I am using Bluex to communicate with a BLE device from an Elixir application. It works great on a Linux host with a Bluetooth dongle, but I have troubles to get it work on a Raspberry Pi with the same dongle.
So far, I have built a custom Nerves system, including Bluetooth support in the Kernel and bluez-utils. Using the CLI utils, I am able to scan and view my device, so the system is OK. Now, the issue is that Bluex is not finding the DBus socket. There is effectively no /run/dbus/system_bus_socket.
Is there something I can do to have the DBus socket present? As Nerves systems are really minimalistic, I assume nothing is started apart from the BEAM, so I should start DBus and bluetoothd myself, isn’t it? If you have hints on how I should do that properly, I’m all ears!
6 Likes
Great to see some bluetooth love.
You are on the right track. You will have to start dbus and bluetoothd manually as part of your supervision tree. Check out muontrap
2 Likes
Yes, I am writing a tiny app to test that a BLE device works properly, and Elixir is a great language to get things done quickly.
Thanks for the tip, I’ll check muontrap and see what I can do from that.
3 Likes
Ooo, this looks awesome, many uses for it! Deserves it’s own post!
I believe it was announced on the forum here a while back. Its really useful for this kind of thing.
1 Like
Hmm, I checked the forum via a quick search and there was no result, did a bigger google search on the site and no result, so not yet. 
hm maybe i made that up then. I’ll poke Frank 
1 Like
Hey there 
. I’m on the footsteps of @ejpcmac in setting up bluetooth an I’m following @ConnorRigby 's advice of using muontrap. I am stuck at creating the /run/dbus directory that dbus-daemon --system needs. I’ve tried multiple approaches:
- Creating the directory in
post-build.sh with mkdir -p /run/dbus
- Creating the directory through
rootfs_overlay
- Creating the directory with a new config file in
/etc/dbus.config with content -d /run/dbus
All of them failed in providing a /run/dbus directory at boot time.
Can someone steer me back on the right path? 
Your help is much appreciated! Thanks!
In your Elixir application’s start function, call File.mkdir("/run/dbus").
Here’s the longer answer:
The default Nerves configuration mounts several different filesystems. The main one, /, is read-only and contains OTP, Linux programs, and your Elixir application and all of its dependencies. You can add to that one via the rootfs_overlay. However, some directories are made writable at runtime. Anything placed in those directories at compile-time isn’t accessible at runtime and that’s what you’re running into. Examples are the /tmp, /dev, /run, and /root directories. In the case of /tmp, /dev, and /run, Nerves mounts small in-memory filesystems. They’ll lose their contents on reboots so if you’re going to create a directory in them, you need to do it each boot. This is generally desirable since if something gets messed up in them, a reboot returns you to a good state. In the case of /root, a writable filesystem is mounted there so that iex prompt history and other things survive reboots.
As a general rule of thumb, I try to avoid using the rootfs_overlay feature. Usually Linux programs have a way to tell them where their configuration files are. That lets me put the config file in a more convenient location - like in my OTP application’s priv directory and tell it to read it from there. If programs need to write data, I can redirect them some place on /data, etc. Of course, it’s nice to see things work once before moving paths around, so this might be something to consider after the dust settles.
3 Likes
Thanks for the great answer! That explains everything. Happy New Year, cheers!
