Create embServe Workload

Services to be deployed in Far-Edge devices powered by embServe can be generated using C code. For that, an SDK is provided by embServe, which services must use to ensure compatibility. This page details a simple example showing how this process works.

embServe SDK Setup

The SDK includes everything needed to generate embServe-compliant files. To setup the SDK:

1. Download the SDK bundle from here

2. Unzip file

unzip embservesdk-*.zip -d embserve-sdk

3. Setup Python venv

cd embserve-sdk
python3 -m venv .venv
. .venv/bin/activate

4. Setup the toolchain

./setup_toolchain.sh

5. Export the toolchain path as instructed by the setup script. As an example:

export ZEPHYR_SDK_PATH=/home/user/Workspace/embserve-sdk/toolchain/zephyr/zephyr-sdk-0.16.1

6. Add the folder embserve-sdk/toolchain/udynlink-ng to the PATH. As an example:

export PATH=/home/user/Workspace/embserve-sdk/toolchain/udynlink-ng:$PATH

note

You need to run the commands above every time you open a new shell. To prevent this, you can add them to the end of your ~/.bashrc or ~/.profile

7. Validate the mkmodule command is available:

mkmodule --help

Creating an embServe Service

1. Create a folder inside the embserve-sdk/ folder

cd embserve-sdk
mkdir service

2. Create the file service.c in the embserve-sdk/ folder with the following:

#include "../include/embserve.h"
#include <string.h>
#include <stdio.h>


static bool running = false;
static int delay_ms;

int setup(void) {
    printf("%s started!", SERVICE_NAME);
    running = true;
    delay_ms = 30000;
    return 0;
}

void loop(void) {
    printf("Loop");

    embserve_sensor_data_t sensor_data = {0};
    int sensor_id = 0;

    while(running) {
        int res = embserve_get_sensor(SENSOR_TEMPERATURE, &sensor_id);
        if(res < 0) {
            printf("Failed to get sensor!");
            return;
        }

        res = embserve_get_sensor_data(sensor_id, &sensor_data);
        if(res < 0) {
            printf("Failed to get sensor data!");
            return;
        }

        embserve_io_t output = {
            .key = "value",
            .type = EMBSERVE_FLOAT,
            .data_len = 1,
            .data = sensor_data.values
        };
        embserve_output(&output);
        sleep(delay_ms);
    }
}

void on_input(embserve_io_t * input) {
    if(strcmp(input->key, "rate") == 0) {
        int new_delay_ms = *(int *)input->data;
        if(new_delay_ms < 1000) {
            printf("on_input(): Bad input. Rate too low (%d)", new_delay_ms);
            return;
        }

        delay_ms = new_delay_ms;
        printf("on_input(): New rate %d ms", delay_ms);
    }
}

void on_quit(void) {
    running = false;
}

info

This service collects Temperature data and sends it as an output ("value"). The sample rate is configurable with the input "rate".

3. Build the service:

mkmodule service.c ../src/*.c --name temperature_sensor

4. Get the base64 of the generated binary

base64 temperature_sensor.bin -w 0

5. Create the JSON manifest, service.json. Replace <service binary in base64> with the output of the previous command :

{
    "name": "temperature_sensor",
    "version": 1,
    "service": <service binary in base64>,
    "dependencies": {
        "sdk_api": 1
    },
    "inputs": [
        {
            "key": "rate",
            "type": 0
        }
    ],
    "outputs": [
        {
            "key": "value",
            "type": 1
        }
    ]
}

warning

The inputs and outputs section must match your service. Any output or input not matching the manifest will be discarded by embServe!

6. Service manifest is now ready to be deployed

For more information about embServe and its services, refer to this article.

Refer to Deploy Workloads on Far-edge Devices to learn how to deploy this service using FITA.