Testing And Benchmark

OceanBase Testing and Benchmarking

Before resolving or acquiring any test prerequisite, read the shared deployment package closure and expand the selected test into its complete controller-side toolchain and runtime dependencies.

Before any package network request, select the effective artifact source before the acquisition mechanism and apply the shared fixed package-source workflow. This applies to every test prerequisite. A .repo file is configuration rather than a package source; prove the effective package URL, and keep operating-system dependency repositories separate. For a missing prerequisite such as ob-sysbench, remain on the current required source while trying applicable controller-local OBD/tool, curl, wget, or package-manager paths. Verify and import/register the exact artifact locally before retrying obd test. Do not move OBD or test execution off the selected controller, and never use obd mirror as a network downloader.

Read common-workflow.md for every test. Then read exactly the selected tool reference:

Shared Gates

• Before selecting syntax, plugins, or artifacts, read product and capability resolution.
• Before any live controller/host/deployment query, SSH/SQL/API/network access, tool discovery that may install, data preparation, parameter change, external action, or load execution, read the operation contract.
• Before reporting a test successful, read the completion criteria.
• After a failed, timed-out, interrupted, or mixed run, read failure recovery and evidence before retrying or cleaning.
• Before removing a dataset, database object, generated file, tool, cache, or result, read cleanup and ownership boundaries.

Hard Boundaries

• Benchmark preparation and workload execution can create, update, or delete tenant data. Use an approved test tenant and dataset; do not infer permission to load a production tenant.
• For Sysbench, TPC-H, and TPC-C, pass the installed command's long optimization option explicitly. Use the no-OBD-optimization level for the default baseline; any parameter mutation or restart requires separate authorization. Do not add this benchmark-only option to mysqltest.
• For Sysbench, TPC-H, and TPC-C, keep the built-in cluster status check enabled unless the user explicitly accepts a reviewed equivalent check. MySQLtest has a different option set. Never reuse a short option from a deployment command in an obd test command.
• Track tool availability, prerequisite installation, test preparation, execution, result collection, and cleanup as distinct stages and results.
• Check the test toolchain before or during the supported OBD test workflow. If OBD installs a missing tool, JRE, OBClient, link, or profile entry as part of the requested test, record and verify that prerequisite change instead of treating it as a separate discovery gate.
• For mysqltest, keep the installed long option equivalent to --disable-reboot enabled unless the user separately authorizes the exact snapshot/redeploy behavior. In verified current plugins, an ordinary failed case can otherwise lead to a forced redeploy even without auto-retry.
• Do not put database passwords in reusable commands, output, or reports.

For tenant creation or backup, use tenant-management. For deployment, component, or cluster lifecycle work, use cluster-management.