Sorry for my poor English. If you can help with improving the English in this documentation, please contact me.
This package contains MySQL client API written entirely in Go. It is designed to work with the MySQL protocol version 4.1 or greater. It definitely works well with MySQL server version 5.0 and 5.1 (I use these versions of MySQL servers for my applications). Some people claim that mymysql works with older versions of MySQL protocol too.
v1.5.4: Bugs fixed in native and godrv packages.
v1.5.3: Bugs fixed in new godrv code.
v1.5.1: Conn.NetConn method added.
v1.5: Needs Go 1.1 (time.ParseInLocation and net.Dialer) to compile.
v1.4: Stmt.ResetParams, Stmt.Map and Stmt.NumFields methods disappeared.
New Stmt.Fields method. godrv implements driver.Queryer interface which
improves performance when compiled with Go tip.
v1.3: Overall performance improved by factor 1.5 to 1.8. All Encode* functions
now accept properly sized []byte slice as first argument.
v1.2: Faster execution of simple queries in mymysql/godrv. EscapeString
method renamed to Escape.
v1.1: Client error codes moved from mymysql/native pacage to mymysql/mysql.
v1.0: Transactions added to autorc, new Transaction.IsValid method. I think this library is mature enough to release it as v1.0
v0.4.11: Add Reconnect, Register, SetMaxPktSize, Bind to autorc.
v0.4.10: New Clone method for create connection from other connection.
v0.4.9: New method for create connection from configuration in file: NewFromCF.
v0.4.8: New methods for obtain only first/last row from result set. Better implementation of discarding rows in End method.
v0.4.7: ScanRow and MakeRow methods addad. ScanRow is more efficient than GetRow because it doesn't allocate memory for every row received from the server. godrv Value.Next method now uses the new ScanRow method.
v0.4.6: StatusOnly method added to mysql.Result.
v0.4.5: New autorc.Conn.PrepareOnce method.
v0.4.4:
- Row.Int, Row.Uint, Row.Int64, ... methods now panic in case of error.
- New Row.Float method.
v0.4.3:
- Fixed issue with panic when the server returns MYSQL_TYPE_NEWDECIMAL.
- Decimals are returned as float64 (previously they were returned as []byte).
v0.4.2:
- A lot of changes with MySQL time handling:
- Datetime type replaced by time.Time.
- Time type replaced by time.Duration.
- Support for time.Time type added to godrv.
-
row.Int64/row.Uint64 methods added.
-
Rename BindParams to Bind.
v0.4.1:
BindParams supports Go bool type.
v0.4:
- Modular design:
- MySQL wire protocol handling moved to mymysql/native
- Thread safe wrapper of native engine in separate mymysql/thrsafe
- mymysql/mysql package contains definitions of interfaces to engines and common (engine-independent) functions.
- Automatic reconnect interface moved to mymysql/autorc.
-
mysql.New and other functions returns mostly interface types. So all previously exported members were converted to methods (with except mysql.Row and mysql.Field - their definition didn't changed).
-
Transactions added. If you use *mymysql/thrsafe" engine transactions are fully thread safe.
-
Driver for exp/sql.
To install all subpackages of mymysql you need to goinstal three of them:
$ go get github.com/ziutek/mymysql/thrsafe
$ go get github.com/ziutek/mymysql/autorc
$ go get github.com/ziutek/mymysql/godrv
go get automagically selects the proper version of mymysql for your Go release. After this command mymysql is ready to use.
For testing you will need to create the test database and a test user:
mysql> create database test;
mysql> grant all privileges on test.* to testuser@localhost;
mysql> set password for testuser@localhost = password("TestPasswd9");
Make sure that MySQL max_allowed_packet variable in my.cnf is equal or greater than 34M (In order to test long packets).
The default MySQL server address is 127.0.0.1:3306.
Next run tests:
$ cd $GOPATH/src/github.com/ziutek/mymysql
$ ./all.bash test
package main
import (
"os"
"github.com/ziutek/mymysql/mysql"
_ "github.com/ziutek/mymysql/native" // Native engine
// _ "github.com/ziutek/mymysql/thrsafe" // Thread safe engine
)
func main() {
db := mysql.New("tcp", "", "127.0.0.1:3306", user, pass, dbname)
err := db.Connect()
if err != nil {
panic(err)
}
rows, res, err := db.Query("select * from X where id > %d", 20)
if err != nil {
panic(err)
}
for _, row := range rows {
for _, col := range row {
if col == nil {
// col has NULL value
} else {
// Do something with text in col (type []byte)
}
}
// You can get specific value from a row
val1 := row[1].([]byte)
// You can use it directly if conversion isn't needed
os.Stdout.Write(val1)
// You can get converted value
number := row.Int(0) // Zero value
str := row.Str(1) // First value
bignum := row.MustUint(2) // Second value
// You may get values by column name
first := res.Map("FirstColumn")
second := res.Map("SecondColumn")
val1, val2 := row.Int(first), row.Str(second)
}
}
If you do not want to load the entire result into memory you may use Start and GetRow methods:
res, err := db.Start("select * from X")
checkError(err)
// Print fields names
for _, field := range res.Fields() {
fmt.Print(field.Name, " ")
}
fmt.Println()
// Print all rows
for {
row, err := res.GetRow()
checkError(err)
if row == nil {
// No more rows
break
}
// Print all cols
for _, col := range row {
if col == nil {
fmt.Print("<NULL>")
} else {
os.Stdout.Write(col.([]byte))
}
fmt.Print(" ")
}
fmt.Println()
}
GetRow method allocates a new chunk of memory for every received row. If your query returns hundreds of rows you should opt for the ScanRow method to avoid unnecessary allocations:
// Print all rows
row := res.MakeRow()
for {
err := res.ScanRow(row)
if err == io.EOF {
// No more rows
break
}
checkError(err)
// Print all cols
// [...]
}
You can use Run or Exec method for prepared statements:
stmt, err := db.Prepare("insert into X values (?, ?)")
checkError(err)
type Data struct {
Id int
Tax *float32 // nil means NULL
}
data = new(Data)
for {
err := getData(data)
if err == endOfData {
break
}
checkError(err)
_, err = stmt.Run(data.Id, data.Tax)
checkError(err)
}
getData is the function which retrieves data from somewhere and set Id and Tax fields of the Data struct. In the case of Tax field getData may assign a pointer the retrieved variable or nil if NULL should be stored in database.
If you pass parameters to Run or Exec method, the data is rebound on every method call. This isn't efficient if the statement will be executed more than once. You can bind parameters and use Run or Exec method without parameters, to avoid these unnecessary rebinds. Warning! If you use Bind in multithreaded applications, you should ensure that no other thread will use Bind for the same statement, until you no longer need bound parameters.
The simplest way to bind parameters is:
stmt.Bind(data.Id, data.Tax)
but you can't use it in our example, because parameters bound this way can't be changed by getData function. You may modify bindings like this:
stmt.Bind(&data.Id, &data.Tax)
and now it should work properly. But in our example there is better solution:
stmt.Bind(data)
If Bind method has one parameter, and this parameter is a struct or a pointer to the struct, it treats all fields of this struct as parameters and binds them.
This is the improved code of the previous example:
data = new(Data)
stmt.Bind(data)
for {
err := getData(data)
if isEndOfData(error) {
break
}
checkError(err)
_, err = stmt.Run()
checkError(err)
}
_, err = db.Start("CREATE TABLE web (url VARCHAR(80), content LONGBLOB)")
checkError(err)
ins, err := db.Prepare("INSERT INTO web VALUES (?, ?)")
checkError(err)
var url string
ins.Bind(&url, []byte(nil)) // []byte(nil) for properly type binding
for {
// Read URL from stdin
url = ""
fmt.Scanln(&url)
if len(url) == 0 {
// Stop reading if URL is blank line
break
}
// Make a connection
resp, err := http.Get(url)
checkError(err)
// We can retrieve response directly into database because
// the resp.Body implements io.Reader. Use 8 kB buffer.
err = ins.SendLongData(1, resp.Body, 8192)
checkError(err)
// Execute insert statement
_, err = ins.Run()
checkError(err)
}
res, err := db.Start("select id from M; select name from M")
checkError(err)
// Get result from first select
for {
row, err := res.GetRow()
checkError(err)
if row == nil {
// End of first result
break
}
// Do something with with the data
functionThatUseId(row.Int(0))
}
// Get result from second select
res, err = res.NextResult()
checkError(err)
if res == nil {
panic("Hmm, there is no result. Why?!")
}
for {
row, err := res.GetRow()
checkError(err)
if row == nil {
// End of second result
break
}
// Do something with with the data
functionThatUseName(row.Str(0))
}
import (
"github.com/ziutek/mymysql/mysql"
_ "github.com/ziutek/mymysql/thrsafe" // for thread safe transactions
)
// [...]
// Statement prepared before transaction begins
ins, err := db.Prepare("insert A values (?, ?)")
checkError(err)
// Begin a new transaction
tr, err := db.Begin()
checkError(err)
// Now db is locked, so any method that uses db and sends commands to
// MySQL server will be blocked until Commit or Rollback is called.
// Commands in transaction are thread safe to
go func() {
_, err = tr.Start("insert A values (1, 'jeden')")
checkError(err)
} ()
_, err = tr.Start("insert A values (2, 'dwa')")
checkError(err)
// You can't use statements prepared before transaction in the usual way,
// because the connection is locked by the Begin method. You must bind the statement
// to the transaction before using it.
_, err = tr.Do(ins).Run(3, "three")
checkError(err)
// For a greater number of calls
ti := tr.Do(ins)
_, err = ti.Run(4, "four")
checkError(err)
_, err = ti.Run(5, "five")
checkError(err)
// At the end you can Commit or Rollback. tr is invalidated and using it
// after Commit/Rollback will cause a panic.
tr.Commit()
import (
"github.com/ziutek/mymysql/autorc"
_ "github.com/ziutek/mymysql/thrsafe" // You may also use the native engine
)
// [...]
db := autorc.New("tcp", "", "127.0.0.1:3306", user, pass, dbname)
// Initilisation commands. They will be executed after each connect.
db.Register("set names utf8")
// There is no need to explicity connect to the MySQL server
rows, res, err := db.Query("SELECT * FROM R")
checkError(err)
// Now we are connected.
// It does not matter if connection will be interrupted during sleep, eg
// due to server reboot or network down.
time.Sleep(9e9)
// If we can reconnect in no more than db.MaxRetries attempts this
// statement will be prepared.
sel, err := db.Prepare("SELECT name FROM R where id > ?")
checkError(err)
// We can destroy our connection server side
_, _, err = db.Query("kill %d", db.Raw.ThreadId())
checkError(err)
// But it doesn't matter
sel.Bind(2)
rows, res, err = sel.Exec()
checkError(err)
import (
"database/sql"
_"github.com/ziutek/mymysql/godrv"
)
// [...]
// Open new connection. The uri need to have the following syntax:
//
// [PROTOCOL_SPECFIIC*]DBNAME/USER/PASSWD
//
// where protocol specific part may be empty (this means connection to
// local server using default protocol). Currently possible forms:
// DBNAME/USER/PASSWD
// unix:SOCKPATH*DBNAME/USER/PASSWD
// unix:SOCKPATH,OPTIONS*DBNAME/USER/PASSWD
// tcp:ADDR*DBNAME/USER/PASSWD
// tcp:ADDR,OPTIONS*DBNAME/USER/PASSWD
//
// OPTIONS can contain comma separated list of options in form:
// opt1=VAL1,opt2=VAL2,boolopt3,boolopt4
// Currently implemented options:
// laddr - local address/port (eg. 1.2.3.4:0)
// timeout - connect timeout in format accepted by time.ParseDuration
// Register initialisation commands
// (workaround, see http://codereview.appspot.com/5706047)
godrv.Register("SET NAMES latin2") // Overrides default utf8
godrv.Register("CREATE TABLE IF NOT EXISTS my_table ( ... )")
// Create a connection handler
db, err := sql.Open("mymysql", "test/testuser/TestPasswd9")
checkErr(err)
// For other information about database/sql see its documentation.
ins, err := db.Prepare("INSERT my_table SET txt=?")
checkErr(err)
res, err := ins.Exec("some text")
checkErr(err)
id, err := res.LastInsertId()
checkErr(err)
checkErr(ins.Close(ins))
rows, err := db.Query("SELECT * FROM go")
checkErr(err)
for rows.Next() {
var id int
var txt string
checkErr(rows.Scan(&id, &txt))
// Do something with id and txt
}
checkErr(db.Close())
import (
"github.com/ziutek/mymysql/mysql"
_ "github.com/ziutek/mymysql/thrsafe" // or native
)
// [...]
res, err := my.Start("CALL MyProcedure(1, 2, 3)")
checkErr(err)
// Procedure can return more than one result set so we have to read all
// results up to the result that doesn't include result set (status only
// result).
for !res.StatusOnly() {
rows, err := res.GetRows()
checkErr(err)
useRows(rows)
res, err := res.NextResult()
checkErr(err)
if res == nil {
panic("nil result from procedure")
}
}
import (
"github.com/ziutek/mymysql/autorc"
_ "github.com/ziutek/mymysql/thrsafe" // You may also use the native engine
)
// [...]
db := autorc.New("tcp", "", "127.0.0.1:3306", user, pass, dbname)
var stmt1, stmt2 autorc.Stmt
func updateDb() {
err := db.PrepareOnce(&stmt1, someSQL1)
checkDbErr(err)
err = db.PrepareOnce(&stmt2, someSQL2)
checkDbErr(err)
err = db.Begin(func(tr mysql.Transaction, args ...interface{}) error {
// This function will be called again if returns a recoverable error
s1 := tr.Do(stmt1.Raw)
s2 := tr.Do(stmt2.Raw)
if _, err := s1.Run(); err != nil {
return err
}
if _, err := s2.Run(); err != nil {
return err
}
// You have to commit or rollback before return
return tr.Commit()
})
checkDbErr(err)
}
Additional examples are in examples directory.
In the case of classic text queries, all variables that are sent to the MySQL server are embedded in the text query. Thus you always convert them to a string and send them embedded in an SQL query:
rows, res, err := db.Query("select * from X where id > %d", id)
After text query you always receive a text result. Mysql text result corresponds to []byte type in mymysql. It isn't string type due to avoidance of unnecessary type conversions. You can always convert []byte to string yourself:
fmt.Print(string(rows[0][1].([]byte)))
or using Str helper method:
fmt.Print(rows[0].Str(1))
There are other helper methods for data conversion like Int or Uint:
fmt.Print(rows[0].Int(1))
All three above examples return value received in row 0 column 1. If you prefer to use the column names, you can use res.Map which maps result field names to corresponding indexes:
name := res.Map("name")
fmt.Print(rows[0].Str(name))
In case of prepared statements, the type mapping is slightly more complicated. For parameters sent from the client to the server, Go/mymysql types are mapped for MySQL protocol types as below:
string --> MYSQL_TYPE_STRING
[]byte --> MYSQL_TYPE_VAR_STRING
int8, uint8 --> MYSQL_TYPE_TINY
int16, uint16 --> MYSQL_TYPE_SHORT
int32, uint32 --> MYSQL_TYPE_LONG
int64, uint64 --> MYSQL_TYPE_LONGLONG
int, uint --> protocol integer type which match size of int
bool --> MYSQL_TYPE_TINY
float32 --> MYSQL_TYPE_FLOAT
float64 --> MYSQL_TYPE_DOUBLE
time.Time --> MYSQL_TYPE_DATETIME
mysql.Timestamp --> MYSQL_TYPE_TIMESTAMP
mysql.Date --> MYSQL_TYPE_DATE
time.Duration --> MYSQL_TYPE_TIME
mysql.Blob --> MYSQL_TYPE_BLOB
nil --> MYSQL_TYPE_NULL
The MySQL server maps/converts them to a particular MySQL storage type.
For received results MySQL storage types are mapped to Go/mymysql types as below:
TINYINT --> int8
UNSIGNED TINYINT --> uint8
SMALLINT --> int16
UNSIGNED SMALLINT --> uint16
MEDIUMINT, INT --> int32
UNSIGNED MEDIUMINT, UNSIGNED INT --> uint32
BIGINT --> int64
UNSIGNED BIGINT --> uint64
FLOAT --> float32
DOUBLE --> float64
DECIMAL --> float64
TIMESTAMP, DATETIME --> time.Time
DATE --> mysql.Date
TIME --> time.Duration
YEAR --> int16
CHAR, VARCHAR, BINARY, VARBINARY --> []byte
TEXT, TINYTEXT, MEDIUMTEXT, LONGTEX --> []byte
BLOB, TINYBLOB, MEDIUMBLOB, LONGBLOB --> []byte
BIT --> []byte
SET, ENUM --> []byte
NULL --> nil
This package can send and receive MySQL data packets that are biger than 16 MB. This means that you can receive response rows biger than 16 MB and can execute prepared statements with parameter data bigger than 16 MB without using SendLongData method. If you want to use this feature you need to change the default mymysql setting using the Conn.SetMaxPktSize method and change max_allowed_packet value in your MySQL server configuration.
If you import "mymysql/thrsafe" engine instead of "mymysql/native" engine all methods are thread safe, unless the description of the method says something else.
If one thread is calling Query or Exec method, other threads will be blocked if they call Query, Start, Exec, Run or other method which send data to the server, until Query/Exec return in first thread.
If one thread is calling Start or Run method, other threads will be blocked if they call Query, Start, Exec, Run or other method which send data to the server, until all results and all rows will be readed from the connection in first thread.
In most of my web applications I use the autorecon interface with thrsafe engine. For any new connection, one gorutine is created. There is one persistant connection to MySQL server shared by all gorutines. Applications are usually running on dual-core machines with GOMAXPROCS=2. I use siege to test any application befor put it into production. There is example output from siege:
# siege my.httpserver.pl -c25 -d0 -t 30s
** SIEGE 2.69
** Preparing 25 concurrent users for battle.
The server is now under siege...
Lifting the server siege.. done.
Transactions: 3212 hits
Availability: 100.00 %
Elapsed time: 29.83 secs
Data transferred: 3.88 MB
Response time: 0.22 secs
Transaction rate: 107.68 trans/sec
Throughput: 0.13 MB/sec
Concurrency: 23.43
Successful transactions: 3218
Failed transactions: 0
Longest transaction: 9.28
Shortest transaction: 0.01
- Complete documentation
- There is MySQL "bug" in the SUM function. If you use prepared statements SUM returns DECIMAL value, even if you sum integer column. mymysql returns decimals as float64 so cast result from sum to integer (or use Row.Int) causes panic.