New Features in MongoDB Atlas
0 likes868 views
The document details the implementation journey of MongoDB Atlas across three cloud providers: AWS, Azure, and GCP, starting from its MVP in June 2016 to its current status in 2018. It covers hardware specifications, account structures, software tools, and the challenges faced in adapting features for Azure and GCP. Key comparisons of disk and instance sizes across the cloud platforms are also highlighted.
New Features in MongoDB Atlas
- 1. MongoDB Atlas - A Tale of Three Cloud Providers
- 2. Narrator Implementation Deep Dive 4+ years at MongoDB 2+ years: Lead for Ops Manager Last 18+ months: Lead for Atlas
- 3. Flashback December 2015 Implementation Deep Dive Cloud Team was 4 years old Define an Atlas MVP Worked through Winter/Spring Atlas released June 2016 at MongoDB World
- 4. Flashback December 2016 Implementation Deep Dive AWS is great, but we want to be on GCP/Azure too Worked through Winter/Spring Here we are at MongoDB World 2017 releasing on GCP/Azure.
- 5. Present Day Implementation Deep Dive Chapter 1: Hardware Chapter 2: Account Layout Chapter 3: Software Epilogue: MongoDB 2018
- 7. 7 A Customer on AWS AWS Account X - Region Y VPC for Customer N Availability Zone A Availability Zone B Availability Zone C Subnet A Subnet B Subnet C mongod - 27107 mongod - 27107 mongod - 27107
- 8. 8 A Customer on Azure Azure Subscription X Resource Group for Customer Y VNET Availability Set mongod - 27107mongod - 27107mongod - 27107
- 9. 9 A Customer on Google Cloud Platform GCP Organization Project for Customer N Zone A Zone B Zone C Network Subnet mongod - 27107 mongod - 27107 mongod - 27107
- 10. 10 Disk Comparison Cloud Storage Type Capacity Performance AWS GP2 10GB - 16TB 3 IOPS/GB AWS IO1 10GB - 16TB 15/30 IOPS/GB Azure P-Series 128GB 512GB 1024GB 500 IOPS 2300 IOPS 5000 IOPS GCP SSD Persistent Storage 1GB-64TB 30 IOPS/GB
- 11. 11 Instance Size Comparison Cloud AWS Azure GCP AWS T2-Series (M10/M20) ⅕ Core ⅖ Core 2GB 4GB AWS M4-Series (M30+) 2 Core 4 Core ... 8GB 16GB ... Azure DS_v2 1 Core 2 Core ... 3.5GB 7GB ... GCP G1 Series 1/2 1.7GB GCP N1 Series 1 Core 2 Core ... 3.75GB 7.5GB ...
- 12. Chapter 2: Account Structure
- 13. Goals Implementation Deep Dive Consolidated Authentication Consolidated Billing Scales
- 14. 14 AWS Account Layout One root account 1000 customers per Account Service account per Account Consolidated billing to root account Delegated auth from root account AWS Root Account Dev Account QA Account Prod Account Dev VPC N
- 15. 15 Azure Account Layout Azure AD holds users / applications Customer assigned to Subscription 400 customers per Subscription Active Directory Dev Sub. QA Sub. Prod Sub Dev Resource Group N
- 16. 16 GCP Account Layout One GCP Organization A folder per environment Customer gets a project in a folder Service account per folder Organization Dev Folder QA Folder Prod Folder Dev Project N
- 17. Chapter 3: Software What was easy and what was not
- 18. 18 Cloud Agents - Easy! ● Monitoring Agent, Backup Agent, Automation Agent ● Our tools, written in Go. ● Talk to / Manage MongoDB and report back to Cloud infrastructure
- 19. 19 Machine Images - Easy! ● All CentOS 7 ● Minor updates to Chef cookbooks ● Scripts to build Azure/GCP Images
- 20. 20 Split Horizon DNS - More Difficult AWS provided this for free Need to build for Azure/GCP Used multiple Zones (Route53) Used dnsmasq on the VMs mongodb.net (Authoritative Zone) mongodb.net (Private Zone) Customer Atlas Server dnsmasq Local DNS Server Local DNS Server Rest of Internet
- 21. 21 Azure Availability Sets - More Difficult AWS has Availability Zones GCP has Zones Azure has Availability Sets (FD, UD) ● Availability Set per shard ● Can not assign FD, UD ● Managed disk copies Availability Set - Shard 1 FD0, UP0 FD1, UD1 FD2, UD2 Availability Set - Shard 2 FD0, UP0 FD2, UD2 FD1, UD1
- 22. 22 GCP is Under Construction - More Difficult ● Lots of feature we found critical were beta, or even alpha. ● Had to use beta endpoints for some changes ● Evaluate on a case-by-case basis if something was really needed
- 23. Epilogue
- 24. 24 MongoDB 2018 Preview ● Optimized disk and instance size upgrades on Azure/GCP ● Cross-region Clusters ● Cross-cloud Cluster
- 25. Questions