Golang Apache Impala Driver

The actively supported Apache Impala driver for Go's database/sql package

This driver started as a fork of github.com/bippio/go-impala, which hasn't been updated in over four years and appears to be abandoned. Several issues have been fixed since then — some quite severe. The original codebase also didn't support Go modules.

Install

Add impala-go to your Go module:

go get github.com/sclgo/impala-go

Alternatively, see below how to use it as a CLI. impala-go does not use CGO.

Connection Parameters and DSN

The data source name (DSN; connection string) uses a URL format: impala://username:password@host:port?param1=value&param2=value

Driver name is impala.

Parameters:

• auth - string. Authentication mode. Supported values: noauth, ldap.
• tls - boolean. Enable TLS
• ca-cert - The file that contains the public key certificate of the CA that signed the Impala certificate
• batch-size - integer value (default: 1024). Maximum number of rows fetched per request.
• buffer-size- in bytes (default: 4096). Buffer size for the Thrift transport.
• mem-limit - string value (example: 3m). Memory limit for query, as a share of available RAM or a fixed value. See https://impala.apache.org/docs/build/html/topics/impala_mem_limit.html for details.
• query-timeout - integer value in seconds. Query timeout - see https://impala.apache.org/docs/build/html/topics/impala_query_timeout_s.html for details.
• socket-timeout - integer or string value (default: 5s). The maximum socket idle time, expressed as a time duration in this syntax. If the value is an integer without a time unit, milliseconds are assumed.
• connect-timeout - integer or string value (default: 10s). The max wait for initial connection to server, expressed as a time duration in this syntax. If the value is an integer without a time unit, milliseconds are assumed.
• tls-insecure-skip-verify - boolean. Disables TLS certificate verification by enabling the tls.Config.InsecureSkipVerify option. Behaves the same way as AllowSelfSignedCerts in the official JDBC driver.
• reuse-session - boolean. Disables resetting the session when database/sql requests it. When this setting is enabled, this driver behaves consistently with the other DB drivers in the ecosystem but diverges somewhat from documented database/sql behavior. This setting is disabled by default for backward compatibility and alignment with published Go documentation. It must be enabled when this driver is used in github.com/xo/usql. usql returns the connection to the pool after each statement, relying on the typical driver behavior.

A string of this format can be constructed using the URL type in the net/url package.

  query := url.Values{}
  query.Add("auth", "ldap")

  u := &url.URL{
      Scheme:   "impala",
      User:     url.UserPassword(username, password),
      Host:     net.JoinHostPort(hostname, port),
      RawQuery: query.Encode(),
  }
  db, err := sql.Open("impala", u.String())

Also, you can bypass the string-based data source name by using sql.OpenDB:

  opts := impala.DefaultOptions
  opts.Host = hostname
  opts.UseLDAP = true
  opts.Username = username
  opts.Password = password

  connector := impala.NewConnector(&opts)
  db, err := sql.OpenDB(connector)

Impala supports numerous other session options which can be configured with the SET statement. mem-limit and query-timeout are the only two such options the driver supports as part of the DSN. Those DSN fields for those are an exception for backwards compatibility. The preferred way to set any session option is issuing SET statements to a SQL connection. Users may find it useful to wrap the driver.Connector returned by impala.NewConnector so that a set of session options are automatically applied to all created connections.

CLI

impala-go is included in xo/usql - the universal SQL CLI, inspired by psql.

Install usql, start it, then on its prompt, run:

\connect impala DSN

where DSN is a data source name in the format above. Review the usql documentation for other options.

The latest version of usql typically comes with the latest version of impala-go but if you need to use a different one, you can prepare a custom build using usqlgen. For example, the following command builds a usql binary in the working directory using impala-go from master branch:

go run github.com/sclgo/usqlgen@latest build --get github.com/sclgo/impala-go@master -- -tags impala

usql with impala-go is arguably a better CLI for Impala than the official impala-shell. For one, usql is much easier to install.

Example Go code

package main

// Simple program to list databases and the tables

import (
	"context"
	"database/sql"
	"log"

	"github.com/sclgo/impala-go"
)

func main() {
	opts := impala.DefaultOptions

	opts.Host = "localhost" // impala host
	opts.Port = "21050"

	// enable LDAP authentication:
	//opts.UseLDAP = true
	//opts.Username = "<ldap username>"
	//opts.Password = "<ldap password>"
	//
	// enable TLS
	//opts.UseTLS = true
	//opts.CACertPath = "/path/to/cacert"

	connector := impala.NewConnector(&opts)
	db := sql.OpenDB(connector)
	defer func() {
		_ = db.Close()
	}()

	ctx := context.Background()

	rows, err := db.QueryContext(ctx, "SHOW DATABASES")
	if err != nil {
		log.Fatal(err)
	}

	var name, comment string
	databases := make([]string, 0) // databases will contain all the DBs to enumerate later
	for rows.Next() {
		if err := rows.Scan(&name, &comment); err != nil {
			log.Fatal(err)
		}
		databases = append(databases, name)
	}
	if err := rows.Err(); err != nil {
		log.Fatal(err)
	}
	log.Println("List of Databases", databases)

	tables, err := impala.NewMetadata(db).GetTables(ctx, "%", "%")
	if err != nil {
		log.Fatal(err)
	}
	log.Println("List of Tables", tables)
}

Check out also an open data end-to-end demo.

Data types

Impala data types are mapped to Go types as expected, with the following exceptions:

• "Complex" types - MAP, STRUCT, ARRAY - are not supported. Impala itself has limited support for those. As a workaround, select individual fields or flatten such values within select statements.
• Decimals are converted to strings by the Impala server API. Either parse the decimal value after Rows.Scan, or use a custom sql.Scanner implementation in Row(s).Scan e.g. Decimal from github.com/cockroachdb/apd. Note that the processing of sql.Scanner within Row(s).Scan is a feature of the database/sql package, and not the driver. The ScanType of such values is string, while the DatabaseTypeName is DECIMAL. Retrieving precision and scale using the DecimalSize API is supported.

Context support

The driver methods recognize Context and support early cancellation in most cases. Additionally, the Query methods return early before all rows are retrieved. Exec methods return after the operation completes (this may be configurable in the future). Exec methods can still be stopped early by cancelling the context from another goroutine.

It is also supported to use a QueryContext method on a sql.Conn for a DDL/DML statement if you need the method to return before the statement completes. In that case, calling Rows.Next will wait for the statement to complete and then return false.

Compatibility and Support

The library is actively tested with Impala 4.4 and 3.4. All 3.x and 4.x minor versions should work well. 2.x is also supported on a best-effort basis.

While Impala shares the majority of its API with Apache Hive, this driver doesn't support Hive. Instead, it is recommended to use a dedicated Hive driver or client. Please file an issue if you find it more valuable to use this driver with Hive compared to the existing drivers.

The library is not compatible with TinyGo because Thrift for Go doesn't support it. The Thrift code incompatible with TinyGo is not referenced by impala-go but compilation fails nonetheless. Last checked with tinygo 0.41.1, thrift 0.23, and Go 1.26 on 2026-05-10. (Dev note: Any new release of Thrift or TinyGo may resolve the issue. Run make test-tinygo after updates to check again.) Thrift team tracks progress on TinyGo support at https://issues.apache.org/jira/browse/THRIFT-5209.

File any issues that you encounter as GitHub issues.

Versioning

The library follows semantic versioning, as specified in SemVer 2.0.0, and will use semantic import versioning (SIV), if a 2.0 version is ever needed. However, the rare exceptions to the semantic versioning rules that are allowed for the Go language and standard library, apply to this library as well. For example, a minor release can include a breaking change if the change was required to fix a security issue. Review the rest of the exceptions at https://go.dev/doc/go1compat#expectations.

gorelease tool is included in CI to automate the detection of most semantic versioning violations.

The minimum Go version may increase in minor, not patch, releases following general practice. The last two Go minor releases will always be supported.

Copyright and acknowledgements

This library started as a fork of github.com/bippio/go-impala, under the MIT license. This library retains the same license.

The project logo combines the Golang Gopher from github.com/golang-samples/gopher-vector with the Apache Impala logo, licensed under the Apache 2 license.