Mesos Deep Dive on Persistent Storage for Apps

1© Copyright 2016 EMC Corporation. All rights reserved. 1© Copyright 2016 EMC Corporation. All rights reserved.
DEEP DIVE WITH MESOS & PERSISTENT
STORAGE FOR APPLICATIONS
DAVID VONTHENEN, EMC {code}

2© Copyright 2016 EMC Corporation. All rights reserved.
• David vonThenen, Developer Advocate - EMC {code}
– Twitter: @dvonthenen
– Blog: http://dvonthenen.com
– Github: https://github.com/dvonthenen
PRESENTER

• Quick Review of Mesos
• Frameworks and Two Layer Scheduling
• Production Applications
• Demo
• Q&A
AGENDA

QUICK REVIEW OF MESOS

• From a resource perspective, it’s a cluster manager:
– Pools resources to be centrally managed as a single unit
• From an application perspective, it’s a scheduler:
– Dispatches workloads to consume pooled resources
• Often described as a Data Center Operating System
(DCOS)
– Static versus Dynamic Partition of Resource
– Efficient utilization of CPU, Memory, Disk
• Proven resource scheduler for production workloads
QUICK REVIEW OF MESOS

SCHEDULERS ARE DIFFICULT TO BUILD

• Master
– Manages Mesos Agents
– Enables fine-grained sharing of resources (CPU, RAM, …)
across frameworks in the form of resource offers.
• Zookeeper
– Enables highly reliable distributed coordination (cluster
state)
• Agent (aka Slave)
– The entity that will be running the task
MESOS COMPONENTS

MESOS ARCHITECTURE

• Generic
– Anything that can be run on a standard Linux shell
– Provides a mechanism for deploying applications packages
– Supports Docker workloads
• Framework
– A plug-in to Mesos that allows you to implement a sub-
scheduler
– specialize a workload by providing custom placement and
resource allocation
APPLICATION SUPPORT

GENERIC APPLICATIONS

FRAMEWORKS AND
TWO LAYER SCHEDULING

• On the surface…
• Allows specialization of
workload by providing
custom placement and
resource allocation
• Provides monitoring of
these workloads to protect
and react when things fail
MESOS FRAMEWORK

• Scheduler
– a scheduler that registers with the master to be offered
resources
• Executor
– an executor process that is launched on slave nodes to run
the framework’s tasks
FRAMEWORK COMPONENTS

FRAMEWORK ARCHITECTURE

• Odds are you have been using at least 1 framework
without realizing it…
AS IT TURNS OUT…
• Marathon is a framework
– A framework for performing container
orchestration for long running tasks
• Chronos is a framework
– A framework for performing batch or
scheduled tasks

• Zooming into the “Marathon” box…
• Marathon is a Scheduler
• Launching a Marathon task, the “generic”
applications are created inside a “generic” (or
default) executor
• Dogfooding!
MARATHON DISSECTED

MESOS FRAMEWORKS

• 2 layer scheduling or sub-
scheduling of resources
• Manages the lifecycle of your
workload through application
specific monitoring
• What does this mean
specifically?
DIGGING DEEPER

BUILDING HADOOP CLUSTERS

MANAGING HADOOP CLUSTERS

APPLICATIONS USING HADOOP

THIS SEEMS SILLY

• Allows you to subsume the default scheduler such
that:
• You can install and configure your application
• Resources can be allocated in a way that optimizes
your applications
• Allows operation real-time automated improvements
for your applications
– Scaling up for load balancing peak times
– Scaling back to free up resources
2 LAYER SCHEDULING

• The details of your applications are hidden from the
end user!
– Deployment
– Operations
– Growth
– Disaster recovery
– Performance
• Just a consumer of the application as a service!
THE NON-TECHNICAL VERSION

• CouchBase Framework (doesn’t exist, let’s imagine it
does)
– CouchBase is a NoSQL DB
• Monitor the CouchBase instances, when nearing full
– Dynamically add an node using “server-add”
– Then call “rebalance”
• Reactive and Proactive but fully automated
• I think we had all seen this movie…
FOR EXAMPLE

SKYNET

PRODUCTION APPLICATIONS

• Containers are ephemeral therefore stateless
• Quickly found the need for state
– Configuration
– Data
– Logging / Auditing
• Containerizers like Mesos and Docker introduced
exporting to local storage
• Failure of a node? hardware? A better way?
EVOLUTION OF STATEFUL

• Persistent
– Containers can come and go,
but continue where they left off
– Application Data, Logs, Audit
Trail, etc
• External
– Storage lives outside the Mesos
Agent node
– High Availability
– Tolerates node failure
PERSISTENT EXTERNAL STORAGE

• REX-Ray
– Provides a vendor agnostic storage orchestration engine
– AWS, GCE, ScaleIO, XtremIO, many many more
– Github: https://github.com/emccode/rexray
• mesos-module-dvdi
– Provides a hook to Mesos agent nodes to provision, attach,
and detach external storage
– Github: https://github.com/emccode/mesos-module-dvdi
HOW WE ACHIEVE THIS?

• We can run “real world” applications
• Persistence – DBs
• Highly Available, Fault Tolerant
• Audit trail – Transaction Logs
• Not worry about the “2am call”
STATELESS TO STATEFUL

LET’S FACE IT…

• Trouble is… this still represents the “old way” of
thinking
• PREVENT THE CRASH OR FAILURE!
• Problem is you can’t
• What is the alternative?
STATEFUL 2.0?

• Mesos frameworks can proactively monitor for thing
going sideways
• Leveraging external persistent storage and
frameworks fail fast and recover quicker!
• Skynet!
FRAMEWORKS + STATEFUL 2.0

DEMO

• Running on Amazon EC2
– Mesos
– Marathon
– Mesos DNS
• Production HA Cluster
– 3 x Master + Zookeeper Nodes (t2.micros)
– 5 x Slave Nodes with Docker Support (t2.medium)
DEMO CONFIGURATION

• Elastic Search Framework
– Create and mount external volume
– Pour some data into Elastic Search
– Github: https://github.com/mesos/elasticsearch
• Elastic Search Instances backed by Persistent
External Storage
– Using REX-Ray
– Using mesos-module-dvdi
DEMO: FRAMEWORK + EXTERNAL STORAGE

• Deploy Elastic Search Framework using REST API
– Observe Scheduler creation
– Using browser, log into the Elastic Search Framework UI
– Observe Executor nodes get created and attaching to EBS
volumes
• Simulate Failure
– Kill EC2 Instance
– Watch executor recreation and reattach to EBS volumes
DEMO: FRAMEWORK + EXTERNAL STORAGE

DEMO: AUTO DEPLOY ELASTIC SEARCH

Data Persistence in the New Container World
Wednesday 3PM
Joshua Bernstein
VP of Technology for ETD
Tobi Knaup
CEO & Co-Founder of Mesosphere
Guru Session

@EMCcode
@dvonthenen
emccode.com
community.emccode.com
Come visit us at
Booth #1044
or in the vLab
Questions?

Mesos Deep Dive on Persistent Storage for Apps

Mesos Deep Dive on Persistent Storage for Apps

Empfohlen

Empfohlen

Weitere ähnliche Inhalte

Was ist angesagt?

Was ist angesagt? (20)

Andere mochten auch

Andere mochten auch (15)

Ähnlich wie Mesos Deep Dive on Persistent Storage for Apps

Ähnlich wie Mesos Deep Dive on Persistent Storage for Apps (20)

Kürzlich hochgeladen

Kürzlich hochgeladen (20)

Mesos Deep Dive on Persistent Storage for Apps