sqlcore

type ColumnTyper ¶

type ColumnTyper interface {
	DatabaseTypeName() string
}

ColumnTyper provides type name information for a database column. *sql.ColumnType satisfies this interface.

type CopyFromStdinExecutor ¶

type CopyFromStdinExecutor interface {
	CopyFromStdin(ctx context.Context, request CopyFromStdinRequest, data io.Reader) (rowCount int64, err error)
}

CopyFromStdinExecutor is an optional capability implemented by executors that need the COPY input bytes shipped to a different filesystem than the one the COPY FROM STDIN handler is running on. The remote (Flight) executor implements this to spool the bytes onto the worker pod, where the worker can then run COPY FROM <path>. Local executors do not need to implement it; the standard local-tempfile path works for them since the executor and the COPY FROM SQL run in the same process / host.

request.SQLTemplate must contain a path placeholder that the receiver substitutes with the destination spool path before executing. The request also carries exact binary column types when scanner validation is required. The placeholder string is defined alongside the implementation (flightclient.CopyFromStdinPathPlaceholder).

type CopyFromStdinRequest ¶

type CopyFromStdinRequest struct {
	SQLTemplate                     string   `json:"sql_template"`
	PostgresBinaryDatabaseTypeNames []string `json:"postgres_binary_database_type_names,omitempty"`
}

CopyFromStdinRequest carries the worker-local SQL template plus optional lossless PostgreSQL binary COPY schema metadata. The latter lets the worker validate tuple width and normalize scanner-sensitive values without parsing generated SQL or reconstructing types from lossy Arrow physical types.

type ExactColumnTyper ¶

type ExactColumnTyper interface {
	ColumnTyper
	ExactDatabaseTypeName() (name string, ok bool)
}

ExactColumnTyper exposes the database's original logical type name when a transport has to project it onto a less expressive result type. Arrow, for example, represents DuckDB UUID as String and HUGEINT as Decimal128. Callers making encoding decisions must use this lossless name and fail closed when it is unavailable; DatabaseTypeName remains the projected result type used by existing pgwire result handling.

type ExecResult ¶

type ExecResult interface {
	RowsAffected() (int64, error)
}

ExecResult represents the result of a non-query execution.

type QueryExecutor ¶

type QueryExecutor interface {
	QueryContext(ctx context.Context, query string, args ...any) (RowSet, error)
	ExecContext(ctx context.Context, query string, args ...any) (ExecResult, error)
	Query(query string, args ...any) (RowSet, error)
	Exec(query string, args ...any) (ExecResult, error)
	ConnContext(ctx context.Context) (RawConn, error)
	PingContext(ctx context.Context) error
	Close() error

	// LastProfilingOutput returns the JSON profiling output from the last
	// executed query, or "" if profiling is not enabled or not available
	// (e.g. Flight SQL mode where the query ran on a remote worker).
	LastProfilingOutput() string
}

QueryExecutor abstracts database query execution, allowing both local (*sql.DB) and remote (Arrow Flight SQL) backends.

type RawConn ¶

type RawConn interface {
	Raw(func(any) error) error
	Close() error
}

RawConn provides access to the underlying driver connection. *sql.Conn satisfies this interface.

type RowSet ¶

type RowSet interface {
	Columns() ([]string, error)
	ColumnTypes() ([]ColumnTyper, error)
	Next() bool
	Scan(dest ...any) error
	Close() error
	Err() error
}

RowSet represents a set of rows from a query result.