Mesos is an open source cluster management framework that provides efficient sharing of computing resources across large clusters of servers. It was first released in 2009 at the University of California, Berkeley and uses a distributed systems kernel to combine CPU, storage, memory and other resources into one big shared pool. Mesos works by having a master node that manages slave nodes, with frameworks running applications that use executors to run tasks on the slaves.

1. marc@sloppy.io
Introduction to Apache Mesos

2. marc@sloppy.io
Overview
● Introduction
● Architecture
● Security
● Container
● High availability

3. marc@sloppy.io
Introduction
● First release in 2009 at the Berkley University
● Framework to use datacenter resources
efficiently
● Combine Cpu, storage, memory etc. to one big
shared virtual resource
● A distributed systems kernel
● 10.000 lines C++ code

4. marc@sloppy.io
Introduction - Definitions
● Master - Scheduler
● Slaves – Working Nodes
● Frameworks – Application running on Mesos
● Executors – Run tasks on the slaves
● Executor-Task - Running job on the slave
● Resource Offer - Slave resources which could
be used by the frameworks

5. marc@sloppy.io
Architecture

6. marc@sloppy.io
Introduction - Frameworks
https://docs.mesosphere.com/frameworks/

7. marc@sloppy.io
Resource Allocation

8. marc@sloppy.io
Resource Allocation
1) Slave 1 reports to the master that it has 4
CPUs and 4 GB of memory free. The master
then invokes the allocation policy module, which
tells it that framework 1 should be offered all
available resources.
2) The master sends a resource offer describing
what is available on slave 1 to framework 1.

9. marc@sloppy.io
Resource Allocation
3) The framework’s scheduler replies to the
master with information about two tasks to run
on the slave, using <2 CPUs, 1 GB RAM> for
the first task, and <1 CPUs, 2 GB RAM> for the
second task.
2) Finally, the master sends the tasks to the
slave, which allocates appropriate resources
to the framework’s executor, which in turn
launches the two tasks

10. marc@sloppy.io
Resource Allocation - DRF
● Resource offer decision are made by the
Resource Allocation Modul in the master
● In a heterogeneous environment resource
allocation is difficult
● What is a fair share, when:
User a require 1 CPU, 4GB RAM
User b require 3 CPUs, 1 GB RAM
● Mesos: Dominant Resource Fairness

11. marc@sloppy.io
DRF
● A modified fair share algorithm
● The goal is that each framework receives a fair
share of the the resources most needed by the
framework
● Dominant resource: Resource most demand by
the framework
● Dominant Share: The highest percentage of
shares owned across all resources of a
framework

12. marc@sloppy.io
DRF - Example
● Resource offer: 9 Cpu, 18GB RAM
● Tasks User A: 1CPU, 4 GB RAM - RAM=DR
● Tasks User B: 3CPUs, 1GB RAM – CPU=DR
● Each Framework has 2/3 of its DS

13. marc@sloppy.io
DRF - Example
● Framework1: 1CPU, 4GB RAM
● Framework2: 3CPU, 1GB RAM
● Buggy tasks could be killed by mesos
● Framework can have guaranteed allocation, non
of its tasks should be killed

14. marc@sloppy.io
RA Master Configuration
Name Default Example
allocation_interval 1s
framework_sorter drf
user_sorter drf
offer_timeout 5 minutes
roles - marathon,jenkins
weights - marathon=2,jenkins=1

15. marc@sloppy.io
RA Slave Configuration
Name Default Example
attributes ssd:true,rack:2
default_role *
resources cpus(jenkins):1;disk(jenkins):10000;
cpus(marathon):3;mem(marathon):2000

16. marc@sloppy.io
Mesos Security
● Default configuration = No security
Name Example
Master authenticate_slaves true
credentials /etc/mesos.pw
authenticators crammd5
authenticate true
Slave credential /etc/mesos.pw

17. marc@sloppy.io
Framework Security
1) Framework to (re-)register with authorized roles
2)Framework to launch task/executors as authorized
users
3)Authorized principals to shutdown frameworks
through “/shutdown” HTTP endpoint

18. marc@sloppy.io
Security ACLs
Subjects Action Object
principals register_framework roles
usernames run_tasks users
shutdown_frameworks framework_principals
● A set of subjects can perform an action
on a set of objects

19. marc@sloppy.io
Security ACLs Example

20. marc@sloppy.io
Extract of the mesos api
URL Function
master:5050/help REST Documentation
master:5050/metrics/snapshot Metrics of the cluster
master:5050/master/tasks.json List mesos tasks
master:5050/master/redirect 307 to the leading master
master:5050/master/shutdown Shutdown Framework
master:5050/registrar(1)/registry Content of the current registry
slave:5051/files/browse.json?path=pathOnSlave Browse files in sandbox
slave:5051/files/read.json?path=stdoutOnSlave Read stdout from sandbox
slave:5051/system/stats.json Local system metrics

21. marc@sloppy.io
Resource Isolation
● Mesos supports Docker - and Mesos Container
● Resource isolation with cgroups or posix

22. marc@sloppy.io
Mesos and Docker

23. marc@sloppy.io
Mesos HA

24. marc@sloppy.io
Mesos Tasks States
TaskState Int Description
TASK_STARTING 0
TASK_RUNNING 1 Task
TASK_FINISHED 2 TERMINAL: The task finished successfully
TASK_FAILED 3 TERMINAL: The task failed to finished
TASK_KILLED 4 TERMINAL: The task was killed by executor
TASK_LOST 5 TERMINAL: The task was failed but can
rescheduled
TASK_STAGING 6 Initial State
TASK_ERROR 7 TERMINAL: Task description contains an error

25. marc@sloppy.io
References
● http://mesos.apache.org/documentation/latest/
● Mesos: A Platform for Fine-Grained Resource Sharing in the Data Center
● Dominant Resource Fairness: Fair Allocation of Multiple Resource Types
● playing-traffic-cop-resource-allocation-in-apache-mesos
● https://mesosphere.com/

26. marc@sloppy.io
Thank you