Fault-tolerance at scale
made easy with Mesosphere and Google Cloud Platform

Sunil Shah
sunil@mesosphere.io

We are Mesosphere

Status quo is static partitioning
and use of virtual machines

Add some virtual machines

Provision Hadoop

Provision a web service

Moar data, moar Hadoop

Mesos let us treat a cluster of nodes...

As one big computer

Mesos is...

The Mesos ecosystem is
growing

High Level View of Mesos

Framework = scheduler + executor

Schedulers get resource offers

Executors run tasks

We love containers

Most modern applications are a web of containers

Now what?

Say hi to Marathon

a self-serve interface to your cluster

distributed "init" for long-running services

a private fault-tolerant PaaS

a container orchestration platform

Marathon does it!

• Start, stop, scale, update apps
• Nice web interface, API
• Highly available, no SPoF
• Native Docker support (as of Mesos 0.20.0)
• Pluggable event bus
• Rolling deploy / restart
• Application health checks
• Artifact staging

Service Discovery

Set environment variables

Read config from device (rsync'ed to fs)

Read from K-V store

Use DNS

HAProxy works pretty well

Marathon REST

POST /v2/apps

GET /v2/apps

PUT /v2/apps/{appId}

GET /v2/apps/{appId}/tasks

DELETE /v2/apps/{appId}/tasks/{taskId}

...

Run a Docker container

http -v POST http://10.53.20.188:8080/v2/apps @app-ruby.json

{
  "container": {
    "type": "DOCKER",
    "docker": {
      "image": "superguenter/demo-app"
    }
  },
  "cmd": "rails server -p $PORT",
  "id": "rails-demo",
  "instances": 1,
  "cpus": 0.01,
  "mem": 256,
  "ports": [3000]
}

          

Scale!

http -v PUT http://10.53.20.188:8080/v2/apps/demo @scale-app.json

{
  "instances": 3
}

          

Deploy a new Python version

http -v PUT http://10.53.20.188:8080/v2/apps/demo @app-python.json

{
  "container": {
    "type": "DOCKER",
    "docker": {
      "image": "superguenter/demo-app"
    }
  },
  "cmd":  "python -m SimpleHTTPServer $PORT",
  "id": "demo",
  "instances": 3,
  "cpus": 0.01,
  "mem": 256,
  "ports": [3000]
}
          

Use HTTP healthchecks

http -v POST http://10.53.20.188:8080/v2/apps @app-with-healthcheck.json

{
  "container": {
    "type": "DOCKER",
    "docker": {
      "image": "superguenter/demo-app"
    }
  },
  "cmd": "rails server -p $PORT",
  "id": "healthcheck-demo",
  "instances": 1,
  "cpus": 0.01,
  "mem": 256,
  "ports": [3000],
  "healthChecks": [
    {
      "path": "/",
      "portIndex": 0,
      "protocol": "HTTP",
      "gracePeriodSeconds": 30,
      "intervalSeconds": 30,
      "timeoutSeconds": 30,
      "maxConsecutiveFailures": 0
    }
  ],
  "constraints": [
      ["hostname", "UNIQUE"]
  ]
}
          

Add a UNIQUE hostname constraint

http -v POST http://10.53.20.188:8080/v2/apps @app-with-constraint.json

{
  "container": {
    "type": "DOCKER",
    "docker": {
      "image": "superguenter/demo-app"
    }
  },
  "cmd": "python -m SimpleHTTPServer $PORT",
  "id": "constraint-demo",
  "instances": 6,
  "cpus": 0.01,
  "mem": 256,
  "ports": [3000],
  "healthChecks": [
    {
      "path": "/",
      "portIndex": 0,
      "protocol": "HTTP"
    }
  ],
  "constraints": [
      ["hostname", "UNIQUE"]
  ]
}
          

Mesos for all the containers

Mesos for all the things

Thanks!

Come and talk to us