CREATE TABLE IF NOT EXISTS puppers (\n  `id` BIGINT(22),\n  `name` VARCHAR(256),\n  `image` VARCHAR(256),\n  PRIMARY KEY(id)\n);\n

","language","sql",[],["SingleFetchClassInstance",234],{"_33":34,"_35":36,"_37":235,"_39":236},{},[237],"Here is an example of a simple VSchema for a sharded database for the table above that illustrates the three elements of a VSchema:",["SingleFetchClassInstance",239],{"_33":34,"_35":226,"_37":240,"_39":243},{"_228":241,"_230":242},"

{\n  \"sharded\": true,\n  \"vindexes\": {\n    \"hash_vdx\": {\n      \"type\": \"hash\"\n    }\n  },\n  \"tables\": {\n    \"puppers\": {\n      \"column_Vindexes\": [\n        {\n          \"column\": \"id\",\n          \"name\": \"hash_vdx\"\n        }\n      ]\n    }\n  }\n}\n

","json",[],["SingleFetchClassInstance",245],{"_33":34,"_35":96,"_37":246,"_39":248},{"_98":247},"what-is-a-vindex",[249],["SingleFetchClassInstance",250],{"_33":34,"_35":49,"_37":251,"_39":253},{"_51":252},"#what-is-a-vindex",[254],"What is a Vindex?",["SingleFetchClassInstance",256],{"_33":34,"_35":36,"_37":257,"_39":258},{},[259],"A relational database has a schema that consists of tables, columns and indexes amongst other elements. In the same way an index makes it efficient to access a given row in a given table in an unsharded database, a Vindex for a given table in a sharded database allows you to access a row quickly by allowing you to determine which shard a row lives in. Like indexes, Vindexes can be primary or secondary and a given table can only have a single primary Vindex, but can have multiple secondary Vindexes.",["SingleFetchClassInstance",261],{"_33":34,"_35":96,"_37":262,"_39":264},{"_98":263},"how-does-a-vindex-map-a-row-to-a-shard",[265],["SingleFetchClassInstance",266],{"_33":34,"_35":49,"_37":267,"_39":269},{"_51":268},"#how-does-a-vindex-map-a-row-to-a-shard",[270],"How does a Vindex map a row to a shard?",["SingleFetchClassInstance",272],{"_33":34,"_35":36,"_37":273,"_39":274},{},[275],"To understand this, you will have to understand the concept of keyspaces and keyspace_ids. In the Vitess world every row in every sharded database has a keyspace_id. The keyspace_id is not stored, but is computed by applying a specific sharding function to the value of a specific column in that row. The keyspace_ids range from 0x00 - 0xFF and this range represents the entire keyspace. Vitess shards cover this entire range. This is why a sharded database in Vitess is called a keyspace. Each shard spans a range in the keyspace and shards are named by starting and ending keyspace_id values for the shard. Let us take the example of a 4-shard keyspace. If we divide the range 0x00 to 0xFF (hexadecimal values) in four equal ranges, we get:",["SingleFetchClassInstance",277],{"_33":34,"_35":226,"_37":278,"_39":280},{"_228":279,"_230":-7},"

0x00-0x40\n0x40-0x80\n0x80-0xC0\n0xC0-0xFF\n

",[],["SingleFetchClassInstance",282],{"_33":34,"_35":36,"_37":283,"_39":284},{},[285],"Thus the four shards would be named: “00-40”, “40-80”, “80-C0”, “C0-FF”. Vitess drops the “00” at the beginning of the keyspace range and the FF at the end, so the shards are called “-40”, “40-80”, “80-C0”, “C0-” instead.",["SingleFetchClassInstance",287],{"_33":34,"_35":36,"_37":288,"_39":289},{},[290],"Vindexes allow you to map the value of a particular column in a given row to one or more keyspace_ids. Each shard has a starting and ending keyspace_id, thus given a keyspace_id, you can deterministically tell which shard a row belongs to.",["SingleFetchClassInstance",292],{"_33":34,"_35":36,"_37":293,"_39":294},{},[295,296,302],"Just like an index, a Vindex is applied to a column in a table, but a Vindex has an additional property called sharding function. Vitess gives you 15 ",["SingleFetchClassInstance",297],{"_33":34,"_35":49,"_37":298,"_39":300},{"_51":299},"https://vitess.io/docs/reference/features/vindexes/",[301],"predefined sharding functions"," and certain sharding functions go well with columns of certain types, for example you would typically use “hash” sharding function with a numeric column, or you would use “unicode_loose_md5” for a varchar or varbinary column. You can also write custom sharding functions and use those instead.",["SingleFetchClassInstance",304],{"_33":34,"_35":96,"_37":305,"_39":307},{"_98":306},"what-is-a-vitess-sequence",[308],["SingleFetchClassInstance",309],{"_33":34,"_35":49,"_37":310,"_39":312},{"_51":311},"#what-is-a-vitess-sequence",[313],"What is a Vitess sequence?",["SingleFetchClassInstance",315],{"_33":34,"_35":36,"_37":316,"_39":317},{},[318],"In an unsharded database when you need to assign a monotonically increasing value to a row, you can define a column to be of type “autoincrement”. In the sharded world, if you do this, you would end up having duplicate values in the same column across shards. Vitess solves this problem by allowing you to create ids which are monotonically increasing and unique across shards for a given table by defining sequences in the VSchema. The implementation of Vitess Sequences is backed by a row in a table a secondary database. Vitess limits the number of writes needed by allowing you to cache a certain number of values.",["SingleFetchClassInstance",320],{"_33":34,"_35":36,"_37":321,"_39":322},{},[323],"Let us make the id column in the puppers table above of type Vitess sequence by adding the following segment to table entry for puppers in our VSchema:",["SingleFetchClassInstance",325],{"_33":34,"_35":226,"_37":326,"_39":328},{"_228":327,"_230":242},"

\"auto_increment\": {\n\t\"column\": \"id\",\n\t\"sequence\": \"pupper_seq\"\n}\n

",[],["SingleFetchClassInstance",330],{"_33":34,"_35":36,"_37":331,"_39":332},{},[333],"Which gives the following VSchema:",["SingleFetchClassInstance",335],{"_33":34,"_35":226,"_37":336,"_39":338},{"_228":337,"_230":242},"

{\n  \"sharded\": true,\n  \"vindexes\": {\n    \"hash\": {\n      \"type\": \"hash\"\n    }\n  },\n  \"tables\": {\n    \"puppers\": {\n      \"column_Vindexes\": [\n        {\n          \"column\": \"id\",\n          \"name\": \"hash\"\n        }\n      ],\n      \"auto_increment\": {\n        \"column\": \"id\",\n        \"sequence\": \"pupper_seq\"\n      }\n    }\n  }\n}\n

",[],["SingleFetchClassInstance",340],{"_33":34,"_35":96,"_37":341,"_39":343},{"_98":342},"how-to-define-sequences",[344],["SingleFetchClassInstance",345],{"_33":34,"_35":49,"_37":346,"_39":348},{"_51":347},"#how-to-define-sequences",[349],"How to define sequences?",["SingleFetchClassInstance",351],{"_33":34,"_35":36,"_37":352,"_39":353},{},[354],"You will notice that we are using the term pupper_seq without defining it anywhere. Here is how that process works. You typically define sequences in an unsharded keyspace that lives alongside your main keyspace. In the Vitess world this keyspace is typically called a lookup keyspace. Here is how you would define the sequence in the lookup keyspace. You first create a table named puppers_seq and you insert one row in it initializing the sequence. Note the comment vitess_sequence associated with the CREATE TABLE statement. It’s important for you to keep that because that is used by Vitess to treat this table distinctly from other tables which hold real data.",["SingleFetchClassInstance",356],{"_33":34,"_35":226,"_37":357,"_39":359},{"_228":358,"_230":231},"

CREATE TABLE IF NOT EXISTS pupper_seq (\n  `id` INT,\n  `next_id` BIGINT,\n  `cache` BIGINT,\n  PRIMARY KEY(id)\n) comment 'vitess_sequence';\n\nINSERT INTO pupper_seq (id, next_id, cache) VALUES (0, 1, 3);\n

",[],["SingleFetchClassInstance",361],{"_33":34,"_35":36,"_37":362,"_39":363},{},[364],"After defining this table, you apply the following VSchema to the lookup keyspace:",["SingleFetchClassInstance",366],{"_33":34,"_35":226,"_37":367,"_39":369},{"_228":368,"_230":242},"

{\n  \"sharded\": false,\n  \"tables\": {\n    \"pupper_seq\": {\n      \"type\": \"sequence\"\n    }\n  }\n}\n

",[],["SingleFetchClassInstance",371],{"_33":34,"_35":36,"_37":372,"_39":373},{},[374],"These two steps define pupper_seq as a Vitess sequence and it can now be used by the VSchema for the sharded keyspace as we have used it above.",["SingleFetchClassInstance",376],{"_33":34,"_35":96,"_37":377,"_39":379},{"_98":378},"putting-all-this-together",[380],["SingleFetchClassInstance",381],{"_33":34,"_35":49,"_37":382,"_39":384},{"_51":383},"#putting-all-this-together",[385],"Putting all this together",["SingleFetchClassInstance",387],{"_33":34,"_35":36,"_37":388,"_39":389},{},[390],"So, for the Goodest Doggo dog rating application, here is how we will organize our data. We will create two keyspaces, one called lookup, unsharded and one called puppers which is sharded. We will start with two shards and we can reshard as needed as we go. The lookup keyspace will hold tables we need for sequences and for lookup Vindexes. The doggers keyspace will hold the actual data. Here is the schema and VSchema for these two keyspaces:",["SingleFetchClassInstance",392],{"_33":34,"_35":36,"_37":393,"_39":394},{},[395],"The keyspace doggers has three tables: puppers, ratings, and users. We want the id column to be of type autoincrement, but as you can see below, because this is a sharded database so we do not specify the column as autoincrement in the schema, but instead we define a Vitess sequence on that column backed by the puppers_seq table in the lookup schema. This is the same for the id column for the ratings table.",["SingleFetchClassInstance",397],{"_33":34,"_35":36,"_37":398,"_39":399},{},[400],"We want the table puppers sharded by its id column, but the table ratings sharded by the user_id column. This is expressed in the VSchema.",["SingleFetchClassInstance",402],{"_33":34,"_35":36,"_37":403,"_39":404},{},[405,406,411],"Here is the Schema and VSchema for the ",["SingleFetchClassInstance",407],{"_33":34,"_35":43,"_37":408,"_39":409},{},[410],"puppers"," database:",["SingleFetchClassInstance",413],{"_33":34,"_35":226,"_37":414,"_39":416},{"_228":415,"_230":231},"

CREATE TABLE `puppers` (\n  `id` bigint(22) NOT NULL,\n  `name` varchar(256) DEFAULT NULL,\n  `image` varchar(256) DEFAULT NULL,\n  PRIMARY KEY (`id`)\n) ENGINE = InnoDB DEFAULT CHARSET = utf8;\n\nCREATE TABLE `ratings` (\n  `id` bigint(22) DEFAULT NULL,\n  `user_id` bigint(22) DEFAULT NULL,\n  `rating` bigint(20) DEFAULT NULL,\n  `pupper_id` bigint(22) DEFAULT NULL,\n  KEY `pupper_id` (`pupper_id`)\n) ENGINE = InnoDB DEFAULT CHARSET = utf8;\n\nCREATE TABLE `users` (\n  `id` bigint(22) NOT NULL,\n  `email` varchar(64) DEFAULT NULL,\n  `password` varbinary(256) DEFAULT NULL,\n  PRIMARY KEY (`id`)\n) ENGINE = InnoDB DEFAULT CHARSET = utf8;\n

",[],["SingleFetchClassInstance",418],{"_33":34,"_35":226,"_37":419,"_39":421},{"_228":420,"_230":242},"

{\n  \"sharded\": true,\n  \"vindexes\": {\n    \"binary_md5_vdx\": { \"type\": \"binary_md5\" },\n    \"hash_vdx\": { \"type\": \"hash\" }\n  },\n  \"tables\": {\n    \"puppers\": {\n      \"columnVindexes\": [{ \"column\": \"id\", \"name\": \"hash_vdx\" }],\n      \"autoIncrement\": { \"column\": \"id\", \"sequence\": \"pupper_seq\" }\n    },\n    \"ratings\": {\n      \"columnVindexes\": [{ \"column\": \"pupper_id\", \"name\": \"hash_vdx\" }],\n      \"autoIncrement\": { \"column\": \"id\", \"sequence\": \"rating_seq\" }\n    },\n    \"users\": {\n      \"columnVindexes\": [{ \"column\": \"id\", \"name\": \"binary_md5_vdx\" }]\n    }\n  }\n}\n

",[],["SingleFetchClassInstance",423],{"_33":34,"_35":36,"_37":424,"_39":425},{},[405,426,411],["SingleFetchClassInstance",427],{"_33":34,"_35":43,"_37":428,"_39":429},{},[430],"lookup",["SingleFetchClassInstance",432],{"_33":34,"_35":226,"_37":433,"_39":435},{"_228":434,"_230":231},"

CREATE TABLE IF NOT EXISTS pupper_seq (\n  id INT,\n  next_id BIGINT,\n  cache BIGINT,\n  PRIMARY KEY(id)\n) comment 'vitess_sequence';\n\nINSERT INTO pupper_seq (id, next_id, cache) VALUES (0, 1, 3);\n\nCREATE TABLE IF NOT EXISTS rating_seq (\n  id INT,\n  next_id BIGINT,\n  cache BIGINT,\n  PRIMARY KEY(id)\n) comment 'vitess_sequence';\n\nINSERT INTO rating_seq (id, next_id, cache) VALUES (0, 1, 3);\n

",[],["SingleFetchClassInstance",437],{"_33":34,"_35":226,"_37":438,"_39":440},{"_228":439,"_230":242},"

{\n  \"sharded\": false,\n  \"tables\": {\n    \"pupper_seq\": {\n      \"type\": \"sequence\"\n    },\n    \"rating_seq\": {\n      \"type\": \"sequence\"\n    }\n  }\n}\n

",[],["SingleFetchClassInstance",442],{"_33":34,"_35":96,"_37":443,"_39":445},{"_98":444},"want-to-see-this-in-action",[446],["SingleFetchClassInstance",447],{"_33":34,"_35":49,"_37":448,"_39":450},{"_51":449},"#want-to-see-this-in-action",[451],"Want to see this in action?",["SingleFetchClassInstance",453],{"_33":34,"_35":36,"_37":454,"_39":455},{},[456],"We have created a Quickstart Demo on PlanetScaleDB that creates the databases we described above and populates them with data from our sample dog rating application.",["SingleFetchClassInstance",458],{"_33":34,"_35":36,"_37":459,"_39":460},{},[461],"To get started, you will need to:",["SingleFetchClassInstance",463],{"_33":34,"_35":178,"_37":464,"_39":465},{},[466,477,487,492,504,516,521,526,531],["SingleFetchClassInstance",467],{"_33":34,"_35":183,"_37":468,"_39":469},{},[470,471,69],"Create a ",["SingleFetchClassInstance",472],{"_33":34,"_35":49,"_37":473,"_39":475},{"_51":474},"https://auth.planetscale.com/sign-up",[476],"PlanetScaleDB account",["SingleFetchClassInstance",478],{"_33":34,"_35":183,"_37":479,"_39":480},{},[481,482,69],"Select the Quickstart Demo ",["SingleFetchClassInstance",483],{"_33":34,"_35":43,"_37":484,"_39":485},{},[486],"VSchema",["SingleFetchClassInstance",488],{"_33":34,"_35":183,"_37":489,"_39":490},{},[491],"This will spin up a cluster and two databases and when successful will present you with a database URL.",["SingleFetchClassInstance",493],{"_33":34,"_35":183,"_37":494,"_39":495},{},[496,497,503],"In the meantime, ",["SingleFetchClassInstance",498],{"_33":34,"_35":49,"_37":499,"_39":501},{"_51":500},"https://github.com/planetscale/quickstarts/tree/master/goodest-doggo",[502],"download and start"," the application for your platform.",["SingleFetchClassInstance",505],{"_33":34,"_35":183,"_37":506,"_39":507},{},[508,509,515],"Once the application is running, it will provide you with a URL (such as ",["SingleFetchClassInstance",510],{"_33":34,"_35":49,"_37":511,"_39":513},{"_51":512},"http://localhost:8000/",[514],"http://localhost:8000",") on your local host that you can browse to.",["SingleFetchClassInstance",517],{"_33":34,"_35":183,"_37":518,"_39":519},{},[520],"This web application will prompt you for the connection string for the cluster you created with the Quickstart Demo on PlanetScaleDB.",["SingleFetchClassInstance",522],{"_33":34,"_35":183,"_37":523,"_39":524},{},[525],"Input the connection string into the app and start rating the puppies.",["SingleFetchClassInstance",527],{"_33":34,"_35":183,"_37":528,"_39":529},{},[530],"You can look at how the data is distributed across shards by clicking on the tab “Show Data”.",["SingleFetchClassInstance",532],{"_33":34,"_35":183,"_37":533,"_39":534},{},[535],"You can also connect to the database to a MySQL client and try running queries as you run the app to see how the data is distributed across shards.","body_text","Since writing this blog we have released a new version of PlanetScale. Learn more about what we’ve built and give it a try, and be sure to check out our docs.\nPlease note, this blog refers to PlanetScaleDB v1 and is not applicable to our latest product.\nAt PlanetScale, we have built PlanetScaleDB, a fully managed database-as-a-service on top of open source Vitess that enables horizontal scaling of MySQL far beyond what you can do with a single instance. In this blog, we’ll explain how sharding works in Vitess and on PlanetScaleDB.\nA sharded database is a collection of multiple databases (shards) with identical relational schemas. Vitess allows your application to treat a sharded database as though it is a humongous monolithic database without having to worry about the complexities of sharding. Because of this, you can start with a small database on PlanetScaleDB and grow to massive scale without changing your application logic.\nIn this blog post, we’ll explore the Vitess sharding concepts: VSchema, Vindex, and Vitess Sequences using a sample dog rating application called “Goodest Doggo”. This sample app allows users to rate puppies and as you can imagine, needs to be designed to grow to humungous scale.\nHow is Vitess sharding different?\nA sharded database is a collection of multiple databases (shards) with identical relational schemas. Many database systems that use shards for scaling shard the data in a table without consideration for co-locating the data that belongs together. This results in inefficiencies around writing the data in a transactionally consistent fashion as well as reading the data. Vitess, in contrast, allows you to co-locate the related data.\nAssume you have a users table with the id column as the primary key and an orders table with its own id column as its primary key and user_id column as a secondary key. Vitess allows you to shard the user table using its id column and shard the orders table using the user_id column. This ensures both the user row for a user and the order rows for that user live in the same shard. A VSchema allows you to express this information.\nIn other words, just as a relational schema tells us how the data is organized within a single database, using tables, columns and indexes, Vitess uses a VSchema to define how the data is organized in shards across multiple databases. This allows us to define mechanisms which make it more efficient to access the data from these shards.\nWhat are the elements of a VSchema?\nJust as you would write a SQL statement that defines your relational database schema by creating table definitions which define columns and indexes, you define a VSchema (or a sharding scheme as it is called in PlanetScaleDB) by populating a JSON document that has the following information:\n\"sharded\": whether the database is sharded or not\n\"vindexes\": definition of all vindex types used in the VSchema\n“tables”: one entry for each table, each table entry has the following information:\na. Primary Vindex applied to a column in the table b. (optional) Secondary Vindexes applied to columns c. (optional) Sequences definitions columns for ids\nAs an example, consider the following table named “puppers” that represents the dogs in our application:CREATE TABLE IF NOT EXISTS puppers (\n `id` BIGINT(22),\n `name` VARCHAR(256),\n `image` VARCHAR(256),\n PRIMARY KEY(id)\n);\n\nHere is an example of a simple VSchema for a sharded database for the table above that illustrates the three elements of a VSchema:{\n \"sharded\": true,\n \"vindexes\": {\n \"hash_vdx\": {\n \"type\": \"hash\"\n }\n },\n \"tables\": {\n \"puppers\": {\n \"column_Vindexes\": [\n {\n \"column\": \"id\",\n \"name\": \"hash_vdx\"\n }\n ]\n }\n }\n}\n\nWhat is a Vindex?\nA relational database has a schema that consists of tables, columns and indexes amongst other elements. In the same way an index makes it efficient to access a given row in a given table in an unsharded database, a Vindex for a given table in a sharded database allows you to access a row quickly by allowing you to determine which shard a row lives in. Like indexes, Vindexes can be primary or secondary and a given table can only have a single primary Vindex, but can have multiple secondary Vindexes.\nHow does a Vindex map a row to a shard?\nTo understand this, you will have to understand the concept of keyspaces and keyspace_ids. In the Vitess world every row in every sharded database has a keyspace_id. The keyspace_id is not stored, but is computed by applying a specific sharding function to the value of a specific column in that row. The keyspace_ids range from 0x00 - 0xFF and this range represents the entire keyspace. Vitess shards cover this entire range. This is why a sharded database in Vitess is called a keyspace. Each shard spans a range in the keyspace and shards are named by starting and ending keyspace_id values for the shard. Let us take the example of a 4-shard keyspace. If we divide the range 0x00 to 0xFF (hexadecimal values) in four equal ranges, we get:0x00-0x40\n0x40-0x80\n0x80-0xC0\n0xC0-0xFF\n\nThus the four shards would be named: “00-40”, “40-80”, “80-C0”, “C0-FF”. Vitess drops the “00” at the beginning of the keyspace range and the FF at the end, so the shards are called “-40”, “40-80”, “80-C0”, “C0-” instead.\nVindexes allow you to map the value of a particular column in a given row to one or more keyspace_ids. Each shard has a starting and ending keyspace_id, thus given a keyspace_id, you can deterministically tell which shard a row belongs to.\nJust like an index, a Vindex is applied to a column in a table, but a Vindex has an additional property called sharding function. Vitess gives you 15 predefined sharding functions and certain sharding functions go well with columns of certain types, for example you would typically use “hash” sharding function with a numeric column, or you would use “unicode_loose_md5” for a varchar or varbinary column. You can also write custom sharding functions and use those instead.\nWhat is a Vitess sequence?\nIn an unsharded database when you need to assign a monotonically increasing value to a row, you can define a column to be of type “autoincrement”. In the sharded world, if you do this, you would end up having duplicate values in the same column across shards. Vitess solves this problem by allowing you to create ids which are monotonically increasing and unique across shards for a given table by defining sequences in the VSchema. The implementation of Vitess Sequences is backed by a row in a table a secondary database. Vitess limits the number of writes needed by allowing you to cache a certain number of values.\nLet us make the id column in the puppers table above of type Vitess sequence by adding the following segment to table entry for puppers in our VSchema:\"auto_increment\": {\n\t\"column\": \"id\",\n\t\"sequence\": \"pupper_seq\"\n}\n\nWhich gives the following VSchema:{\n \"sharded\": true,\n \"vindexes\": {\n \"hash\": {\n \"type\": \"hash\"\n }\n },\n \"tables\": {\n \"puppers\": {\n \"column_Vindexes\": [\n {\n \"column\": \"id\",\n \"name\": \"hash\"\n }\n ],\n \"auto_increment\": {\n \"column\": \"id\",\n \"sequence\": \"pupper_seq\"\n }\n }\n }\n}\n\nHow to define sequences?\nYou will notice that we are using the term pupper_seq without defining it anywhere. Here is how that process works. You typically define sequences in an unsharded keyspace that lives alongside your main keyspace. In the Vitess world this keyspace is typically called a lookup keyspace. Here is how you would define the sequence in the lookup keyspace. You first create a table named puppers_seq and you insert one row in it initializing the sequence. Note the comment vitess_sequence associated with the CREATE TABLE statement. It’s important for you to keep that because that is used by Vitess to treat this table distinctly from other tables which hold real data.CREATE TABLE IF NOT EXISTS pupper_seq (\n `id` INT,\n `next_id` BIGINT,\n `cache` BIGINT,\n PRIMARY KEY(id)\n) comment 'vitess_sequence';\n\nINSERT INTO pupper_seq (id, next_id, cache) VALUES (0, 1, 3);\n\nAfter defining this table, you apply the following VSchema to the lookup keyspace:{\n \"sharded\": false,\n \"tables\": {\n \"pupper_seq\": {\n \"type\": \"sequence\"\n }\n }\n}\n\nThese two steps define pupper_seq as a Vitess sequence and it can now be used by the VSchema for the sharded keyspace as we have used it above.\nPutting all this together\nSo, for the Goodest Doggo dog rating application, here is how we will organize our data. We will create two keyspaces, one called lookup, unsharded and one called puppers which is sharded. We will start with two shards and we can reshard as needed as we go. The lookup keyspace will hold tables we need for sequences and for lookup Vindexes. The doggers keyspace will hold the actual data. Here is the schema and VSchema for these two keyspaces:\nThe keyspace doggers has three tables: puppers, ratings, and users. We want the id column to be of type autoincrement, but as you can see below, because this is a sharded database so we do not specify the column as autoincrement in the schema, but instead we define a Vitess sequence on that column backed by the puppers_seq table in the lookup schema. This is the same for the id column for the ratings table.\nWe want the table puppers sharded by its id column, but the table ratings sharded by the user_id column. This is expressed in the VSchema.\nHere is the Schema and VSchema for the puppers database:CREATE TABLE `puppers` (\n `id` bigint(22) NOT NULL,\n `name` varchar(256) DEFAULT NULL,\n `image` varchar(256) DEFAULT NULL,\n PRIMARY KEY (`id`)\n) ENGINE = InnoDB DEFAULT CHARSET = utf8;\n\nCREATE TABLE `ratings` (\n `id` bigint(22) DEFAULT NULL,\n `user_id` bigint(22) DEFAULT NULL,\n `rating` bigint(20) DEFAULT NULL,\n `pupper_id` bigint(22) DEFAULT NULL,\n KEY `pupper_id` (`pupper_id`)\n) ENGINE = InnoDB DEFAULT CHARSET = utf8;\n\nCREATE TABLE `users` (\n `id` bigint(22) NOT NULL,\n `email` varchar(64) DEFAULT NULL,\n `password` varbinary(256) DEFAULT NULL,\n PRIMARY KEY (`id`)\n) ENGINE = InnoDB DEFAULT CHARSET = utf8;\n{\n \"sharded\": true,\n \"vindexes\": {\n \"binary_md5_vdx\": { \"type\": \"binary_md5\" },\n \"hash_vdx\": { \"type\": \"hash\" }\n },\n \"tables\": {\n \"puppers\": {\n \"columnVindexes\": [{ \"column\": \"id\", \"name\": \"hash_vdx\" }],\n \"autoIncrement\": { \"column\": \"id\", \"sequence\": \"pupper_seq\" }\n },\n \"ratings\": {\n \"columnVindexes\": [{ \"column\": \"pupper_id\", \"name\": \"hash_vdx\" }],\n \"autoIncrement\": { \"column\": \"id\", \"sequence\": \"rating_seq\" }\n },\n \"users\": {\n \"columnVindexes\": [{ \"column\": \"id\", \"name\": \"binary_md5_vdx\" }]\n }\n }\n}\n\nHere is the Schema and VSchema for the lookup database:CREATE TABLE IF NOT EXISTS pupper_seq (\n id INT,\n next_id BIGINT,\n cache BIGINT,\n PRIMARY KEY(id)\n) comment 'vitess_sequence';\n\nINSERT INTO pupper_seq (id, next_id, cache) VALUES (0, 1, 3);\n\nCREATE TABLE IF NOT EXISTS rating_seq (\n id INT,\n next_id BIGINT,\n cache BIGINT,\n PRIMARY KEY(id)\n) comment 'vitess_sequence';\n\nINSERT INTO rating_seq (id, next_id, cache) VALUES (0, 1, 3);\n{\n \"sharded\": false,\n \"tables\": {\n \"pupper_seq\": {\n \"type\": \"sequence\"\n },\n \"rating_seq\": {\n \"type\": \"sequence\"\n }\n }\n}\n\nWant to see this in action?\nWe have created a Quickstart Demo on PlanetScaleDB that creates the databases we described above and populates them with data from our sample dog rating application.\nTo get started, you will need to:\nCreate a PlanetScaleDB account.\nSelect the Quickstart Demo VSchema.\nThis will spin up a cluster and two databases and when successful will present you with a database URL.\nIn the meantime, download and start the application for your platform.\nOnce the application is running, it will provide you with a URL (such as http://localhost:8000) on your local host that you can browse to.\nThis web application will prompt you for the connection string for the cluster you created with the Quickstart Demo on PlanetScaleDB.\nInput the connection string into the app and start rating the puppies.\nYou can look at how the data is distributed across shards by clicking on the tab “Show Data”.\nYou can also connect to the database to a MySQL client and try running queries as you run the app to see how the data is distributed across shards.","aside","toc",[541,546,548,550,552,554,556,558],{"_39":542,"_98":99,"_543":544,"_545":106},[],"level",2,"title",{"_39":547,"_98":163,"_543":544,"_545":170},[],{"_39":549,"_98":247,"_543":544,"_545":254},[],{"_39":551,"_98":263,"_543":544,"_545":270},[],{"_39":553,"_98":306,"_543":544,"_545":313},[],{"_39":555,"_98":342,"_543":544,"_545":349},[],{"_39":557,"_98":378,"_543":544,"_545":385},[],{"_39":559,"_98":444,"_543":544,"_545":451},[],"Learn Horizontal Scaling on PlanetScaleDB with Vitess — Rate Puppies in a Rust app with Sharded MySQL Database","authors",[563],{"_27":564,"_35":565,"_566":-7},"jiten","Jiten Vaidya","x","categories",[569],"engineering","excerpt","Rate Puppies in a Rust app with Sharded MySQL Database","createdAt","2020-08-14T19:45:00.000Z","meta",{"_576":577,"_578":571,"_579":580,"_545":560},"canonical","https://planetscale.com/blog/learn-horizontal-scaling-on-planetscaledb-with-vitess-rate-puppies-in-a-rust-app-with-sharded-mysql-database","description","image","/assets/social-DtnN8vYG.png","category","pagination",{"_584":585,"_586":587,"_588":8,"_589":585,"_590":591},"page",6,"hasPrev",true,"hasNext","totalPages","url","https://planetscale.com/blog/category/engineering","https://planetscale.com","actionData","errors"]

Blog

Learn Horizontal Scaling on PlanetScaleDB with Vitess — Rate Puppies in a Rust app with Sharded MySQL Database