lf-edge / ekuiper

Lightweight data stream processing engine for IoT edge

LF Edge eKuiper - An edge lightweight IoT data analytics software

English | 简体中文

Overview

LF Edge eKuiper is a lightweight IoT data analytics and stream processing engine running on resource-constraint edge devices. The major goal for eKuiper is to provide a streaming software framework (similar to Apache Flink) in edge side. eKuiper's rule engine allows user to provide either SQL based or graph based (similar to Node-RED) rules to create IoT edge analytics applications within few minutes.

User scenarios

It can be run at various IoT edge user scenarios, such as,

• Real-time processing of production line data in the IIoT
• Gateway of connected vehicle analyze the data from CAN in IoV
• Real-time analysis of wind turbines and smart bulk energy storage data in smart energy

eKuiper processing at the edge can greatly reduce system response latency, save network bandwidth and storage costs and improve system security.

Features

• Lightweight

  • Core server package is only about 4.5M, memory footprint is about 10MB

• Cross-platform

  • CPU Arch：X86 AMD * 32/64; ARM * 32/64; PPC
  • Popular Linux distributions, OpenWrt Linux, MacOS and Docker
  • Industrial PC, Raspberry Pi, industrial gateway, home gateway, MEC edge cloud server

• Data analysis support

  • Support data ETL
  • Data order, group, aggregation and join with different data sources (the data from databases and files)
  • 60+ functions, includes mathematical, string, aggregate and hash etc
  • 4 time windows & count window

• Highly extensible

  It supports to extend at Source, Functions and Sink with Golang or Python.

  • Source: allows users to add more data source for analytics.
  • Sink: allows users to send analysis result to different customized systems.
  • UDF functions: allow users to add customized functions for data analysis (for example, AI/ML function invocation)

• Management

  • A free web based management dashboard for visualized management
  • Plugins, streams and rules management through CLI, REST API and config maps(Kubernetes)
  • Easily be integrated with Kubernetes frameworks KubeEdge, OpenYurt, K3s Baetyl

• Integration with EMQX products

  Seamless integration with EMQX, Neuron & NanoMQ, and provided an end-to-end solution from IIoT, IoV

Quick start

• 5 minutes quick start
• Getting started
• EdgeX rule engine tutorial

Community

Join our Slack, and then join ekuiper or ekuiper-user channel.

Meeting

Subscribe to community events calendar.

Weekly community meeting at Friday 10:30AM GMT+8:

• Zoom meeting link
• Meeting minutes

Contributing

Thank you for your contribution! Please refer to the CONTRIBUTING.md for more information.

Performance test result

MQTT throughput test

• Using JMeter MQTT plugin to send IoT data to EMQX Broker, such as: {"temperature": 10, "humidity" : 90}, the value of temperature and humidity are random integer between 0 - 100.
• eKuiper subscribe from EMQX Broker, and analyze data with SQL: SELECT * FROM demo WHERE temperature > 50
• The analysis result are wrote to local file by using file sink plugin.

Devices	Message # per second	CPU usage	Memory usage
Raspberry Pi 3B+	12k	sys+user: 70%	20M
AWS t2.micro( 1 Core * 1 GB) Ubuntu18.04	10k	sys+user: 25%	20M

EdgeX throughput test

• A Go application is written to send data to ZeroMQ message bus, the data is as following.

  {
    "Device": "demo", "Created": 000, …
    "readings": 
    [
       {"Name": "Temperature", value: "30", "Created":123 …},
       {"Name": "Humidity", value: "20", "Created":456 …}
    ]
  }
  

• eKuiper subscribe from EdgeX ZeroMQ message bus, and analyze data with SQL: SELECT * FROM demo WHERE temperature > 50. 90% of data will be filtered by the rule.

• The analysis result are sent to nop sink, so all the result data will be ignored.

	Message # per second	CPU usage	Memory usage
AWS t2.micro( 1 Core * 1 GB) Ubuntu18.04	11.4 k	sys+user: 75%	32M

Max number of rules support

• 8000 rules with 800 message/second in total
• Configurations
  • 2 core * 4GB memory in AWS
  • Ubuntu
• Resource usage
  • Memory: 89% ~ 72%
  • CPU: 25%
  • 400KB - 500KB / rule
• Rule
  • Source: MQTT
  • SQL: SELECT temperature FROM source WHERE temperature > 20 (90% data are filtered)
  • Sink: Log

Multiple rules with shared source instance

• 300 rules with a shared MQTT stream instance.
  • 500 messages/second in the MQTT source
  • 150,000 message processing per second in total
• Configurations:
  • 2 Core * 2GB memory in AWS
  • Ubuntu
• Resource usage
  • Memory: 95MB
  • CPU: 50%
• Rule
  • Source: MQTT
  • SQL: SELECT temperature FROM source WHERE temperature > 20, (90% data are filtered)
  • Sink: 90% nop and 10% MQTT

To run the benchmark by yourself, please check the instruction.

Documents

Check out the latest document in official website.

Build from source

Preparation

• Go version >= 1.22

Compile

• Binary:

  • Binary: $ make

  • Binary files that support EdgeX: $ make build_with_edgex

  • Minimal binary file with core runtime only: $ make build_core

• Packages: $ make pkg

  • Packages: $ make pkg

  • Package files that support EdgeX: $ make pkg_with_edgex

• Docker images: $ make docker

  Docker images support EdgeX by default

Prebuilt binaries are provided in the release assets. If using os or arch which does not have prebuilt binaries, please use cross-compilation, refer to this doc.

During compilation, features can be selected through go build tags so that users can build a customized product with only the desired feature set to reduce binary size. This is critical when the target deployment environment has resource constraint. Please refer to features for more detail.

Open source license

Apache 2.0