Global Cluster Topologies in MongoDB Atlas - Andrew Davidson
0 likes212 views
Andrew Davidson discusses the need for global clusters in MongoDB Atlas to meet the expectations of a rapidly growing internet user base. He highlights the importance of low-latency experiences, global operational data, and the impact of latency on sales and customer engagement. The document outlines various strategies for deploying scalable and efficient global databases while ensuring data accessibility across regions.
![Data includes location attribute
globalColl:
{
_id: 123,
firstName: "Johannes" ,
lastName: "Doe”
activities" : [
{ "title": "Skiing",
"attributes": ["downhill","jumps"]
},
{ "title": "Coding",
"attributes": ["mongodb","python"]
}
],
}
globalColl:
{
_id: 123,
location : "DE",
firstName: "Johannes" ,
lastName: "Doe”
activities" : [
{ "title": "Skiing",
"attributes": ["downhill","jumps"]
},
{ "title": "Coding",
"attributes": ["mongodb","python"]
}
],
}](https://image.slidesharecdn.com/globalclustertopologiesinmongodbatlas-andrewdavidson-181113145644/85/Global-Cluster-Topologies-in-MongoDB-Atlas-Andrew-Davidson-38-320.jpg)
![For example {location : 1, _id : 1}
globalColl:
{
_id: 123,
location : "CA",
firstName: "Jane" ,
lastName: "Doe”
activities" : [
{ "title": "Mountaineering",
"attributes": ["glaciers"]
},
{ "title": "Coding",
"attributes": ["java","mongodb"]
}
],
}](https://image.slidesharecdn.com/globalclustertopologiesinmongodbatlas-andrewdavidson-181113145644/85/Global-Cluster-Topologies-in-MongoDB-Atlas-Andrew-Davidson-42-320.jpg)
Global Cluster Topologies in MongoDB Atlas - Andrew Davidson
- 1. Global Clusters for a Global World Andrew Davidson Lead Product Manager, MongoDB Atlas
- 3. 3.15 Billion new internet users since first MongoDB code commit illion
- 5. Every year a US-sized population is coming online for the first time
- 6. You don’t have to be a household-name company to meaningfully reach them
- 8. reach yes, a CDN can reach
- 9. meaningfully reach them but users expect more
- 10. Global Operational Data What users want • Low-latency reads, writes, updates
- 11. Global Operational Data What users want • Low-latency reads, writes, updates • Interactive experiences
- 12. Global Operational Data What users want • Instant gratification
- 13. Global Operational Data What users want • Instant gratification • Relentless improvement
- 14. Global Operational Data What users want • Instant gratification • Relentless improvement • Personalization
- 15. Global Operational Data What users want • Instant gratification • Relentless improvement • Personalization • Yadda...yadda
- 16. Global Operational Data What users want • Instant gratification • Relentless improvement What you want • Happy users (see above)
- 17. Global Operational Data What users want • Instant gratification • Relentless improvement What you want • Happy users • Stay ahead of competition • low-latency experiences • constant agile iteration
- 18. Global Operational Data What users want • Instant gratification • Relentless improvement What you want • Happy users • Stay ahead of competition • Reporting • Single connection string • Aggregations
- 19. Global Operational Data = meaningful reach What users want • Instant gratification • Relentless improvement What you want • Happy users • Stay ahead of competition • Global awareness
- 20. 100ms
- 21. 100ms is associated with 1% drop in sales
- 22. 100ms is associated with 1% drop in sales … from some of the most beloved e-commerce sites on the internet
- 23. 100ms is associated with You missing the chance to engage a customer meaningfully
- 24. 100ms is associated with You missing the chance to engage a customer meaningfully … and missing the chance to find your market fit
- 25. Global Write latencies from US East US East – US East: 2ms US East – Frankfurt: 89ms US East – Singapore: 225ms
- 26. Options
- 29. Global Multi-Region Replication Advantages • Single connection string globally • Low latency reads anywhere • Cross-region High Availability • Great for reference data Disadvantages • All writes go to preferred region (a large part of the world experiences slow writes) • Low latency reads require secondary read concern (You could even have your preferred region follow the sun to achieve part-time global writes)
- 31. Cluster Per Region Advantages • Schema doesn’t need to reflect data affinity • Easy today in MongoDB Atlas • Independently scalable Disadvantages • No global aggregation • Application tier manages end user to connection string affinity
- 32. Yes. You, too, can have instant gratification. Can we get the best of both worlds?
- 33. Global Clusters with Global Writes
- 35. Scale into regions over time
- 37. How Global Writes work
- 38. Data includes location attribute globalColl: { _id: 123, firstName: "Johannes" , lastName: "Doe” activities" : [ { "title": "Skiing", "attributes": ["downhill","jumps"] }, { "title": "Coding", "attributes": ["mongodb","python"] } ], } globalColl: { _id: 123, location : "DE", firstName: "Johannes" , lastName: "Doe” activities" : [ { "title": "Skiing", "attributes": ["downhill","jumps"] }, { "title": "Coding", "attributes": ["mongodb","python"] } ], }
- 39. What is a location attribute? Generally... an ISO 3166-1 country code • like US, DE, IN, or JP Or optionally and more specifically an ISO-3166-2 subdivision code • like US-CA or US-NY Examples Description Potential Region Affinity US United States US East DE Germany Frankfurt IN India Mumbai JP Japan Tokyo US-CA California US West US-NY New York US East
- 41. MongoDB Atlas will be very opinionated... In the Data Explorer, you’ll be guided to set a shard key like • {location : 1, <identifier> : 1} You’ll receive global-targeting query examples • db.globalColl.find_one( { location : "XY", <identifier> : 123 }) • etc.
- 42. For example {location : 1, _id : 1} globalColl: { _id: 123, location : "CA", firstName: "Jane" , lastName: "Doe” activities" : [ { "title": "Mountaineering", "attributes": ["glaciers"] }, { "title": "Coding", "attributes": ["java","mongodb"] } ], }
- 43. MongoDB’s sharding handles the rest
- 44. US East Application Driver mongos mongos Repl0 Repl1 Repl2 Repl0 Repl1 Repl2 Repl0 Repl1 Repl2 ••• US East Zone Shard Europe Zone Shard Singapore Zone Shard Frankfurt Application Driver Singapore Application Driver mongos ••• mongos ••• ••• Americas Users EMEA Users APAC Users Global Load Balancer MongoDB Atlas Global Cluster •••
- 46. Concise and global, it will look like mongodb+srv://user:pass@globalCluster-rzeyq.mongodb.net
- 47. Concise and global, it will look like • Your driver will use this address to discover every mongos in the cluster globally • Taking advantage of the MongoDB 3.6 driver support for SRV records … so you can deploy the same application code everywhere mongodb+srv://user:pass@globalCluster-rzeyq.mongodb.net
- 48. Nice to have
- 49. What if I want low latency global reads?
- 50. ...configure local reads in all zones
- 51. And you get low latency global reads with the secondary read concern
- 52. Limits and where this is all going
- 53. Massive reach MongoDB Atlas • in-country regions for 27% of the world population • with a much higher percent served by regions in nearby countries Available in 56 global cloud regions • AWS: 15 regions (38 availability zones) • Azure: 26 regions (26 availability zones w/ 60 Fault Domains) • GCP: 15 regions (38 Availability zones) And the big three cloud providers MongoDB Atlas delivers on are expanding fast
- 54. There are gaps Least served regions from a Cloud perspective Africa • S. Africa announced by Azure; a long way from most of Africa’s population, however • Biggest economies in Africa: Nigeria, South Africa, Egypt; with Ethiopia now the fastest growing Middle East • We’ll see China • It’s complicated... Russia • It’s also complicated… • Frankfurt remains best bet but also Finland on Google Cloud Special / Government / Compliance contexts
- 55. But MongoDB runs everywhere Over time we should expect more and more of the world’s population to be well served by major cloud infrastructure But you can always run your own MongoDB outside of where the big public cloud platforms and MongoDB Atlas are operational • MongoDB Ops Manager / MongoDB Cloud Manager • Live Migrate to MongoDB Atlas when ready So you can have a seamless development experience in any context
- 56. Imagine 2030
- 57. 2030 Projection GDP & relative position compared with today 1 China ↑ 17 Spain ↓ 2 United States ↓ 18 Canada ↓ 3 India ↑↑↑↑ 19 Egypt ↑↑↑ 4 Japan ↓ 20 Pakistan ↑↑↑ 5 Indonesia ↑↑↑ 21 Nigeria ↑↑ 6 Russia ↑ 22 Thailand ↑↑ 7 Germany ↓ 23 Australia ↓ 8 Brazil ↑ 24 Philippines ↑↑ 9 Mexico ↑ 25 Malaysia ↑↑ 10 United Kingdom ↓ 26 Poland ↓ 11 France ↓ 27 Argentina ↓ 12 Turkey ↑ 28 Bangladesh ↑↑↑ 13 Saudi Arabia ↑↑ 29 Vietnam ↑↑↑ 14 South Korea ↓ 30 South Africa ↑↑ 15 Italy ↓ 31 Colombia ↑↑ 16 Iran ↑↑ 32 Netherlands ↓
- 58. You can go global incrementally
- 59. Path to Global Over Time 1. Cluster – Single Region ← deploy today for free on Atlas a. Focus on finding traction b. Scale up when needed in a few minutes on Atlas 2. Cluster – Multi Region HA a. Get your application HA as well b. Scale up when needed in a few minutes on Atlas 3. Cluster – add Global Read Regions a. Route users to nearby application instance; use secondary reads b. Scale up when needed in a few minutes on Atlas 4. Enable Global Writes a. add a second write zone b. add location attribute to your data & shard your global collections in the Atlas UI 5. Expand Reach of Global Cluster a. add more write zones b. or add shards to existing zones c. scale up any time
- 60. And you can follow your users...
- 61. Now go out and reach those 4+ billion internet users
- 62. (or reach a small subset of them with my workflow processing automation framework app. Remember, it’s still cool if it’s global)
- 63. Thanks!
- 64. Andrew Davidson Lead Product Manager, MongoDB Atlas andrew.davidson@mongodb.com