with non-overlapping data. The number of shards in a keyspace can vary depending on the use-case and load characteristics, some Vitess users have hundreds of shards in some keyspaces. Shard Naming Shard throttling is not needed, but in my experience it’s still wise to use, since replication lag can imply load on primary, and it’s best to not overload the primary. • We issue a DELETE FROM my_table LIMIT 50 does not matter. It’s important to note that the DELETE statement actually causes table pages to load into the buffer pool, which works against our objective. Once all rows are purged from a table, we

Growing with Vitess Vitess app server app server app server Master Replica Replica routing Load balancing Master promotion with Orchestrator Growing with Vitess Vitess app server app server app

throttling is not needed, but in my experience it’s still wise to use, since replication lag can imply load on primary, and it’s best to not overload the primary. • We issue a DELETE FROM my_table LIMIT 50 does not matter. 47 It’s important to note that the DELETE statement actually causes table pages to load into the buffer pool, which works against our objective. Once all rows are purged from a table, we timestamp is older than, say, 2 days. As mentioned above, purging the table actually caused the table to load onto the buffer pool, and we need to wait for it to naturally get evicted, before dropping it. Suggested

with non-overlapping data. The number of shards in a keyspace can vary depending on the use-case and load characteristics, some Vitess users have hundreds of shards in some keyspaces. Shard Naming Shard throttling is not needed, but in my experience it’s still wise to use, since replication lag can imply load on primary, and it’s best to not overload the primary. • We issue a DELETE FROM my_table LIMIT 50 does not matter. It’s important to note that the DELETE statement actually causes table pages to load into the buffer pool, which works against our objective. Once all rows are purged from a table, we

with non-overlapping data. The number of shards in a keyspace can vary depending on the use-case and load characteristics, some Vitess users have hundreds of shards in some keyspaces. Shard Naming Shard throttling is not needed, but in my experience it’s still wise to use, since replication lag can imply load on primary, and it’s best to not overload the primary. • We issue a DELETE FROM my_table LIMIT 50 does not matter. It’s important to note that the DELETE statement actually causes table pages to load into the buffer pool, which works against our objective. Once all rows are purged from a table, we

ks as the default keyspace with all queries being sent to the master. Consequently ks@replica will load balance requests across all REPLICA tablet types, and ks@rdonly will choose RDONLY. You can also To revert this, type unalias mysql && unalias vtctlclient or close your session. Create Schema Load our initial schema: vtctlclient ApplySchema -sql="$(cat create_commerce_schema.sql)" commerce vtctlclient If any of the above fields are missing, vitess will fail to load the table. No operation will be allowed on a table that has failed to load. 59 Enqueuing messages The application can enqueue messages

ks as the default keyspace with all queries being sent to the master. Consequently ks@replica will load balance requests across all REPLICA tablet types, and ks@rdonly will choose RDONLY. You can also If any of the above fields are missing, vitess will fail to load the table. No operation will be allowed on a table that has failed to load. Enqueuing messages The application can enqueue messages using specific shards or groups of shards while requesting messages. This is useful for partitioning or load balancing. The MessageStream API allows you to specify these constraints. The request parameters are as

ks as the default keyspace with all queries being sent to the master. Consequently ks@replica will load balance requests across all REPLICA tablet types, and ks@rdonly will choose RDONLY. You can also If any of the above fields are missing, vitess will fail to load the table. No operation will be allowed on a table that has failed to load. 49 Enqueuing messages The application can enqueue messages specific shards or groups of shards while requesting messages. This is useful for partitioning or load balancing. The MessageStream API allows you to specify these constraints. The request parameters are as

ks as the default keyspace with all queries being sent to the master. Consequently ks@replica will load balance requests across all REPLICA tablet types, and ks@rdonly will choose RDONLY. You can also To revert this, type unalias mysql && unalias vtctlclient or close your session. Create Schema Load our initial schema: vtctlclient ApplySchema -sql="$(cat create_commerce_schema.sql)" commerce vtctlclient If any of the above fields are missing, vitess will fail to load the table. No operation will be allowed on a table that has failed to load. 47 Enqueuing messages The application can enqueue messages