Intelligently connect data sources with data sinks in block-based balena applications.
The connector
block is a docker image that runs telegraf and code to find other services running on the device, and intelligently connect them.
- Automatically finds HTTP data sources on the device
- Automatically finds supported data storage services on the device
- Configurable HTTP listener
- Configurable external HTTP data source(s)
- Configurable HTTP output
- Device metric support (CPU and Memory usage)
To use this image, create a container in your docker-compose.yml
file as shown below:
version: '2.1'
services:
connector:
image: bh.cr/balenalabs/connector-<arch>
restart: always
labels:
io.balena.features.balena-api: '1' # necessary to discover services
io.balena.features.supervisor-api: 1 # necessary to discover services in local mode
privileged: true # necessary to change container hostname
ports:
- "8080" # only necessary if using ExternalHttpListener (see below)
You can also set your docker-compose.yml
to build a dockerfile.template
file, and use the build variable %%BALENA_ARCH%%
so that the correct image is automatically built for your device type (see supported devices):
docker-compose.yml:
version: '2.1'
services:
connector:
build: ./
restart: always
labels:
io.balena.features.balena-api: '1' # necessary to discover services
io.balena.features.supervisor-api: 1 # necessary to discover services in local mode
privileged: true # necessary to change container hostname
ports:
- "8080" # only necessary if using ExternalHttpListener (see below)
dockerfile.template
FROM bh.cr/balenalabs/connector-%%BALENA_ARCH%%
The connector
block has been tested to work on the following devices:
Device Type | Status |
---|---|
Raspberry Pi 3b+ | ✔ |
Raspberry Pi 4 | ✔ |
Intel NUC | ✔ |
Generic AMD64 | ✔ |
This type of data source runs it's own HTTP server and provides data readings as json
strings. The service must expose port 7575
like this:
sensor:
build: ./sensor
expose:
- '7575'
The connector
block will find this service and configure telegraf to periodically pull from it via HTTP.
The default timeout for retrieving data is 2 seconds. You can change this by setting INTERNAL_HTTP_TIMEOUT
to the number of seconds (e.g. 4
).
By adding an MQTT broker to an application, you can push data into the connector
block. Add your broker such as:
mqtt:
image: arm32v6/eclipse-mosquitto
ports:
- "1883:1883"
restart: always
As long as you call the service mqtt
the connector
block will automatically find it and configure telegraf to pull data from the broker. Ensure the data is formatted as json
strings. Telegraf will be configured to only pull from the sensors
topic, so any other data you may wish to put onto the MQTT broker will not be stored (e.g. control or signalling messages).
Example code:
client = mqtt.Client("1")
client.connect("localhost")
while(True):
value = GetReading() # code omitted for brevity
client.publish("sensors",json.dumps(value))
time.sleep(5)
By default any string fields recieved from MQTT are ignored. For any fields you want to be brought in you will need to specify them in a variable called MQTT_INPUT_STRINGS_FIELDS
as a comma-separated list. See the section HTTP section below for a worked example.
This type of source is pulled from a provide via the internet. It is enabled by adding an environment variable to the connector
service called EXTERNAL_HTTP_PULL_URL
and setting it to the URL of the source:
Setting the vaiable EXTERNAL_HTTP_PULL_NAME
(as above) allows you to rename the resulting data source, otherwise it will appear in your data sinks (see below) as inputs.http
.
Some HTTP APIs that you might like to use with EXTERNAL_HTTP_PULL
will require authorization. For that reason you can pass additional parameters using the format EXTERNAL_HTTP_PULL_HEADER_<header-name>
. For example: EXTERNAL_HTTP_PULL_HEADER_Authorization
could be set to Basic: YWxhZGRpbjpvcGVuc2VzYW1l
.
By default any string fields recieved from a HTTP API are ignored. For any fields you want to be brought in you will need to specify them in a variable called EXTERNAL_HTTP_PULL_STRINGS_FIELDS
as a comma-separated list. Here's a worked example:
Say my weather HTTP API brings in the following JSON:
{
"timezone":3600,
"id":2643743,
"name":"London",
"cod":200
"sys":{
"type":1,
"id":1414,
"country":"GB",
"sunrise":1597726289,
"sunset":1597778246
},
"weather":[
{
"id":802,
"main":"Clouds",
"description":"scattered clouds",
"icon":"03d"
}
]
}
In that example, to bring in the "name" field so that "London" appears in my data, I need to add name
to my EXTERNAL_HTTP_PULL_STRINGS_FIELDS
variable.
However, because the "country" element is nested within the "sys" element, I need to using some notation to specify that JSON path, like this sys_country
.
Notice also that the "weather" element has an array of nested elements, including a description of the weather. To get that description I'll need to specify the path (like above), but I also need to specify the index of the array element, in this case "0". So I'll add weather_0_description
to my environment variable. All together that will look like this:
This type of data source pushes to your device. It is configured by enabling a built-in HTTP listener with the environment variable ENABLE_EXTERNAL_HTTP_LISTENER
set to 1
:
Again, the resulting data source can be given a custom name (as above) by setting the EXTERNAL_HTTP_LISTENER_NAME
variable.
Additionally, you sometimes need to specify a json_query
path - which effectively limits the portion of the JSON document being parsed. This path can be specified with the EXTERNAL_HTTP_LISTENER_JSON_QUERY
variable.
This data source provides the in-built telegraf metrics for the CPU and Memory usage of the device. It is enabled by setting the environment variable ENABLE_DEVICE_METRICS
to 1
.
This data source is useful for testing connector
or simply to allow device resource monitoring as part of your application.
Adding an influx timeseries database to your application will cause the connector
block to configure telegraf to push data into it. You must name the service influxdb
for it to be automatically discovered, such as:
influxdb:
image: influxdb@sha256:73f876e0c3bd02900f829d4884f53fdfffd7098dd572406ba549eed955bf821f
container_name: influxdb
restart: always
By default the database used will be called balena
, however you can set a custom database name by setting the INFLUXDB_DB
environment variable.
This data sink will send the data to a URL specified with the environment variable HTTP_PUSH_URL
.
This data sink pushes data into the Azure Monitor service, using an Azure Application Insights account. In order to use this sink, login (or create) to your Microsoft Azure account, create an Application Insights resource and copy the instrumentation key. Guide here: https://docs.microsoft.com/en-us/azure/azure-monitor/app/create-new-resource
Place this key into an environment variable on your balena device called APPLICATION_INSIGHTS_KEY
.
You can now view the data by pointing Azure Monitor to your Application Insights account and charting the correct metrics:
You can change the pull interval for internal and external HTTP endpoints by setting the PULL_INTERVAL
env variable.
The default value is 10s
.
For a list of valid inputs, refer to the telegraf documentation.
By default the connector
block creates a telegraf configuration file from the combination of discovered services and device environment variables. However for custom configurations you can overload the CMD
directive, as such:
dockerfile.template
FROM bh.cr/balenalabs/connector-%%BALENA_ARCH%%
COPY customTelegraf.conf .
CMD ["--config customTelegraf.conf"]
This will stop the auto-wiring code from running and cause telegraf to be run purely from the supplied configuration file.
You can turn on telegraf debugging by setting the environment variable DEBUG
to 1
. This turns on debug logging to the console.