back to listing index

chrislusf/glow

[web search]
Original source (github.com)
Tags: golang go distributed-computing spark map-reduce
Clipped on: 2016-03-07

Skip to content
Glow is an easy-to-use distributed computation system written in Go, similar to Hadoop Map Reduce, Spark, Flink, Storm, etc.
Go Protocol Buffer

README.md

glow

Image (Asset 3/5) alt= Image (Asset 4/5) alt=

Purpose

Glow is providing a library to easily compute in parallel threads or distributed to clusters of machines.

Installation

normalgo get github.com/chrislusf/glow
go get github.com/chrislusf/glow/flow
normal

One minute tutorial

Simple Start

Here is a simple full example:

normalpackage main

import (
    "flag"
    "strings"

    "github.com/chrislusf/glow/flow"
)

func main() {
    flag.Parse()

    flow.New().TextFile(
        "/etc/passwd", 3,
    ).Filter(func(line string) bool {
        return !strings.HasPrefix(line, "#")
    }).Map(func(line string, ch chan string) {
        for _, token := range strings.Split(line, ":") {
            ch <- token
        }
    }).Map(func(key string) int {
        return 1
    }).Reduce(func(x int, y int) int {
        return x + y
    }).Map(func(x int) {
        println("count:", x)
    }).Run()
}

normal

Try it.

normal  > ./word_count
normal

It will run the input text file, '/etc/passwd', in 3 go routines, filter/map/map, and then reduced to one number in one goroutine (not exactly one goroutine, but let's skip the details for now.) and print it out.

This is useful already, saving lots of idiomatic but repetitive code on channels, sync wait, etc, to fully utilize more CPU cores.

However, there is one more thing! It can run across a Glow cluster, which can be run multiple servers/racks/data centers!

Scale it out

To setup the Glow cluster, we do not need experts on Zookeeper/HDFS/Mesos/YARN etc. Just build or download one binary file.

Setup the cluster

normal  // fetch and install via go, or just download it from somewhere
  > go get github.com/chrislusf/glow
  // start a master on one computer
  > glow master
  // run one or more agents on computers
  > glow agent --dir . --max.executors=16 --memory=2048 --master="localhost:8930" --port 8931
normal

Glow Master and Glow Agent run very efficiently. They take about 6.5MB and 5.5MB memory respectively in my environments. I would recommend set up agents on any server you can find. You can tap into the computing power whenever you need to.

Start the driver program

To leap from one computer to clusters of computers, add this line to the import list:

normal    _ "github.com/chrislusf/glow/driver"
normal

This will "steroidize" the code to run in cluster mode!

normal> ./word_count -glow -glow.leader="localhost:8930"
normal

The word_count program will become a driver program, dividing the execution into a directed acyclic graph(DAG), and send tasks to agents.

Visualize the flow

To understand how each executor works, you can visualize the flow by generating a dot file of the flow, and render it to png file via "dot" command provided from graphviz.

normal> ./word_count -glow -glow.flow.plot > x.dot
> dot -Tpng -otestSelfJoin.png x.dot
normal

Image (Asset 5/5) alt=

Read More

  1. Wiki page: https://github.com/chrislusf/glow/wiki
  2. Mailing list: https://groups.google.com/forum/#!forum/glow-user-discussion
  3. Examples: https://github.com/chrislusf/glow/tree/master/examples/

Docker container

Docker is not required. But if you like docker, here are instructions.

normal// Cross compile artefact for docker
GOOS=linux GOARCH=amd64 CGO_ENABLED=0 go build .
// build container
docker build -t glow .
normal

See  examples/  directory for docker-compose setups.

Contribution

Start using it! And report or fix any issue you have seen, add any feature you want.

Fork it, code it, and send pull requests. Better first discuss about the feature you want on the mailing list. https://groups.google.com/forum/#!forum/glow-user-discussion

License

http://www.apache.org/licenses/LICENSE-2.0