Documentation
¶
Overview ¶
Package csv implements an ephemeral Stroppy driver that writes generator output to CSV files on the local filesystem instead of a database. It exists to (a) benchmark pure generation throughput without database I/O, (b) produce reference output for cross-tool comparisons, and (c) feed downstream systems that bulk-load from delimited files (ClickHouse, DuckDB, PostgreSQL COPY, etc.).
Configuration is entirely URL-driven. The path component of the URL selects the output directory (defaults to the current working directory when absent) and the query string carries the small set of supported knobs: ?merge=true|false, ?separator=comma|tab, ?header=true|false.
The driver implements only the typed InsertRequest NATIVE path. Every other InsertMethod is rejected with ErrUnsupportedInsertMethod; runtime query execution is rejected with ErrCsvDriverNoQuery. DDL emitted by the drop_schema and create_schema workload steps is accepted and processed out-of-band: DROP clauses delete the workload's output directory for idempotent reruns, CREATE is a noop.
Index ¶
- Variables
- type Driver
- func (d *Driver) Begin(_ context.Context, _ stroppyconfig.TxIsolationLevel) (driver.Tx, error)
- func (*Driver) ClassifyError(err error) driver.ErrorFacts
- func (d *Driver) Insert(ctx context.Context, req *driver.InsertRequest) (_ *stats.Query, retErr error)
- func (d *Driver) RunQuery(_ context.Context, sqlStr string, _ map[string]any) (*driver.QueryResult, error)
- func (d *Driver) Teardown(ctx context.Context) error
Constants ¶
This section is empty.
Variables ¶
var ErrCsvDriverNoQuery = errors.New("csv: driver does not execute queries")
ErrCsvDriverNoQuery is returned when a non-DDL query reaches the CSV driver. CSV is write-only: it has no result set to produce and no transaction to run under. DDL emitted by the drop_schema and create_schema workload steps is recognized and handled out-of-band (DROP removes the workload's output directory; CREATE is a noop), so these steps remain runnable alongside load_data.
var ErrIncompleteLoad = errors.New("csv: output generation is incomplete")
ErrIncompleteLoad prevents final output from representing an incomplete CSV generation.
var ErrInvalidOption = errors.New("csv: invalid URL option")
ErrInvalidOption is the static parent error for any invalid URL query value. The concrete per-option message wraps it.
var (
ErrUnsupportedInsertMethod = fmt.Errorf("csv: %w", driver.ErrInsertMethodNotSupported)
)
ErrUnsupportedInsertMethod is returned when an InsertRequest requests anything other than NATIVE. CSV is write-only: PLAIN_BULK and PLAIN_QUERY imply SQL-shaped emission, which the CSV driver does not synthesize. Matches the rejection pattern used by the other drivers.
Functions ¶
This section is empty.
Types ¶
type Driver ¶
type Driver struct {
// contains filtered or unexported fields
}
Driver emits generator rows to CSV files. One Driver instance is scoped to one benchmark run; tables accumulate under a single output directory and are either merged or left as worker shards at Teardown.
func NewDriver ¶
NewDriver parses opts.Config.URL and returns a ready-to-use Driver. The output directory is created lazily (on first write) so Setup succeeds even when dir is a prefix that does not yet exist.
func (*Driver) Begin ¶
func (d *Driver) Begin(_ context.Context, _ stroppyconfig.TxIsolationLevel) (driver.Tx, error)
Begin refuses to start a transaction. CSV writes have no rollback semantics and workloads that call tx.* are not supported.
func (*Driver) ClassifyError ¶
func (*Driver) ClassifyError(err error) driver.ErrorFacts
func (*Driver) Insert ¶
func (d *Driver) Insert( ctx context.Context, req *driver.InsertRequest, ) (_ *stats.Query, retErr error)
Insert runs a typed driver.InsertRequest through the CSV driver by draining each worker's gen.Cursor partition into a distinct shard file. NATIVE is the only method: CSV is write-only and does not synthesize SQL-shaped emission for PLAIN_BULK/PLAIN_QUERY.
func (*Driver) RunQuery ¶
func (d *Driver) RunQuery( _ context.Context, sqlStr string, _ map[string]any, ) (*driver.QueryResult, error)
RunQuery accepts DDL (CREATE/DROP/TRUNCATE/ALTER/COMMENT) as a noop so workload drop_schema and create_schema steps stay valid with the CSV driver selected. DROP is treated as a directive to wipe the workload's output directory; everything else silently succeeds. Non-DDL queries return ErrCsvDriverNoQuery.