TPC-B workload
TPC-B (spec §1) with deterministic plain-Go data generation. The workload loads
three tables through Bench.Insert and runs the canonical five-statement
transaction.
Variants
tpcb/tx runs the DML statements in one client-side transaction and supports
PostgreSQL, MySQL, Picodata, and YDB.
tpcb/procs calls tpcb_transaction once per iteration and supports
PostgreSQL, MySQL, CockroachDB, and YugabyteDB. Each invocation uses the selected
transaction isolation and reports transaction metrics. MySQL procedures own
rollback as well as commit and use tpcb_transaction_atomic; install the new
definition before reusing a loaded database. PostgreSQL-compatible functions use the statement's implicit transaction.
Run it
./build/stroppy run tpcb/tx -d pg \
-D url=postgres://postgres:postgres@localhost:5432/stroppy
./build/stroppy run tpcb/tx -d mysql \
-D 'url=root:pass@tcp(127.0.0.1:3306)/stroppy'
./build/stroppy run tpcb/tx -d pico
./build/stroppy run tpcb/tx -d ydb -D url=grpc://localhost:2136/local
# Stored-procedure variant (PostgreSQL or MySQL).
./build/stroppy run tpcb/procs -d pg
# Write the generated tables to CSV without a database.
./build/stroppy run tpcb/tx \
-D driverType=csv \
-D url='/tmp/tpcb-csv?merge=true&workload=tpcb' \
--scale-factor 1 \
--steps drop_schema,create_schema,load_data
Typed workload parameters are listed by stroppy run tpcb/tx --help and
stroppy run tpcb/procs --help:
--scale-factor sets the branch count; each branch has 10 tellers and
100,000 accounts.
--load-workers sets the workers used to load each table.
--retry-attempts sets the maximum transaction attempts.
--tx-isolation overrides the driver-specific isolation default.
--sql-file overrides the dialect SQL file.
Setup and execution
Setup runs these gatable steps in order:
drop_schema
create_schema
create_procedures for tpcb/procs only
load_data, using typed driver.InsertRequest values backed by
gen.IndexedSource batches for branches, tellers, and accounts
create_indexes
create_foreign_keys
analyze
Each workload iteration updates an account, reads its balance, updates a
teller and branch, and inserts a history row. The tx variant executes those
statements in one driver transaction; the procs variant performs one
server-side procedure call. Retry behavior and metrics are shared.
history starts empty and is populated by measured transactions. Filler
columns use fixed-width ASCII, which the specification permits.
Run shapes and two-pass runs
Use typed run flags directly:
# Fixed work shared across virtual users.
./build/stroppy run tpcb/tx --executor shared-iterations --vus 4 --iterations 100
# Fixed-duration throughput.
./build/stroppy run tpcb/tx --executor constant-vus --vus 64 --duration 1h
# Load first, then measure the existing data.
./build/stroppy run tpcb/tx --scale-factor 10 --no-steps workload
./build/stroppy run tpcb/tx --executor constant-vus --vus 64 --duration 1h \
--steps workload
A normal run with no step filter performs setup and measurement together.
Environment variables and -e remain compatibility inputs; direct flags and
typed config are preferred.
Integration test
Build the binary and start the baseline services before running the tagged test:
make tmpfs-up
make build
go test -tags=integration -count=1 -run TestTpcbWorkloadEndToEnd ./test/integration
make tmpfs-down
TiDB SQL
TiDB speaks the mysql protocol, but it has no stored procedures, so it gets its own
variant — tidb.sql: the mysql asset without the DROP PROCEDURE, the
create_procedures section and the workload_procs call. -d tidb selects it.
Measured: without it, tpcb/tx on TiDB fails at the first statement of
mysql.sql with Error 8108 (HY000): Unsupported type *resolve.NodeW, because that
DROP PROCEDURE runs during setup even for the raw-transaction variant. The procedure
variant (tpcb/procs) is therefore not available on TiDB, and the workload's
required-query check reports that rather than failing somewhere mid-run.
CockroachDB SQL
Run with the crdb driver — -d crdb — and the workload selects its CockroachDB
asset itself. TPC-B ships one because CockroachDB rejects a bare ANALYZE: the
variant names each table explicitly and supports both tx and procs.
Note that --sql-file takes a filesystem path, not the name of an embedded
asset, so it cannot select a shipped variant — earlier versions of this file said
--sql-file crdb.sql, which fails with open crdb.sql: no such file or directory.
Use -d crdb, or extract the asset yourself and pass its path.