GoHive is a driver for Hive and the Spark Distributed SQL Engine in go that supports connection mechanisms KERBEROS(Gssapi Sasl), NONE(Plain Sasl), LDAP, CUSTOM and NOSASL, both for binary and HTTP transport, with and without SSL. The kerberos mechanism will pick a different authentication level depending on hive.server2.thrift.sasl.qop
.
GoHive also offers support to query the Hive metastore with various connection mechanisms, including KERBEROS.
GoHive can be installed with:
go get github.com/beltran/gohive
To add kerberos support GoHive requires header files to build against the GSSAPI C library. They can be installed with:
- Ubuntu:
sudo apt-get install libkrb5-dev
- MacOS:
brew install homebrew/dupes/heimdal --without-x11
- Debian:
yum install -y krb5-devel
Then:
go get -tags kerberos github.com/beltran/gohive
connection, errConn := gohive.Connect("hs2.example.com", 10000, "KERBEROS", configuration)
if errConn != nil {
log.Fatal(errConn)
}
cursor := connection.Cursor()
cursor.Exec(ctx, "INSERT INTO myTable VALUES(1, '1'), (2, '2'), (3, '3'), (4, '4')")
if cursor.Err != nil {
log.Fatal(cursor.Err)
}
cursor.Exec(ctx, "SELECT * FROM myTable")
if cursor.Err != nil {
log.Fatal(cursor.Err)
}
var i int32
var s string
for cursor.HasMore(ctx) {
cursor.FetchOne(ctx, &i, &s)
if cursor.Err != nil {
log.Fatal(cursor.Err)
}
log.Println(i, s)
}
cursor.Close()
connection.Close()
cursor.HasMore
may query Hive for more rows if not all of them have been received. Once the row is
read is discarded from memory so as long as the fetch size is not too big there's no limit to how much
data can be queried.
The thrift client is directly exposed, so the API exposed by the Hive metastore can be called directly.
configuration := gohive.NewMetastoreConnectConfiguration()
connection, err := gohive.ConnectToMetastore("hm.example.com", 9083, "KERBEROS", configuration)
if err != nil {
log.Fatal(err)
}
database := hive_metastore.Database{
Name: "my_new_database",
LocationUri: "/"}
err = connection.Client.CreateDatabase(context.Background(), &database)
if err != nil {
log.Fatal(err)
}
databases, err := connection.Client.GetAllDatabases(context.Background())
if err != nil {
log.Fatal(err)
}
log.Println("databases ", databases)
connection.Close()
The rest of these docs are pertinent to connect to Hive.
configuration := NewConnectConfiguration()
configuration.Service = "hive"
// Previously kinit should have done: kinit -kt ./secret.keytab hive/[email protected]
connection, errConn := Connect("hs2.example.com", 10000, "KERBEROS", configuration)
This implies setting in hive-site.xml:
hive.server2.authentication = KERBEROS
hive.server2.authentication.kerberos.principal = hive/[email protected]
hive.server2.authentication.kerberos.keytab = path/to/keytab.keytab
configuration := NewConnectConfiguration()
// If it's not set it will be picked up from the logged user
configuration.Username = "myUsername"
// This may not be necessary
configuration.Password = "myPassword"
connection, errConn := Connect("hs2.example.com", 10000, "NONE", configuration)
This implies setting in hive-site.xml:
hive.server2.authentication = NONE
connection, errConn := Connect("hs2.example.com", 10000, "NOSASL", NewConnectConfiguration())
This implies setting in hive-site.xml:
hive.server2.authentication = NOSASL
Binary transport mode is supported for auth mechanisms PLAIN, KERBEROS and NOSASL. Http transport mode is supported for PLAIN and KERBEROS:
configuration := NewConnectConfiguration()
configuration.HttpPath = "cliservice" // this is the default path in Hive configuration.
configuration.TransportMode = "http"
configuration.Service = "hive"
connection, errConn := Connect("hs2.example.com", 10000, "KERBEROS", configuration)
This implies setting in hive-site.xml:
hive.server2.authentication = KERBEROS
, orNONE
hive.server2.transport.mode = http
hive.server2.thrift.http.port = 10001
A connection can be made using zookeeper:
connection, errConn := ConnectZookeeper("zk1.example.com:2181,zk2.example.com:2181", "NONE", configuration)
The last two parameters determine how the connection to Hive will be made once the Hive hosts are retrieved from zookeeper.
For example if a NULL
value is in a row, the following operations would put 0
into i
:
var i int32
cursor.FetchOne(context.Background(), &i)
To differentiate between these two values (NULL
and 0
) the following will set i
to nil
or *i
to 0
:
var i *int32 = new(int32)
cursor.FetchOne(context.Background(), &i)
which will produce the same result as:
var i *int32
cursor.FetchOne(context.Background(), &i)
Alternatively, using the rowmap API, m := cursor.RowMap(context.Background())
,
m
would be map[string]interface{}{"table_name.column_name": nil}
for a NULL
value. It will return a map
where the keys are table_name.column_name
. This works fine with Hive but using Spark Thirft SQL server table_name
is not present and the keys are column_name
and it can lead to problems if two tables have the same column name so the FetchOne
API should be used in this case.
Tests can be run with:
./scripts/integration
This uses dhive and it will start three docker instances with Hive, the Hive metastore, and Kerberos. kinit
, klist
, kdestroy
have to be installed locally. hs2.example.com
and hm.example.com
will have to be an alias for 127.0.0.1 in /etc/hosts
. The krb5 configuration file should be created with bash scripts/create_krbconf.sh
. Overall the steps used in the travis CI can be followed.