Once upon a time, there was gorp. Gorp was good, and Let's Encrypt's adopted it for Boulder's database persistence layer. The maintainers became busy with other projects, and gorp stopped getting updated. However, it's still very useful to Boulder, and there are a few tweaks we'd like to make for our own purposes, so we've forked it.
We maintain this primarily for Boulder's own use, and intend to make some API breaking changes during 2023. You are welcome to use our fork if you like it, but we do not expect to spend a lot of time maintaining it for non-Boulder uses.
In particular we maintain this to use with MariaDB 10.5+ and InnoDB.
package main
import (
"database/sql"
"github.com/letsencrypt/borp"
_ "github.com/mattn/go-sqlite3"
"log"
"time"
)
func main() {
// initialize the DbMap
dbmap := initDb()
defer dbmap.Db.Close()
// delete any existing rows
err := dbmap.TruncateTables()
checkErr(err, "TruncateTables failed")
// create two posts
p1 := newPost("Go 1.1 released!", "Lorem ipsum lorem ipsum")
p2 := newPost("Go 1.2 released!", "Lorem ipsum lorem ipsum")
// insert rows - auto increment PKs will be set properly after the insert
err = dbmap.Insert(&p1, &p2)
checkErr(err, "Insert failed")
// use convenience SelectInt
count, err := dbmap.SelectInt("select count(*) from posts")
checkErr(err, "select count(*) failed")
log.Println("Rows after inserting:", count)
// update a row
p2.Title = "Go 1.2 is better than ever"
count, err = dbmap.Update(&p2)
checkErr(err, "Update failed")
log.Println("Rows updated:", count)
// fetch one row - note use of "post_id" instead of "Id" since column is aliased
//
// Postgres users should use $1 instead of ? placeholders
// See 'Known Issues' below
err = dbmap.SelectOne(&p2, "select * from posts where post_id=?", p2.Id)
checkErr(err, "SelectOne failed")
log.Println("p2 row:", p2)
// fetch all rows
var posts []Post
_, err = dbmap.Select(&posts, "select * from posts order by post_id")
checkErr(err, "Select failed")
log.Println("All rows:")
for x, p := range posts {
log.Printf(" %d: %v\n", x, p)
}
// delete row by PK
count, err = dbmap.Delete(&p1)
checkErr(err, "Delete failed")
log.Println("Rows deleted:", count)
// delete row manually via Exec
_, err = dbmap.Exec("delete from posts where post_id=?", p2.Id)
checkErr(err, "Exec failed")
// confirm count is zero
count, err = dbmap.SelectInt("select count(*) from posts")
checkErr(err, "select count(*) failed")
log.Println("Row count - should be zero:", count)
log.Println("Done!")
}
type Post struct {
// db tag lets you specify the column name if it differs from the struct field
Id int64 `db:"post_id"`
Created int64
Title string `db:",size:50"` // Column size set to 50
Body string `db:"article_body,size:1024"` // Set both column name and size
}
func newPost(title, body string) Post {
return Post{
Created: time.Now().UnixNano(),
Title: title,
Body: body,
}
}
func initDb() *borp.DbMap {
// connect to db using standard Go database/sql API
// use whatever database/sql driver you wish
db, err := sql.Open("sqlite3", "/tmp/post_db.bin")
checkErr(err, "sql.Open failed")
// construct a borp DbMap
dbmap := &borp.DbMap{Db: db, Dialect: borp.SqliteDialect{}}
// add a table, setting the table name to 'posts' and
// specifying that the Id property is an auto incrementing PK
dbmap.AddTableWithName(Post{}, "posts").SetKeys(true, "Id")
// create the table. in a production system you'd generally
// use a migration tool, or create the tables via scripts
err = dbmap.CreateTablesIfNotExists()
checkErr(err, "Create tables failed")
return dbmap
}
func checkErr(err error, msg string) {
if err != nil {
log.Fatalln(msg, err)
}
}
First define some types:
type Invoice struct {
Id int64
Created int64
Updated int64
Memo string
PersonId int64
}
type Person struct {
Id int64
Created int64
Updated int64
FName string
LName string
}
// Example of using tags to alias fields to column names
// The 'db' value is the column name
//
// A hyphen will cause borp to skip this field, similar to the
// Go json package.
//
// This is equivalent to using the ColMap methods:
//
// table := dbmap.AddTableWithName(Product{}, "product")
// table.ColMap("Id").Rename("product_id")
// table.ColMap("Price").Rename("unit_price")
// table.ColMap("IgnoreMe").SetTransient(true)
//
// You can optionally declare the field to be a primary key and/or autoincrement
//
type Product struct {
Id int64 `db:"product_id, primarykey, autoincrement"`
Price int64 `db:"unit_price"`
IgnoreMe string `db:"-"`
}
Then create a mapper, typically you'd do this one time at app startup:
// connect to db using standard Go database/sql API
// use whatever database/sql driver you wish
db, err := sql.Open("mymysql", "tcp:localhost:3306*mydb/myuser/mypassword")
// construct a borp DbMap
dbmap := &borp.DbMap{Db: db, Dialect: borp.MySQLDialect{"InnoDB", "UTF8"}}
// register the structs you wish to use with borp
// you can also use the shorter dbmap.AddTable() if you
// don't want to override the table name
//
// SetKeys(true) means we have a auto increment primary key, which
// will get automatically bound to your struct post-insert
//
t1 := dbmap.AddTableWithName(Invoice{}, "invoice_test").SetKeys(true, "Id")
t2 := dbmap.AddTableWithName(Person{}, "person_test").SetKeys(true, "Id")
t3 := dbmap.AddTableWithName(Product{}, "product_test").SetKeys(true, "Id")
borp supports embedding structs. For example:
type Names struct {
FirstName string
LastName string
}
type WithEmbeddedStruct struct {
Id int64
Names
}
es := &WithEmbeddedStruct{-1, Names{FirstName: "Alice", LastName: "Smith"}}
err := dbmap.Insert(es)
See the TestWithEmbeddedStruct
function in borp_test.go
for a full example.
Automatically create / drop registered tables. This is useful for unit tests but is entirely optional. You can of course use borp with tables created manually, or with a separate migration tool (like sql-migrate, goose or migrate).
// create all registered tables
dbmap.CreateTables()
// same as above, but uses "if not exists" clause to skip tables that are
// already defined
dbmap.CreateTablesIfNotExists()
// drop
dbmap.DropTables()
Optionally you can pass in a logger to trace all SQL statements. I recommend enabling this initially while you're getting the feel for what borp is doing on your behalf.
Borp defines a GorpLogger
interface that Go's built in log.Logger
satisfies.
However, you can write your own GorpLogger
implementation, or use a package such
as glog
if you want more control over how statements are logged.
// Will log all SQL statements + args as they are run
// The first arg is a string prefix to prepend to all log messages
dbmap.TraceOn("[gorp]", log.New(os.Stdout, "myapp:", log.Lmicroseconds))
// Turn off tracing
dbmap.TraceOff()
// Must declare as pointers so optional callback hooks
// can operate on your data, not copies
inv1 := &Invoice{0, 100, 200, "first order", 0}
inv2 := &Invoice{0, 100, 200, "second order", 0}
// Insert your rows
err := dbmap.Insert(inv1, inv2)
// Because we called SetKeys(true) on Invoice, the Id field
// will be populated after the Insert() automatically
fmt.Printf("inv1.Id=%d inv2.Id=%d\n", inv1.Id, inv2.Id)
Continuing the above example, use the Update
method to modify an Invoice:
// count is the # of rows updated, which should be 1 in this example
count, err := dbmap.Update(inv1)
If you have primary key(s) defined for a struct, you can use the Delete
method to remove rows:
count, err := dbmap.Delete(inv1)
Use the Get
method to fetch a single row by primary key. It returns
nil if no row is found.
// fetch Invoice with Id=99
obj, err := dbmap.Get(Invoice{}, 99)
inv := obj.(*Invoice)
Select()
and SelectOne()
provide a simple way to bind arbitrary queries to a slice
or a single struct.
// Select a slice - first return value is not needed when a slice pointer is passed to Select()
var posts []Post
_, err := dbmap.Select(&posts, "select * from post order by id")
// You can also use primitive types
var ids []string
_, err := dbmap.Select(&ids, "select id from post")
// Select a single row.
// Returns an error if no row found, or if more than one row is found
var post Post
err := dbmap.SelectOne(&post, "select * from post where id=?", id)
Want to do joins? Just write the SQL and the struct. Borp will bind them:
// Define a type for your join
// It *must* contain all the columns in your SELECT statement
//
// The names here should match the aliased column names you specify
// in your SQL - no additional binding work required. simple.
//
type InvoicePersonView struct {
InvoiceId int64
PersonId int64
Memo string
FName string
}
// Create some rows
p1 := &Person{0, 0, 0, "bob", "smith"}
err = dbmap.Insert(p1)
checkErr(err, "Insert failed")
// notice how we can wire up p1.Id to the invoice easily
inv1 := &Invoice{0, 0, 0, "xmas order", p1.Id}
err = dbmap.Insert(inv1)
checkErr(err, "Insert failed")
// Run your query
query := "select i.Id InvoiceId, p.Id PersonId, i.Memo, p.FName " +
"from invoice_test i, person_test p " +
"where i.PersonId = p.Id"
// pass a slice to Select()
var list []InvoicePersonView
_, err := dbmap.Select(&list, query)
// this should test true
expected := InvoicePersonView{inv1.Id, p1.Id, inv1.Memo, p1.FName}
if reflect.DeepEqual(list[0], expected) {
fmt.Println("Woot! My join worked!")
}
Borp provides a few convenience methods for selecting a single string or int64.
// select single int64 from db (use $1 instead of ? for postgresql)
i64, err := dbmap.SelectInt("select count(*) from foo where blah=?", blahVal)
// select single string from db:
s, err := dbmap.SelectStr("select name from foo where blah=?", blahVal)
You may use a map or struct to bind parameters by name. This is supported in the Exec and Select* functions on DbMap and Transaction. It is not supported by Prepare, Query, or QueryRow.
_, err := dbm.Select(&dest, "select * from Foo where name = :name and age = :age", map[string]interface{}{
"name": "Rob",
"age": 31,
})
You can execute raw SQL if you wish. Particularly good for batch operations.
res, err := dbmap.Exec("delete from invoice_test where PersonId=?", 10)
You can batch operations into a transaction:
func InsertInv(dbmap *DbMap, inv *Invoice, per *Person) error {
// Start a new transaction
trans, err := dbmap.Begin()
if err != nil {
return err
}
err = trans.Insert(per)
checkErr(err, "Insert failed")
inv.PersonId = per.Id
err = trans.Insert(inv)
checkErr(err, "Insert failed")
// if the commit is successful, a nil error is returned
return trans.Commit()
}
Use hooks to update data before/after saving to the db. Good for timestamps:
// implement the PreInsert and PreUpdate hooks
func (i *Invoice) PreInsert(s borp.SqlExecutor) error {
i.Created = time.Now().UnixNano()
i.Updated = i.Created
return nil
}
func (i *Invoice) PreUpdate(s borp.SqlExecutor) error {
i.Updated = time.Now().UnixNano()
return nil
}
// You can use the SqlExecutor to cascade additional SQL
// Take care to avoid cycles. Borp won't prevent them.
//
// Here's an example of a cascading delete
//
func (p *Person) PreDelete(s borp.SqlExecutor) error {
query := "delete from invoice_test where PersonId=?"
_, err := s.Exec(query, p.Id)
if err != nil {
return err
}
return nil
}
Full list of hooks that you can implement:
PostGet
PreInsert
PostInsert
PreUpdate
PostUpdate
PreDelete
PostDelete
All have the same signature. for example:
func (p *MyStruct) PostUpdate(s borp.SqlExecutor) error
Note that this behaviour has changed in v2. See Migration Guide.
Borp provides a simple optimistic locking feature, similar to Java's
JPA, that will raise an error if you try to update/delete a row whose
version
column has a value different than the one in memory. This
provides a safe way to do "select then update" style operations
without explicit read and write locks.
// Version is an auto-incremented number, managed by borp
// If this property is present on your struct, update
// operations will be constrained
//
// For example, say we defined Person as:
type Person struct {
Id int64
Created int64
Updated int64
FName string
LName string
// automatically used as the Version col
// use table.SetVersionCol("columnName") to map a different
// struct field as the version field
Version int64
}
p1 := &Person{0, 0, 0, "Bob", "Smith", 0}
err = dbmap.Insert(p1) // Version is now 1
checkErr(err, "Insert failed")
obj, err := dbmap.Get(Person{}, p1.Id)
p2 := obj.(*Person)
p2.LName = "Edwards"
_,err = dbmap.Update(p2) // Version is now 2
checkErr(err, "Update failed")
p1.LName = "Howard"
// Raises error because p1.Version == 1, which is out of date
count, err := dbmap.Update(p1)
_, ok := err.(borp.OptimisticLockError)
if ok {
// should reach this statement
// in a real app you might reload the row and retry, or
// you might propegate this to the user, depending on the desired
// semantics
fmt.Printf("Tried to update row with stale data: %v\n", err)
} else {
// some other db error occurred - log or return up the stack
fmt.Printf("Unknown db err: %v\n", err)
}
Indexes are frequently critical for performance. Here is how to add them to your tables.
In the example below we put an index both on the Id field, and on the AcctId field.
type Account struct {
Id int64
AcctId string // e.g. this might be a long uuid for portability
}
// indexType (the 2nd param to AddIndex call) is "Btree" or "Hash" for MySQL.
// demonstrate adding a second index on AcctId, and constrain that field to have unique values.
dbm.AddTable(iptab.Account{}).SetKeys(true, "Id").AddIndex("AcctIdIndex", "Btree", []string{"AcctId"}).SetUnique(true)
err = dbm.CreateTablesIfNotExists()
checkErr(err, "CreateTablesIfNotExists failed")
err = dbm.CreateIndex()
checkErr(err, "CreateIndex failed")
Check the effect of the CreateIndex() call in mysql:
$ mysql
MariaDB [test]> show create table Account;
+---------+--------------------------+
| Account | CREATE TABLE `Account` (
`Id` bigint(20) NOT NULL AUTO_INCREMENT,
`AcctId` varchar(255) DEFAULT NULL,
PRIMARY KEY (`Id`),
UNIQUE KEY `AcctIdIndex` (`AcctId`) USING BTREE <<<--- yes! index added.
) ENGINE=InnoDB DEFAULT CHARSET=utf8
+---------+--------------------------+
borp uses the Go 1 database/sql
package. A full list of compliant
drivers is available here:
http://code.google.com/p/go-wiki/wiki/SQLDrivers
Sadly, SQL databases differ on various issues. borp provides a Dialect interface that should be implemented per database vendor. Dialects are provided for:
- MySQL
- PostgreSQL
- sqlite3
Each of these three databases pass the test suite. See borp_test.go
for example DSNs for these three databases.
In order to use sqlite3 extensions you need to first register a custom driver:
import (
"database/sql"
// use whatever database/sql driver you wish
sqlite "github.com/mattn/go-sqlite3"
)
func customDriver() (*sql.DB, error) {
// create custom driver with extensions defined
sql.Register("sqlite3-custom", &sqlite.SQLiteDriver{
Extensions: []string{
"mod_spatialite",
},
})
// now you can then connect using the 'sqlite3-custom' driver instead of 'sqlite3'
return sql.Open("sqlite3-custom", "/tmp/post_db.bin")
}
Different databases use different strings to indicate variable
placeholders in prepared SQL statements. Unlike some database
abstraction layers (such as JDBC), Go's database/sql
does not
standardize this.
SQL generated by borp in the Insert
, Update
, Delete
, and Get
methods delegates to a Dialect implementation for each database, and
will generate portable SQL.
Raw SQL strings passed to Exec
, Select
, SelectOne
, SelectInt
,
etc will not be parsed. Consequently you may have portability issues
if you write a query like this:
// works on MySQL and Sqlite3, but not with Postgresql err :=
dbmap.SelectOne(&val, "select * from foo where id = ?", 30)
In Select
and SelectOne
you can use named parameters to work
around this. The following is portable:
err := dbmap.SelectOne(&val, "select * from foo where id = :id",
map[string]interface{} { "id": 30})
Additionally, when using Postgres as your database, you should utilize
$1
instead of ?
placeholders as utilizing ?
placeholders when
querying Postgres will result in pq: operator does not exist
errors. Alternatively, use dbMap.Dialect.BindVar(varIdx)
to get the
proper variable binding for your dialect.
borp will pass time.Time
fields through to the database/sql
driver, but note that the behavior of this type varies across database
drivers.
MySQL users should be especially cautious. See: ziutek/mymysql#77
To avoid any potential issues with timezone/DST, consider:
- Using an integer field for time data and storing UNIX time.
- Using a custom time type that implements some SQL types:
The included tests may be run against MySQL, Postgres, or sqlite3. You must set two environment variables so the test code knows which driver to use, and how to connect to your database.
# MySQL example:
export GORP_TEST_DSN=gomysql_test/gomysql_test/abc123
export GORP_TEST_DIALECT=mysql
# run the tests
go test
# run the tests and benchmarks
go test -bench="Bench" -benchtime 10
Valid GORP_TEST_DIALECT
values are: "mysql"(for mymysql),
"gomysql"(for go-sql-driver), "postgres", or "sqlite" See the
test_all.sh
script for examples of all 3 databases. This is the
script I run locally to test the library.
- matthias-margush - column aliasing via tags
- Rob Figueiredo - @robfig
- Quinn Slack - @sqs