switch from github.com/lib/pq to github.com/jackc/pgx/v5HEAD20240905master

lib/pq is out of maintenance for some time now, so switch to the newer
more active library. Looks like it finally stabilized after a long time.

1 files changed, 2346 insertions, 0 deletions

diff --git a/vendor/github.com/jackc/pgx/v5/pgconn/pgconn.go b/vendor/github.com/jackc/pgx/v5/pgconn/pgconn.go
new file mode 100644
index 0000000..7efb522
--- /dev/null
+++ b/vendor/github.com/jackc/pgx/v5/pgconn/pgconn.go
@@ -0,0 +1,2346 @@
+package pgconn
+
+import (
+	"context"
+	"crypto/md5"
+	"crypto/tls"
+	"encoding/binary"
+	"encoding/hex"
+	"errors"
+	"fmt"
+	"io"
+	"math"
+	"net"
+	"strconv"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/jackc/pgx/v5/internal/iobufpool"
+	"github.com/jackc/pgx/v5/internal/pgio"
+	"github.com/jackc/pgx/v5/pgconn/ctxwatch"
+	"github.com/jackc/pgx/v5/pgconn/internal/bgreader"
+	"github.com/jackc/pgx/v5/pgproto3"
+)
+
+const (
+	connStatusUninitialized = iota
+	connStatusConnecting
+	connStatusClosed
+	connStatusIdle
+	connStatusBusy
+)
+
+// Notice represents a notice response message reported by the PostgreSQL server. Be aware that this is distinct from
+// LISTEN/NOTIFY notification.
+type Notice PgError
+
+// Notification is a message received from the PostgreSQL LISTEN/NOTIFY system
+type Notification struct {
+	PID     uint32 // backend pid that sent the notification
+	Channel string // channel from which notification was received
+	Payload string
+}
+
+// DialFunc is a function that can be used to connect to a PostgreSQL server.
+type DialFunc func(ctx context.Context, network, addr string) (net.Conn, error)
+
+// LookupFunc is a function that can be used to lookup IPs addrs from host. Optionally an ip:port combination can be
+// returned in order to override the connection string's port.
+type LookupFunc func(ctx context.Context, host string) (addrs []string, err error)
+
+// BuildFrontendFunc is a function that can be used to create Frontend implementation for connection.
+type BuildFrontendFunc func(r io.Reader, w io.Writer) *pgproto3.Frontend
+
+// PgErrorHandler is a function that handles errors returned from Postgres. This function must return true to keep
+// the connection open. Returning false will cause the connection to be closed immediately. You should return
+// false on any FATAL-severity errors. This will not receive network errors. The *PgConn is provided so the handler is
+// aware of the origin of the error, but it must not invoke any query method.
+type PgErrorHandler func(*PgConn, *PgError) bool
+
+// NoticeHandler is a function that can handle notices received from the PostgreSQL server. Notices can be received at
+// any time, usually during handling of a query response. The *PgConn is provided so the handler is aware of the origin
+// of the notice, but it must not invoke any query method. Be aware that this is distinct from LISTEN/NOTIFY
+// notification.
+type NoticeHandler func(*PgConn, *Notice)
+
+// NotificationHandler is a function that can handle notifications received from the PostgreSQL server. Notifications
+// can be received at any time, usually during handling of a query response. The *PgConn is provided so the handler is
+// aware of the origin of the notice, but it must not invoke any query method. Be aware that this is distinct from a
+// notice event.
+type NotificationHandler func(*PgConn, *Notification)
+
+// PgConn is a low-level PostgreSQL connection handle. It is not safe for concurrent usage.
+type PgConn struct {
+	conn              net.Conn
+	pid               uint32            // backend pid
+	secretKey         uint32            // key to use to send a cancel query message to the server
+	parameterStatuses map[string]string // parameters that have been reported by the server
+	txStatus          byte
+	frontend          *pgproto3.Frontend
+	bgReader          *bgreader.BGReader
+	slowWriteTimer    *time.Timer
+	bgReaderStarted   chan struct{}
+
+	customData map[string]any
+
+	config *Config
+
+	status byte // One of connStatus* constants
+
+	bufferingReceive    bool
+	bufferingReceiveMux sync.Mutex
+	bufferingReceiveMsg pgproto3.BackendMessage
+	bufferingReceiveErr error
+
+	peekedMsg pgproto3.BackendMessage
+
+	// Reusable / preallocated resources
+	resultReader      ResultReader
+	multiResultReader MultiResultReader
+	pipeline          Pipeline
+	contextWatcher    *ctxwatch.ContextWatcher
+	fieldDescriptions [16]FieldDescription
+
+	cleanupDone chan struct{}
+}
+
+// Connect establishes a connection to a PostgreSQL server using the environment and connString (in URL or keyword/value
+// format) to provide configuration. See documentation for [ParseConfig] for details. ctx can be used to cancel a
+// connect attempt.
+func Connect(ctx context.Context, connString string) (*PgConn, error) {
+	config, err := ParseConfig(connString)
+	if err != nil {
+		return nil, err
+	}
+
+	return ConnectConfig(ctx, config)
+}
+
+// Connect establishes a connection to a PostgreSQL server using the environment and connString (in URL or keyword/value
+// format) and ParseConfigOptions to provide additional configuration. See documentation for [ParseConfig] for details.
+// ctx can be used to cancel a connect attempt.
+func ConnectWithOptions(ctx context.Context, connString string, parseConfigOptions ParseConfigOptions) (*PgConn, error) {
+	config, err := ParseConfigWithOptions(connString, parseConfigOptions)
+	if err != nil {
+		return nil, err
+	}
+
+	return ConnectConfig(ctx, config)
+}
+
+// Connect establishes a connection to a PostgreSQL server using config. config must have been constructed with
+// [ParseConfig]. ctx can be used to cancel a connect attempt.
+//
+// If config.Fallbacks are present they will sequentially be tried in case of error establishing network connection. An
+// authentication error will terminate the chain of attempts (like libpq:
+// https://www.postgresql.org/docs/11/libpq-connect.html#LIBPQ-MULTIPLE-HOSTS) and be returned as the error.
+func ConnectConfig(ctx context.Context, config *Config) (*PgConn, error) {
+	// Default values are set in ParseConfig. Enforce initial creation by ParseConfig rather than setting defaults from
+	// zero values.
+	if !config.createdByParseConfig {
+		panic("config must be created by ParseConfig")
+	}
+
+	var allErrors []error
+
+	connectConfigs, errs := buildConnectOneConfigs(ctx, config)
+	if len(errs) > 0 {
+		allErrors = append(allErrors, errs...)
+	}
+
+	if len(connectConfigs) == 0 {
+		return nil, &ConnectError{Config: config, err: fmt.Errorf("hostname resolving error: %w", errors.Join(allErrors...))}
+	}
+
+	pgConn, errs := connectPreferred(ctx, config, connectConfigs)
+	if len(errs) > 0 {
+		allErrors = append(allErrors, errs...)
+		return nil, &ConnectError{Config: config, err: errors.Join(allErrors...)}
+	}
+
+	if config.AfterConnect != nil {
+		err := config.AfterConnect(ctx, pgConn)
+		if err != nil {
+			pgConn.conn.Close()
+			return nil, &ConnectError{Config: config, err: fmt.Errorf("AfterConnect error: %w", err)}
+		}
+	}
+
+	return pgConn, nil
+}
+
+// buildConnectOneConfigs resolves hostnames and builds a list of connectOneConfigs to try connecting to. It returns a
+// slice of successfully resolved connectOneConfigs and a slice of errors. It is possible for both slices to contain
+// values if some hosts were successfully resolved and others were not.
+func buildConnectOneConfigs(ctx context.Context, config *Config) ([]*connectOneConfig, []error) {
+	// Simplify usage by treating primary config and fallbacks the same.
+	fallbackConfigs := []*FallbackConfig{
+		{
+			Host:      config.Host,
+			Port:      config.Port,
+			TLSConfig: config.TLSConfig,
+		},
+	}
+	fallbackConfigs = append(fallbackConfigs, config.Fallbacks...)
+
+	var configs []*connectOneConfig
+
+	var allErrors []error
+
+	for _, fb := range fallbackConfigs {
+		// skip resolve for unix sockets
+		if isAbsolutePath(fb.Host) {
+			network, address := NetworkAddress(fb.Host, fb.Port)
+			configs = append(configs, &connectOneConfig{
+				network:          network,
+				address:          address,
+				originalHostname: fb.Host,
+				tlsConfig:        fb.TLSConfig,
+			})
+
+			continue
+		}
+
+		ips, err := config.LookupFunc(ctx, fb.Host)
+		if err != nil {
+			allErrors = append(allErrors, err)
+			continue
+		}
+
+		for _, ip := range ips {
+			splitIP, splitPort, err := net.SplitHostPort(ip)
+			if err == nil {
+				port, err := strconv.ParseUint(splitPort, 10, 16)
+				if err != nil {
+					return nil, []error{fmt.Errorf("error parsing port (%s) from lookup: %w", splitPort, err)}
+				}
+				network, address := NetworkAddress(splitIP, uint16(port))
+				configs = append(configs, &connectOneConfig{
+					network:          network,
+					address:          address,
+					originalHostname: fb.Host,
+					tlsConfig:        fb.TLSConfig,
+				})
+			} else {
+				network, address := NetworkAddress(ip, fb.Port)
+				configs = append(configs, &connectOneConfig{
+					network:          network,
+					address:          address,
+					originalHostname: fb.Host,
+					tlsConfig:        fb.TLSConfig,
+				})
+			}
+		}
+	}
+
+	return configs, allErrors
+}
+
+// connectPreferred attempts to connect to the preferred host from connectOneConfigs. The connections are attempted in
+// order. If a connection is successful it is returned. If no connection is successful then all errors are returned. If
+// a connection attempt returns a [NotPreferredError], then that host will be used if no other hosts are successful.
+func connectPreferred(ctx context.Context, config *Config, connectOneConfigs []*connectOneConfig) (*PgConn, []error) {
+	octx := ctx
+	var allErrors []error
+
+	var fallbackConnectOneConfig *connectOneConfig
+	for i, c := range connectOneConfigs {
+		// ConnectTimeout restricts the whole connection process.
+		if config.ConnectTimeout != 0 {
+			// create new context first time or when previous host was different
+			if i == 0 || (connectOneConfigs[i].address != connectOneConfigs[i-1].address) {
+				var cancel context.CancelFunc
+				ctx, cancel = context.WithTimeout(octx, config.ConnectTimeout)
+				defer cancel()
+			}
+		} else {
+			ctx = octx
+		}
+
+		pgConn, err := connectOne(ctx, config, c, false)
+		if pgConn != nil {
+			return pgConn, nil
+		}
+
+		allErrors = append(allErrors, err)
+
+		var pgErr *PgError
+		if errors.As(err, &pgErr) {
+			const ERRCODE_INVALID_PASSWORD = "28P01"                    // wrong password
+			const ERRCODE_INVALID_AUTHORIZATION_SPECIFICATION = "28000" // wrong password or bad pg_hba.conf settings
+			const ERRCODE_INVALID_CATALOG_NAME = "3D000"                // db does not exist
+			const ERRCODE_INSUFFICIENT_PRIVILEGE = "42501"              // missing connect privilege
+			if pgErr.Code == ERRCODE_INVALID_PASSWORD ||
+				pgErr.Code == ERRCODE_INVALID_AUTHORIZATION_SPECIFICATION && c.tlsConfig != nil ||
+				pgErr.Code == ERRCODE_INVALID_CATALOG_NAME ||
+				pgErr.Code == ERRCODE_INSUFFICIENT_PRIVILEGE {
+				return nil, allErrors
+			}
+		}
+
+		var npErr *NotPreferredError
+		if errors.As(err, &npErr) {
+			fallbackConnectOneConfig = c
+		}
+	}
+
+	if fallbackConnectOneConfig != nil {
+		pgConn, err := connectOne(ctx, config, fallbackConnectOneConfig, true)
+		if err == nil {
+			return pgConn, nil
+		}
+		allErrors = append(allErrors, err)
+	}
+
+	return nil, allErrors
+}
+
+// connectOne makes one connection attempt to a single host.
+func connectOne(ctx context.Context, config *Config, connectConfig *connectOneConfig,
+	ignoreNotPreferredErr bool,
+) (*PgConn, error) {
+	pgConn := new(PgConn)
+	pgConn.config = config
+	pgConn.cleanupDone = make(chan struct{})
+	pgConn.customData = make(map[string]any)
+
+	var err error
+
+	newPerDialConnectError := func(msg string, err error) *perDialConnectError {
+		err = normalizeTimeoutError(ctx, err)
+		e := &perDialConnectError{address: connectConfig.address, originalHostname: connectConfig.originalHostname, err: fmt.Errorf("%s: %w", msg, err)}
+		return e
+	}
+
+	pgConn.conn, err = config.DialFunc(ctx, connectConfig.network, connectConfig.address)
+	if err != nil {
+		return nil, newPerDialConnectError("dial error", err)
+	}
+
+	if connectConfig.tlsConfig != nil {
+		pgConn.contextWatcher = ctxwatch.NewContextWatcher(&DeadlineContextWatcherHandler{Conn: pgConn.conn})
+		pgConn.contextWatcher.Watch(ctx)
+		tlsConn, err := startTLS(pgConn.conn, connectConfig.tlsConfig)
+		pgConn.contextWatcher.Unwatch() // Always unwatch `netConn` after TLS.
+		if err != nil {
+			pgConn.conn.Close()
+			return nil, newPerDialConnectError("tls error", err)
+		}
+
+		pgConn.conn = tlsConn
+	}
+
+	pgConn.contextWatcher = ctxwatch.NewContextWatcher(config.BuildContextWatcherHandler(pgConn))
+	pgConn.contextWatcher.Watch(ctx)
+	defer pgConn.contextWatcher.Unwatch()
+
+	pgConn.parameterStatuses = make(map[string]string)
+	pgConn.status = connStatusConnecting
+	pgConn.bgReader = bgreader.New(pgConn.conn)
+	pgConn.slowWriteTimer = time.AfterFunc(time.Duration(math.MaxInt64),
+		func() {
+			pgConn.bgReader.Start()
+			pgConn.bgReaderStarted <- struct{}{}
+		},
+	)
+	pgConn.slowWriteTimer.Stop()
+	pgConn.bgReaderStarted = make(chan struct{})
+	pgConn.frontend = config.BuildFrontend(pgConn.bgReader, pgConn.conn)
+
+	startupMsg := pgproto3.StartupMessage{
+		ProtocolVersion: pgproto3.ProtocolVersionNumber,
+		Parameters:      make(map[string]string),
+	}
+
+	// Copy default run-time params
+	for k, v := range config.RuntimeParams {
+		startupMsg.Parameters[k] = v
+	}
+
+	startupMsg.Parameters["user"] = config.User
+	if config.Database != "" {
+		startupMsg.Parameters["database"] = config.Database
+	}
+
+	pgConn.frontend.Send(&startupMsg)
+	if err := pgConn.flushWithPotentialWriteReadDeadlock(); err != nil {
+		pgConn.conn.Close()
+		return nil, newPerDialConnectError("failed to write startup message", err)
+	}
+
+	for {
+		msg, err := pgConn.receiveMessage()
+		if err != nil {
+			pgConn.conn.Close()
+			if err, ok := err.(*PgError); ok {
+				return nil, newPerDialConnectError("server error", err)
+			}
+			return nil, newPerDialConnectError("failed to receive message", err)
+		}
+
+		switch msg := msg.(type) {
+		case *pgproto3.BackendKeyData:
+			pgConn.pid = msg.ProcessID
+			pgConn.secretKey = msg.SecretKey
+
+		case *pgproto3.AuthenticationOk:
+		case *pgproto3.AuthenticationCleartextPassword:
+			err = pgConn.txPasswordMessage(pgConn.config.Password)
+			if err != nil {
+				pgConn.conn.Close()
+				return nil, newPerDialConnectError("failed to write password message", err)
+			}
+		case *pgproto3.AuthenticationMD5Password:
+			digestedPassword := "md5" + hexMD5(hexMD5(pgConn.config.Password+pgConn.config.User)+string(msg.Salt[:]))
+			err = pgConn.txPasswordMessage(digestedPassword)
+			if err != nil {
+				pgConn.conn.Close()
+				return nil, newPerDialConnectError("failed to write password message", err)
+			}
+		case *pgproto3.AuthenticationSASL:
+			err = pgConn.scramAuth(msg.AuthMechanisms)
+			if err != nil {
+				pgConn.conn.Close()
+				return nil, newPerDialConnectError("failed SASL auth", err)
+			}
+		case *pgproto3.AuthenticationGSS:
+			err = pgConn.gssAuth()
+			if err != nil {
+				pgConn.conn.Close()
+				return nil, newPerDialConnectError("failed GSS auth", err)
+			}
+		case *pgproto3.ReadyForQuery:
+			pgConn.status = connStatusIdle
+			if config.ValidateConnect != nil {
+				// ValidateConnect may execute commands that cause the context to be watched again. Unwatch first to avoid
+				// the watch already in progress panic. This is that last thing done by this method so there is no need to
+				// restart the watch after ValidateConnect returns.
+				//
+				// See https://github.com/jackc/pgconn/issues/40.
+				pgConn.contextWatcher.Unwatch()
+
+				err := config.ValidateConnect(ctx, pgConn)
+				if err != nil {
+					if _, ok := err.(*NotPreferredError); ignoreNotPreferredErr && ok {
+						return pgConn, nil
+					}
+					pgConn.conn.Close()
+					return nil, newPerDialConnectError("ValidateConnect failed", err)
+				}
+			}
+			return pgConn, nil
+		case *pgproto3.ParameterStatus, *pgproto3.NoticeResponse:
+			// handled by ReceiveMessage
+		case *pgproto3.ErrorResponse:
+			pgConn.conn.Close()
+			return nil, newPerDialConnectError("server error", ErrorResponseToPgError(msg))
+		default:
+			pgConn.conn.Close()
+			return nil, newPerDialConnectError("received unexpected message", err)
+		}
+	}
+}
+
+func startTLS(conn net.Conn, tlsConfig *tls.Config) (net.Conn, error) {
+	err := binary.Write(conn, binary.BigEndian, []int32{8, 80877103})
+	if err != nil {
+		return nil, err
+	}
+
+	response := make([]byte, 1)
+	if _, err = io.ReadFull(conn, response); err != nil {
+		return nil, err
+	}
+
+	if response[0] != 'S' {
+		return nil, errors.New("server refused TLS connection")
+	}
+
+	return tls.Client(conn, tlsConfig), nil
+}
+
+func (pgConn *PgConn) txPasswordMessage(password string) (err error) {
+	pgConn.frontend.Send(&pgproto3.PasswordMessage{Password: password})
+	return pgConn.flushWithPotentialWriteReadDeadlock()
+}
+
+func hexMD5(s string) string {
+	hash := md5.New()
+	io.WriteString(hash, s)
+	return hex.EncodeToString(hash.Sum(nil))
+}
+
+func (pgConn *PgConn) signalMessage() chan struct{} {
+	if pgConn.bufferingReceive {
+		panic("BUG: signalMessage when already in progress")
+	}
+
+	pgConn.bufferingReceive = true
+	pgConn.bufferingReceiveMux.Lock()
+
+	ch := make(chan struct{})
+	go func() {
+		pgConn.bufferingReceiveMsg, pgConn.bufferingReceiveErr = pgConn.frontend.Receive()
+		pgConn.bufferingReceiveMux.Unlock()
+		close(ch)
+	}()
+
+	return ch
+}
+
+// ReceiveMessage receives one wire protocol message from the PostgreSQL server. It must only be used when the
+// connection is not busy. e.g. It is an error to call ReceiveMessage while reading the result of a query. The messages
+// are still handled by the core pgconn message handling system so receiving a NotificationResponse will still trigger
+// the OnNotification callback.
+//
+// This is a very low level method that requires deep understanding of the PostgreSQL wire protocol to use correctly.
+// See https://www.postgresql.org/docs/current/protocol.html.
+func (pgConn *PgConn) ReceiveMessage(ctx context.Context) (pgproto3.BackendMessage, error) {
+	if err := pgConn.lock(); err != nil {
+		return nil, err
+	}
+	defer pgConn.unlock()
+
+	if ctx != context.Background() {
+		select {
+		case <-ctx.Done():
+			return nil, newContextAlreadyDoneError(ctx)
+		default:
+		}
+		pgConn.contextWatcher.Watch(ctx)
+		defer pgConn.contextWatcher.Unwatch()
+	}
+
+	msg, err := pgConn.receiveMessage()
+	if err != nil {
+		err = &pgconnError{
+			msg:         "receive message failed",
+			err:         normalizeTimeoutError(ctx, err),
+			safeToRetry: true,
+		}
+	}
+	return msg, err
+}
+
+// peekMessage peeks at the next message without setting up context cancellation.
+func (pgConn *PgConn) peekMessage() (pgproto3.BackendMessage, error) {
+	if pgConn.peekedMsg != nil {
+		return pgConn.peekedMsg, nil
+	}
+
+	var msg pgproto3.BackendMessage
+	var err error
+	if pgConn.bufferingReceive {
+		pgConn.bufferingReceiveMux.Lock()
+		msg = pgConn.bufferingReceiveMsg
+		err = pgConn.bufferingReceiveErr
+		pgConn.bufferingReceiveMux.Unlock()
+		pgConn.bufferingReceive = false
+
+		// If a timeout error happened in the background try the read again.
+		var netErr net.Error
+		if errors.As(err, &netErr) && netErr.Timeout() {
+			msg, err = pgConn.frontend.Receive()
+		}
+	} else {
+		msg, err = pgConn.frontend.Receive()
+	}
+
+	if err != nil {
+		// Close on anything other than timeout error - everything else is fatal
+		var netErr net.Error
+		isNetErr := errors.As(err, &netErr)
+		if !(isNetErr && netErr.Timeout()) {
+			pgConn.asyncClose()
+		}
+
+		return nil, err
+	}
+
+	pgConn.peekedMsg = msg
+	return msg, nil
+}
+
+// receiveMessage receives a message without setting up context cancellation
+func (pgConn *PgConn) receiveMessage() (pgproto3.BackendMessage, error) {
+	msg, err := pgConn.peekMessage()
+	if err != nil {
+		return nil, err
+	}
+	pgConn.peekedMsg = nil
+
+	switch msg := msg.(type) {
+	case *pgproto3.ReadyForQuery:
+		pgConn.txStatus = msg.TxStatus
+	case *pgproto3.ParameterStatus:
+		pgConn.parameterStatuses[msg.Name] = msg.Value
+	case *pgproto3.ErrorResponse:
+		err := ErrorResponseToPgError(msg)
+		if pgConn.config.OnPgError != nil && !pgConn.config.OnPgError(pgConn, err) {
+			pgConn.status = connStatusClosed
+			pgConn.conn.Close() // Ignore error as the connection is already broken and there is already an error to return.
+			close(pgConn.cleanupDone)
+			return nil, err
+		}
+	case *pgproto3.NoticeResponse:
+		if pgConn.config.OnNotice != nil {
+			pgConn.config.OnNotice(pgConn, noticeResponseToNotice(msg))
+		}
+	case *pgproto3.NotificationResponse:
+		if pgConn.config.OnNotification != nil {
+			pgConn.config.OnNotification(pgConn, &Notification{PID: msg.PID, Channel: msg.Channel, Payload: msg.Payload})
+		}
+	}
+
+	return msg, nil
+}
+
+// Conn returns the underlying net.Conn. This rarely necessary. If the connection will be directly used for reading or
+// writing then SyncConn should usually be called before Conn.
+func (pgConn *PgConn) Conn() net.Conn {
+	return pgConn.conn
+}
+
+// PID returns the backend PID.
+func (pgConn *PgConn) PID() uint32 {
+	return pgConn.pid
+}
+
+// TxStatus returns the current TxStatus as reported by the server in the ReadyForQuery message.
+//
+// Possible return values:
+//
+//	'I' - idle / not in transaction
+//	'T' - in a transaction
+//	'E' - in a failed transaction
+//
+// See https://www.postgresql.org/docs/current/protocol-message-formats.html.
+func (pgConn *PgConn) TxStatus() byte {
+	return pgConn.txStatus
+}
+
+// SecretKey returns the backend secret key used to send a cancel query message to the server.
+func (pgConn *PgConn) SecretKey() uint32 {
+	return pgConn.secretKey
+}
+
+// Frontend returns the underlying *pgproto3.Frontend. This rarely necessary.
+func (pgConn *PgConn) Frontend() *pgproto3.Frontend {
+	return pgConn.frontend
+}
+
+// Close closes a connection. It is safe to call Close on an already closed connection. Close attempts a clean close by
+// sending the exit message to PostgreSQL. However, this could block so ctx is available to limit the time to wait. The
+// underlying net.Conn.Close() will always be called regardless of any other errors.
+func (pgConn *PgConn) Close(ctx context.Context) error {
+	if pgConn.status == connStatusClosed {
+		return nil
+	}
+	pgConn.status = connStatusClosed
+
+	defer close(pgConn.cleanupDone)
+	defer pgConn.conn.Close()
+
+	if ctx != context.Background() {
+		// Close may be called while a cancellable query is in progress. This will most often be triggered by panic when
+		// a defer closes the connection (possibly indirectly via a transaction or a connection pool). Unwatch to end any
+		// previous watch. It is safe to Unwatch regardless of whether a watch is already is progress.
+		//
+		// See https://github.com/jackc/pgconn/issues/29
+		pgConn.contextWatcher.Unwatch()
+
+		pgConn.contextWatcher.Watch(ctx)
+		defer pgConn.contextWatcher.Unwatch()
+	}
+
+	// Ignore any errors sending Terminate message and waiting for server to close connection.
+	// This mimics the behavior of libpq PQfinish. It calls closePGconn which calls sendTerminateConn which purposefully
+	// ignores errors.
+	//
+	// See https://github.com/jackc/pgx/issues/637
+	pgConn.frontend.Send(&pgproto3.Terminate{})
+	pgConn.flushWithPotentialWriteReadDeadlock()
+
+	return pgConn.conn.Close()
+}
+
+// asyncClose marks the connection as closed and asynchronously sends a cancel query message and closes the underlying
+// connection.
+func (pgConn *PgConn) asyncClose() {
+	if pgConn.status == connStatusClosed {
+		return
+	}
+	pgConn.status = connStatusClosed
+
+	go func() {
+		defer close(pgConn.cleanupDone)
+		defer pgConn.conn.Close()
+
+		deadline := time.Now().Add(time.Second * 15)
+
+		ctx, cancel := context.WithDeadline(context.Background(), deadline)
+		defer cancel()
+
+		pgConn.CancelRequest(ctx)
+
+		pgConn.conn.SetDeadline(deadline)
+
+		pgConn.frontend.Send(&pgproto3.Terminate{})
+		pgConn.flushWithPotentialWriteReadDeadlock()
+	}()
+}
+
+// CleanupDone returns a channel that will be closed after all underlying resources have been cleaned up. A closed
+// connection is no longer usable, but underlying resources, in particular the net.Conn, may not have finished closing
+// yet. This is because certain errors such as a context cancellation require that the interrupted function call return
+// immediately, but the error may also cause the connection to be closed. In these cases the underlying resources are
+// closed asynchronously.
+//
+// This is only likely to be useful to connection pools. It gives them a way avoid establishing a new connection while
+// an old connection is still being cleaned up and thereby exceeding the maximum pool size.
+func (pgConn *PgConn) CleanupDone() chan (struct{}) {
+	return pgConn.cleanupDone
+}
+
+// IsClosed reports if the connection has been closed.
+//
+// CleanupDone() can be used to determine if all cleanup has been completed.
+func (pgConn *PgConn) IsClosed() bool {
+	return pgConn.status < connStatusIdle
+}
+
+// IsBusy reports if the connection is busy.
+func (pgConn *PgConn) IsBusy() bool {
+	return pgConn.status == connStatusBusy
+}
+
+// lock locks the connection.
+func (pgConn *PgConn) lock() error {
+	switch pgConn.status {
+	case connStatusBusy:
+		return &connLockError{status: "conn busy"} // This only should be possible in case of an application bug.
+	case connStatusClosed:
+		return &connLockError{status: "conn closed"}
+	case connStatusUninitialized:
+		return &connLockError{status: "conn uninitialized"}
+	}
+	pgConn.status = connStatusBusy
+	return nil
+}
+
+func (pgConn *PgConn) unlock() {
+	switch pgConn.status {
+	case connStatusBusy:
+		pgConn.status = connStatusIdle
+	case connStatusClosed:
+	default:
+		panic("BUG: cannot unlock unlocked connection") // This should only be possible if there is a bug in this package.
+	}
+}
+
+// ParameterStatus returns the value of a parameter reported by the server (e.g.
+// server_version). Returns an empty string for unknown parameters.
+func (pgConn *PgConn) ParameterStatus(key string) string {
+	return pgConn.parameterStatuses[key]
+}
+
+// CommandTag is the status text returned by PostgreSQL for a query.
+type CommandTag struct {
+	s string
+}
+
+// NewCommandTag makes a CommandTag from s.
+func NewCommandTag(s string) CommandTag {
+	return CommandTag{s: s}
+}
+
+// RowsAffected returns the number of rows affected. If the CommandTag was not
+// for a row affecting command (e.g. "CREATE TABLE") then it returns 0.
+func (ct CommandTag) RowsAffected() int64 {
+	// Find last non-digit
+	idx := -1
+	for i := len(ct.s) - 1; i >= 0; i-- {
+		if ct.s[i] >= '0' && ct.s[i] <= '9' {
+			idx = i
+		} else {
+			break
+		}
+	}
+
+	if idx == -1 {
+		return 0
+	}
+
+	var n int64
+	for _, b := range ct.s[idx:] {
+		n = n*10 + int64(b-'0')
+	}
+
+	return n
+}
+
+func (ct CommandTag) String() string {
+	return ct.s
+}
+
+// Insert is true if the command tag starts with "INSERT".
+func (ct CommandTag) Insert() bool {
+	return strings.HasPrefix(ct.s, "INSERT")
+}
+
+// Update is true if the command tag starts with "UPDATE".
+func (ct CommandTag) Update() bool {
+	return strings.HasPrefix(ct.s, "UPDATE")
+}
+
+// Delete is true if the command tag starts with "DELETE".
+func (ct CommandTag) Delete() bool {
+	return strings.HasPrefix(ct.s, "DELETE")
+}
+
+// Select is true if the command tag starts with "SELECT".
+func (ct CommandTag) Select() bool {
+	return strings.HasPrefix(ct.s, "SELECT")
+}
+
+type FieldDescription struct {
+	Name                 string
+	TableOID             uint32
+	TableAttributeNumber uint16
+	DataTypeOID          uint32
+	DataTypeSize         int16
+	TypeModifier         int32
+	Format               int16
+}
+
+func (pgConn *PgConn) convertRowDescription(dst []FieldDescription, rd *pgproto3.RowDescription) []FieldDescription {
+	if cap(dst) >= len(rd.Fields) {
+		dst = dst[:len(rd.Fields):len(rd.Fields)]
+	} else {
+		dst = make([]FieldDescription, len(rd.Fields))
+	}
+
+	for i := range rd.Fields {
+		dst[i].Name = string(rd.Fields[i].Name)
+		dst[i].TableOID = rd.Fields[i].TableOID
+		dst[i].TableAttributeNumber = rd.Fields[i].TableAttributeNumber
+		dst[i].DataTypeOID = rd.Fields[i].DataTypeOID
+		dst[i].DataTypeSize = rd.Fields[i].DataTypeSize
+		dst[i].TypeModifier = rd.Fields[i].TypeModifier
+		dst[i].Format = rd.Fields[i].Format
+	}
+
+	return dst
+}
+
+type StatementDescription struct {
+	Name      string
+	SQL       string
+	ParamOIDs []uint32
+	Fields    []FieldDescription
+}
+
+// Prepare creates a prepared statement. If the name is empty, the anonymous prepared statement will be used. This
+// allows Prepare to also to describe statements without creating a server-side prepared statement.
+//
+// Prepare does not send a PREPARE statement to the server. It uses the PostgreSQL Parse and Describe protocol messages
+// directly.
+func (pgConn *PgConn) Prepare(ctx context.Context, name, sql string, paramOIDs []uint32) (*StatementDescription, error) {
+	if err := pgConn.lock(); err != nil {
+		return nil, err
+	}
+	defer pgConn.unlock()
+
+	if ctx != context.Background() {
+		select {
+		case <-ctx.Done():
+			return nil, newContextAlreadyDoneError(ctx)
+		default:
+		}
+		pgConn.contextWatcher.Watch(ctx)
+		defer pgConn.contextWatcher.Unwatch()
+	}
+
+	pgConn.frontend.SendParse(&pgproto3.Parse{Name: name, Query: sql, ParameterOIDs: paramOIDs})
+	pgConn.frontend.SendDescribe(&pgproto3.Describe{ObjectType: 'S', Name: name})
+	pgConn.frontend.SendSync(&pgproto3.Sync{})
+	err := pgConn.flushWithPotentialWriteReadDeadlock()
+	if err != nil {
+		pgConn.asyncClose()
+		return nil, err
+	}
+
+	psd := &StatementDescription{Name: name, SQL: sql}
+
+	var parseErr error
+
+readloop:
+	for {
+		msg, err := pgConn.receiveMessage()
+		if err != nil {
+			pgConn.asyncClose()
+			return nil, normalizeTimeoutError(ctx, err)
+		}
+
+		switch msg := msg.(type) {
+		case *pgproto3.ParameterDescription:
+			psd.ParamOIDs = make([]uint32, len(msg.ParameterOIDs))
+			copy(psd.ParamOIDs, msg.ParameterOIDs)
+		case *pgproto3.RowDescription:
+			psd.Fields = pgConn.convertRowDescription(nil, msg)
+		case *pgproto3.ErrorResponse:
+			parseErr = ErrorResponseToPgError(msg)
+		case *pgproto3.ReadyForQuery:
+			break readloop
+		}
+	}
+
+	if parseErr != nil {
+		return nil, parseErr
+	}
+	return psd, nil
+}
+
+// Deallocate deallocates a prepared statement.
+//
+// Deallocate does not send a DEALLOCATE statement to the server. It uses the PostgreSQL Close protocol message
+// directly. This has slightly different behavior than executing DEALLOCATE statement.
+//   - Deallocate can succeed in an aborted transaction.
+//   - Deallocating a non-existent prepared statement is not an error.
+func (pgConn *PgConn) Deallocate(ctx context.Context, name string) error {
+	if err := pgConn.lock(); err != nil {
+		return err
+	}
+	defer pgConn.unlock()
+
+	if ctx != context.Background() {
+		select {
+		case <-ctx.Done():
+			return newContextAlreadyDoneError(ctx)
+		default:
+		}
+		pgConn.contextWatcher.Watch(ctx)
+		defer pgConn.contextWatcher.Unwatch()
+	}
+
+	pgConn.frontend.SendClose(&pgproto3.Close{ObjectType: 'S', Name: name})
+	pgConn.frontend.SendSync(&pgproto3.Sync{})
+	err := pgConn.flushWithPotentialWriteReadDeadlock()
+	if err != nil {
+		pgConn.asyncClose()
+		return err
+	}
+
+	for {
+		msg, err := pgConn.receiveMessage()
+		if err != nil {
+			pgConn.asyncClose()
+			return normalizeTimeoutError(ctx, err)
+		}
+
+		switch msg := msg.(type) {
+		case *pgproto3.ErrorResponse:
+			return ErrorResponseToPgError(msg)
+		case *pgproto3.ReadyForQuery:
+			return nil
+		}
+	}
+}
+
+// ErrorResponseToPgError converts a wire protocol error message to a *PgError.
+func ErrorResponseToPgError(msg *pgproto3.ErrorResponse) *PgError {
+	return &PgError{
+		Severity:            msg.Severity,
+		SeverityUnlocalized: msg.SeverityUnlocalized,
+		Code:                string(msg.Code),
+		Message:             string(msg.Message),
+		Detail:              string(msg.Detail),
+		Hint:                msg.Hint,
+		Position:            msg.Position,
+		InternalPosition:    msg.InternalPosition,
+		InternalQuery:       string(msg.InternalQuery),
+		Where:               string(msg.Where),
+		SchemaName:          string(msg.SchemaName),
+		TableName:           string(msg.TableName),
+		ColumnName:          string(msg.ColumnName),
+		DataTypeName:        string(msg.DataTypeName),
+		ConstraintName:      msg.ConstraintName,
+		File:                string(msg.File),
+		Line:                msg.Line,
+		Routine:             string(msg.Routine),
+	}
+}
+
+func noticeResponseToNotice(msg *pgproto3.NoticeResponse) *Notice {
+	pgerr := ErrorResponseToPgError((*pgproto3.ErrorResponse)(msg))
+	return (*Notice)(pgerr)
+}
+
+// CancelRequest sends a cancel request to the PostgreSQL server. It returns an error if unable to deliver the cancel
+// request, but lack of an error does not ensure that the query was canceled. As specified in the documentation, there
+// is no way to be sure a query was canceled. See https://www.postgresql.org/docs/11/protocol-flow.html#id-1.10.5.7.9
+func (pgConn *PgConn) CancelRequest(ctx context.Context) error {
+	// Open a cancellation request to the same server. The address is taken from the net.Conn directly instead of reusing
+	// the connection config. This is important in high availability configurations where fallback connections may be
+	// specified or DNS may be used to load balance.
+	serverAddr := pgConn.conn.RemoteAddr()
+	var serverNetwork string
+	var serverAddress string
+	if serverAddr.Network() == "unix" {
+		// for unix sockets, RemoteAddr() calls getpeername() which returns the name the
+		// server passed to bind(). For Postgres, this is always a relative path "./.s.PGSQL.5432"
+		// so connecting to it will fail. Fall back to the config's value
+		serverNetwork, serverAddress = NetworkAddress(pgConn.config.Host, pgConn.config.Port)
+	} else {
+		serverNetwork, serverAddress = serverAddr.Network(), serverAddr.String()
+	}
+	cancelConn, err := pgConn.config.DialFunc(ctx, serverNetwork, serverAddress)
+	if err != nil {
+		// In case of unix sockets, RemoteAddr() returns only the file part of the path. If the
+		// first connect failed, try the config.
+		if serverAddr.Network() != "unix" {
+			return err
+		}
+		serverNetwork, serverAddr := NetworkAddress(pgConn.config.Host, pgConn.config.Port)
+		cancelConn, err = pgConn.config.DialFunc(ctx, serverNetwork, serverAddr)
+		if err != nil {
+			return err
+		}
+	}
+	defer cancelConn.Close()
+
+	if ctx != context.Background() {
+		contextWatcher := ctxwatch.NewContextWatcher(&DeadlineContextWatcherHandler{Conn: cancelConn})
+		contextWatcher.Watch(ctx)
+		defer contextWatcher.Unwatch()
+	}
+
+	buf := make([]byte, 16)
+	binary.BigEndian.PutUint32(buf[0:4], 16)
+	binary.BigEndian.PutUint32(buf[4:8], 80877102)
+	binary.BigEndian.PutUint32(buf[8:12], pgConn.pid)
+	binary.BigEndian.PutUint32(buf[12:16], pgConn.secretKey)
+
+	if _, err := cancelConn.Write(buf); err != nil {
+		return fmt.Errorf("write to connection for cancellation: %w", err)
+	}
+
+	// Wait for the cancel request to be acknowledged by the server.
+	// It copies the behavior of the libpq: https://github.com/postgres/postgres/blob/REL_16_0/src/interfaces/libpq/fe-connect.c#L4946-L4960
+	_, _ = cancelConn.Read(buf)
+
+	return nil
+}
+
+// WaitForNotification waits for a LISTEN/NOTIFY message to be received. It returns an error if a notification was not
+// received.
+func (pgConn *PgConn) WaitForNotification(ctx context.Context) error {
+	if err := pgConn.lock(); err != nil {
+		return err
+	}
+	defer pgConn.unlock()
+
+	if ctx != context.Background() {
+		select {
+		case <-ctx.Done():
+			return newContextAlreadyDoneError(ctx)
+		default:
+		}
+
+		pgConn.contextWatcher.Watch(ctx)
+		defer pgConn.contextWatcher.Unwatch()
+	}
+
+	for {
+		msg, err := pgConn.receiveMessage()
+		if err != nil {
+			return normalizeTimeoutError(ctx, err)
+		}
+
+		switch msg.(type) {
+		case *pgproto3.NotificationResponse:
+			return nil
+		}
+	}
+}
+
+// Exec executes SQL via the PostgreSQL simple query protocol. SQL may contain multiple queries. Execution is
+// implicitly wrapped in a transaction unless a transaction is already in progress or SQL contains transaction control
+// statements.
+//
+// Prefer ExecParams unless executing arbitrary SQL that may contain multiple queries.
+func (pgConn *PgConn) Exec(ctx context.Context, sql string) *MultiResultReader {
+	if err := pgConn.lock(); err != nil {
+		return &MultiResultReader{
+			closed: true,
+			err:    err,
+		}
+	}
+
+	pgConn.multiResultReader = MultiResultReader{
+		pgConn: pgConn,
+		ctx:    ctx,
+	}
+	multiResult := &pgConn.multiResultReader
+	if ctx != context.Background() {
+		select {
+		case <-ctx.Done():
+			multiResult.closed = true
+			multiResult.err = newContextAlreadyDoneError(ctx)
+			pgConn.unlock()
+			return multiResult
+		default:
+		}
+		pgConn.contextWatcher.Watch(ctx)
+	}
+
+	pgConn.frontend.SendQuery(&pgproto3.Query{String: sql})
+	err := pgConn.flushWithPotentialWriteReadDeadlock()
+	if err != nil {
+		pgConn.asyncClose()
+		pgConn.contextWatcher.Unwatch()
+		multiResult.closed = true
+		multiResult.err = err
+		pgConn.unlock()
+		return multiResult
+	}
+
+	return multiResult
+}
+
+// ExecParams executes a command via the PostgreSQL extended query protocol.
+//
+// sql is a SQL command string. It may only contain one query. Parameter substitution is positional using $1, $2, $3,
+// etc.
+//
+// paramValues are the parameter values. It must be encoded in the format given by paramFormats.
+//
+// paramOIDs is a slice of data type OIDs for paramValues. If paramOIDs is nil, the server will infer the data type for
+// all parameters. Any paramOID element that is 0 that will cause the server to infer the data type for that parameter.
+// ExecParams will panic if len(paramOIDs) is not 0, 1, or len(paramValues).
+//
+// paramFormats is a slice of format codes determining for each paramValue column whether it is encoded in text or
+// binary format. If paramFormats is nil all params are text format. ExecParams will panic if
+// len(paramFormats) is not 0, 1, or len(paramValues).
+//
+// resultFormats is a slice of format codes determining for each result column whether it is encoded in text or
+// binary format. If resultFormats is nil all results will be in text format.
+//
+// ResultReader must be closed before PgConn can be used again.
+func (pgConn *PgConn) ExecParams(ctx context.Context, sql string, paramValues [][]byte, paramOIDs []uint32, paramFormats []int16, resultFormats []int16) *ResultReader {
+	result := pgConn.execExtendedPrefix(ctx, paramValues)
+	if result.closed {
+		return result
+	}
+
+	pgConn.frontend.SendParse(&pgproto3.Parse{Query: sql, ParameterOIDs: paramOIDs})
+	pgConn.frontend.SendBind(&pgproto3.Bind{ParameterFormatCodes: paramFormats, Parameters: paramValues, ResultFormatCodes: resultFormats})
+
+	pgConn.execExtendedSuffix(result)
+
+	return result
+}
+
+// ExecPrepared enqueues the execution of a prepared statement via the PostgreSQL extended query protocol.
+//
+// paramValues are the parameter values. It must be encoded in the format given by paramFormats.
+//
+// paramFormats is a slice of format codes determining for each paramValue column whether it is encoded in text or
+// binary format. If paramFormats is nil all params are text format. ExecPrepared will panic if
+// len(paramFormats) is not 0, 1, or len(paramValues).
+//
+// resultFormats is a slice of format codes determining for each result column whether it is encoded in text or
+// binary format. If resultFormats is nil all results will be in text format.
+//
+// ResultReader must be closed before PgConn can be used again.
+func (pgConn *PgConn) ExecPrepared(ctx context.Context, stmtName string, paramValues [][]byte, paramFormats []int16, resultFormats []int16) *ResultReader {
+	result := pgConn.execExtendedPrefix(ctx, paramValues)
+	if result.closed {
+		return result
+	}
+
+	pgConn.frontend.SendBind(&pgproto3.Bind{PreparedStatement: stmtName, ParameterFormatCodes: paramFormats, Parameters: paramValues, ResultFormatCodes: resultFormats})
+
+	pgConn.execExtendedSuffix(result)
+
+	return result
+}
+
+func (pgConn *PgConn) execExtendedPrefix(ctx context.Context, paramValues [][]byte) *ResultReader {
+	pgConn.resultReader = ResultReader{
+		pgConn: pgConn,
+		ctx:    ctx,
+	}
+	result := &pgConn.resultReader
+
+	if err := pgConn.lock(); err != nil {
+		result.concludeCommand(CommandTag{}, err)
+		result.closed = true
+		return result
+	}
+
+	if len(paramValues) > math.MaxUint16 {
+		result.concludeCommand(CommandTag{}, fmt.Errorf("extended protocol limited to %v parameters", math.MaxUint16))
+		result.closed = true
+		pgConn.unlock()
+		return result
+	}
+
+	if ctx != context.Background() {
+		select {
+		case <-ctx.Done():
+			result.concludeCommand(CommandTag{}, newContextAlreadyDoneError(ctx))
+			result.closed = true
+			pgConn.unlock()
+			return result
+		default:
+		}
+		pgConn.contextWatcher.Watch(ctx)
+	}
+
+	return result
+}
+
+func (pgConn *PgConn) execExtendedSuffix(result *ResultReader) {
+	pgConn.frontend.SendDescribe(&pgproto3.Describe{ObjectType: 'P'})
+	pgConn.frontend.SendExecute(&pgproto3.Execute{})
+	pgConn.frontend.SendSync(&pgproto3.Sync{})
+
+	err := pgConn.flushWithPotentialWriteReadDeadlock()
+	if err != nil {
+		pgConn.asyncClose()
+		result.concludeCommand(CommandTag{}, err)
+		pgConn.contextWatcher.Unwatch()
+		result.closed = true
+		pgConn.unlock()
+		return
+	}
+
+	result.readUntilRowDescription()
+}
+
+// CopyTo executes the copy command sql and copies the results to w.
+func (pgConn *PgConn) CopyTo(ctx context.Context, w io.Writer, sql string) (CommandTag, error) {
+	if err := pgConn.lock(); err != nil {
+		return CommandTag{}, err
+	}
+
+	if ctx != context.Background() {
+		select {
+		case <-ctx.Done():
+			pgConn.unlock()
+			return CommandTag{}, newContextAlreadyDoneError(ctx)
+		default:
+		}
+		pgConn.contextWatcher.Watch(ctx)
+		defer pgConn.contextWatcher.Unwatch()
+	}
+
+	// Send copy to command
+	pgConn.frontend.SendQuery(&pgproto3.Query{String: sql})
+
+	err := pgConn.flushWithPotentialWriteReadDeadlock()
+	if err != nil {
+		pgConn.asyncClose()
+		pgConn.unlock()
+		return CommandTag{}, err
+	}
+
+	// Read results
+	var commandTag CommandTag
+	var pgErr error
+	for {
+		msg, err := pgConn.receiveMessage()
+		if err != nil {
+			pgConn.asyncClose()
+			return CommandTag{}, normalizeTimeoutError(ctx, err)
+		}
+
+		switch msg := msg.(type) {
+		case *pgproto3.CopyDone:
+		case *pgproto3.CopyData:
+			_, err := w.Write(msg.Data)
+			if err != nil {
+				pgConn.asyncClose()
+				return CommandTag{}, err
+			}
+		case *pgproto3.ReadyForQuery:
+			pgConn.unlock()
+			return commandTag, pgErr
+		case *pgproto3.CommandComplete:
+			commandTag = pgConn.makeCommandTag(msg.CommandTag)
+		case *pgproto3.ErrorResponse:
+			pgErr = ErrorResponseToPgError(msg)
+		}
+	}
+}
+
+// CopyFrom executes the copy command sql and copies all of r to the PostgreSQL server.
+//
+// Note: context cancellation will only interrupt operations on the underlying PostgreSQL network connection. Reads on r
+// could still block.
+func (pgConn *PgConn) CopyFrom(ctx context.Context, r io.Reader, sql string) (CommandTag, error) {
+	if err := pgConn.lock(); err != nil {
+		return CommandTag{}, err
+	}
+	defer pgConn.unlock()
+
+	if ctx != context.Background() {
+		select {
+		case <-ctx.Done():
+			return CommandTag{}, newContextAlreadyDoneError(ctx)
+		default:
+		}
+		pgConn.contextWatcher.Watch(ctx)
+		defer pgConn.contextWatcher.Unwatch()
+	}
+
+	// Send copy from query
+	pgConn.frontend.SendQuery(&pgproto3.Query{String: sql})
+	err := pgConn.flushWithPotentialWriteReadDeadlock()
+	if err != nil {
+		pgConn.asyncClose()
+		return CommandTag{}, err
+	}
+
+	// Send copy data
+	abortCopyChan := make(chan struct{})
+	copyErrChan := make(chan error, 1)
+	signalMessageChan := pgConn.signalMessage()
+	var wg sync.WaitGroup
+	wg.Add(1)
+
+	go func() {
+		defer wg.Done()
+		buf := iobufpool.Get(65536)
+		defer iobufpool.Put(buf)
+		(*buf)[0] = 'd'
+
+		for {
+			n, readErr := r.Read((*buf)[5:cap(*buf)])
+			if n > 0 {
+				*buf = (*buf)[0 : n+5]
+				pgio.SetInt32((*buf)[1:], int32(n+4))
+
+				writeErr := pgConn.frontend.SendUnbufferedEncodedCopyData(*buf)
+				if writeErr != nil {
+					// Write errors are always fatal, but we can't use asyncClose because we are in a different goroutine. Not
+					// setting pgConn.status or closing pgConn.cleanupDone for the same reason.
+					pgConn.conn.Close()
+
+					copyErrChan <- writeErr
+					return
+				}
+			}
+			if readErr != nil {
+				copyErrChan <- readErr
+				return
+			}
+
+			select {
+			case <-abortCopyChan:
+				return
+			default:
+			}
+		}
+	}()
+
+	var pgErr error
+	var copyErr error
+	for copyErr == nil && pgErr == nil {
+		select {
+		case copyErr = <-copyErrChan:
+		case <-signalMessageChan:
+			// If pgConn.receiveMessage encounters an error it will call pgConn.asyncClose. But that is a race condition with
+			// the goroutine. So instead check pgConn.bufferingReceiveErr which will have been set by the signalMessage. If an
+			// error is found then forcibly close the connection without sending the Terminate message.
+			if err := pgConn.bufferingReceiveErr; err != nil {
+				pgConn.status = connStatusClosed
+				pgConn.conn.Close()
+				close(pgConn.cleanupDone)
+				return CommandTag{}, normalizeTimeoutError(ctx, err)
+			}
+			msg, _ := pgConn.receiveMessage()
+
+			switch msg := msg.(type) {
+			case *pgproto3.ErrorResponse:
+				pgErr = ErrorResponseToPgError(msg)
+			default:
+				signalMessageChan = pgConn.signalMessage()
+			}
+		}
+	}
+	close(abortCopyChan)
+	// Make sure io goroutine finishes before writing.
+	wg.Wait()
+
+	if copyErr == io.EOF || pgErr != nil {
+		pgConn.frontend.Send(&pgproto3.CopyDone{})
+	} else {
+		pgConn.frontend.Send(&pgproto3.CopyFail{Message: copyErr.Error()})
+	}
+	err = pgConn.flushWithPotentialWriteReadDeadlock()
+	if err != nil {
+		pgConn.asyncClose()
+		return CommandTag{}, err
+	}
+
+	// Read results
+	var commandTag CommandTag
+	for {
+		msg, err := pgConn.receiveMessage()
+		if err != nil {
+			pgConn.asyncClose()
+			return CommandTag{}, normalizeTimeoutError(ctx, err)
+		}
+
+		switch msg := msg.(type) {
+		case *pgproto3.ReadyForQuery:
+			return commandTag, pgErr
+		case *pgproto3.CommandComplete:
+			commandTag = pgConn.makeCommandTag(msg.CommandTag)
+		case *pgproto3.ErrorResponse:
+			pgErr = ErrorResponseToPgError(msg)
+		}
+	}
+}
+
+// MultiResultReader is a reader for a command that could return multiple results such as Exec or ExecBatch.
+type MultiResultReader struct {
+	pgConn   *PgConn
+	ctx      context.Context
+	pipeline *Pipeline
+
+	rr *ResultReader
+
+	closed bool
+	err    error
+}
+
+// ReadAll reads all available results. Calling ReadAll is mutually exclusive with all other MultiResultReader methods.
+func (mrr *MultiResultReader) ReadAll() ([]*Result, error) {
+	var results []*Result
+
+	for mrr.NextResult() {
+		results = append(results, mrr.ResultReader().Read())
+	}
+	err := mrr.Close()
+
+	return results, err
+}
+
+func (mrr *MultiResultReader) receiveMessage() (pgproto3.BackendMessage, error) {
+	msg, err := mrr.pgConn.receiveMessage()
+	if err != nil {
+		mrr.pgConn.contextWatcher.Unwatch()
+		mrr.err = normalizeTimeoutError(mrr.ctx, err)
+		mrr.closed = true
+		mrr.pgConn.asyncClose()
+		return nil, mrr.err
+	}
+
+	switch msg := msg.(type) {
+	case *pgproto3.ReadyForQuery:
+		mrr.closed = true
+		if mrr.pipeline != nil {
+			mrr.pipeline.expectedReadyForQueryCount--
+		} else {
+			mrr.pgConn.contextWatcher.Unwatch()
+			mrr.pgConn.unlock()
+		}
+	case *pgproto3.ErrorResponse:
+		mrr.err = ErrorResponseToPgError(msg)
+	}
+
+	return msg, nil
+}
+
+// NextResult returns advances the MultiResultReader to the next result and returns true if a result is available.
+func (mrr *MultiResultReader) NextResult() bool {
+	for !mrr.closed && mrr.err == nil {
+		msg, err := mrr.receiveMessage()
+		if err != nil {
+			return false
+		}
+
+		switch msg := msg.(type) {
+		case *pgproto3.RowDescription:
+			mrr.pgConn.resultReader = ResultReader{
+				pgConn:            mrr.pgConn,
+				multiResultReader: mrr,
+				ctx:               mrr.ctx,
+				fieldDescriptions: mrr.pgConn.convertRowDescription(mrr.pgConn.fieldDescriptions[:], msg),
+			}
+
+			mrr.rr = &mrr.pgConn.resultReader
+			return true
+		case *pgproto3.CommandComplete:
+			mrr.pgConn.resultReader = ResultReader{
+				commandTag:       mrr.pgConn.makeCommandTag(msg.CommandTag),
+				commandConcluded: true,
+				closed:           true,
+			}
+			mrr.rr = &mrr.pgConn.resultReader
+			return true
+		case *pgproto3.EmptyQueryResponse:
+			return false
+		}
+	}
+
+	return false
+}
+
+// ResultReader returns the current ResultReader.
+func (mrr *MultiResultReader) ResultReader() *ResultReader {
+	return mrr.rr
+}
+
+// Close closes the MultiResultReader and returns the first error that occurred during the MultiResultReader's use.
+func (mrr *MultiResultReader) Close() error {
+	for !mrr.closed {
+		_, err := mrr.receiveMessage()
+		if err != nil {
+			return mrr.err
+		}
+	}
+
+	return mrr.err
+}
+
+// ResultReader is a reader for the result of a single query.
+type ResultReader struct {
+	pgConn            *PgConn
+	multiResultReader *MultiResultReader
+	pipeline          *Pipeline
+	ctx               context.Context
+
+	fieldDescriptions []FieldDescription
+	rowValues         [][]byte
+	commandTag        CommandTag
+	commandConcluded  bool
+	closed            bool
+	err               error
+}
+
+// Result is the saved query response that is returned by calling Read on a ResultReader.
+type Result struct {
+	FieldDescriptions []FieldDescription
+	Rows              [][][]byte
+	CommandTag        CommandTag
+	Err               error
+}
+
+// Read saves the query response to a Result.
+func (rr *ResultReader) Read() *Result {
+	br := &Result{}
+
+	for rr.NextRow() {
+		if br.FieldDescriptions == nil {
+			br.FieldDescriptions = make([]FieldDescription, len(rr.FieldDescriptions()))
+			copy(br.FieldDescriptions, rr.FieldDescriptions())
+		}
+
+		values := rr.Values()
+		row := make([][]byte, len(values))
+		for i := range row {
+			if values[i] != nil {
+				row[i] = make([]byte, len(values[i]))
+				copy(row[i], values[i])
+			}
+		}
+		br.Rows = append(br.Rows, row)
+	}
+
+	br.CommandTag, br.Err = rr.Close()
+
+	return br
+}
+
+// NextRow advances the ResultReader to the next row and returns true if a row is available.
+func (rr *ResultReader) NextRow() bool {
+	for !rr.commandConcluded {
+		msg, err := rr.receiveMessage()
+		if err != nil {
+			return false
+		}
+
+		switch msg := msg.(type) {
+		case *pgproto3.DataRow:
+			rr.rowValues = msg.Values
+			return true
+		}
+	}
+
+	return false
+}
+
+// FieldDescriptions returns the field descriptions for the current result set. The returned slice is only valid until
+// the ResultReader is closed. It may return nil (for example, if the query did not return a result set or an error was
+// encountered.)
+func (rr *ResultReader) FieldDescriptions() []FieldDescription {
+	return rr.fieldDescriptions
+}
+
+// Values returns the current row data. NextRow must have been previously been called. The returned [][]byte is only
+// valid until the next NextRow call or the ResultReader is closed.
+func (rr *ResultReader) Values() [][]byte {
+	return rr.rowValues
+}
+
+// Close consumes any remaining result data and returns the command tag or
+// error.
+func (rr *ResultReader) Close() (CommandTag, error) {
+	if rr.closed {
+		return rr.commandTag, rr.err
+	}
+	rr.closed = true
+
+	for !rr.commandConcluded {
+		_, err := rr.receiveMessage()
+		if err != nil {
+			return CommandTag{}, rr.err
+		}
+	}
+
+	if rr.multiResultReader == nil && rr.pipeline == nil {
+		for {
+			msg, err := rr.receiveMessage()
+			if err != nil {
+				return CommandTag{}, rr.err
+			}
+
+			switch msg := msg.(type) {
+			// Detect a deferred constraint violation where the ErrorResponse is sent after CommandComplete.
+			case *pgproto3.ErrorResponse:
+				rr.err = ErrorResponseToPgError(msg)
+			case *pgproto3.ReadyForQuery:
+				rr.pgConn.contextWatcher.Unwatch()
+				rr.pgConn.unlock()
+				return rr.commandTag, rr.err
+			}
+		}
+	}
+
+	return rr.commandTag, rr.err
+}
+
+// readUntilRowDescription ensures the ResultReader's fieldDescriptions are loaded. It does not return an error as any
+// error will be stored in the ResultReader.
+func (rr *ResultReader) readUntilRowDescription() {
+	for !rr.commandConcluded {
+		// Peek before receive to avoid consuming a DataRow if the result set does not include a RowDescription method.
+		// This should never happen under normal pgconn usage, but it is possible if SendBytes and ReceiveResults are
+		// manually used to construct a query that does not issue a describe statement.
+		msg, _ := rr.pgConn.peekMessage()
+		if _, ok := msg.(*pgproto3.DataRow); ok {
+			return
+		}
+
+		// Consume the message
+		msg, _ = rr.receiveMessage()
+		if _, ok := msg.(*pgproto3.RowDescription); ok {
+			return
+		}
+	}
+}
+
+func (rr *ResultReader) receiveMessage() (msg pgproto3.BackendMessage, err error) {
+	if rr.multiResultReader == nil {
+		msg, err = rr.pgConn.receiveMessage()
+	} else {
+		msg, err = rr.multiResultReader.receiveMessage()
+	}
+
+	if err != nil {
+		err = normalizeTimeoutError(rr.ctx, err)
+		rr.concludeCommand(CommandTag{}, err)
+		rr.pgConn.contextWatcher.Unwatch()
+		rr.closed = true
+		if rr.multiResultReader == nil {
+			rr.pgConn.asyncClose()
+		}
+
+		return nil, rr.err
+	}
+
+	switch msg := msg.(type) {
+	case *pgproto3.RowDescription:
+		rr.fieldDescriptions = rr.pgConn.convertRowDescription(rr.pgConn.fieldDescriptions[:], msg)
+	case *pgproto3.CommandComplete:
+		rr.concludeCommand(rr.pgConn.makeCommandTag(msg.CommandTag), nil)
+	case *pgproto3.EmptyQueryResponse:
+		rr.concludeCommand(CommandTag{}, nil)
+	case *pgproto3.ErrorResponse:
+		rr.concludeCommand(CommandTag{}, ErrorResponseToPgError(msg))
+	}
+
+	return msg, nil
+}
+
+func (rr *ResultReader) concludeCommand(commandTag CommandTag, err error) {
+	// Keep the first error that is recorded. Store the error before checking if the command is already concluded to
+	// allow for receiving an error after CommandComplete but before ReadyForQuery.
+	if err != nil && rr.err == nil {
+		rr.err = err
+	}
+
+	if rr.commandConcluded {
+		return
+	}
+
+	rr.commandTag = commandTag
+	rr.rowValues = nil
+	rr.commandConcluded = true
+}
+
+// Batch is a collection of queries that can be sent to the PostgreSQL server in a single round-trip.
+type Batch struct {
+	buf []byte
+	err error
+}
+
+// ExecParams appends an ExecParams command to the batch. See PgConn.ExecParams for parameter descriptions.
+func (batch *Batch) ExecParams(sql string, paramValues [][]byte, paramOIDs []uint32, paramFormats []int16, resultFormats []int16) {
+	if batch.err != nil {
+		return
+	}
+
+	batch.buf, batch.err = (&pgproto3.Parse{Query: sql, ParameterOIDs: paramOIDs}).Encode(batch.buf)
+	if batch.err != nil {
+		return
+	}
+	batch.ExecPrepared("", paramValues, paramFormats, resultFormats)
+}
+
+// ExecPrepared appends an ExecPrepared e command to the batch. See PgConn.ExecPrepared for parameter descriptions.
+func (batch *Batch) ExecPrepared(stmtName string, paramValues [][]byte, paramFormats []int16, resultFormats []int16) {
+	if batch.err != nil {
+		return
+	}
+
+	batch.buf, batch.err = (&pgproto3.Bind{PreparedStatement: stmtName, ParameterFormatCodes: paramFormats, Parameters: paramValues, ResultFormatCodes: resultFormats}).Encode(batch.buf)
+	if batch.err != nil {
+		return
+	}
+
+	batch.buf, batch.err = (&pgproto3.Describe{ObjectType: 'P'}).Encode(batch.buf)
+	if batch.err != nil {
+		return
+	}
+
+	batch.buf, batch.err = (&pgproto3.Execute{}).Encode(batch.buf)
+	if batch.err != nil {
+		return
+	}
+}
+
+// ExecBatch executes all the queries in batch in a single round-trip. Execution is implicitly transactional unless a
+// transaction is already in progress or SQL contains transaction control statements. This is a simpler way of executing
+// multiple queries in a single round trip than using pipeline mode.
+func (pgConn *PgConn) ExecBatch(ctx context.Context, batch *Batch) *MultiResultReader {
+	if batch.err != nil {
+		return &MultiResultReader{
+			closed: true,
+			err:    batch.err,
+		}
+	}
+
+	if err := pgConn.lock(); err != nil {
+		return &MultiResultReader{
+			closed: true,
+			err:    err,
+		}
+	}
+
+	pgConn.multiResultReader = MultiResultReader{
+		pgConn: pgConn,
+		ctx:    ctx,
+	}
+	multiResult := &pgConn.multiResultReader
+
+	if ctx != context.Background() {
+		select {
+		case <-ctx.Done():
+			multiResult.closed = true
+			multiResult.err = newContextAlreadyDoneError(ctx)
+			pgConn.unlock()
+			return multiResult
+		default:
+		}
+		pgConn.contextWatcher.Watch(ctx)
+	}
+
+	batch.buf, batch.err = (&pgproto3.Sync{}).Encode(batch.buf)
+	if batch.err != nil {
+		multiResult.closed = true
+		multiResult.err = batch.err
+		pgConn.unlock()
+		return multiResult
+	}
+
+	pgConn.enterPotentialWriteReadDeadlock()
+	defer pgConn.exitPotentialWriteReadDeadlock()
+	_, err := pgConn.conn.Write(batch.buf)
+	if err != nil {
+		multiResult.closed = true
+		multiResult.err = err
+		pgConn.unlock()
+		return multiResult
+	}
+
+	return multiResult
+}
+
+// EscapeString escapes a string such that it can safely be interpolated into a SQL command string. It does not include
+// the surrounding single quotes.
+//
+// The current implementation requires that standard_conforming_strings=on and client_encoding="UTF8". If these
+// conditions are not met an error will be returned. It is possible these restrictions will be lifted in the future.
+func (pgConn *PgConn) EscapeString(s string) (string, error) {
+	if pgConn.ParameterStatus("standard_conforming_strings") != "on" {
+		return "", errors.New("EscapeString must be run with standard_conforming_strings=on")
+	}
+
+	if pgConn.ParameterStatus("client_encoding") != "UTF8" {
+		return "", errors.New("EscapeString must be run with client_encoding=UTF8")
+	}
+
+	return strings.Replace(s, "'", "''", -1), nil
+}
+
+// CheckConn checks the underlying connection without writing any bytes. This is currently implemented by doing a read
+// with a very short deadline. This can be useful because a TCP connection can be broken such that a write will appear
+// to succeed even though it will never actually reach the server. Reading immediately before a write will detect this
+// condition. If this is done immediately before sending a query it reduces the chances a query will be sent that fails
+// without the client knowing whether the server received it or not.
+//
+// Deprecated: CheckConn is deprecated in favor of Ping. CheckConn cannot detect all types of broken connections where
+// the write would still appear to succeed. Prefer Ping unless on a high latency connection.
+func (pgConn *PgConn) CheckConn() error {
+	ctx, cancel := context.WithTimeout(context.Background(), 1*time.Millisecond)
+	defer cancel()
+
+	_, err := pgConn.ReceiveMessage(ctx)
+	if err != nil {
+		if !Timeout(err) {
+			return err
+		}
+	}
+
+	return nil
+}
+
+// Ping pings the server. This can be useful because a TCP connection can be broken such that a write will appear to
+// succeed even though it will never actually reach the server. Pinging immediately before sending a query reduces the
+// chances a query will be sent that fails without the client knowing whether the server received it or not.
+func (pgConn *PgConn) Ping(ctx context.Context) error {
+	return pgConn.Exec(ctx, "-- ping").Close()
+}
+
+// makeCommandTag makes a CommandTag. It does not retain a reference to buf or buf's underlying memory.
+func (pgConn *PgConn) makeCommandTag(buf []byte) CommandTag {
+	return CommandTag{s: string(buf)}
+}
+
+// enterPotentialWriteReadDeadlock must be called before a write that could deadlock if the server is simultaneously
+// blocked writing to us.
+func (pgConn *PgConn) enterPotentialWriteReadDeadlock() {
+	// The time to wait is somewhat arbitrary. A Write should only take as long as the syscall and memcpy to the OS
+	// outbound network buffer unless the buffer is full (which potentially is a block). It needs to be long enough for
+	// the normal case, but short enough not to kill performance if a block occurs.
+	//
+	// In addition, on Windows the default timer resolution is 15.6ms. So setting the timer to less than that is
+	// ineffective.
+	if pgConn.slowWriteTimer.Reset(15 * time.Millisecond) {
+		panic("BUG: slow write timer already active")
+	}
+}
+
+// exitPotentialWriteReadDeadlock must be called after a call to enterPotentialWriteReadDeadlock.
+func (pgConn *PgConn) exitPotentialWriteReadDeadlock() {
+	if !pgConn.slowWriteTimer.Stop() {
+		// The timer starts its function in a separate goroutine. It is necessary to ensure the background reader has
+		// started before calling Stop. Otherwise, the background reader may not be stopped. That on its own is not a
+		// serious problem. But what is a serious problem is that the background reader may start at an inopportune time in
+		// a subsequent query. For example, if a subsequent query was canceled then a deadline may be set on the net.Conn to
+		// interrupt an in-progress read. After the read is interrupted, but before the deadline is cleared, the background
+		// reader could start and read a deadline error. Then the next query would receive the an unexpected deadline error.
+		<-pgConn.bgReaderStarted
+		pgConn.bgReader.Stop()
+	}
+}
+
+func (pgConn *PgConn) flushWithPotentialWriteReadDeadlock() error {
+	pgConn.enterPotentialWriteReadDeadlock()
+	defer pgConn.exitPotentialWriteReadDeadlock()
+	err := pgConn.frontend.Flush()
+	return err
+}
+
+// SyncConn prepares the underlying net.Conn for direct use. PgConn may internally buffer reads or use goroutines for
+// background IO. This means that any direct use of the underlying net.Conn may be corrupted if a read is already
+// buffered or a read is in progress. SyncConn drains read buffers and stops background IO. In some cases this may
+// require sending a ping to the server. ctx can be used to cancel this operation. This should be called before any
+// operation that will use the underlying net.Conn directly. e.g. Before Conn() or Hijack().
+//
+// This should not be confused with the PostgreSQL protocol Sync message.
+func (pgConn *PgConn) SyncConn(ctx context.Context) error {
+	for i := 0; i < 10; i++ {
+		if pgConn.bgReader.Status() == bgreader.StatusStopped && pgConn.frontend.ReadBufferLen() == 0 {
+			return nil
+		}
+
+		err := pgConn.Ping(ctx)
+		if err != nil {
+			return fmt.Errorf("SyncConn: Ping failed while syncing conn: %w", err)
+		}
+	}
+
+	// This should never happen. Only way I can imagine this occurring is if the server is constantly sending data such as
+	// LISTEN/NOTIFY or log notifications such that we never can get an empty buffer.
+	return errors.New("SyncConn: conn never synchronized")
+}
+
+// CustomData returns a map that can be used to associate custom data with the connection.
+func (pgConn *PgConn) CustomData() map[string]any {
+	return pgConn.customData
+}
+
+// HijackedConn is the result of hijacking a connection.
+//
+// Due to the necessary exposure of internal implementation details, it is not covered by the semantic versioning
+// compatibility.
+type HijackedConn struct {
+	Conn              net.Conn
+	PID               uint32            // backend pid
+	SecretKey         uint32            // key to use to send a cancel query message to the server
+	ParameterStatuses map[string]string // parameters that have been reported by the server
+	TxStatus          byte
+	Frontend          *pgproto3.Frontend
+	Config            *Config
+	CustomData        map[string]any
+}
+
+// Hijack extracts the internal connection data. pgConn must be in an idle state. SyncConn should be called immediately
+// before Hijack. pgConn is unusable after hijacking. Hijacking is typically only useful when using pgconn to establish
+// a connection, but taking complete control of the raw connection after that (e.g. a load balancer or proxy).
+//
+// Due to the necessary exposure of internal implementation details, it is not covered by the semantic versioning
+// compatibility.
+func (pgConn *PgConn) Hijack() (*HijackedConn, error) {
+	if err := pgConn.lock(); err != nil {
+		return nil, err
+	}
+	pgConn.status = connStatusClosed
+
+	return &HijackedConn{
+		Conn:              pgConn.conn,
+		PID:               pgConn.pid,
+		SecretKey:         pgConn.secretKey,
+		ParameterStatuses: pgConn.parameterStatuses,
+		TxStatus:          pgConn.txStatus,
+		Frontend:          pgConn.frontend,
+		Config:            pgConn.config,
+		CustomData:        pgConn.customData,
+	}, nil
+}
+
+// Construct created a PgConn from an already established connection to a PostgreSQL server. This is the inverse of
+// PgConn.Hijack. The connection must be in an idle state.
+//
+// hc.Frontend is replaced by a new pgproto3.Frontend built by hc.Config.BuildFrontend.
+//
+// Due to the necessary exposure of internal implementation details, it is not covered by the semantic versioning
+// compatibility.
+func Construct(hc *HijackedConn) (*PgConn, error) {
+	pgConn := &PgConn{
+		conn:              hc.Conn,
+		pid:               hc.PID,
+		secretKey:         hc.SecretKey,
+		parameterStatuses: hc.ParameterStatuses,
+		txStatus:          hc.TxStatus,
+		frontend:          hc.Frontend,
+		config:            hc.Config,
+		customData:        hc.CustomData,
+
+		status: connStatusIdle,
+
+		cleanupDone: make(chan struct{}),
+	}
+
+	pgConn.contextWatcher = ctxwatch.NewContextWatcher(hc.Config.BuildContextWatcherHandler(pgConn))
+	pgConn.bgReader = bgreader.New(pgConn.conn)
+	pgConn.slowWriteTimer = time.AfterFunc(time.Duration(math.MaxInt64),
+		func() {
+			pgConn.bgReader.Start()
+			pgConn.bgReaderStarted <- struct{}{}
+		},
+	)
+	pgConn.slowWriteTimer.Stop()
+	pgConn.bgReaderStarted = make(chan struct{})
+	pgConn.frontend = hc.Config.BuildFrontend(pgConn.bgReader, pgConn.conn)
+
+	return pgConn, nil
+}
+
+// Pipeline represents a connection in pipeline mode.
+//
+// SendPrepare, SendQueryParams, and SendQueryPrepared queue requests to the server. These requests are not written until
+// pipeline is flushed by Flush or Sync. Sync must be called after the last request is queued. Requests between
+// synchronization points are implicitly transactional unless explicit transaction control statements have been issued.
+//
+// The context the pipeline was started with is in effect for the entire life of the Pipeline.
+//
+// For a deeper understanding of pipeline mode see the PostgreSQL documentation for the extended query protocol
+// (https://www.postgresql.org/docs/current/protocol-flow.html#PROTOCOL-FLOW-EXT-QUERY) and the libpq pipeline mode
+// (https://www.postgresql.org/docs/current/libpq-pipeline-mode.html).
+type Pipeline struct {
+	conn *PgConn
+	ctx  context.Context
+
+	expectedReadyForQueryCount int
+	pendingSync                bool
+
+	err    error
+	closed bool
+}
+
+// PipelineSync is returned by GetResults when a ReadyForQuery message is received.
+type PipelineSync struct{}
+
+// CloseComplete is returned by GetResults when a CloseComplete message is received.
+type CloseComplete struct{}
+
+// StartPipeline switches the connection to pipeline mode and returns a *Pipeline. In pipeline mode requests can be sent
+// to the server without waiting for a response. Close must be called on the returned *Pipeline to return the connection
+// to normal mode. While in pipeline mode, no methods that communicate with the server may be called except
+// CancelRequest and Close. ctx is in effect for entire life of the *Pipeline.
+//
+// Prefer ExecBatch when only sending one group of queries at once.
+func (pgConn *PgConn) StartPipeline(ctx context.Context) *Pipeline {
+	if err := pgConn.lock(); err != nil {
+		return &Pipeline{
+			closed: true,
+			err:    err,
+		}
+	}
+
+	pgConn.pipeline = Pipeline{
+		conn: pgConn,
+		ctx:  ctx,
+	}
+	pipeline := &pgConn.pipeline
+
+	if ctx != context.Background() {
+		select {
+		case <-ctx.Done():
+			pipeline.closed = true
+			pipeline.err = newContextAlreadyDoneError(ctx)
+			pgConn.unlock()
+			return pipeline
+		default:
+		}
+		pgConn.contextWatcher.Watch(ctx)
+	}
+
+	return pipeline
+}
+
+// SendPrepare is the pipeline version of *PgConn.Prepare.
+func (p *Pipeline) SendPrepare(name, sql string, paramOIDs []uint32) {
+	if p.closed {
+		return
+	}
+	p.pendingSync = true
+
+	p.conn.frontend.SendParse(&pgproto3.Parse{Name: name, Query: sql, ParameterOIDs: paramOIDs})
+	p.conn.frontend.SendDescribe(&pgproto3.Describe{ObjectType: 'S', Name: name})
+}
+
+// SendDeallocate deallocates a prepared statement.
+func (p *Pipeline) SendDeallocate(name string) {
+	if p.closed {
+		return
+	}
+	p.pendingSync = true
+
+	p.conn.frontend.SendClose(&pgproto3.Close{ObjectType: 'S', Name: name})
+}
+
+// SendQueryParams is the pipeline version of *PgConn.QueryParams.
+func (p *Pipeline) SendQueryParams(sql string, paramValues [][]byte, paramOIDs []uint32, paramFormats []int16, resultFormats []int16) {
+	if p.closed {
+		return
+	}
+	p.pendingSync = true
+
+	p.conn.frontend.SendParse(&pgproto3.Parse{Query: sql, ParameterOIDs: paramOIDs})
+	p.conn.frontend.SendBind(&pgproto3.Bind{ParameterFormatCodes: paramFormats, Parameters: paramValues, ResultFormatCodes: resultFormats})
+	p.conn.frontend.SendDescribe(&pgproto3.Describe{ObjectType: 'P'})
+	p.conn.frontend.SendExecute(&pgproto3.Execute{})
+}
+
+// SendQueryPrepared is the pipeline version of *PgConn.QueryPrepared.
+func (p *Pipeline) SendQueryPrepared(stmtName string, paramValues [][]byte, paramFormats []int16, resultFormats []int16) {
+	if p.closed {
+		return
+	}
+	p.pendingSync = true
+
+	p.conn.frontend.SendBind(&pgproto3.Bind{PreparedStatement: stmtName, ParameterFormatCodes: paramFormats, Parameters: paramValues, ResultFormatCodes: resultFormats})
+	p.conn.frontend.SendDescribe(&pgproto3.Describe{ObjectType: 'P'})
+	p.conn.frontend.SendExecute(&pgproto3.Execute{})
+}
+
+// Flush flushes the queued requests without establishing a synchronization point.
+func (p *Pipeline) Flush() error {
+	if p.closed {
+		if p.err != nil {
+			return p.err
+		}
+		return errors.New("pipeline closed")
+	}
+
+	err := p.conn.flushWithPotentialWriteReadDeadlock()
+	if err != nil {
+		err = normalizeTimeoutError(p.ctx, err)
+
+		p.conn.asyncClose()
+
+		p.conn.contextWatcher.Unwatch()
+		p.conn.unlock()
+		p.closed = true
+		p.err = err
+		return err
+	}
+
+	return nil
+}
+
+// Sync establishes a synchronization point and flushes the queued requests.
+func (p *Pipeline) Sync() error {
+	if p.closed {
+		if p.err != nil {
+			return p.err
+		}
+		return errors.New("pipeline closed")
+	}
+
+	p.conn.frontend.SendSync(&pgproto3.Sync{})
+	err := p.Flush()
+	if err != nil {
+		return err
+	}
+
+	p.pendingSync = false
+	p.expectedReadyForQueryCount++
+
+	return nil
+}
+
+// GetResults gets the next results. If results are present, results may be a *ResultReader, *StatementDescription, or
+// *PipelineSync. If an ErrorResponse is received from the server, results will be nil and err will be a *PgError. If no
+// results are available, results and err will both be nil.
+func (p *Pipeline) GetResults() (results any, err error) {
+	if p.closed {
+		if p.err != nil {
+			return nil, p.err
+		}
+		return nil, errors.New("pipeline closed")
+	}
+
+	if p.expectedReadyForQueryCount == 0 {
+		return nil, nil
+	}
+
+	return p.getResults()
+}
+
+func (p *Pipeline) getResults() (results any, err error) {
+	for {
+		msg, err := p.conn.receiveMessage()
+		if err != nil {
+			p.closed = true
+			p.err = err
+			p.conn.asyncClose()
+			return nil, normalizeTimeoutError(p.ctx, err)
+		}
+
+		switch msg := msg.(type) {
+		case *pgproto3.RowDescription:
+			p.conn.resultReader = ResultReader{
+				pgConn:            p.conn,
+				pipeline:          p,
+				ctx:               p.ctx,
+				fieldDescriptions: p.conn.convertRowDescription(p.conn.fieldDescriptions[:], msg),
+			}
+			return &p.conn.resultReader, nil
+		case *pgproto3.CommandComplete:
+			p.conn.resultReader = ResultReader{
+				commandTag:       p.conn.makeCommandTag(msg.CommandTag),
+				commandConcluded: true,
+				closed:           true,
+			}
+			return &p.conn.resultReader, nil
+		case *pgproto3.ParseComplete:
+			peekedMsg, err := p.conn.peekMessage()
+			if err != nil {
+				p.conn.asyncClose()
+				return nil, normalizeTimeoutError(p.ctx, err)
+			}
+			if _, ok := peekedMsg.(*pgproto3.ParameterDescription); ok {
+				return p.getResultsPrepare()
+			}
+		case *pgproto3.CloseComplete:
+			return &CloseComplete{}, nil
+		case *pgproto3.ReadyForQuery:
+			p.expectedReadyForQueryCount--
+			return &PipelineSync{}, nil
+		case *pgproto3.ErrorResponse:
+			pgErr := ErrorResponseToPgError(msg)
+			return nil, pgErr
+		}
+
+	}
+}
+
+func (p *Pipeline) getResultsPrepare() (*StatementDescription, error) {
+	psd := &StatementDescription{}
+
+	for {
+		msg, err := p.conn.receiveMessage()
+		if err != nil {
+			p.conn.asyncClose()
+			return nil, normalizeTimeoutError(p.ctx, err)
+		}
+
+		switch msg := msg.(type) {
+		case *pgproto3.ParameterDescription:
+			psd.ParamOIDs = make([]uint32, len(msg.ParameterOIDs))
+			copy(psd.ParamOIDs, msg.ParameterOIDs)
+		case *pgproto3.RowDescription:
+			psd.Fields = p.conn.convertRowDescription(nil, msg)
+			return psd, nil
+
+		// NoData is returned instead of RowDescription when there is no expected result. e.g. An INSERT without a RETURNING
+		// clause.
+		case *pgproto3.NoData:
+			return psd, nil
+
+		// These should never happen here. But don't take chances that could lead to a deadlock.
+		case *pgproto3.ErrorResponse:
+			pgErr := ErrorResponseToPgError(msg)
+			return nil, pgErr
+		case *pgproto3.CommandComplete:
+			p.conn.asyncClose()
+			return nil, errors.New("BUG: received CommandComplete while handling Describe")
+		case *pgproto3.ReadyForQuery:
+			p.conn.asyncClose()
+			return nil, errors.New("BUG: received ReadyForQuery while handling Describe")
+		}
+	}
+}
+
+// Close closes the pipeline and returns the connection to normal mode.
+func (p *Pipeline) Close() error {
+	if p.closed {
+		return p.err
+	}
+
+	p.closed = true
+
+	if p.pendingSync {
+		p.conn.asyncClose()
+		p.err = errors.New("pipeline has unsynced requests")
+		p.conn.contextWatcher.Unwatch()
+		p.conn.unlock()
+
+		return p.err
+	}
+
+	for p.expectedReadyForQueryCount > 0 {
+		_, err := p.getResults()
+		if err != nil {
+			p.err = err
+			var pgErr *PgError
+			if !errors.As(err, &pgErr) {
+				p.conn.asyncClose()
+				break
+			}
+		}
+	}
+
+	p.conn.contextWatcher.Unwatch()
+	p.conn.unlock()
+
+	return p.err
+}
+
+// DeadlineContextWatcherHandler handles canceled contexts by setting a deadline on a net.Conn.
+type DeadlineContextWatcherHandler struct {
+	Conn net.Conn
+
+	// DeadlineDelay is the delay to set on the deadline set on net.Conn when the context is canceled.
+	DeadlineDelay time.Duration
+}
+
+func (h *DeadlineContextWatcherHandler) HandleCancel(ctx context.Context) {
+	h.Conn.SetDeadline(time.Now().Add(h.DeadlineDelay))
+}
+
+func (h *DeadlineContextWatcherHandler) HandleUnwatchAfterCancel() {
+	h.Conn.SetDeadline(time.Time{})
+}
+
+// CancelRequestContextWatcherHandler handles canceled contexts by sending a cancel request to the server. It also sets
+// a deadline on a net.Conn as a fallback.
+type CancelRequestContextWatcherHandler struct {
+	Conn *PgConn
+
+	// CancelRequestDelay is the delay before sending the cancel request to the server.
+	CancelRequestDelay time.Duration
+
+	// DeadlineDelay is the delay to set on the deadline set on net.Conn when the context is canceled.
+	DeadlineDelay time.Duration
+
+	cancelFinishedChan             chan struct{}
+	handleUnwatchAfterCancelCalled func()
+}
+
+func (h *CancelRequestContextWatcherHandler) HandleCancel(context.Context) {
+	h.cancelFinishedChan = make(chan struct{})
+	var handleUnwatchedAfterCancelCalledCtx context.Context
+	handleUnwatchedAfterCancelCalledCtx, h.handleUnwatchAfterCancelCalled = context.WithCancel(context.Background())
+
+	deadline := time.Now().Add(h.DeadlineDelay)
+	h.Conn.conn.SetDeadline(deadline)
+
+	go func() {
+		defer close(h.cancelFinishedChan)
+
+		select {
+		case <-handleUnwatchedAfterCancelCalledCtx.Done():
+			return
+		case <-time.After(h.CancelRequestDelay):
+		}
+
+		cancelRequestCtx, cancel := context.WithDeadline(handleUnwatchedAfterCancelCalledCtx, deadline)
+		defer cancel()
+		h.Conn.CancelRequest(cancelRequestCtx)
+
+		// CancelRequest is inherently racy. Even though the cancel request has been received by the server at this point,
+		// it hasn't necessarily been delivered to the other connection. If we immediately return and the connection is
+		// immediately used then it is possible the CancelRequest will actually cancel our next query. The
+		// TestCancelRequestContextWatcherHandler Stress test can produce this error without the sleep below. The sleep time
+		// is arbitrary, but should be sufficient to prevent this error case.
+		time.Sleep(100 * time.Millisecond)
+	}()
+}
+
+func (h *CancelRequestContextWatcherHandler) HandleUnwatchAfterCancel() {
+	h.handleUnwatchAfterCancelCalled()
+	<-h.cancelFinishedChan
+
+	h.Conn.conn.SetDeadline(time.Time{})
+}