EMC World 2016 - code.04 Extending Mesos for Storage and External Resources
Download as PPTX, PDF0 likes972 views
The document discusses the integration of storage as a global resource within the Mesos cluster management framework, highlighting the challenges of managing distributed systems and scheduler optimization. It outlines the need for tracking storage resources and managing failures, proposing strategies for integrating global resources either through the Mesos offer cycle or out-of-band handling. The presentation also emphasizes the flexibility required in large-scale installations with multiple storage providers.
EMC World 2016 - code.04 Extending Mesos for Storage and External Resources
- 1. © Copyright 2016 EMC Corporation. All rights reserved.© Copyright 2016 EMC Corporation. All rights reserved. 1 EXTENDING MESOS FOR STORAGE AS A GLOBAL RESOURCE STEVE WONG DEVELOPER ADVOCATE – EMC{code} GREG MANN SOFTWARE ENGINEER – Mesosphere
- 2. © Copyright 2016 EMC Corporation. All rights reserved. 2 GOT CONTAINERS AT SCALE? YOU NEED A SCHEDULER…
- 3. © Copyright 2016 EMC Corporation. All rights reserved. 3 ASSUME 1000’S OF SERVICES • How do you place an assortment of containers onto your servers efficiently?
- 4. © Copyright 2016 EMC Corporation. All rights reserved. 4 WHAT’S A SCHEDULER? • Manages compute nodes – Takes reports from each node regarding their health and available capacity • Manages incoming work requests – Accepts job requests – Based on requirements, decides which node should fulfill each request – May attempt to optimize job placement to maximize locality, utilization, etc. on demand scheduled batch
- 5. © Copyright 2016 EMC Corporation. All rights reserved. 5 When you build a cluster, there’s no getting around it: you are building a distributed system • Difficult to design, build and operate • Introduces complexity Mesos takes on some of the "heavy lifting" of deploying and operating a distributed system MESOS AS A CLUSTER MANAGER
- 6. © Copyright 2016 EMC Corporation. All rights reserved. 6 A distributed systems kernel • Abstracts many independent nodes into a single pool • Negotiates among multiple schedulers – Competing schedulers can suffer from resource starvation – Mesos offers resources to frameworks to ensure fairness WHAT IS MESOS?
- 7. © Copyright 2016 EMC Corporation. All rights reserved. 7 WHAT IS MESOS? Marathon
- 8. © Copyright 2016 EMC Corporation. All rights reserved. 8 • Where it started • Recent enhancements • Roadmap MESOS AND STORAGE
- 9. © Copyright 2016 EMC Corporation. All rights reserved. 9 CLUSTER NODES OFFER LOCAL RESOURCES
- 10. © Copyright 2016 EMC Corporation. All rights reserved. 10 MESOS MASTER PRESENTS OFFERS TO FRAMEWORKS Marathon
- 11. © Copyright 2016 EMC Corporation. All rights reserved. 11 • Where it started • Recent enhancements • Roadmap MESOS AND STORAGE
- 12. © Copyright 2016 EMC Corporation. All rights reserved. 12 RESERVE CLUSTER NODE HOSTED VOLUME
- 13. © Copyright 2016 EMC Corporation. All rights reserved. 13 External volume mounts were added in 2015 • Provides Availability + Scale MOUNT A VOLUME FROM EXTERNAL STORAGE
- 14. © Copyright 2016 EMC Corporation. All rights reserved. 14 • Where it started • Recent enhancements • Roadmap MESOS AND STORAGE
- 15. © Copyright 2016 EMC Corporation. All rights reserved. 15 NOT ALL RESOURCES ARE LOCAL
- 16. © Copyright 2016 EMC Corporation. All rights reserved. 16 AGENT FAILURE • Agents are the source of truth for resource information
- 17. © Copyright 2016 EMC Corporation. All rights reserved. 17 AGENT FAILURE • When an agent fails this information is lost, but so are the resources!
- 18. © Copyright 2016 EMC Corporation. All rights reserved. 18 AGENT FAILURE • When a failed agent was using a global resource, the resource is still available!
- 19. © Copyright 2016 EMC Corporation. All rights reserved. 19 AGENT FAILURE • The state of this resource must be stored off the agent
- 20. © Copyright 2016 EMC Corporation. All rights reserved. 20 NEED A NEW ABSTRACTION • We need a third entity to track the use of global resources
- 21. © Copyright 2016 EMC Corporation. All rights reserved. 21 • Option #1: Integrate global resources into the Mesos offer cycle HOW TO INTEGRATE GLOBAL RESOURCES?
- 22. © Copyright 2016 EMC Corporation. All rights reserved. 22 • Option #2: Handle global resources “out-of-band” HOW TO INTEGRATE GLOBAL RESOURCES?
- 23. © Copyright 2016 EMC Corporation. All rights reserved. 23 Mesos was originally designed to manage compute node resources • Storage in this context meant DAS A JIRA proposal exists to include global resources in the Mesos offer model, but this may not be necessary. • Feedback is welcome! HOW TO INTEGRATE GLOBAL RESOURCES?
- 24. © Copyright 2016 EMC Corporation. All rights reserved. 24 HOW DO YOU TRACK WHAT STORAGE YOU HAVE? … AND WHERE IT’S IN USE?
- 25. © Copyright 2016 EMC Corporation. All rights reserved. Create Utilize Retire MESOS INTERFACES WITH STORAGE Lifecycle management with governance Mount managementAssignment to applications
- 26. © Copyright 2016 EMC Corporation. All rights reserved. 26 A LARGE SCALE INSTALLATION NEEDS TO ALLOW FOR MULTIPLE STORAGE PROVIDERS What does this remind you of?
- 27. © Copyright 2016 EMC Corporation. All rights reserved. 27 MESOS = 2 LEVEL CONTAINER SCHEDULER
- 28. © Copyright 2016 EMC Corporation. All rights reserved. 28 TWO LAYER EXTERNAL STORAGE SCHEDULING
- 29. © Copyright 2016 EMC Corporation. All rights reserved. 29 STORAGE SCHEDULER = DELIVER SCALE OUT Scale out is not just about adding capacity. A Storage Scheduler will let you: • Add, remove, replace storage provider platforms – in or out of public cloud • Add, remove, replace schedulers • This is about giving you flexibility and shifting the burden of keeping up to vendors
- 30. © Copyright 2016 EMC Corporation. All rights reserved. 30 POLLY AS A MESOS FRAMEWORK
- 31. © Copyright 2016 EMC Corporation. All rights reserved. 31 DCOS: THE DATACENTER OPERATING SYSTEM learn more at mesosphere.com
- 32. © Copyright 2016 EMC Corporation. All rights reserved. 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
- 33. © Copyright 2016 EMC Corporation. All rights reserved. @EMCcode @cantbewong mesosphere.com emccode.com community.emccode.com Come visit us at Booth #1044 or in the vLab Questions?
- 34. © Copyright 2016 EMC Corporation. All rights reserved.© Copyright 2016 EMC Corporation. All rights reserved. 34
Editor's Notes
- #4: Playing Tetris with manual load placement might seem fun for a minute or two, but this gets old quick Container placement ABOLUTELY must be automated,
- #30: Schedulers can also enable end use self service, With controls, audit, and health monitoring
- #32: Schedulers can also enable end use self service, With controls, audit, and health monitoring