This repository generates the main docker-compose.yml (follow the steps to generate this here) file that powers the Nebra miners.
The docker-compose.yml file is pushed to Balena (using GitHub Actions), which in turn pulls down the various Docker images outlined below.
There are currently five different services running within this device, which are all outlined below.
Repo: github.com/NebraLtd/hm-diag
The diagnostics container is designed for local troubleshooting. It runs a local web server that displays various diagnostics data.
Note that this container is also responsible for serving content to the Hotspot-Production-Tool and also contains the gateway-mfr-rs tool which configures the ECC Key in production.
Repo: github.com/NebraLtd/hm-pktfwd
This container is responsible for configuring packet forwarder's region and starts the radio module.
Repo: github.com/NebraLtd/hm-config
This container is (partially) responsible for the device onboarding and provides the Bluetooth LE to allow the hotspot to be configured via the Helium App. It is also responsible for configuring WiFi.
As of 27th Jan 2022, we have also moved the UPnP functionality into this repo instead of the (now archived) UPnP repo. This attempts to use UPnP to set up a port forwarding rule if your router supports it and the function is turned on in your router settings. This is possible in hm-config because we already use "host networking" in this container.
Removing the UPnP container reduces CPU overhead and redundant code.
Repo: github.com/NebraLtd/hm-miner
This container is the actual Helium Miner software (from upstream), with the required configuration files added.
Repo: github.com/balenablocks/dbus
This container configures a DBus session bus instance that is used by the gateway config container to communicate with the miner code (note that there is also a separate system bus running on the host OS which is used by gateway config to communicate with bluez for configuring Bluetooth services). This removes the need to have a custom com.helium.Miner.conf dbus config file on the host OS as was done previously (and meant we had to run a custom balena device type).
This is a guide to help you get started with the repository and get it running on your local device. This guide is focused on pushing the repository onto a Raspberry Pi using Balena.
Prerequisites:
- Local Test Device (Ex: Raspberry Pi)
- Computer for development and pushing to the device
- Git installed - download here
- Balena CLI (Install located on Balena step in quick start steps below)
This is used for deploying locally, it automatically generates one for each device as part of the github workflow.
You may notice that after cloning the repo that you are missing a docker-compose.yml file which is required to push the containers to balena. This is a result of us generating the docker-compose file based on a settings.ini file in the root directory. You can also find the latest docker-compose files used in production in the device-compose-files folder (generated via github workflow). Here are the steps of generating a new docker-compose.yml file:
Note for generating the file for local tests: The repo's short Git SHA is fetched from GitHub pipeline automatically while working with standard procedure. But if you need to test your working copy locally, you need to create an environment variable first via running following command on Linux:
$ export FIRMWARE_SHORT_HASH=$(git rev-parse --short HEAD)Then generate the docker compose file in the same console with following procedures.
Step 1: Update the settings.ini file to accept the new values which you wish to set in the generation of the docker-compose.yml file
Step 2: Generate your respective .yml file based on the device you will be pushing to:
- RPI (Raspberry Pi based device):
python gen_docker_compose.py rpi -o device-compose-files/docker-compose-rpi.yml - ROCKPI (ROCK Pi based device):
python gen_docker_compose.py rockpi -o device-compose-files/docker-compose-rockpi.yml
Step 3: Copy device compose file into root directory as docker-compose.yml:
- RPI:
cp device-compose-files/docker-compose-rpi.yml ./docker-compose.yml - ROCKPI:
cp device-compose-files/docker-compose-rockpi.yml ./docker-compose.yml
Step 4: Validate that the changed values you applied to settings.ini were applied in the docker-compose.yml file
Step 5: Continue with the steps below to push the containers to your fleet via balena
Step 1: Clone the repository to your local machine using one of the following commands:
- HTTP:
git clone https://github.com/NebraLtd/helium-miner-software.git - SSH:
git clone [email protected]:NebraLtd/helium-miner-software.git
Step 2: Follow the getting started guide for Balena to help you install Balena on your local test device and get a fleet setup so you can start pushing code to it.
Step 3: Once you've gone through the steps and have Balena setup with your device in your fleet, open your cli terminal and navigate to the root directory of the cloned repository (Ex: /usr/name/documents/helium-miner-software).
Step 4: Once you're at the root directory. You want to push the code by running the following command:
$ balena push <fleet-name>Step 5: Once complete check your fleet on the Balena dashboard and all modules should be running on the local test device.
For some Nebra Hotspots that use spidev1.2 you may need to add a DT overlay in the device or fleet configuration section on balenaCloud to enable spi1.
Additionally, for the SPI ethernet based Nebra Light Hotspot you need to add the DT overlay for the enc28j60. And for UART based GPS that is on non-standard UART pins you need to add the uart dtoverlay.
To do this you need to find the "Define DT Overlays" section, click activate and then add "enc28j60","spi1-3cs","uart0,txd0_pin=32,rxd0_pin=33,pin_func=7"
GitHub Actions is used to deploy directly to Balena.
The master branch is mapped to "Testnet" (i.e. our staging environment), and the production branch is mapped to the production environments.
Do note that the Docker Tag is hard coded in the docker-compose.yml file.
Whilst in theory you can run matrix builds to build to several balena fleets simultaneously using GitHub Actions we found (at the time of writing - December 2021) that this did not perform satisfactorily in practice. We deploy on a production release now to 16 separate fleets meaning 16 separate builds triggered via the balena API simultaneously. Somewhere between 50 and 100% of the builds would fail each time.
We have discovered a useful fix for this using the following two workarounds:
- use multiple API keys within the action to avoid any rate limiting or timeout issues.
- add random periods of delay to each matrix action run using using a random sleep on the command line within the action.
This has allowed our builds to continue deploying successfully.
On occasion (as originally described in this issue) CI/CD builds via GitHub actions will fail to push correctly to balena. We have found that as a workaround, it is often possible to force push these to balena as draft releases using the --draft tag in the balena deploy command of the balena CLI tool. For example:
balena deploy nebraltd/helium-indoor-868 --logs --debug --nocache --build --draft
This can either be done manually by someone with the right privileges on our balenaCloud account, or it can be done by triggering the workflow_dispatch: event on the push-to-prod-draft.yml github action from the GitHub actions menu (making sure to use the production branch). There is also an equivalent action for our testnet fleet.
Once this has completed and the draft builds have been pushed to balena, you will need to either finalise the releases using balena CLI locally, or visit the releases menu of each fleet with a draft release, click on the release and then select Finalize Release from the dropdown menu.
@TODO:
- Write more comprehensive QA/QC checklist, including how to QA the onboarding part
- Write automated health checks for post deploys
Typically before merging into production the following checks are performed:
- Containers are updated on the master branch to versions ready to be deployed.
- This is deployed automatically to devices on the testnet.
- Open a PR from master to production branch.
- Typically leave in testnet for at least 3 Hours to see if any issues are reported from testnet users.
- Check to see if units on testnet are still synced, beaconing & witnessing as expected. (Typically Teeny Felt Rook & Gigantic Basil Turtle are good choices to check due to their locations)
- If all looks good, merge into production branch and monitor updates on balena dashboard.
Please refer to this guide for more information.
This software uses a mixture of information from: