trino-go-client

README

Trino Go client

A Trino client for the Go programming language. It is a driver for the standard database/sql package: your program opens a *sql.DB and sends SQL statements to Trino over HTTP or HTTPS, and the driver turns the results into Go values.

The driver authenticates with HTTP Basic credentials, Kerberos, JSON web tokens or OAuth2, reads results through the spooling protocol when the server offers it, runs transactions, and converts every Trino type, including nested ARRAY, MAP, ROW and VARIANT values, into Go values.

Requirements

• Go 1.27 or newer
• Trino 372 or newer

Installation

go get github.com/trinodb/trino-go-client/trino

Quick start

Import the package for its side effect of registering the trino driver, then open a database with a DSN:

import (
	"database/sql"

	_ "github.com/trinodb/trino-go-client/trino"
)

db, err := sql.Open("trino", "http://user@localhost:8080?catalog=tpch&schema=sf1")
if err != nil {
	return err
}
defer db.Close()

rows, err := db.QueryContext(ctx, "SELECT name, regionkey FROM nation WHERE nationkey < ?", 5)
if err != nil {
	return err
}
defer rows.Close()
for rows.Next() {
	var name string
	var regionKey int64
	if err := rows.Scan(&name, &regionKey); err != nil {
		return err
	}
}
return rows.Err()

sql.Open returns an error for a DSN that does not parse, and does not contact the server. Settings that a DSN cannot carry, such as an http.Client, go in a trino.Config passed to trino.NewConnector and sql.OpenDB.

Every request the driver sends carries a User-Agent such as trino-go-client/v0.333.0 os=linux arch=amd64 lang/go=go1.27.1, with the driver version taken from the module build information (unknown when it is not available).

Connecting

DSN

The Data Source Name is a URL with a mandatory username, an optional password, an optional catalog and schema path, and optional query string parameters:

http[s]://user[:pass]@host[:port][/catalog[/schema]][?parameters]

The path sets the catalog, or the catalog and schema, with the same syntax as a JDBC URL: http://user@localhost:8080/tpch/sf1 is the same as http://user@localhost:8080?catalog=tpch&schema=sf1. An empty path or a lone / sets neither, and one trailing slash is ignored. A path with more than two segments, or with an empty catalog or schema such as //sf1, is an error. Set each of the catalog and schema either in the path or with the catalog and schema parameters, not both: /tpch?schema=sf1 is valid, but /tpch?catalog=tpch is an error. The path is not a prefix for a server behind a reverse proxy; requests always go to /v1/statement on the host.

A password enables HTTP Basic authentication and requires HTTPS. The parameters are case-sensitive. Values with characters that have a meaning in a URL, such as /, : or ;, must be URL-encoded, which Config.FormatDSN does for you:

dsn, err := (&trino.Config{
	ServerURI:         "https://user@localhost:8443",
	Catalog:           "tpch",
	SessionProperties: map[string]string{"query_max_run_time": "10m"},
}).FormatDSN()

Examples:

http://user@localhost:8080?source=hello&catalog=default&schema=foobar
https://user:secret@localhost:8443?session_properties=query_max_run_time:10m;query_priority:2
http://user@localhost:8080/tpch/sf1?roles=catalog1:role1;catalog2:role2

Config and Connector

Every DSN parameter has a field on trino.Config, and a few settings exist only there, because a string cannot carry them: an http.Client, a logged-in Kerberos client, and the callbacks of external authentication. Pass the Config to trino.NewConnector and open the database with sql.OpenDB:

connector, err := trino.NewConnector(&trino.Config{
	ServerURI:  "https://user@localhost:8443",
	Catalog:    "default",
	HTTPClient: &http.Client{Timeout: time.Minute},
})
if err != nil {
	return err
}
db := sql.OpenDB(connector)

NewConnector validates the Config and copies it, so later changes to it have no effect. Config.ServerURI must not have a path; set Config.Catalog and Config.Schema instead.

HTTPClient is used for every request. Its transport decides how to reach and verify the coordinator, so it cannot be combined with custom_client, the certificate parameters, SSLVerification, httpProxy or socksProxy; Config.TLSConfig builds the tls.Config those settings would produce, for the client's own transport. As with the default client, the driver does not follow redirects with HTTPClient, except for spooled segment downloads. Config.FormatDSN returns an error when a field that a DSN cannot carry is set.

Parameters

Parameter	Config field	Default	Purpose
source	Source	empty	Name of the application, for query tracking
catalog	Catalog	empty	Default catalog
schema	Schema	empty	Default schema
session_properties	SessionProperties	empty	Session properties
extra_credentials	ExtraCredentials	empty	Credentials for connectors
roles	Roles	empty	Catalog roles
clientTags	ClientTags	empty	Tags for resource groups
resourceEstimates	ResourceEstimates	empty	Expected resource usage, for resource groups
trace_token	TraceToken	empty	Correlation token for the server logs
client_info	ClientInfo	empty	Free-form description of the client
language	Language	server default	Locale for locale-sensitive functions
timezone	TimeZone	local zone	Session time zone
query_timeout	QueryTimeout	10h	Timeout for queries without a context deadline
request_retry_timeout	RequestRetryTimeout	2m	How long one HTTP request is retried
request_retry_max_attempts	RequestRetryMaxAttempts	20	How many times one HTTP request is sent
heartbeat_interval	HeartbeatInterval	30s	Spooling heartbeat interval
explicitPrepare	DisableExplicitPrepare	true	Prepared statements through headers or EXECUTE IMMEDIATE
accessToken	AccessToken	empty	JWT sent as a bearer token
forwardAuthorizationHeader	ForwardAuthorizationHeader	false	Allow a per-query accessToken argument
externalAuthentication	ExternalAuthentication	false	Log in through the browser
externalAuthenticationTimeout	ExternalAuthenticationTimeout	2m	How long to wait for the login
oauth2ClientId, oauth2ClientSecret	OAuth2ClientID, OAuth2ClientSecret	empty	OAuth2 client credentials
oauth2TokenEndpoint	OAuth2TokenEndpoint	advertised by the server	The only token endpoint the client secret is sent to
oauth2Scope	OAuth2Scope	advertised by the server	Scope requested with the client credentials
KerberosEnabled	KerberosEnabled	false	Kerberos authentication
KerberosKeytabPath	KerberosKeytabPath	empty	Keytab to log in with
KerberosCredentialCachePath	KerberosCredentialCachePath	KRB5CCNAME	Credential cache to reuse a ticket from
KerberosPrincipal	KerberosPrincipal	empty	Principal to authenticate as
KerberosRealm	KerberosRealm	empty	Realm of the principal
KerberosConfigPath	KerberosConfigPath	empty	krb5 configuration file
KerberosRemoteServiceName	KerberosRemoteServiceName	trino	Service name of the coordinator
KerberosServicePrincipalPattern	KerberosServicePrincipalPattern	${SERVICE}@${HOST}	Service principal of the coordinator
KerberosUseCanonicalHostname	KerberosDisableCanonicalHostname	true	Resolve the coordinator host before building the principal
SSLCertPath, SSLCert	SSLCertPath, SSLCert	system trust store	Additional CA certificate
SSLClientCertPath, SSLClientCert, SSLClientKeyPath, SSLClientKey	same names	empty	Client certificate for mutual TLS
SSLVerification	SSLVerification	FULL	How the server certificate is verified
httpProxy, socksProxy	HTTPProxy, SOCKSProxy	environment	Proxy for every request
custom_client	CustomClientName	empty	A registered http.Client

source

Type:           string
Valid values:   string describing the source of the connection to Trino
Default:        empty

The source parameter is sent as the X-Trino-Source header and names the application in the Trino web UI, the query history and event listeners, so administrators can trace a query back to the program that ran it.

catalog

Type:           string
Valid values:   the name of a catalog configured in the Trino server
Default:        empty

The catalog parameter defines the Trino catalog where schemas exist to organize tables. It can also be set as the first segment of the DSN path.

schema

Type:           string
Valid values:   the name of an existing schema in the catalog
Default:        empty

The schema parameter defines the Trino schema where tables exist. This is also known as namespace in some environments. It can also be set as the second segment of the DSN path, after the catalog.

session_properties

Type:           string
Valid values:   semicolon-separated list of key:value session properties
Default:        empty

The session_properties parameter must contain valid parameters accepted by the Trino server. Run SHOW SESSION in Trino to get the current list. Property names, like extra_credentials and roles keys, must be non-empty printable ASCII without spaces, = or ,; values are URL-encoded, so they may contain = and ,. SET SESSION changes a property for the rest of the session, see Session state and the connection pool.

extra_credentials

Type:           string
Valid values:   semicolon-separated list of key:value pairs
Default:        empty

The extra_credentials parameter passes credentials to connectors that authenticate to the underlying system with the user's identity, as described for each connector in the Trino documentation. Each pair is sent as an X-Trino-Extra-Credential header with the statement request only, not with the requests that fetch result pages. Keys follow the same rules as session property names.

roles

Type:           string
Valid values:   semicolon-separated list of catalog:role pairs
Default:        empty

The roles parameter selects a role to assume in each listed catalog for the whole session. A role selected with SET ROLE lasts until database/sql resets the connection for its next caller, and no roles are sent while a SET SESSION AUTHORIZATION switch is active. Roles can also be set for one query with the X-Trino-Role named argument.

c := &trino.Config{
	ServerURI: "https://foobar@localhost:8443",
	Roles:     map[string]string{"catalog1": "role1", "catalog2": "role2"},
}
dsn, err := c.FormatDSN()

clientTags

Type:           string
Valid values:   comma-separated list of tags (e.g. tag1,tag2)
Default:        empty

The clientTags parameter is sent as the X-Trino-Client-Tags header. Resource group selectors can match on the tags, so they route queries to resource groups and help with query tracking. A tag cannot contain ,. The X-Trino-Client-Tags named argument overrides the tags for one query.

config := &trino.Config{
	ServerURI:  "http://foobar@localhost:8080",
	ClientTags: []string{"tag1", "tag2", "tag3"},
}
dsn, err := config.FormatDSN()

resourceEstimates

Type:           string
Valid values:   semicolon-separated list of resource:estimate pairs
Default:        empty

The resourceEstimates parameter tells the coordinator how much each query is expected to use, and is sent as one X-Trino-Resource-Estimate header per estimate with every statement. Resource group selectors can match on these estimates to route queries, for example to keep long or memory-hungry queries out of an interactive group. The server accepts these resources, ignoring case:

• EXECUTION_TIME and CPU_TIME, a duration such as 90s, 10m or 1.5h (units ns, us, ms, s, m, h, d)
• PEAK_MEMORY, a data size such as 512MB or 1.5GB (units B, kB, MB, GB, TB, PB, all powers of 1024)

The values are sent as given. The server rejects a query with an unknown resource or a value it cannot parse, so a Go time.Duration must not be passed through String(), which produces 1h30m0s. Use trino.FormatDuration and trino.FormatDataSize instead, which produce values the server accepts, such as 1.50h and 1.50GB. They use the largest unit that represents the value exactly with two decimal places, so they never round: time.Hour + time.Second becomes 3601.00s. The driver rejects names that are empty, contain = or ,, or contain characters other than printable ASCII. The X-Trino-Resource-Estimate named argument adds to or replaces the estimates for one query.

config := &trino.Config{
	ServerURI: "http://foobar@localhost:8080",
	ResourceEstimates: map[string]string{
		"EXECUTION_TIME": "10m",
		"CPU_TIME":       trino.FormatDuration(90 * time.Minute), // 1.50h
		"PEAK_MEMORY":    trino.FormatDataSize(1 << 30),          // 1GB
	},
}
dsn, err := config.FormatDSN()

trace_token

Type:           string
Valid values:   any string
Default:        empty

The trace_token parameter is sent as the X-Trino-Trace-Token header and is recorded by the coordinator, so queries made through the connection can be correlated with the server logs and event listeners.

client_info

Type:           string
Valid values:   any string
Default:        empty

The client_info parameter is sent as the X-Trino-Client-Info header. It is free-form metadata about the client, shown in the Trino web UI and passed to event listeners.

language

Type:           string
Valid values:   a language tag, e.g. en-US
Default:        empty (the server default)

The language parameter is sent as the X-Trino-Language header and selects the locale used by locale-sensitive functions.

timezone

Type:           string
Valid values:   an IANA time zone name (e.g. Europe/Warsaw), UTC, or an offset
                like +05:30
Default:        the local time zone of the client

The timezone parameter is sent as the X-Trino-Time-Zone header. The server uses it to evaluate current_timestamp, now(), date_trunc and every other function that depends on the session time zone, and the driver uses the same zone to read DATE, TIME and TIMESTAMP values that do not carry a zone of their own. Without the parameter the local zone is used, taken from the TZ environment variable or from /etc/localtime; when neither names a zone, the current UTC offset is sent instead. Set the parameter explicitly to make the results independent of where the client runs, for example timezone=UTC.

Running SET TIME ZONE changes the zone for the rest of the session, and the driver follows it when reading values. The X-Trino-Time-Zone named argument overrides the zone for one query.

config := &trino.Config{
	ServerURI: "http://foobar@localhost:8080",
	TimeZone:  "Asia/Tokyo",
}
dsn, err := config.FormatDSN()

query_timeout

Type:           time.Duration
Valid values:   duration string (e.g. 30s, 5m)
Default:        unset (10h, trino.DefaultQueryTimeout, for queries whose
                context has no deadline)

The query_timeout parameter bounds how long a query may run before the driver cancels it on the server and returns an error. When set, it applies to every query, even one whose context has a deadline. When unset, the context deadline applies, and a context without one gets trino.DefaultQueryTimeout. See Cancellation and timeouts.

request_retry_timeout

Type:           time.Duration
Valid values:   duration string
Default:        2m (trino.DefaultRequestRetryTimeout)

The request_retry_timeout parameter sets how long a single HTTP request is retried on a 429, 502, 503 or 504 response or a network error before the query fails. Only idempotent requests are retried on network errors after the connection is established. A Retry-After header on a 429 or 503 response, in seconds or as an HTTP date, sets the wait before the next attempt instead of the default backoff; the wait never extends past this timeout. The value must be positive.

request_retry_max_attempts

Type:           int
Valid values:   positive integer
Default:        20 (trino.DefaultRequestRetryMaxAttempts)

The request_retry_max_attempts parameter sets how many times a single HTTP request is sent before the query fails. Retrying stops at whichever of request_retry_timeout and request_retry_max_attempts is reached first; with the defaults, the timeout is reached first.

heartbeat_interval

Type:           time.Duration
Valid values:   positive duration string (e.g. 30s, 1m)
Default:        unset (30s)

The heartbeat_interval parameter sets how often the driver sends a HEAD heartbeat to the current nextUri while a spooled query is in progress, so the coordinator does not abandon a query whose results are consumed slowly. It applies to the whole connection and is only used with the spooling protocol.

explicitPrepare

Type:           bool
Valid values:   true, false
Default:        true

The explicitPrepare parameter controls how statements with arguments reach the server. With true, the driver prepares the statement through the X-Trino-Prepared-Statement header and runs EXECUTE. With false, it runs EXECUTE IMMEDIATE, which carries the statement text in the request body instead of a header, so statements larger than the server's header size limit still work. The Config field is DisableExplicitPrepare, whose zero value keeps explicit prepare on.

accessToken

Type:           string
Valid values:   a JSON web token
Default:        empty

The accessToken parameter is sent as a bearer token in the Authorization header of every request, see JSON web token authentication.

forwardAuthorizationHeader

Type:           bool
Valid values:   true, false
Default:        false

The forwardAuthorizationHeader parameter allows the accessToken named argument to set the bearer token of one query, see Authorization header forwarding.

externalAuthentication

Type:           bool
Valid values:   true, false
Default:        false

The externalAuthentication parameter enables external authentication. It requires HTTPS.

externalAuthenticationTimeout

Type:           time.Duration
Valid values:   positive duration string (e.g. 5m)
Default:        unset (2m)

The externalAuthenticationTimeout parameter sets how long the driver waits for the user to log in.

oauth2ClientId / oauth2ClientSecret

Type:           string
Valid values:   the client ID and secret registered with the identity provider
Default:        empty

The oauth2ClientId and oauth2ClientSecret parameters enable OAuth2 client credentials authentication. They must be set together and require HTTPS.

oauth2TokenEndpoint

Type:           string
Valid values:   an https:// URL
Default:        empty (the token endpoint the server advertises)

The oauth2TokenEndpoint parameter pins the token endpoint the client credentials are sent to. A different token endpoint advertised by the server is rejected. It requires oauth2ClientId.

oauth2Scope

Type:           string
Valid values:   space separated scopes
Default:        empty (the scope the server advertises)

The oauth2Scope parameter sets the scope requested from the token endpoint, overriding the scope the server advertises. It requires oauth2ClientId.

KerberosEnabled

Type:           bool
Valid values:   true, false
Default:        false

The KerberosEnabled parameter turns on Kerberos authentication.

KerberosKeytabPath

Type:           string
Valid values:   a filesystem path
Default:        empty (a credential cache is used)

The KerberosKeytabPath parameter names the keytab the driver logs in with, as KerberosPrincipal in KerberosRealm. It cannot be combined with KerberosCredentialCachePath.

KerberosCredentialCachePath

Type:           string
Valid values:   a filesystem path, optionally prefixed with FILE:
Default:        empty (KRB5CCNAME, then /tmp/krb5cc_<uid>, when no keytab is given)

A Kerberos credential cache holding a ticket for the user, used instead of a keytab; it cannot be combined with KerberosKeytabPath. Only file caches can be read: when KRB5CCNAME names another type, such as KEYRING:, KCM: or macOS's default API:, run kinit -c FILE:/path and set this parameter. The cache is read when a connection opens, so connections opened after kinit refreshes it use the new ticket.

KerberosPrincipal

Type:           string
Valid values:   a Kerberos principal name, without the realm
Default:        empty (the principal of the credential cache)

The KerberosPrincipal parameter names the user to authenticate as. With a keytab, the driver logs in as this principal. With a credential cache or a KerberosClient it is optional, and when set must match the principal the cache or client holds.

KerberosRealm

Type:           string
Valid values:   a Kerberos realm
Default:        empty (the realm of the credential cache)

The KerberosRealm parameter is the realm of KerberosPrincipal. With a credential cache or a KerberosClient it must match their realm when set.

KerberosConfigPath

Type:           string
Valid values:   a filesystem path
Default:        empty

The KerberosConfigPath parameter names the krb5 configuration file, usually /etc/krb5.conf, which lists the KDCs and maps host names to realms.

KerberosRemoteServiceName

Type:           string
Valid values:   a Kerberos service name
Default:        trino

The KerberosRemoteServiceName parameter is the service name of the coordinator, substituted for ${SERVICE} in KerberosServicePrincipalPattern.

KerberosServicePrincipalPattern

Type:           string
Valid values:   a service principal with optional ${SERVICE} and ${HOST} placeholders
Default:        ${SERVICE}@${HOST}

The service principal of the coordinator. ${SERVICE} is replaced with KerberosRemoteServiceName and ${HOST} with the lowercased host of the request URL. A service@host result names the Kerberos principal service/host, so the default asks for a ticket for trino/<host>; a result without @, such as HTTP/${HOST}, is used as the principal name. The realm of the service comes from domain_realm in the krb5 configuration, or is the user's realm, and cannot be part of the pattern.

KerberosUseCanonicalHostname

Type:           bool
Valid values:   true, false
Default:        true

${HOST} in KerberosServicePrincipalPattern is the canonical name of the coordinator host instead of the host in the URL, for a coordinator reached through a DNS alias or a load balancer. The host is resolved to an address, following CNAME records, and the first address is looked up in reverse DNS; without a reverse record the address itself is used. localhost and loopback addresses are replaced with the canonical name of the local machine, and requests fail with Fully qualified name of localhost should not resolve to 'localhost' when that resolves to localhost too, as on many laptops. The name is resolved once per connection and host. Set it to false, or set Config.KerberosDisableCanonicalHostname, to use the URL host as is.

Earlier versions of this driver always used the URL host. If you reach the coordinator through a DNS alias and its service principal is registered for that alias, set KerberosUseCanonicalHostname=false to keep asking for the same ticket.

SSLCertPath / SSLCert

Type:           string
Valid values:   a filesystem path (SSLCertPath) or PEM-encoded certificate content (SSLCert)
Default:        empty (the system trust store is used)

An additional CA certificate the driver should trust, as a file path or as inline PEM content; only one of the pair may be set. Requires HTTPS, and cannot be combined with custom_client or HTTPClient, whose transports have their own TLS configuration.

SSLClientCertPath / SSLClientCert and SSLClientKeyPath / SSLClientKey

Type:           string
Valid values:   a filesystem path (...Path) or PEM-encoded content
Default:        empty (no client certificate is presented)

A PEM client certificate and its private key, for TLS client authentication (mutual TLS). The certificate and the key must each be given as either a path or inline content, not both, and the certificate and key must be set together. Requires HTTPS, and cannot be combined with custom_client or HTTPClient.

db, err := sql.Open("trino", "https://user@localhost:8443"+
	"?SSLClientCertPath=/path/to/client-cert.pem"+
	"&SSLClientKeyPath=/path/to/client-key.pem")

SSLVerification

Type:           string
Valid values:   "FULL", "CA", "NONE"
Default:        "FULL"

Controls how the driver validates the coordinator's TLS certificate. FULL validates the certificate chain and the hostname. CA validates the certificate chain but not the hostname. NONE disables certificate validation entirely and must only be used for development, since it also allows a network attacker to intercept the connection. Requires HTTPS, and cannot be combined with custom_client or HTTPClient.

httpProxy / socksProxy

Type:           string
Valid values:   host:port
Default:        empty (the proxy environment variables apply)

Sends every request through an HTTP proxy (httpProxy) or a SOCKS5 proxy (socksProxy). Only one of the two may be set. HTTPS requests reach the coordinator through an HTTP CONNECT tunnel, and the SOCKS5 proxy resolves the coordinator's host name. Neither can be combined with custom_client or HTTPClient; see Proxy.

db, err := sql.Open("trino", "http://user@localhost:8080?socksProxy=localhost:1080")

custom_client

Type:           string
Valid values:   the name of a client previously registered to the driver
Default:        empty (defaults to http.DefaultClient)

The custom_client parameter selects an http.Client registered with trino.RegisterCustomClient for every request, see Custom HTTP client. Its transport decides how to verify the coordinator and how to reach it, so the driver rejects SSLCert, SSLCertPath, the client certificate parameters, SSLVerification, httpProxy and socksProxy combined with it, instead of ignoring them.

Authentication

HTTP Basic authentication

A password in the DSN, or in Config.ServerURI, enables HTTP Basic authentication: the driver sets the Authorization header in every request to Trino. It requires HTTPS; a password over http is an error.

Kerberos authentication

Kerberos authentication is enabled with KerberosEnabled=true and configured with the Kerberos* parameters, on the DSN or on Config. The coordinator only offers Kerberos authentication over HTTPS.

The driver logs in with the keytab in KerberosKeytabPath, or reuses a ticket from a credential cache, such as the one kinit writes. Without a keytab it reads the cache in KerberosCredentialCachePath, then the one named by the KRB5CCNAME environment variable, then /tmp/krb5cc_<uid>. The principal and realm come from the cache; when KerberosPrincipal or KerberosRealm are set, they must match it.

db, err := sql.Open("trino", "https://user@localhost:8443"+
	"?KerberosEnabled=true"+
	"&KerberosConfigPath=/etc/krb5.conf"+
	"&KerberosCredentialCachePath=/tmp/krb5cc_1000")

A program that already holds a logged-in gokrb5 client, for example one that authenticates its own users, can pass it in Config.KerberosClient with trino.NewConnector. All connections of the Connector share it and the driver never logs it in, renews or destroys it. It requires KerberosEnabled and cannot be combined with KerberosKeytabPath or KerberosCredentialCachePath. When KerberosPrincipal or KerberosRealm are set, they must match the client's credentials.

connector, err := trino.NewConnector(&trino.Config{
	ServerURI:       "https://user@localhost:8443",
	KerberosEnabled: true,
	KerberosClient:  kerberosClient,
})

The ticket is requested for the service principal that KerberosServicePrincipalPattern and KerberosUseCanonicalHostname produce. See Kerberos authentication in the Trino documentation for the server side.

JSON web token authentication

A JWT in the accessToken parameter, or in Config.AccessToken, is sent as a bearer token in the Authorization header of every request. See JWT authentication in the Trino documentation for the server side.

External authentication

With externalAuthentication=true, a coordinator that uses OAuth2 authentication logs the user in through the browser. When the coordinator rejects a request, the driver passes the login URL to Config.RedirectHandler, waits for the token until externalAuthenticationTimeout (default 2m) passes, and retries the request with it, so a query running when the token expires continues. The token is kept in Config.TokenCache and shared by all connections of a Connector. By default, RedirectHandler is trino.OpenBrowser, and the cache holds the token in memory.

connector, err := trino.NewConnector(&trino.Config{
	ServerURI:              "https://trino.example.com:8443",
	ExternalAuthentication: true,
	RedirectHandler: func(ctx context.Context, redirectURL *url.URL) error {
		fmt.Println("Log in at", redirectURL)
		return nil
	},
})

Implement trino.TokenCache to keep the token between runs of a program. External authentication requires HTTPS, and cannot be combined with a password, Kerberos, or forwardAuthorizationHeader. An AccessToken is sent until a token is cached.

OAuth2 client credentials

A service without a user to log in can authenticate to a coordinator that uses OAuth2 authentication with the OAuth2 client credentials grant, by setting oauth2ClientId and oauth2ClientSecret to a client registered with the coordinator's identity provider. A request without a valid token is rejected by the coordinator, and Trino 484 and later advertise the identity provider's token endpoint and the scope to request in the WWW-Authenticate challenge of that response. The driver then exchanges the client ID and secret for an access token at that endpoint, by posting grant_type=client_credentials with the client ID and secret as form fields (client_secret_post), and sends the rejected request again with the access token as a bearer token.

connector, err := trino.NewConnector(&trino.Config{
	ServerURI:          "https://trino.example.com:8443",
	OAuth2ClientID:     "trino-client",
	OAuth2ClientSecret: secret,
})

The same settings in a DSN:

https://trino.example.com:8443?oauth2ClientId=trino-client&oauth2ClientSecret=...

The token is shared by all connections of a Connector and is sent with every request until 30 seconds before the expires_in the token endpoint returned, or 5 minutes when it returned none; the next request then gets a new token. A token the coordinator rejects is replaced once per request.

The client secret is never sent to the coordinator, and only to an https:// token endpoint, which does not follow redirects. Set oauth2TokenEndpoint to send it only to an endpoint you trust: the driver then rejects a different endpoint the coordinator advertises, so a compromised coordinator cannot collect the secret. It also lets the driver use the client credentials grant with coordinators older than Trino 484, which do not advertise an endpoint. Set oauth2Scope to request another scope than the one the coordinator advertises. The token endpoint is reached with the same HTTP client, TLS settings and proxy as the coordinator.

Client credentials authentication requires HTTPS, and cannot be combined with external authentication, a password, an accessToken, Kerberos, or forwardAuthorizationHeader.

An identity provider that needs another way to authenticate the client, like HTTP Basic authentication (client_secret_basic), can be used through an HTTPClient from golang.org/x/oauth2, which adds the token to every request and renews it when it expires:

import "golang.org/x/oauth2/clientcredentials"

credentials := clientcredentials.Config{
	ClientID:     "trino-client",
	ClientSecret: secret,
	TokenURL:     "https://idp.example.com/oauth2/token",
}
connector, err := trino.NewConnector(&trino.Config{
	ServerURI:  "https://trino.example.com:8443",
	HTTPClient: credentials.Client(context.Background()),
})

Authorization header forwarding

A program that holds a different token for each of its users can send it with each query: set forwardAuthorizationHeader=true, or Config.ForwardAuthorizationHeader, and pass the token as the accessToken named argument. It overrides the AccessToken of the connection for that query.

rows, err := db.Query("SELECT * FROM foobar", sql.Named("accessToken", userToken))

Using the accessToken named argument without enabling forwardAuthorizationHeader returns an error, so the token is never sent as part of the query text, where Trino would persist it in the query history.

Per-query user and session authorization

A query can run as a user other than the one that authenticated to the coordinator, when system access control allows the impersonation. Pass the user as the X-Trino-User named argument; it is sent as a header and is not part of the query:

db.Query("SELECT * FROM foobar WHERE id=?", 1, sql.Named("X-Trino-User", "Alice"))

Trino 426 and newer also let a session switch its authorization user with SET SESSION AUTHORIZATION <user>, reverted with RESET SESSION AUTHORIZATION. Statements after the switch run as the new user until the reset. The switch is held by the underlying connection, and database/sql resets a connection when it takes it from the pool, so run the switch and every statement that should run as the new user on one sql.Conn:

conn, err := db.Conn(ctx)
if err != nil {
	return err
}
defer conn.Close()

if _, err := conn.ExecContext(ctx, "SET SESSION AUTHORIZATION bob"); err != nil {
	return err
}
// Runs as bob. A db.Query here would run as the DSN user instead.
rows, err := conn.QueryContext(ctx, "SELECT current_user")

While the switch is active, the roles from the roles parameter and any role selected with SET ROLE are not sent; they apply again after the reset. Trino rejects both statements inside a transaction.

Per-query settings

database/sql has no way to pass settings for one query, so the driver reads them from sql.Named arguments. They are consumed by the driver and never reach the query text, and their position among the other arguments does not matter. Any argument whose name starts with X-Trino- is sent as a request header with that name, which overrides the connection's value of that header for the query. The arguments with a special meaning are:

Name	Value	Purpose
X-Trino-User	string	Run the query as another user, see Per-query user
X-Trino-Role	map[string]string	Catalog roles for the query, formatted like roles
X-Trino-Resource-Estimate	map[string]string	Estimates added to or replacing the connection's resourceEstimates
X-Trino-Client-Tags	string	Comma-separated tags replacing clientTags
X-Trino-Time-Zone	string	Time zone replacing timezone, also for reading the results
X-Trino-Progress-Callback	trino.ProgressUpdater	Query id and statistics, see Query id and progress
X-Trino-Progress-Callback-Period	time.Duration	How often the callback runs
warnings	*trino.Warnings	Collects the warnings of the query, see Query warnings
accessToken	string	Bearer token for the query, see Authorization header forwarding
encoding	string	Spooled result encodings, see Spooling protocol
spooling_worker_count	string	Parallel segment downloads, a number
max_out_of_order_segments	string	Buffered out-of-order segments, a number

Query id and progress

database/sql has no way to expose the Trino query id or the statistics the coordinator reports while a query runs. The driver makes them available through a callback passed as a pair of named arguments:

• X-Trino-Progress-Callback - a value implementing the trino.ProgressUpdater interface
• X-Trino-Progress-Callback-Period - a time.Duration limiting how often the callback is invoked while the query state does not change

Both must be set together. The callback receives a trino.QueryProgressInfo with the QueryId and the current query statistics. It is invoked when the query is submitted, when a result page is received, and once more when the query finishes.

The callback runs on its own goroutine, one invocation at a time and in the order the updates arrived. The driver waits for it to accept every update rather than dropping any, so a change of the query state is never missed, but a callback slower than the result pages delays fetching the next page. Keep it quick, or hand the update off to another goroutine if it does more than record the statistics.

For a query using the spooling protocol, FailedSegmentAcknowledgments counts the segments the driver read but could not acknowledge. Those segments stay in storage until they expire. Acknowledgments can finish after the query does, so closing the rows waits for them and, if any failed since the last invocation, invokes the callback once more with the final count.

type queryLogger struct{}

func (queryLogger) Update(info trino.QueryProgressInfo) {
	log.Printf("query %s is %s, %.0f%% done", info.QueryId, info.QueryStats.State, info.QueryStats.ProgressPercentage)
}

rows, err := db.Query("SELECT * FROM foobar",
	sql.Named("X-Trino-Progress-Callback", queryLogger{}),
	sql.Named("X-Trino-Progress-Callback-Period", time.Second),
)

Query warnings

The coordinator can attach warnings to a query, such as deprecation notices, which database/sql also has no way to expose. Pass a *trino.Warnings as the warnings named argument; once the rows are consumed, All returns every distinct warning the coordinator reported, in the order it first reported it. Without a warnings argument, warnings the coordinator reports are dropped.

var warnings trino.Warnings
rows, err := db.Query("SELECT * FROM foobar", sql.Named("warnings", &warnings))
// ... consume rows ...
for _, w := range warnings.All() {
	log.Printf("warning %s: %s", w.Name, w.Message)
}

Sessions and transactions

Session state and the connection pool

Statements that change the session, such as USE, SET PATH, SET SESSION, SET TIME ZONE, SET ROLE, PREPARE and SET SESSION AUTHORIZATION, change it on the underlying connection only. When database/sql hands that connection to the next caller, the driver puts back the user, catalog, schema, session properties, roles and time zone from the DSN, and drops the path and prepared statements. To run several statements in one session, use one sql.Conn:

conn, err := db.Conn(ctx)
if err != nil {
	return err
}
defer conn.Close()

if _, err := conn.ExecContext(ctx, "USE tpch.tiny"); err != nil {
	return err
}
// Reads tpch.tiny.nation. A db.Query here would use the DSN catalog and schema.
rows, err := conn.QueryContext(ctx, "SELECT * FROM nation")

Transactions

Use db.Begin or db.BeginTx to run several statements in a single Trino transaction:

tx, err := db.BeginTx(ctx, nil)
if err != nil {
	return err
}
defer tx.Rollback()

if _, err := tx.ExecContext(ctx, "DELETE FROM reports WHERE day = DATE '2025-01-01'"); err != nil {
	return err
}
if _, err := tx.ExecContext(ctx, "INSERT INTO reports SELECT * FROM staging"); err != nil {
	return err
}
return tx.Commit()

sql.TxOptions is honoured. Isolation accepts sql.LevelDefault, sql.LevelReadUncommitted, sql.LevelReadCommitted, sql.LevelRepeatableRead and sql.LevelSerializable; any other level is rejected. ReadOnly starts a READ ONLY transaction.

Commit and Rollback carry the values of the context passed to BeginTx, such as trace spans, but not its cancellation or deadline, so the rollback database/sql runs when that context is cancelled still reaches the server.

A statement that fails aborts the whole transaction on the server. Rollback reports success in that case, while Commit returns an error.

Closing rows before the last page cancels the query on the server, and Trino aborts the transaction when a query inside it is cancelled. A tx.QueryRow over a result that spans more than one page, or a rows.Close before the end, therefore leaves Commit failing with TRANSACTION_ALREADY_ABORTED. Inside a transaction, read the rows to the end or add a LIMIT to the query.

Transaction control statements must go through the returned *sql.Tx. Sending them directly, as in db.Exec("START TRANSACTION"), is rejected.

[!NOTE] Transaction support depends on the connector. Connectors that only support writes in autocommit mode, such as memory, reject writes inside a multi-statement transaction, and the isolation levels a connector accepts vary.

Connection validation and server version

Opening a connection does not contact the server. db.PingContext fetches the coordinator's /v1/info and fails when the server cannot be reached, answers with an error, or is still starting up. Since that endpoint does not require authentication, the ping then sends HEAD /v1/statement with the connection's credentials and fails when the server rejects them. No query is started. Servers older than Trino 469 do not support that request, so for them a successful ping does not prove the credentials are accepted. With external authentication, that request starts the login flow when no valid token is cached, so a ping can open the browser.

Before returning a connection to the pool, database/sql asks the driver whether it can be reused, without contacting the server. A connection whose Kerberos client failed to authenticate a request is dropped, so the next one reloads the keytab or credential cache, for example after kinit renewed an expired ticket; a statement that failed that way before reaching the server is retried on a new connection.

The same information, including the server version, is available from trino.Conn.ServerInfo through sql.Conn.Raw:

conn, err := db.Conn(ctx)
if err != nil {
	return err
}
defer conn.Close()

var info trino.ServerInfo
err = conn.Raw(func(driverConn any) error {
	info, err = driverConn.(*trino.Conn).ServerInfo(ctx)
	return err
})
if err != nil {
	return err
}
log.Printf("Trino %s in %s, up for %s", info.NodeVersion, info.Environment, info.Uptime)

Data types

Query arguments

When passing arguments to queries, the driver supports the following Go data types:

• integers
• float32 and float64 - passed to Trino as REAL and DOUBLE literals
• bool
• string
• []byte
• slices - passed to Trino as an ARRAY
• maps - passed to Trino as a MAP, built as MAP(ARRAY[keys...], ARRAY[values...]); an empty or nil map is passed as MAP()
• trino.Numeric - a string representation of a number
• time.Time - passed to Trino as a timestamp with a time zone
• the result of trino.Date(year, month, day) - passed to Trino as a date
• the result of trino.Time(hour, minute, second, nanosecond) - passed to Trino as a time without a time zone
• the result of trino.TimeTz(hour, minute, second, nanosecond, location) - passed to Trino as a time with a time zone
• the result of trino.Timestamp(year, month, day, hour, minute, second, nanosecond) - passed to Trino as a timestamp without a time zone
• time.Duration - passed to Trino as an interval day to second. A whole number of milliseconds works with every Trino version. A finer duration, like 1500 * time.Microsecond, is passed with its precision declared, as in INTERVAL '0.0015' SECOND(1, 4), which needs Trino 484 or later; older servers reject the query instead of truncating the value to milliseconds.
• trino.Interval - passed to Trino as an interval year to month when Months is set, and otherwise as an interval day to second like a time.Duration. Non-zero Picoseconds are passed with as many fractional-second digits as they need, up to 12, which also needs Trino 484 or later. Setting Months together with Duration or Picoseconds is an error, and an Interval that is not Valid is passed as NULL.

It's not yet possible to pass:

• byte
• json.RawMessage
• VARIANT values; pass the JSON text as a string and convert it with CAST(json_parse(?) AS VARIANT)

To use the unsupported types, pass them as strings and use casts in the query, like so:

SELECT *
FROM table
WHERE col_json = CAST(? AS JSON) OR col_timestamp = CAST(? AS TIMESTAMP)

Response rows

The driver decodes each value straight into the Go value of its column type, which is what a column scanned into an interface{} holds:

Trino type	Go type
NULL of any type	nil
BOOLEAN	bool
TINYINT, SMALLINT, INTEGER, BIGINT	int64
REAL, DOUBLE	float64
DECIMAL, NUMBER, CHAR, VARCHAR, JSON, UUID, IPADDRESS, INTERVAL YEAR TO MONTH, INTERVAL DAY TO SECOND, Geometry, SphericalGeography, color	string
VARBINARY and every type without a textual form, like HyperLogLog	[]byte
DATE, TIME, TIMESTAMP, with or without a time zone	time.Time
VARIANT	trino.Variant
ARRAY	[]interface{}
MAP	map[string]interface{}
ROW	trino.Row
BingTile, KdbTree	map[string]interface{}

The elements of an ARRAY, the values of a MAP and the fields of a ROW are decoded the same way, at any depth, so an ARRAY(VARBINARY) holds []byte elements and an ARRAY(TIMESTAMP) holds time.Time elements. The keys of a MAP are strings, the way the server sends them, like "1" for a MAP(INTEGER, VARCHAR); trino.NullMap converts them into its key type. A value that does not match its column type fails the query with an error naming the row and the column.

The details for some of the types:

• time and timestamps with precision - all time types are returned as time.Time. All precisions up to nanoseconds (TIMESTAMP(9) or TIME(9)) are supported (since this is the maximum precision Golang's time.Time supports). If a query returns columns defined with a greater precision, values are trimmed to 9 decimal digits. Use CAST to reduce the returned precision, or convert the value to a string that then can be parsed manually.
• DATE, TIME and TIMESTAMP without a time zone - returned as time.Time in the zone of the connection (see the timezone parameter), which is the zone the server used to produce them, also inside an ARRAY, MAP or ROW. MAP keys of these types are parsed by trino.NullMap, in the zone of its Location field, time.Local by default.
• DECIMAL and NUMBER (Trino 480+) - returned as string; use sql.NullString for nullable columns
• IPADDRESS - returned as string
• INTERVAL YEAR TO MONTH and INTERVAL DAY TO SECOND - returned as string, like 1-2 or -3 04:05:06.789, or scanned into trino.Interval, which holds a year to month interval as Months and a day to second interval as a time.Duration. Scanning a day to second interval longer than about 292 years, which a time.Duration cannot hold, returns an error
• parametric intervals (Trino 484+) - intervals have a leading-field precision and up to 12 fractional-second digits, like INTERVAL SECOND(2, 12) or INTERVAL DAY(3) TO SECOND(9), also inside an ARRAY, MAP or ROW, and a value has exactly as many fractional digits as its type, none for a precision of 0. Older servers send every day to second interval rounded to milliseconds. trino.Interval keeps all 12 digits: Duration holds the value truncated toward zero to nanoseconds, and Picoseconds the remaining picoseconds, from -999 to 999, with the same sign as the value. A value with more than 12 fractional digits is an error. sql.ColumnType.DatabaseTypeName() returns the full type, like INTERVAL SECOND(2, 0), INTERVAL DAY(3) TO SECOND(9) or INTERVAL YEAR(5) TO MONTH, and sql.ColumnType.DecimalSize() reports the leading-field precision as the precision and the fractional-seconds precision as the scale. Older servers report INTERVAL DAY TO SECOND or INTERVAL YEAR TO MONTH, without a precision or scale
• UUID - returned as string
• VARIANT (Trino 481+) - returned as trino.Variant, see VARIANT
• Geometry, SphericalGeography and color - returned as string
• BingTile and KdbTree - returned as map[string]interface{}, the JSON object the server sent, with numbers as float64
• any other type, like HyperLogLog, SetDigest, QDigest, and TDigest - returned as []byte, decoded from the base64 form the server sends, the same as VARBINARY

For reading nullable columns, use trino.NullTime or similar structs from the database/sql package, like sql.NullInt64.

Since Trino does not report whether a column can hold NULL, sql.ColumnType.Nullable reports every column as nullable.

ARRAY, MAP and ROW

To read ARRAY, MAP and ROW values, use the generic scanners, which nest to any depth:

• trino.NullSlice[T] for an ARRAY, with its elements in Slice
• trino.NullMap[K, V] for a MAP, with its entries in Map
• trino.NullRow[T] for a ROW, with its fields stored in the struct T in Row

Each one has a Valid field that is false for a NULL value, at every level, so a NULL inner array is told apart from an empty one:

var tags trino.NullSlice[trino.NullSlice[sql.NullString]]
err := db.QueryRow("SELECT ARRAY[ARRAY['a', NULL], NULL]").Scan(&tags)
// tags.Slice[0].Slice == []sql.NullString{{String: "a", Valid: true}, {}}
// tags.Slice[1].Valid == false

var scores trino.NullMap[string, trino.NullSlice[int64]]
err = db.QueryRow("SELECT MAP(ARRAY['a'], ARRAY[ARRAY[BIGINT '1', 2]])").Scan(&scores)
// scores.Map["a"].Slice == []int64{1, 2}

Elements, map keys and values, and row fields can be bool, string, int64 and the narrower integer types, float64, float32, time.Time, []byte, map[string]interface{}, []interface{} or interface{}, the nullable types from database/sql (sql.NullBool, sql.NullString, sql.NullInt64, sql.NullInt32, sql.NullInt16, sql.NullFloat64, sql.NullTime), trino.NullTime, trino.NullBinary, trino.Variant, another generic scanner, or any other type that implements sql.Scanner, whose Scan receives the element as listed in Response rows. A NULL element scanned into a plain type that cannot hold it, like int64, is an error. An integer element also scans into a floating point type, and so does a DECIMAL element, which is a string otherwise. MAP keys without a time zone, like DATE keys, are parsed in the zone of the Location field of trino.NullMap, or of the trino.NullSlice or trino.NullMap it is nested in, and time.Local when no Location is set.

trino.NullRow[T] maps each ROW field to an exported field of the struct T, by the name in a trino:"name" struct tag, or by the Go field name, preferring an exact match and falling back to one ignoring case. A field tagged trino:"-" is skipped. An anonymous ROW field is named field<i>, where i is its zero-based position, so it maps to a struct field called Field0, Field1, and so on. A ROW field without a matching struct field is an error; a struct field without a matching ROW field keeps its zero value. Use trino.NullRow for nested rows too:

type Point struct {
	X     int64 `trino:"x"`
	Label sql.NullString
}

var points trino.NullSlice[trino.NullRow[Point]]
err := db.QueryRow("SELECT ARRAY[CAST(ROW(1, 'a') AS ROW(x INTEGER, label VARCHAR)), NULL]").Scan(&points)
// points.Slice[0].Row == Point{X: 1, Label: sql.NullString{String: "a", Valid: true}}
// points.Slice[1].Valid == false

A ROW whose fields are not known ahead of time scans into a trino.Row, which keeps the field names and converts each value the way a plain column of that type would be converted, including nested ROW and VARIANT values:

var row trino.Row
err := db.QueryRow("SELECT CAST(ROW(1, 'a') AS ROW(x INTEGER, y VARCHAR))").Scan(&row)
// row.Valid == true
// row.Len() == 2
// row.Name(0) == "x", row.Value(0) == int64(1)
// row.Field("y") == "a", true

An ARRAY or MAP scanned into an interface{} holds a []interface{} or a map[string]interface{} whose elements are converted the way a plain column would be: a VARBINARY element is a []byte, a number an int64 or a float64, a DECIMAL a string and a TIMESTAMP a time.Time.

sql.ColumnType.ScanType() reports the generic scanners, instantiated with the scan types of the elements, keys and values, like trino.NullSlice[sql.NullInt32] for an ARRAY(INTEGER), trino.NullMap[sql.NullString, sql.NullInt64] for a MAP(VARCHAR, BIGINT) and trino.NullSlice[trino.Row] for an ARRAY(ROW(...)), so a value created with reflect.New from it can be passed to Scan(). A ROW reports trino.Row, as trino.NullRow needs a struct type. The reported types cover arrays of up to three dimensions, maps of scalar values, and arrays of such maps; a column nested deeper reports interface{} for the values below the deepest level covered, like trino.NullMap[sql.NullString, interface{}] for a MAP(VARCHAR, ARRAY(BIGINT)). A map key that is not comparable in Go, like VARBINARY, is reported as interface{}.

VARIANT

To read a VARIANT value, scan into a trino.Variant:

var v trino.Variant
err := db.QueryRow(`SELECT CAST(JSON '{"a": 1, "b": [true, null]}' AS VARIANT)`).Scan(&v)
// v.Valid == true
// v.Type() == trino.VariantObject
// v.Value() == map[string]interface{}{"a": int64(1), "b": []interface{}{true, nil}}
// v.String() == `{"a":1,"b":[true,null]}`

The driver announces the VARIANT_BINARY client capability, so the server sends each value in its binary encoding and Variant keeps the value's type. Type returns one of the Variant* constants and Value converts the value into a Go value:

• nil for a VARIANT null
• bool
• int64 for every integer width
• float32 and float64
• trino.Numeric for a decimal, in plain notation like -12.345
• string, and a string like 12151fd2-7586-11e9-8f9e-2a86e4085a59 for a UUID
• []byte
• time.Time for dates, times and timestamps. Those without a time zone are in the zone of the connection, like DATE, TIME and TIMESTAMP columns, and timestamps with a time zone are in UTC
• []interface{} for an array and map[string]interface{} for an object, holding converted values

String and MarshalJSON return the JSON text the server returns for CAST(v AS JSON), writing types JSON has no equivalent for as strings, like "2020-01-02" for a date. A SQL NULL scans with Valid set to false, while a VARIANT null, like CAST(JSON 'null' AS VARIANT), is valid and has the type trino.VariantNull.

A VARIANT column does not scan into a string or sql.NullString, as it did when the driver did not announce the VARIANT capabilities and the server sent VARIANT columns as JSON: database/sql only converts strings, byte slices and numbers into a string. Scan into a trino.Variant and call String, or use CAST(v AS JSON) in the query. An ARRAY(VARIANT) scans into a trino.NullSlice[trino.Variant], or still into a trino.NullSlice[sql.NullString] holding the JSON text of each element.

Errors

A query the server rejects or fails returns a *trino.ErrQueryFailed, whose StatusCode is the HTTP status of the response that reported the failure (200 for a query that failed while running) and which wraps the error details. For a failure the server reports, the wrapped error is a *trino.ErrTrino with the ErrorCode, ErrorName, ErrorType, SqlState, Message, ErrorLocation and FailureInfo from the Trino error:

rows, err := db.QueryContext(ctx, "SELECT 1 FORM dual")
var trinoErr *trino.ErrTrino
if errors.As(err, &trinoErr) {
	log.Printf("%s (%d) at line %d: %s", trinoErr.ErrorName, trinoErr.ErrorCode, trinoErr.ErrorLocation.LineNumber, trinoErr.Message)
}

Other errors the driver returns:

• trino.ErrQueryCancelled when the server reports the query was cancelled, for example from the web UI.
• trino.UnsupportedArgError for a query argument of a type the driver cannot serialize, see Query arguments.
• *trino.SegmentExpiredError when a spooled segment could not be downloaded after its expiration time, see Spooling protocol.
• A *trino.ErrQueryFailed without a *trino.ErrTrino inside for an HTTP error, such as 401 Unauthorized, or when the retries of a request were exhausted, see request_retry_timeout.
• A decoding error when a response is not well-formed JSON, which includes invalid UTF-8 and a member name repeated within one object, or when a value does not match the type of its column, naming the row and the column. A malformed first page fails Query, a later one rows.Err().

Cancellation and timeouts

A query runs until its results are consumed, its context is cancelled or its timeout passes. When the context of a QueryContext or ExecContext call is cancelled or reaches its deadline, the driver cancels the query on the server with a DELETE request, giving it trino.DefaultCancelQueryTimeout (30s), and rows.Err() returns the context error. Closing the rows before the last page was read cancels the query the same way, so a program that stops reading early does not leave the query running. A query whose context has no deadline gets query_timeout, or trino.DefaultQueryTimeout (10h) when that is not set.

Spooling protocol

The driver supports the Trino spooling protocol, which enables efficient retrieval of large result sets by downloading data in segments, optionally in parallel and out-of-order.

If the Trino server has the spooling protocol enabled, the driver uses it by default with the json encoding. You can request other encodings for a query with the encoding named argument, as one encoding or a list in order of preference:

• Supported encodings: json, json+lz4, json+zstd

rows, err := db.Query(query, sql.Named("encoding", "json+zstd"))
rows, err := db.Query(query, sql.Named("encoding", "json+zstd, json+lz4, json"))

While a spooled query is in progress, the driver sends periodic HEAD requests to the current nextUri (same URL used for result pages) so the coordinator treats the client as still active, which helps avoid query abandonment when result consumption is slow. The server must support this endpoint (Trino 475+); see heartbeat_interval.

Segment downloads follow redirects, which the coordinator answers them with in the coordinator_storage_redirect and worker_proxy values of protocol.spooling.retrieval-mode, sending the client to the storage or to a worker. A segment request carries only the headers the server listed for the segment, never the X-Trino-* session headers or credentials, so following the redirect does not leak them; the redirect target needs the segment headers, such as the encryption key, so they are kept. The CheckRedirect policy of an HTTPClient or custom_client applies, and the default client stops after 10 redirects.

Spooled segments stay in storage only until the expiresAt time the server reports for each of them, so rows read too slowly can outlive their segments. When a segment download fails after that time, the error is a *trino.SegmentExpiredError carrying the expiration time and wrapping the download error, so errors.As still finds the *trino.ErrQueryFailed with the HTTP status. A download that fails before the segment expires returns the download error as before.

Configuration options

You can tune the spooling protocol using the following parameters, passed as sql.Named arguments to your query:

• Spooling worker count sql.Named("spooling_worker_count", "N") Sets the number of parallel workers used to download spooled segments. Default: 5 Considerations:

  • Increasing this value can improve throughput for large result sets, especially on high-latency networks.
  • Higher values increase parallelism but may also increase memory usage.

• Max out-of-order segments sql.Named("max_out_of_order_segments", "N") Sets the maximum number of segments that can be downloaded and buffered out-of-order before blocking further downloads. Default: 10 Considerations:

  • Higher values increase the potential memory usage, but actual usage depends on download behavior and may be lower in practice.
  • Higher values reduce the chance that one slow or stalled segment will block the download of additional segments.
  • Lower values reduce memory usage but may limit parallelism and throughput.

It is not allowed to set spooling_worker_count higher than max_out_of_order_segments; doing so results in an error. Each download worker must reserve a slot for the segment it fetches, and a slot is only released when that segment can be processed in order. The total number of slots corresponds to max_out_of_order_segments. If you configure more workers than allowed out-of-order segments, the extra workers would immediately block while waiting for a slot, defeating the purpose of parallelism and potentially wasting resources.

rows, err := db.Query(
	query,
	sql.Named("encoding", "json+zstd"),
	sql.Named("spooling_worker_count", "8"),
	sql.Named("max_out_of_order_segments", "20"),
)

Networking

Custom HTTP client

Beyond the TLS and proxy parameters, the driver can send every request through an http.Client of your own, with tuned connection pools, timeouts, instrumentation or an authenticating transport. Pass it as Config.HTTPClient to NewConnector, or register it under a name with trino.RegisterCustomClient and select it with the custom_client DSN parameter:

foobarClient := &http.Client{
	Transport: &http.Transport{
		Proxy: http.ProxyFromEnvironment,
		DialContext: (&net.Dialer{
			Timeout:   30 * time.Second,
			KeepAlive: 30 * time.Second,
		}).DialContext,
		MaxIdleConns:          100,
		IdleConnTimeout:       90 * time.Second,
		TLSHandshakeTimeout:   10 * time.Second,
		ExpectContinueTimeout: 1 * time.Second,
		TLSClientConfig: &tls.Config{
			// your config here...
		},
	},
}
trino.RegisterCustomClient("foobar", foobarClient)
db, err := sql.Open("trino", "https://user@localhost:8443?custom_client=foobar")

The default client does not follow HTTP redirects, because a redirect would carry the X-Trino-* headers, including extra credentials, to a host other than the one in the DSN, and a 301, 302 or 303 response would turn the statement POST into a GET without the query. A redirect response fails the query instead. An HTTPClient is treated the same way, while a registered custom_client follows its own CheckRedirect policy, which allows redirects unless set. Spooled segment downloads follow redirects with every client, see Spooling protocol.

A custom client can also add OpenTelemetry instrumentation. The otelhttp package provides a transport wrapper that creates spans for HTTP requests and propagates the trace ID in HTTP headers:

otelClient := &http.Client{
	Transport: otelhttp.NewTransport(http.DefaultTransport),
}
trino.RegisterCustomClient("otel", otelClient)
db, err := sql.Open("trino", "https://user@localhost:8443?custom_client=otel")

Proxy

By default the driver takes its proxy from the HTTP_PROXY, HTTPS_PROXY and NO_PROXY environment variables, as described for http.ProxyFromEnvironment. Requests to localhost and loopback addresses never use them. To send every request through a specific proxy instead, set the httpProxy or socksProxy parameter, or Config.HTTPProxy or Config.SOCKSProxy:

db, err := sql.Open("trino", "https://user@trino.example.com:8443?httpProxy=proxy.example.com:3128")

An HTTPClient or a custom_client uses its own transport, so its proxy must be configured there, for example with http.ProxyURL; the driver rejects httpProxy and socksProxy combined with either.

DNS resolution

There is no DSN parameter for DNS resolution. To resolve the coordinator's host name differently, for example with a specific DNS server, give the transport of an HTTPClient a dialer with its own net.Resolver:

resolver := &net.Resolver{
	PreferGo: true,
	Dial: func(ctx context.Context, network, _ string) (net.Conn, error) {
		return (&net.Dialer{}).DialContext(ctx, network, "10.0.0.53:53")
	},
}
transport := http.DefaultTransport.(*http.Transport).Clone()
transport.DialContext = (&net.Dialer{Resolver: resolver}).DialContext
connector, err := trino.NewConnector(&trino.Config{
	ServerURI:  "https://user@trino.example.com:8443",
	HTTPClient: &http.Client{Transport: transport},
})

A dialer that maps host names to addresses on its own can replace DialContext in the same way.

License

Apache License V2.0, as described in the LICENSE file.

Build

You can build the client code locally and run the unit tests, which need no server, with the following command:

go test -v -race ./...

The integration tests, which start Trino in Docker, live in the integration module. See CONTRIBUTING.md for how to run them.

Contributing

For contributing, development, and release guidelines, see CONTRIBUTING.md.