf1

Ok, so we’ll use sdf1 as log and sdf2 as cache.  Now what are they’re corresponding IDs?

sudo ls -l /dev/disk/by-id/|grep wwn|grep sdf|grep part
 lrwxrwxrwx 1 root root 10 Sep 15 11:20 wwn-0x5002538c404ff808-part1 -> ../../sdf1
 lrwxrwxrwx 1 root root 10 Sep 15 11:20 wwn-0x5002538c404ff808-part2 -> ../../sdf2

Great, now we know that wwn-0x5002538c404ff808-part1 will be our log and wwn-0x5002538c404ff808-part2 will be our cache.  Again, ZFS’s commands are simple now that we know what we’re calling.  Here’s how we add our cache and log:

zpool add lxd-data log /dev/disk/by-id/wwn-0x5002538c404ff808-part2
 zpool add lxd-data cache /dev/disk/by-id/wwn-0x5002538c404ff808-part1
 wwn-0x5002538c404ff808-part2 log
 wwn-0x5002538c404ff808-part1 cache

Now we can confirm our full ZFS status:

sudo zpool iostat -v
                                   capacity     operations    bandwidth
pool                            alloc   free   read  write   read  write
------------------------------  -----  -----  -----  -----  -----  -----
lxd-data                        1.37G  1.08T      0     16  3.26K  56.8K
  raidz1                        1.37G  1.08T      0     16  3.25K  56.5K
    sda                             -      -      0      4  1.04K  36.5K
    sdb                             -      -      0      4  1.03K  36.5K
    sdc                             -      -      0      4  1.01K  36.5K
    sdd                             -      -      0      4  1.02K  36.5K
    sde                             -      -      0      4  1.03K  36.5K
logs                                -      -      -      -      -      -
  wwn-0x5002538c404ff808-part2   128K  3.72G      0      0      4    313
cache                               -      -      -      -      -      -
  wwn-0x5002538c404ff808-part1   353M   219G      0      0      0  20.4K
------------------------------  -----  -----  -----  -----  -----  -----

Looking good!  Finally, we need to keep our ZFS pool healthy by scrubbing regularly.  This is the part where ZFS self heals and avoids bit rot. Let’s do this with a once per week job in cron:

0 2 * * 0 /sbin/zpool scrub lxd-data


Now, much thanks to Jason Bayton’s excellent guide, we know how to have our VMs get IPs in on LAN instead of being NATed.  Right now my NIC is enp1s0f0 and getting an IP via DHCP.  Looking in  /etc/network/interfaces I see:

# The loopback network interface
auto lo
iface lo inet loopback

# The primary network interface
auto enp1s0f0
iface enp1s0f0 inet dhcp

But to use a bridge (br0) and give my host the static IP of, we’ll make that file look like this:

# The loopback network interface
auto lo
iface lo inet loopback

# The primary network interface
auto br0 
iface br0 inet static
       bridge_ports enp1s0f0
iface enp1s0f0 inet manual

This will allow the VMs to use br0 to natively bridge up to enp1s0f0 and either get a DHCP IP from that LAN or be assigned a static IP. In order for this change to take effect, reboot the host machine.  Jason’s guide suggests running ifdown and ifup, but I found I just lost connectivity and only a reboot would work.

When you next login, be sure you use the new, static IP.


Set file limits

While you can use your system as is to run containers, you’ll really want to updated per the LXD recommendations.  Specifically, you’ll want to allow for a lot more headroom when it comes to file handling.  Edit sysct.conf:

sudo vim /etc/sysctl.conf

Now add the following lines, as recommended by the LXD project:

fs.inotify.max_queued_events = 1048576
fs.inotify.max_user_instances = 1048576
fs.inotify.max_user_watches = 1048576

As well, edit limits.conf

sudo vim /etc/security/limits.conf

Now add the following lines. 100K should be enough:

* soft nofile 100000
* hard nofile 100000


Now that we have our bare metal provisioned, we have our storage configured, file limits increased and our network bridge in place, we can actually get to the virtual machine part, the LXD part, of this post. Heavily leveraging again Jason Bayton’s still excellent guide, we’ll initialize LXD by running lxd init. This go through the first time LXD guide and ask a number of questions about how you want to run LXD.  This is specific to 2.0 ( >2.1 has different questions).  You’ll see below, but the gist of it is that we want to use our ZFS pool and our network bridge:

lxd init
 Name of the storage backend to use (dir or zfs) [default=zfs]:
 Create a new ZFS pool (yes/no) [default=yes]? no
 Name of the existing ZFS pool or dataset: lxd-data
 Would you like LXD to be available over the network (yes/no) [default=no]? yes
 Address to bind LXD to (not including port) [default=all]:
 Port to bind LXD to [default=8443]:
 Trust password for new clients:
 Do you want to configure the LXD bridge (yes/no) [default=yes]?

Note that we don’t accept the ZFS default and specify our own ZFS pool lxd-data. When you say “yes”, you want to configure the bridge, you’ll then be prompted with two questions.  Specify br0 when prompted. Again, refer to Jason Bayton’s guide for thorough screen shots:

Would you like to setup a network bridge for LXD containers now? NO
Do you want to use an existing bridge? YES
Warning: Stopping lxd.service, but it can still be activated by:
 LXD has been successfully configured

Success!  LXD is all set up now.  Let’s prep the host OS for a bunch of open files now.

Hello world container

Whew!  Now that all the hard parts are done, we can finally create our first container. In this example, we’ll be creating a container called  nexus and put some basic limitations on it. Remember, do not create these as root! One of the massive strengths of LXD is that it’s explicitly designed to be secure and running containers as root would remove a lot of this security. We’ll call lxc init and then pass some raw commands to set the IP to be .52 in our existing /24 subnet on the bridge. If you run lxc list our new container shows up. That all looks like this:

lxc init ubuntu: nexus
echo -e " =\ =\n" | lxc config set nexus raw.lxc -
lxc start nexus

lxc list
| NAME  | STATE   | IPV4              | IPV6 | TYPE       | SNAPSHOTS |
| nexus | RUNNING | (eth0) |      | PERSISTENT | 0         |

Finally, we want to limit this container to have 4 CPUs, have 4GB of RAM and 20GB of disk.  Like before, these commands are not run as root. They are imposed on the container in real time and do not require a restart.  Go LXD, go!

lxc config set nexus limits.cpu 4
lxc config set nexus limits.memory 4GB
lxc config device set nexus root size 20GB

You’re all done!  From here you can trivially create tons more containers.  You can let them have an ephemeral IP via DCHP on the bridge or easily set a static IP. If you have a resource intensive need, you don’t set the limits above, the container will have access to the full host resources.  That’d be 40 cores, 64GB RAM and 1TB of disk. If you created more VMs with out resource limiting, LXC would do the balancing of resources so that each container gets it’s fair share.  There’s a LOT more configuration options available.  As well, even the way I declared a static IP can likely be done via dnsmasq, (see this article), but I had trouble getting that to work on my setup, hence the raw calls.

Next steps

Now that you you’ve bootstrapped your bare metal, dialed in your storage back end, specifically deployed your LAN, you should be all set, right?  Not so fast! To make this more of a production deployment, you really need to know how (and practice!) your backup and restore procedures.  This likely will involve the snapshot command (see post 3 of 12) and then backing those snapshots off of the ZFS pool. Speaking of the ZFS pool, our set up as is doesn’t have any alerting if a disk goes bad.  There’s many solutions out there for this – I’m looking at this bash/cron combo.

Having a more automated system to provision containers integrated with a more DevOps-y set up makes sense here. For me this might mean using Ansible, but for you it might be something else!  I’ve heard good things about cloud-init and Juju charms. As well, you’d need a system to monitor and alert on resource consumption.  Finally, a complete production deployment would need at least a pair, if not more, of servers so that you can run LXD in a more highly available setup.

Given this is my first venture into running LXD, I’d love any feedback from you!  Corrections and input are welcome, and praise too of course if I’ve earned it ;) Thanks for reading this far on what may be my longest post ever!


Howto: Sympa 6.2 on Ubuntu 17.04

This post is a continuation of the my last post on the topic, “Howto: Sympa 6.1 on Ubuntu 16.04“.  It should come as no surprise that this is about installing Sympa on the most recent version of Ubuntu to get the most recent version of Sympa (at the time of this writing).  That’d be Sympa 6.2.16 and Ubuntu 17.04. The steps only vary a little between between the two, but here’s the all them for completeness.


Like the last post, this one assumes you have root on your box.  It assumes you have Apache2 installed.  It assumes you’re running a stock Ubuntu 17.04 install.  It assumes you want to run Sympa on your server.  It also assumes you’ll be using Postfix as the lists MTA. It assumes you have a DNS entry (A record) for the server.  As well it assumes you also have an MX record pointing to the A record or no MX record so the MX defaults to the A record.  If this doesn’t apply to you, caveat emptor!

To recap, that’s:

  • Apache 2 installed and working
  • Postfix as MTA
  • Ubuntu 17.04 server
  • Existing DNS entry
  • Run all commands as root

I also was using this server solely to serve Sympa mail and web traffic so if you have a multi-tenant/multi-use server, it may be more complicated.


These steps assume you’re going to install Sympa on  There’s no reason you couldn’t use instead. Your zero step should be sudo apt-get update&& sudo apt-get upgrade.

  1. First step is to install Sympa. As well, we’ll install some packages that are used to fix a bug in 6.2’s GUI where the drop down nav menus don’t work:
    apt-get install -y sympa javascript-common libjs-jquery-migrate-1
  2. When prompted during this install:
    1. Choose a good mysql root password and enter it when prompted
    2. Please select the mail server configuration type that best meets your needs: Internet Site
    3. System mail name:
    4. Which Web Server(s) are you running?: apache 2
    5. Configure database for sympa with dbconfig-common?: Yes
    6. Database type to be used by sympa: mysql
    7. MySQL application password for sympa: <blank> (will assign random one)
  3. Now it’s time to fix that css bug in 6.2 with a copy:
    cp /usr/share/sympa/default/web_tt2/head_javascript.tt2 /etc/sympa/web_tt2

    And then edit /etc/sympa/web_tt2/head_javascript.tt2 and add the jquery-migrate-1.js file after jquery.js:

    <script src="[% static_content_url %]/external/jquery.js"></script>
    <script src="/javascript/jquery-migrate-1.js"></script>
  4. Edit /etc/sympa/sympa/sympa.conf to match the following values:
    default_home home
    create_list intranet

    The “intranet” value will prevent some one from signing up and requesting a list with any approval. should show the sympa UI, w00t!

  5. Ensure Sympa starts at boot:
    update-rc.d sympa defaults
    update-rc.d sympa enable
  6. ensure postfix is updated in /etc/postfix/ edit these values to match:
    myhostname =
    alias_maps = hash:/etc/aliases,hash:/etc/mail/sympa/aliases
    alias_database = hash:/etc/aliases,hash:/etc/mail/sympa/aliases
    mydestination = $myhostname,,,, localhost
    relay_domains = $mydestination,
    local_recipient_maps =
  7. add default aliases for sympa at the top of /etc/mail/sympa/aliases:
    ## main sympa aliases
    sympa: "| /usr/lib/sympa/bin/queue"
    listmaster: "| /usr/lib/sympa/bin/queue"
    bounce+*: "| /usr/lib/sympa/bin/bouncequeue"
  8. reboot and rebuild aliases:

Sympa should now be up and running at!  All mail in and out should work so you can run your own list server. Please report any problems so I can keep this post updated and accurate – thanks!


Howto: Sympa 6.1 on Ubuntu 16.04

Recently I was tasked at work to get an instance of Sympa set up. Their docs are a bit scattered, but I found a promising post on which suggested I could get away with an apt-get install instead of needing to compile from source. Well, it turns out I did get it working, but only after a lot of trial and error. Given that some one else might be trying to do this, and because I had to document the exact steps for work, here’s a handy dandy blog post which I hope will help some one trying to do the same thing.

Good news for those looking to do this for Sympa 6.2 (latest at time of publishing), I have a post on how to do this exact thing on the soon to be released Ubuntu 17.04 with Sympa 6.2.  Stay tuned!


This post assumes you have root on your box.  It assumes you have Apache2 installed.  It assumes you’re running a stock Ubuntu 16.04 install.  It assumes you want to run Sympa on your server.  It also assumes you’ll be using Postfix as the lists MTA. It assumes you have a DNS entry (A record) for the server.  As well it assumes you also have an MX record pointing to the A record or no MX record so the MX defaults to the A record.  If this doesn’t apply to you, caveat emptor!

To recap, that’s:

  • Apache 2 installed and working
  • Postfix as MTA
  • Ubuntu 16.04 server
  • Existing DNS entry
  • Run all commands as root

I also was using this server solely to serve Sympa mail and web traffic so if you have a multi-tenant/multi-use server, it may be more complicated.


These steps assume you’re going to install Sympa on  There’s no reason you couldn’t use instead.

  1. Install sympa:
    apt-get install -y sympa
  2. When prompted during this install:
    1. Choose a good mysql root password and enter it when prompted
    2. Please select the mail server configuration type that best meets your needs: Internet Site
    3. System mail name:
    4. Which Web Server(s) are you running?: apache 2
    5. Database type to be used by sympa: mysql
    6. MySQL application password for sympa: <blank> (will assign random one)
  3. Sympa 6.1 has a present! It ships with a bug.  Fix the regex on line 126 of /usr/share/sympa/lib/ so it looks like this (note “{” is now “\{” ):
    if ($cookie and $cookie =~ /^\d\{,16}$/) {
  4. Edit /etc/sympa/wwsympa.conf to change line 81 to 1 instead of 0:
     use_fast_cgi 1

    If you don’t do this step, you’ll see full HTML pages show up in /var/logs/syslog and only 500 errors in the browser :( should show the sympa UI, w00t!

  5. Ensure Sympa starts at boot:
    update-rc.d sympa defaults
    update-rc.d sympa enable
  6. ensure postfix is updated in /etc/postfix/ edit these values to match:
    myhostname =
    alias_maps = hash:/etc/aliases,hash:/etc/mail/sympa/aliases
    alias_database = hash:/etc/aliases,hash:/etc/mail/sympa/aliases
    mydestination = $myhostname,, , localhost
    relay_domains = $mydestination,
  7. Update /etc/sympa/sympa.conf so that these values match:


  8. update /etc/sympa/wwsympa.conf so that these values match:
    default_home lists
    create_list intranet

    The “intranet” value will prevent some one from signing up and requesting a list with any approval.

  9. add default aliases for sympa at the top of /etc/mail/sympa/aliases:
    ## main sympa aliases
    sympa: "| /usr/lib/sympa/bin/queue"
    listmaster: "| /usr/lib/sympa/bin/queue"
    bounce+*: "| /usr/lib/sympa/bin/bouncequeue"
  10. reboot and rebuild aliases:

Sympa should now be up and running at!  All mail and and out should work so you can run your own list server. Please report any problems so I can keep this post updated and accurate – thanks!


Happy Histogram

The other day I was working on redoing our old Internet governance meeting page. On it, there was a simple histogram of the meetings across a year. It looked like this:

Seeing this server side generated, non-responsive, raster based image, I thought, “We can make this waaaaay better”. So, I went about looking for a drop in replacement that would fix all that, much like MapTable fixed our server side, raster based non-responsive maps.

I tried a number of full blown graphing libraries (jqPlot, Flot and jQuery sparkline, all awesome in their own right) as well, I looked at the top available JS solutions. They all failed me in one way or another. Some of them didn’t like having hundreds of data points being rendered on them. Some of them added unwanted anti-aliasing on 1 pixel wide elements (I’m looking at you Canvas!). Many of them were going to take a lot of effort to look good on mobile devices at the horizontal scaled I was looking for.

So, decided roll our own. And, of course, I decided to open source it ;) Had I drawn it before I made it, I would have drawn something like this:


The idea is that the whole thing would just be made of DIVs styled by CSS. Since CSS is insanely precise, even allowing for negative margins, you can skooch things like x and y axes labels just where you want them.

Thus, Happy Histogram was born:


What you see there is the exact same graph as above, rendered with a whopping 900 byte JavaScript Library. Check it out on Github!

If you want to customize it, just throw a little CSS at it and make it your own:

#year2 .yearHistogram .emptyTop { background-color: yellow; }
#year2  .yearHistogram .filledBottom:hover {background-color: #ddd;}
HappyHistogram('year2', Months, 'orange');


Or maybe you just want to show one value per month? Easy Peasy:



Likely I won’t do too much more development on this unless as it is fully formed and already deployed. However, part of my job satisfaction right now stems from the fact that all new features always get pushed upstream, so y’all might see more work on this soon!


Untangling odd behavior of stupidtable.js

I’ve been doing a lot of work in the day job working with Joseph McCullough’s wonderful Stupid-Table-Plugin for jQuery. This allows you to take a static table and with a little JS code, make it into a dynamic, client side sortable table. As well it’s HIGHLY configurable and totally awesome!

The Odd Behavior

However, I noticed when I had a table with data like this:

<table class="stupid-sort" >
    <th data-sort="string" data-col="fruit" data-sort-default="desc">Fruit</th>
    <th data-sort="string" data-col="color" data-sort-default="desc">Color</th>

Sorting by “Fruit” worked exactly as expected. However, sorting by “Color” yielded odd results. While the 4 colors always were in the correct order (grouped together before or after “red”), the fruit column would change order if you clicked the “Color” header more than twice. Very confusing! Upon close inspection, on the 1st, 5th, 9th etc. clicks of “Color” the “Fruit” columns would be the same order and the same is true of 2nd, 4th, 6th etc. clicks. It then dawned on me that Stupid-Table-Plugin was reverse sorting rows with identical values each time I clicked the same “Color” column head.

The untangle

The fix, once knowing the problem, was easy. Just declare a data-sort-value for each color and append a random string to it (eg foo, bar, baz):

<table class="stupid-sort" >
    <th data-sort="string" data-col="fruit" data-sort-default="desc">Fruit</th>
    <th data-sort="string" data-col="color" data-sort-default="desc">Color</th>
    <td data-sort-value=''>yellow</td>
    <td data-sort-value=''>yellow</td>
    <td data-sort-value='yellow.baz'>yellow</td>
    <td data-sort-value='red'>red</td>

This way the colors will always be sorted in a consistent order. Point haired bosses rejoice – your underling coder made your table sort right!


I want to again thank Stupid-Table-Plugin for being awesome. Great discussions like Issue #106, “weird behavior when td have empty values” both gave me faith in the author as being open to feedback, and made me trust the plugin to be well thought out. Long term, this of course should be fixed, and hopefully will be with Issue #112 – “Sort on Multiple Columns”.

Finally, if you are sorting a BIG table with a lot of data-sort="int" instead of data-sort="string", like above, you can simply make your values into data-sort="float" and concatenate a random, but unique to that row, decimal onto data-sort-value (eg, “300” => “300.0002” and “90” => “90.0301”). This way the sort will still work. If you tried to sort the float values as a strings, it would put “300” before “90” ;)


Weak Crypto in Star Trek Beyond (SPOILERS!!)

Star-Trek-BeyondLet’s get this out of the way upfront: the most recent installation of the Star Trek reboot, “Beyond”, was exactly what I’d hoped it would be. It was an action packed, summer fun movie. It did not exceed, but definitely met my expectations.

The end however, was totally silly. It was the peak of the action and the sound effects and music (more on this in a second) were so loud I don’t think any one could hear my wife and I laughing. Like, laughing to the point of crying because what we were watching was so funny. But, for me at least, I kept on thinking, “wait – what?” There’s no way modern crypto would allow this scene! Let alone hyper futuristic crypto of 2263! Let me explain.

spock.bonesIn the end of the move (yes, like I said, spoilers) the Enterprise crew is rocking the 99 year old USS Franklin. They are fighting the boss of the movie, Krall (not to be confused with Khan), who has control of a massive swarm of space ships based on alien technology. The swam is all perfectly in sync, just as you’d expect your alien swarm army to be. This, by definition, means that they are communicating with each other in real time to coordinate complex movements in 3d space. And how did they figure out how to beat the boss? Well, simply by “corrupting their communication with radio waves”. Radio waves like Sabotage, by the Beastie Boys. That’s right, the Beasties are literally weaponized to literally blow up the enemy hordes. Here’s the Franklin “surfing” the enemy swarm with them blowing up in their wake (click through to see the the preview where the gif came from):


So, aside from the fact that radio waves travel at the speed of of light and should be radiating out in all directions, not trailing behind them, how exactly was this supposed to be working? According to wikipedia, “they [Spock and Bones] learn that VHF transmissions can disrupt Krall’s communications and destroy his fleet.” But, like I said, this is the year 2263! Surely the alien technology powering this swarm has sufficient encryption to ensure that simple VHF radio waves won’t interfere with communication, right? Beyond things like TLS Handshakes, PKI (or even hash-chains if you wanna get tricky), basic header checksums, like in IPv4 would prevent this type of interference.

MD5But, maybe not? Maybe the aliens implementing the swarm networking were just lazy? For example, the MD5 hashing algorithm was released in 1991. By the late 90’s and early 2000’s it was the de facto way to store passwords in databases. However, as early as 2005 collisions had been proven to be a reality. In 2008 at the 25th Chaos Communication Congress, researchers categorically proved that no one should trust MD5. Later that same year CERT issued a CVE agreeing. Fast forward 4 years to 2012, and what do we find? 43 million hacked passwords in unsalted MD5 is what we find. Well, we found out today (Sep 1, 2016), but the coders responsible for securing tens of millions of accounts back in ’12 should have known better. So yeah, maybe the aliens were more focused on meeting deadlines than they were in defending against VHF interference?


A great, small USB-C charger for your XPS 13 (or MacBook) (Updated 4/16/17 DO NOT BUY)

5/12/18 Update: I’ve found a new charger that’s working well (and seemingly safely) – check it out!

4/17/17 Update: These chargers have been deemed unsafe. Please do not use them. See this post for details.

While I was on vacation, I forgot my laptop charger. I found on amazon that you can get an OEM charger for the XPS 13 (9350) for about $30. This is fine, it will work the prior gen XPS 13 with out USB-C/Thunderbolt too. I got it and used it for the remainder of my trip.

Then, when I got home, I wondered if there were any good, cheap, small USB-C chargers that would work with my laptop. Lo, the RAVPower USB-C charger. Full title, on Amazon, is, “USB C Charger RAVPower 36W Dual Ports USB Wall Charger with USB-C (20V 1.5A, 15V 2A, 9V 2A, 5V 3A max) for MacBook, Dell XPS 13, Nexus 5X/ 6P and iSmart (5V 2.4A max) for iPhone and more Black”. OK, what ever RAVPower, too long on the title. But you know what is not too long? The products it works with:

  • Apple’s 2015 MacBook
  • Google’s second Chromebook Pixel
  • ASUS Transformer book T100HA
  • Dell XPS 13 & 15
  • HP Elite x2 1012 G1
  • Razer Blade Stealth, Blade (2016)
  • Nokia N1, ASUS ZenPad S8, Google’s Pixel C
  • Nexus 5X, Nexus 6P
  • Asus ZenFone 3 Deluxe
  • BLU Vivo XL and BLU Vivo 5
  • HTC 10, LG G5
  • Microsoft Lumia 950, 950 XL
  • Motorola Moto Z, Moto Z Force

Totally awesome list! I got the charger today, so I don’t have any long term usage notes just yet. However, given it can both charge my phone over speedy USB-C, or dual charge my laptop AND my phone, as it has normal USB port to charge with, I’m fairly confident it’s my new fave travel charger. One caveat is that it does not come with a USB-C to USB-C cord. I got a nice looking 10′ one which I’ll put a velcro tie on when it arrives.

If you are an XPS 13 user, or thinking about becoming one, be sure to read my write up on running Ubuntu 16.04 after purchasing the windows version of the 9350.

1/4/17 Update – I’ve been using this charger for 4 months now and it’s awesome! I got solid, braided 10′ USB-C <-> USB-C cable to use with which I recommend as well. It comes in 6′ flavors as well. Cable aside, the only caveat is that the charger does get quite hot – but so far not so much that it has me concerned; I don’t see any signs of the plastic discoloring.