Drivers & Configuration

Stroppy passes CLI or JSON configuration directly to a registered Go driver. Use -d for a preset and -D for field overrides.

stroppy run tpcc/tx -d pg
stroppy run tpcc/tx -d pg -D url=postgres://host:5432/bench
stroppy run tpcc/tx -d pg -D pool.maxConns=100

Presets

Preset	driverType	Default local URL	Notes
pg	postgres	postgres://postgres:postgres@localhost:5432	pgx pool; native COPY.
mysql	mysql	myuser:mypassword@tcp(localhost:3306)/mydb	database/sql; bulk INSERT.
pico	picodata	postgres://admin:T0psecret@localhost:1331	PostgreSQL wire; no transactions.
ydb	ydb	grpc://localhost:2136/local	SQL plus native BulkUpsert.
noop	noop	noop://localhost	Discards I/O to measure framework overhead.

CSV is a driver type without a short preset.

Credentials shown in preset URLs are public local-development defaults compiled into Stroppy, not secrets. Always override them outside disposable local setups.

A preset sets driverType and URL. Overrides keep untouched preset fields:

stroppy run tpcb/tx -d mysql \
  -D 'url=root@tcp(mysql.example:3306)/bench?parseTime=true'

Raw JSON can replace a preset:

stroppy run tpcc/tx \
  -d '{"driverType":"postgres","url":"postgres://db:5432/bench"}'

Raw JSON is validated with the same strict field rules as config files.

Indexed drivers

Use numbered flags for workloads that need multiple drivers:

stroppy run tpcc/tx -d pg -d1 mysql
stroppy run tpcc/tx -d pg -D url=postgres://pg/bench \
  -d1 mysql -D1 'url=root@tcp(mysql:3306)/bench'

-d/-D target index 0; -d1/-D1 target index 1.

Driver fields

Field	Type	Description
url	string	Driver connection URL, DSN, or output path.
driverType	string	postgres, mysql, picodata, ydb, noop, csv.
bulkSize	integer	Rows per multi-row INSERT/native batch; default 2500.
defaultInsertMethod	string	Fallback for requests with no workload-selected method.
pool.*	nested	Portable pool settings.
postgres.*	nested	pgx-specific settings.
sql.*	nested	database/sql settings.
insertProgress.*	nested	Load progress watcher.
caCertFile	string	CA certificate PEM path.
authToken	string	Token credential, especially for YDB.
authUser / authPassword	string	Static credentials.
tlsInsecureSkipVerify	boolean	Skip certificate verification for testing only.

The removed errorMode and defaultTxIsolation fields are rejected. Error policy belongs to workloads; transaction isolation is a typed workload parameter (--tx-isolation).

Pool configuration

Portable pool fields map by driver family:

Portable field	PostgreSQL/Picodata	MySQL/YDB
pool.maxConns	postgres.maxConns	sql.maxOpenConns
pool.minConns	postgres.minConns	sql.maxIdleConns
pool.maxConnLifetime	postgres.maxConnLifetime	sql.connMaxLifetime
pool.maxConnIdleTime	postgres.maxConnIdleTime	sql.connMaxIdleTime

stroppy run tpcc/tx -d pg \
  -D pool.maxConns=100 \
  -D pool.minConns=20 \
  -D pool.maxConnLifetime=1h

Explicit postgres.* or sql.* values take priority over pool.*.

Additional PostgreSQL/Picodata fields:

postgres.minIdleConns
postgres.traceLogLevel
postgres.defaultQueryExecMode
postgres.descriptionCacheCapacity
postgres.statementCacheCapacity

Additional MySQL/YDB fields:

sql.maxOpenConns
sql.maxIdleConns
sql.connMaxLifetime
sql.connMaxIdleTime

Noop and CSV ignore pool settings.

Insert progress

Typed loads can log and export periodic progress:

stroppy run tpcc/tx -d pg \
  -D insertProgress.enabled=true \
  -D insertProgress.interval=30s \
  -D insertProgress.stallAfter=2m \
  -D insertProgress.mode=both

Modes: off, log, metrics, both.

Progress distinguishes generated, in-flight, and confirmed rows where the backend can provide that distinction. A final completion or failure sample is always emitted when tracking is enabled.

Insert capabilities

stroppy probe reports current methods. In v6.0.0:

Driver	plain_query	plain_bulk	columnar	native
PostgreSQL	yes	yes	yes	COPY
MySQL	yes	yes	no	multi-row INSERT
Picodata	yes	yes	no	multi-row INSERT
YDB	yes	yes	BulkUpsert	BulkUpsert
Noop	drain	drain	drain	drain
CSV	no	no	no	files

A workload-selected driver.InsertRequest.Method takes precedence over defaultInsertMethod. The driver default fills only an unset request.

PostgreSQL

PostgreSQL uses pgxpool.

• native: COPY.
• columnar: one array per column expanded with unnest, avoiding the 65535 bound-parameter limit on wide batches.
• plain_bulk: multi-row INSERT.
• plain_query: one-row INSERT batches.

postgres.defaultQueryExecMode accepts exec, cache_statement, cache_describe, describe_exec, or simple_protocol. Stroppy defaults to exec when unset.

stroppy run tpch/tx -d pg \
  -D url=postgres://host:5432/bench \
  -D postgres.defaultQueryExecMode=cache_statement

MySQL

MySQL uses database/sql and Go MySQL DSNs:

stroppy run tpcc/procs -d mysql \
  -D 'url=root@tcp(mysql.example:3306)/bench?parseTime=true'

plain_bulk uses multi-row INSERT. native maps to the same bulk path; Stroppy does not use LOAD DATA LOCAL INFILE. Statement query timeouts also add a server-side MAX_EXECUTION_TIME hint where applicable, reducing the chance that a timed-out SELECT keeps work alive on a pooled connection.

Picodata

Picodata uses its PostgreSQL wire endpoint through the shared SQL driver path. It supports typed loads and SQL queries, subject to sbroad SQL limitations.

Picodata Begin() always returns an error. Transactional workload variants use isolation none by default:

stroppy run tpcc/tx -d pico --tx-isolation none

TPC-H has a dedicated Picodata query port. TPC-DS is load-only on Picodata in v6; run it with --no-steps workload.

YDB

YDB accepts grpc:// and grpcs:// URLs. Native and columnar loads map to BulkUpsert.

stroppy run tpcc/tx -d ydb \
  -D url=grpcs://host:2135/database \
  -D caCertFile=./ca.pem \
  -D authToken="$YDB_TOKEN"

Static credentials are also available through authUser and authPassword. TPC workloads normally use serializable isolation on YDB.

Noop

Noop drives generation, batching, query construction, metrics, and transaction bookkeeping, then discards I/O. Use stroppy baseline for a purpose-built framework-overhead measurement.

CSV

CSV writes typed load requests to files and has no query path. It requires native insertion.

stroppy run tpcb/tx \
  -D driverType=csv \
  -D url='/tmp/tpcb-csv?merge=true&header=true&workload=tpcb' \
  --steps drop_schema,create_schema,load_data

URL options:

Option	Default	Meaning
merge	true	Merge worker shards into one table CSV.
header	true	Include headers; sidecars when shards stay separate.
separator	comma	comma, ,, tab, or \t.
workload	default	Output subdirectory.

Fresh and repeated generations publish shards, merged files, and manifests atomically. Failed or canceled loads retain recoverable shards without exposing partial output as complete.

Error classification and run behavior

Drivers translate backend errors into shared facts such as serialization, deadlock, lock timeout, transient, timeout, cancellation, and unsupported. Workloads choose actions; driver config does not.

Transactional workloads retry serialization conflicts, deadlocks, lock timeouts, and unconditional transient facts by default. Conditional transient retries require an idempotent workload operation.

A terminal nonfatal transaction error fails one iteration and lets the VU continue. Query-set workloads count a failed query and continue. Both paths print bounded warnings and a prominent final error summary while exiting 0. Setup, fatal, validation, and teardown failures remain nonzero.

See Transactions & Errors.

Go driver interface

Adding a driver requires a source build. Current interface:

type Driver interface {
    Insert(context.Context, *InsertRequest) (*stats.Query, error)
    RunQuery(context.Context, string, map[string]any) (*QueryResult, error)
    Begin(context.Context, config.TxIsolationLevel) (Tx, error)
    ClassifyError(error) ErrorFacts
    Teardown(context.Context) error
}

See Extensibility for registration and testing.