ksql

KissSQL

KissSQL or the "Keep it Simple" SQL package was created to offer an actually simple and satisfactory tool for interacting with SQL Databases.

The core idea on ksql is to offer an easy to use interface, the actual comunication with the database is decoupled so we can use ksql on top of pgx, database/sql and possibly other tools. You can even create you own backend adapter for ksql which is useful in some situations.

Why ksql?

Note: If you want numbers see our Benchmark section below

ksql is meant to improve on the existing ecosystem by optimizing for the most interesting use-cases with as little extra baggage possible, offering among other things:

• An easier time setting up and learning it
• Less opportunities for making mistakes, which makes code reviews easier
• A succinct and idiomatic Go idiom reducing the cognitive complexity of your code
• Easy ways of mocking your database when you need to.
• Support for all common databases
• No DSL: Use SQL for your queries

And for a few important use-cases that cannot follow these rules perfectly, we have carefully chosen a few powerful abstractions that might be slightly more complicated to learn, such as:

• The QueryChunks() function which is necessary for the few situations when you might load big amounts of the data in a single query.
• And the possibility of omitting the SELECT ... part of the query which causes ksql to write this part for you saving a lot of work when working with big structs/tables.
• Support for nesting structs when working with JOINs.

Supported Drivers:

ksql is well decoupled from its backend implementation which makes it easy to change the actual technology used, currently we already support the following options:

• Using the database/sql as the backend we support the following drivers:
  • "postgres"
  • "sqlite3"
  • "mysql"
  • "sqlserver"
• We also support pgx (actually pgxpool) as the backend which is a lot faster for Postgres databases.

If you need a new database/sql driver or backend adapter included please open an issue or make your own implementation and submit it as a Pull Request.

Comparing ksql with other tools

ksql was created because of a few insatisfactions with the existing packages for interacting with relational databases in Go. To mention a few:

Low Level Tools:

Tools like database/sql, sqlx and even pgx will usually require you to check errors several times for the same query and also when iterating over several rows you end up with a for rows.Next() {} loop which is often more cognitive complex than desirable.

High Level Tools such as ORMs:

More high level tools such as gorm and bun will often force you and your team to interact with a complicated DSL which requires time to learn it and then ending up still being a little bit harder to read than a regular SQL query would be.

Code Generation tools:

Tools like sqlc and sqlboiler that rely on code generation are good options if performance is your main goal, but they also have some issues that might bother you:

• There is some learning curve that goes beyond just reading a GoDoc as with most packages.
• You will often need to copy to and from custom generated structs instead of using your own.
• Sometimes the generated function will not be as flexible as you'd prefer forcing you to make some tricks with SQL (e.g. that happens with sqlc for partial updates for example).
• And it does add an extra step on your building process.

And finally you might just prefer to avoid codegen when possible, in which case ksql is also for you.

Kiss Interface

The current interface is as follows and we plan on keeping it with as little functions as possible, so don't expect many additions:

// Provider describes the ksql public behavior
//
// The Insert, Update, Delete and QueryOne functions return ksql.ErrRecordNotFound
// if no record was found or no rows were changed during the operation.
type Provider interface {
	Insert(ctx context.Context, table Table, record interface{}) error
	Update(ctx context.Context, table Table, record interface{}) error
	Delete(ctx context.Context, table Table, idOrRecord interface{}) error

	Query(ctx context.Context, records interface{}, query string, params ...interface{}) error
	QueryOne(ctx context.Context, record interface{}, query string, params ...interface{}) error
	QueryChunks(ctx context.Context, parser ChunkParser) error

	Exec(ctx context.Context, query string, params ...interface{}) (rowsAffected int64, _ error)
	Transaction(ctx context.Context, fn func(Provider) error) error
}

Usage examples

This example is also available here if you want to compile it yourself.

Also we have a small feature for building the "SELECT" part of the query if you rather not use SELECT * queries, you may skip to the Select Generator Feature which is recommended.

package main

import (
	"context"
	"fmt"

	"github.com/vingarcia/ksql"
	"github.com/vingarcia/ksql/adapters/ksqlite3"
	"github.com/vingarcia/ksql/nullable"
)

// User ...
type User struct {
	ID   int    `ksql:"id"`
	Name string `ksql:"name"`
	Age  int    `ksql:"age"`

	// This field will be saved as JSON in the database
	Address Address `ksql:"address,json"`
}

// PartialUpdateUser ...
type PartialUpdateUser struct {
	ID      int      `ksql:"id"`
	Name    *string  `ksql:"name"`
	Age     *int     `ksql:"age"`
	Address *Address `ksql:"address,json"`
}

// Address ...
type Address struct {
	State string `json:"state"`
	City  string `json:"city"`
}

// UsersTable informs ksql the name of the table and that it can
// use the default value for the primary key column name: "id"
var UsersTable = ksql.NewTable("users")

func main() {
	ctx := context.Background()

	// The available adapters are:
	// - kpgx.New(ctx, connURL, ksql.Config{})
	// - kmysql.New(ctx, connURL, ksql.Config{})
	// - ksqlserver.New(ctx, connURL, ksql.Config{})
	// - ksqlite3.New(ctx, connURL, ksql.Config{})
	//
	// For more detailed examples see:
	// - `./examples/all_adapters/all_adapters.go`
	//
	// In this example we'll use sqlite3:
	db, err := ksqlite3.New(ctx, "/tmp/hello.sqlite", ksql.Config{
		MaxOpenConns: 1,
	})
	if err != nil {
		panic(err.Error())
	}

	// In the definition below, please note that BLOB is
	// the only type we can use in sqlite for storing JSON.
	_, err = db.Exec(ctx, `CREATE TABLE IF NOT EXISTS users (
	  id INTEGER PRIMARY KEY,
		age INTEGER,
		name TEXT,
		address BLOB
	)`)
	if err != nil {
		panic(err.Error())
	}

	var alison = User{
		Name: "Alison",
		Age:  22,
		Address: Address{
			State: "MG",
		},
	}
	err = db.Insert(ctx, UsersTable, &alison)
	if err != nil {
		panic(err.Error())
	}
	fmt.Println("Alison ID:", alison.ID)

	// Inserting inline:
	err = db.Insert(ctx, UsersTable, &User{
		Name: "Cristina",
		Age:  27,
		Address: Address{
			State: "SP",
		},
	})
	if err != nil {
		panic(err.Error())
	}

	// Deleting Alison:
	err = db.Delete(ctx, UsersTable, alison.ID)
	if err != nil {
		panic(err.Error())
	}

	// Retrieving Cristina:
	var cris User
	err = db.QueryOne(ctx, &cris, "SELECT * FROM users WHERE name = ? ORDER BY id", "Cristina")
	if err != nil {
		panic(err.Error())
	}
	fmt.Printf("Cristina: %#v\n", cris)

	// Updating all fields from Cristina:
	cris.Name = "Cris"
	err = db.Update(ctx, UsersTable, cris)

	// Changing the age of Cristina but not touching any other fields:

	// Partial update technique 1:
	err = db.Update(ctx, UsersTable, struct {
		ID  int `ksql:"id"`
		Age int `ksql:"age"`
	}{ID: cris.ID, Age: 28})
	if err != nil {
		panic(err.Error())
	}

	// Partial update technique 2:
	err = db.Update(ctx, UsersTable, PartialUpdateUser{
		ID:  cris.ID,
		Age: nullable.Int(28),
	})
	if err != nil {
		panic(err.Error())
	}

	// Listing first 10 users from the database
	// (each time you run this example a new Cristina is created)
	//
	// Note: Using this function it is recommended to set a LIMIT, since
	// not doing so can load too many users on your computer's memory or
	// cause an Out Of Memory Kill.
	//
	// If you need to query very big numbers of users we recommend using
	// the `QueryChunks` function.
	var users []User
	err = db.Query(ctx, &users, "SELECT * FROM users LIMIT 10")
	if err != nil {
		panic(err.Error())
	}

	// Making transactions:
	err = db.Transaction(ctx, func(db ksql.Provider) error {
		var cris2 User
		err = db.QueryOne(ctx, &cris2, "SELECT * FROM users WHERE id = ?", cris.ID)
		if err != nil {
			// This will cause an automatic rollback:
			return err
		}

		err = db.Update(ctx, UsersTable, PartialUpdateUser{
			ID:  cris2.ID,
			Age: nullable.Int(29),
		})
		if err != nil {
			// This will also cause an automatic rollback and then panic again
			// so that we don't hide the panic inside the KissSQL library
			panic(err.Error())
		}

		// Commits the transaction
		return nil
	})
	if err != nil {
		panic(err.Error())
	}

	fmt.Printf("Users: %#v\n", users)
}

Query Chunks Feature

It's very unsual for us to need to load a number of records from the database that might be too big for fitting in memory, e.g. load all the users and send them somewhere. But it might happen.

For these cases it's best to load chunks of data at a time so that we can work on a substantial amount of data at a time and never overload our memory capacity. For this use case we have a specific function called QueryChunks:

err = db.QueryChunks(ctx, ksql.ChunkParser{
	Query:     "SELECT * FROM users WHERE type = ?",
	Params:    []interface{}{usersType},
	ChunkSize: 100,
	ForEachChunk: func(users []User) error {
		err := sendUsersSomewhere(users)
		if err != nil {
			// This will abort the QueryChunks loop and return this error
			return err
		}
		return nil
	},
})
if err != nil {
	panic(err.Error())
}

It's signature is more complicated than the other two Query* methods, thus, it is adivisible to always prefer using the other two when possible reserving this one for the rare use-case where you are actually loading big sections of the database into memory.

Select Generator Feature

There are good reasons not to use SELECT * queries the most important of them is that you might end up loading more information than you are actually going to use putting more pressure in your database for no good reason.

To prevent that ksql has a feature specifically for building the SELECT part of the query using the tags from the input struct. Using it is very simple and it works with all the 3 Query* functions:

Querying a single user:

var user User
err = db.QueryOne(ctx, &user, "FROM users WHERE id = ?", userID)
if err != nil {
	panic(err.Error())
}

Querying a page of users:

var users []User
err = db.Query(ctx, &users, "FROM users WHERE type = ? ORDER BY id LIMIT ? OFFSET ?", "Cristina", limit, offset)
if err != nil {
	panic(err.Error())
}

Querying all the users, or any potentially big number of users, from the database (not usual, but supported):

err = db.QueryChunks(ctx, ksql.ChunkParser{
	Query:     "FROM users WHERE type = ?",
	Params:    []interface{}{usersType},
	ChunkSize: 100,
	ForEachChunk: func(users []User) error {
		err := sendUsersSomewhere(users)
		if err != nil {
			// This will abort the QueryChunks loop and return this error
			return err
		}
		return nil
	},
})
if err != nil {
	panic(err.Error())
}

The implementation of this feature is actually simple internally. First we check if the query is starting with the word FROM, if it is then we just get the ksql tags from the struct and then use it for building the SELECT statement.

The SELECT statement is then cached so we don't have to build it again the next time in order to keep the library efficient even when using this feature.

Select Generation with Joins

So there is one use-case that was not covered by ksql so far:

What if you want to JOIN multiple tables for which you already have structs defined? Would you need to create a new struct to represent the joined columns of the two tables? no, we actually have this covered as well.

ksql has a special feature for allowing the reuse of existing structs by using composition in an anonymous struct, and then generating the SELECT part of the query accordingly:

Querying a single joined row:

var row struct{
	User User `tablename:"u"`     // (here the tablename must match the aliased tablename in the query)
	Post Post `tablename:"posts"` // (if no alias is used you should use the actual name of the table)
}
err = db.QueryOne(ctx, &row, "FROM users as u JOIN posts ON u.id = posts.user_id WHERE u.id = ?", userID)
if err != nil {
	panic(err.Error())
}

Querying a page of joined rows:

var rows []struct{
	User User `tablename:"u"`
	Post Post `tablename:"p"`
}
err = db.Query(ctx, &rows,
	"FROM users as u JOIN posts as p ON u.id = p.user_id WHERE name = ? LIMIT ? OFFSET ?",
	"Cristina", limit, offset,
)
if err != nil {
	panic(err.Error())
}

Querying all the users, or any potentially big number of users, from the database (not usual, but supported):

err = db.QueryChunks(ctx, ksql.ChunkParser{
	Query:     "FROM users as u JOIN posts as p ON u.id = p.user_id WHERE type = ?",
	Params:    []interface{}{usersType},
	ChunkSize: 100,
	ForEachChunk: func(rows []struct{
		User User `tablename:"u"`
		Post Post `tablename:"p"`
	}) error {
		err := sendRowsSomewhere(rows)
		if err != nil {
			// This will abort the QueryChunks loop and return this error
			return err
		}
		return nil
	},
})
if err != nil {
	panic(err.Error())
}

As advanced as this feature might seem we don't do any parsing of the query, and all the work is done only once and then cached.

What actually happens is that we use the "tablename" tag to build the SELECT part of the query like this:

• SELECT u.id, u.name, u.age, p.id, p.title

This is then cached, and when we need it again we concatenate it with the rest of the query.

This feature has two important limitations:

1. It is not possible to use tablename tags together with normal ksql tags. Doing so will cause the tablename tags to be ignored in favor of the ksql ones.
2. It is not possible to use it without omitting the SELECT part of the query. While in normal queries we match the selected field with the attribute by name, in queries joining multiple tables we can't use this strategy because different tables might have columns with the same name, and we don't really have access to the full name of these columns making, for example, it impossible to differentiate between u.id and p.id except by the order in which these fields were passed. Thus, it is necessary that the library itself writes the SELECT part of the query when using this technique so that we can control the order or the selected fields.

Ok, but what if I don't want to use this feature?

You are not forced to, and there are a few use-cases where you would prefer not to, e.g.:

var rows []struct{
	UserName string `ksql:"name"`
	PostTitle string `ksql:"title"`
}
err := db.Query(ctx, &rows, "SELECT u.name, p.title FROM users u JOIN posts p ON u.id = p.user_id LIMIT 10")
if err != nil {
	panic(err.Error())
}

In the example above, since we are only interested in a couple of columns it is far simpler and more efficient for the database to only select the columns that we actually care about, so it's better not to use composite structs.

Testing Examples

This library has a few helper functions for helping your tests:

• kstructs.FillStructWith(struct interface{}, dbRow map[string]interface{}) error
• kstructs.FillSliceWith(structSlice interface{}, dbRows []map[string]interface{}) error
• kstructs.StructToMap(struct interface{}) (map[string]interface{}, error)
• kstructs.CallFunctionWithRows(fn interface{}, rows []map[string]interface{}) (map[string]interface{}, error)

If you want to see examples (we have examples for all the public functions) just read the example tests available on our example service

Benchmark Comparison

The results of the benchmark are good: they show that ksql is in practical terms, as fast as sqlx which was our goal from the start.

To understand the benchmark below you must know that all tests are performed using Postgres 12.1 and that we are comparing the following tools:

• ksql using the adapter that wraps database/sql
• ksql using the adapter that wraps pgx
• sql
• sqlx
• pgx (with pgxpool)
• gorm

For each of these tools we are running 3 different queries:

The insert-one query looks like:

INSERT INTO users (name, age) VALUES ($1, $2) RETURNING id

The single-row query looks like:

SELECT id, name, age FROM users OFFSET $1 LIMIT 1

The multiple-rows query looks like:

SELECT id, name, age FROM users OFFSET $1 LIMIT 10

Keep in mind that some of the tools tested actually build the query internally so the actual query might differ a little bit from the example ones above.

Without further ado, here are the results:

$ make bench TIME=5s
cd benchmarks && go test -bench=. -benchtime=5s
goos: linux
goarch: amd64
pkg: github.com/vingarcia/ksql/benchmarks
cpu: Intel(R) Core(TM) i5-3210M CPU @ 2.50GHz
BenchmarkInsert/ksql/sql-adapter/insert-one-4         	    6931	    845240 ns/op
BenchmarkInsert/sql/insert-one-4                      	    6534	    827073 ns/op
BenchmarkInsert/sql/prep-statements/insert-one-4      	    9369	    651082 ns/op
BenchmarkInsert/sqlx/insert-one-4                     	    6112	    825379 ns/op
BenchmarkInsert/ksql/pgx-adapter/insert-one-4         	    9243	    655494 ns/op
BenchmarkInsert/pgxpool/insert-one-4                  	    9312	    633110 ns/op
BenchmarkInsert/gorm/insert-one-4                     	    6433	   1124113 ns/op
BenchmarkQuery/ksql/sql-adapter/single-row-4          	   17823	    335562 ns/op
BenchmarkQuery/ksql/sql-adapter/multiple-rows-4       	   16302	    360614 ns/op
BenchmarkQuery/sql/single-row-4                       	   18807	    341746 ns/op
BenchmarkQuery/sql/multiple-rows-4                    	   18151	    347268 ns/op
BenchmarkQuery/sql/prep-statements/single-row-4       	   40617	    150549 ns/op
BenchmarkQuery/sql/prep-statements/multiple-rows-4    	   36740	    162904 ns/op
BenchmarkQuery/sqlx/single-row-4                      	   18183	    312080 ns/op
BenchmarkQuery/sqlx/multiple-rows-4                   	   17359	    332093 ns/op
BenchmarkQuery/ksql/pgx-adapter/single-row-4          	   35664	    151669 ns/op
BenchmarkQuery/ksql/pgx-adapter/multiple-rows-4       	   33708	    180191 ns/op
BenchmarkQuery/pgxpool/single-row-4                   	   41640	    153285 ns/op
BenchmarkQuery/pgxpool/multiple-rows-4                	   41274	    155219 ns/op
BenchmarkQuery/gorm/single-row-4                      	   38065	    161875 ns/op
BenchmarkQuery/gorm/multiple-rows-4                   	   25285	    227142 ns/op
PASS
ok  	github.com/vingarcia/ksql/benchmarks	183.246s
Benchmark executed at: 2021-11-16
Benchmark executed on commit: fc6a9c2950903139ed7a8432bdcfdb3eb89f1e21

Running the ksql tests (for contributors)

The tests use docker-test for setting up all the supported databases, which means that:

• You need to have docker installed

• You must be able to run docker without sudo, i.e. if you are not root you should add yourself to the docker group, e.g.:

  $ sudo usermod <your_username> -aG docker
  

  And then restart your login session (or just reboot)

After that you can just run the tests by using:

make test

But it is recommended to first download the required images using:

docker pull postgres:14.0
docker pull mysql:8.0.27
docker pull mcr.microsoft.com/mssql/server:2017-latest

Otherwise the first attempt to run the tests will spend a long time downloading these images and then fail because the TestMain() function is configured to kill the containers after 20 seconds.

TODO List

• Add tests for tables using composite keys
• Add support for serializing structs as other formats such as YAML
• Update kstructs.FillStructWith to work with ksql:"..,json" tagged attributes
• Create a way for users to submit user defined dialects
• Improve error messages
• Add support for the update function to work with maps for partial updates
• Add support for the insert function to work with maps
• Add support for a ksql.Array(params ...interface{}) for allowing queries like this: db.Query(ctx, &user, "SELECT * FROM user WHERE id in (?)", ksql.Array(1,2,3))

Optimization Oportunities

• Test if using a pointer on the field info is faster or not
• Consider passing the cached structInfo as argument for all the functions that use it, so that we don't need to get it more than once in the same call.
• Use a cache to store all queries after they are built
• Preload the insert method for all dialects inside ksql.NewTable()