rocm-systems/projects/rocprofiler-systems/tests/rocpd-validation-rules/gpu-connect/amd-smi-rules.json

{
    "required_tables": [
        {
            "min_rows": 1,
            "name_prefix": "rocpd_info_pmc",
            "required_columns": [
                "agent_id",
                "target_arch",
                "name",
                "symbol",
                "description",
                "units",
                "value_type"
            ],
            "validation_queries": [
                {
                    "comparison": "greater_than",
                    "description": "Check for Xgmi amd-smi metrics",
                    "error_message": "Did not find Xgmi data in amd-smi metrics",
                    "expected_result": 1,
                    "query": "SELECT COUNT(*) as count FROM {table_name} WHERE symbol LIKE 'Xgmi%'"
                },
                {
                    "comparison": "greater_than",
                    "description": "Check for Pcie amd-smi metrics",
                    "error_message": "Did not find Pcie data in amd-smi metrics",
                    "expected_result": 1,
                    "query": "SELECT COUNT(*) as count FROM {table_name} WHERE symbol LIKE 'Pcie%'"
                }
            ]
        },
        {
            "min_rows": 500,
            "name_prefix": "rocpd_pmc_event",
            "required_columns": [
                "event_id",
                "pmc_id",
                "value"
            ],
            "validation_queries": [
                {
                    "comparison": "greater_than",
                    "description": "Check for amd-smi monitoring busy times",
                    "error_message": "Less than expected number of captured amd-smi xgmi link speed samples!",
                    "expected_result": 100,
                    "query": "SELECT COUNT(*) as count FROM {table_name} event JOIN rocpd_info_pmc info ON event.pmc_id = info.id WHERE info.name = 'device_xgmi_link_speed'"
                },
                {
                    "comparison": "greater_than",
                    "description": "Check for amd-smi monitoring busy times",
                    "error_message": "Less than expected number of captured amd-smi xgmi link width samples!",
                    "expected_result": 100,
                    "query": "SELECT COUNT(*) as count FROM {table_name} event JOIN rocpd_info_pmc info ON event.pmc_id = info.id WHERE info.name = 'device_xgmi_link_width'"
                },
                {
                    "comparison": "greater_than",
                    "description": "Check for amd-smi monitoring busy times",
                    "error_message": "Less than expected number of captured amd-smi xgmi read data samples!",
                    "expected_result": 100,
                    "query": "SELECT COUNT(*) as count FROM {table_name} event JOIN rocpd_info_pmc info ON event.pmc_id = info.id WHERE info.name LIKE 'device_xgmi_read_data%'"
                },
                {
                    "comparison": "greater_than",
                    "description": "Check for amd-smi monitoring busy times",
                    "error_message": "Less than expected number of captured amd-smi xgmi write data samples!",
                    "expected_result": 100,
                    "query": "SELECT COUNT(*) as count FROM {table_name} event JOIN rocpd_info_pmc info ON event.pmc_id = info.id WHERE info.name LIKE 'device_xgmi_write_data%'"
                },
                {
                    "comparison": "greater_than",
                    "description": "Check for amd-smi monitoring busy times",
                    "error_message": "Less than expected number of captured amd-smi pcie bandwidth instantaneous samples!",
                    "expected_result": 100,
                    "query": "SELECT COUNT(*) as count FROM {table_name} event JOIN rocpd_info_pmc info ON event.pmc_id = info.id WHERE info.name = 'device_pcie_bandwidth_inst'"
                },
                {
                    "comparison": "greater_than",
                    "description": "Check for amd-smi monitoring busy times",
                    "error_message": "Less than expected number of captured amd-smi pcie bandwidth accumulated samples!",
                    "expected_result": 100,
                    "query": "SELECT COUNT(*) as count FROM {table_name} event JOIN rocpd_info_pmc info ON event.pmc_id = info.id WHERE info.name = 'device_pcie_bandwidth_acc'"
                },
                {
                    "comparison": "greater_than",
                    "description": "Check for amd-smi monitoring busy times",
                    "error_message": "Less than expected number of captured amd-smi pcie link speed samples!",
                    "expected_result": 100,
                    "query": "SELECT COUNT(*) as count FROM {table_name} event JOIN rocpd_info_pmc info ON event.pmc_id = info.id WHERE info.name = 'device_pcie_link_speed'"
                },
                {
                    "comparison": "greater_than",
                    "description": "Check for amd-smi monitoring busy times",
                    "error_message": "Less than expected number of captured amd-smi pcie link width samples!",
                    "expected_result": 100,
                    "query": "SELECT COUNT(*) as count FROM {table_name} event JOIN rocpd_info_pmc info ON event.pmc_id = info.id WHERE info.name = 'device_pcie_link_width'"
                }
            ]
        }
    ]
}