Sharding MySQL with Vitess
- 1. Sharding MySQL with Vitess Harun Küçük
- 2. What is Sharding? Sharding is a type of database partitioning that separates very large databases into smaller,faster and more easily managed parts called data shards.
- 3. • Non-Scalable Master Why we need Sharding? Sample Traditional MySQL Replication • Scalable App Layer • Scalable Replicas
- 4. Vitess • Started 2010 , youtube https://vitess.io/ • Open source since 2011 https://github.com/vitessio/vitess • Incubating project in CNCF https://www.cncf.io/projects/
- 5. Vitess Architecture • Lightweight proxy server • Routes traffic to correct vttablet • Returns consolidated results back to the clients
- 6. Vitess Architecture • Proxy server that sits in front of MySQL instance • Protect MySQL from harmful queries • Connection Pooling • Query rewriting • Hot row protection
- 7. Vitess Architecture • Stores metadata (running servers,sharding schema,Replication Graph) • Etcd, Apache Zookeeper or consul could be used for topology
- 8. Vitess Architecture • Vtctl is command line tool, Vtctld is an HTTP server that lets you browse the information stored in the topology.
- 9. Vitess Architecture • Replica tablets: candidates for master tablet , Readonly tables: for batch jobs, resharding,bigdata,backups etc.
- 10. Vitess Key Adaptors • Started 2010 at Youtube and It has been serving all Youtube database traffic since 2011. Youtube had 256 shards and each shards had between 80 and 120 replicas across 20 datacenters all around the world. (Approx. 256K instance) • JD.com is the 2nd largest retailer company in China. JD.com has more than 10.000 instance (master,replicas) in Vitess on kubernetes cluster • Square Cash App fully runs on Vitess. Square has more than 64 shards. • Slack migrated 40% database traffic to Vitess and their goal is 100% • Pinterest’s all of advertising campaign management fully runs on Vitess
- 11. Example: Sakila DVD Rental Company Database Lets suppose we have a DVD rental company and our database diagram live below
- 12. Day 1: Table Rows Query:
- 16. Entity Group for Sharding Payment,Rental and Customer tables have customer_id column. So these three tables should be sharded horizontally by customer_id
- 17. Step 1: Vertical Sharding
- 18. Step 2: Horizontal Sharding
- 20. Prerequisites for Demo • Vitess (https://github.com/vitessio/vitess) • Kubernetes Cluster or Minicube • Etcd operator for topology cluster • Helm for vitess helm charts (vitess/helm/vitess) • NFS client provisioner for persistent NFS volumes (https://github.com/helm/charts/tree/master/stable/nfs-client-provisioner)
- 21. 1.) initiate new cluster, Concepts File: initial_cluster.yaml • Cell : Zone, availability zone, datacenter • KeySpace : Logical Database • Vschema : Vitess Schema, contains metadata about how tables are organized across keyspaces and shards • Vindex : index to find shards
- 22. 1.) initiate new cluster File: initial_cluster.yaml
- 23. 1.) initiate new cluster
- 24. 2.) Create New Keyspace File:create_keyspace.yaml
- 25. 2.) Create New Keyspace File:create_keyspace.yaml
- 26. 3.) Split Keyspace Schema File:split_keyspace_schema.yaml
- 27. 3.) Split Keyspace Schema File:split_keyspace_schema.yaml
- 28. 3.) Split KeySpace Schema
- 29. 4.) Vertical Split Clone File:vertical_split_clone.yaml
- 30. 4.) Vertical Split Clone File:vertical_split_clone.yaml
- 31. 4.) Vertical Split Clone
- 33. 4.) Migrate ReadOnly and Replica
- 35. 5.) Drop Blacklisted Tables
- 39. 6.) Primary Vindex for Horizontal Sharding File: create_vindex.yaml
- 40. Sharding by Hash Function
- 41. Sharding by Hash Function 1001 1000 1002 1003 1004 1005 1006 1007
- 42. Vitess Hash Algorithm (3des hash) 4000 = 4000000000000000 8000 = 8000000000000000
- 43. Vitess Hash Based Sharding (3des hash) 264 combination. theoretically allows us to have an infinite number of shards
- 44. Vitess Hash Based Sharding (3des hash)
- 45. 7.) Create Horizontal Shards File: create_horizontal_shards.yaml
- 46. 7.) Create Horizontal Shards
- 47. 7.) Horizontal SplitClone File: horizontal_split_clone.yaml
- 48. 7.) Horizontal SplitClone File: horizontal_split_clone.yaml
- 49. 7.) Horizontal SplitClone File: horizontal_split_clone.yaml
- 50. 8.) Migrate RdOnly,Replica and Master File: 13_migrate_readonly_replica.yaml 14_migrate_master.yaml
- 51. 8.) Migrate RdOnly,Replica and Master File: 13_migrate_readonly_replica.yaml 14_migrate_master.yaml
- 52. Querying Vitess Cluster - Counts
- 53. Querying Vitess Cluster - Routing
- 54. Querying Vitess Cluster - Routing
- 55. Querying Vitess Cluster – @replica,@rdonly
- 56. Querying Vitess Cluster – Cross Shard Join
- 57. Querying Vitess Cluster – Cross Shard Join
- 58. Querying Vitess Cluster – Cross Shard Join = + N * N = row count of 2. query’s result 1 2 3
- 60. Querying Vitess Cluster –Cross Shard Join,VReplication
- 61. Querying Vitess Cluster –Scatter Query
- 62. Querying Vitess Cluster –Scatter Query LookupVindex
- 63. Drawbacks: Application Transactions • Best Effort Commit -> Consistency Problems • 2 Phase Commit -> Performance Cost
- 64. Drawbacks: Distributed Deadlocks Shard Level Deadlock
- 65. Drawbacks: Distributed Deadlocks Database Level Deadlock
- 68. Thank You