コード例 #1
0
 def test_column_group_measurement(self):
     measurement = Measurement(
         metric=Metric.MIN,
         column_name='AGE',
         group_values=[GroupValue(group={'country': 'US'}, value=3.4)])
     self.assertEqual("min(AGE): \n  group{'country': 'US'} = 3.4",
                      str(measurement))
     JsonHelper.to_json(JsonHelper.to_jsonnable(measurement.to_dict()))
コード例 #2
0
    def _run_sql_metric_default_and_run_tests(self,
                                              sql_metric: SqlMetricYml,
                                              resolved_sql: str,
                                              scan_column: Optional[ScanColumn] = None):
        try:
            row_tuple, description = self.warehouse.sql_fetchone_description(resolved_sql)
            self.queries_executed += 1

            test_variables = self._get_test_variables(scan_column)
            column_name = scan_column.column_name if scan_column is not None else None

            measurements = []
            for i in range(len(row_tuple)):
                metric_name = sql_metric.metric_names[i] if sql_metric.metric_names is not None else description[i][0]
                metric_value = row_tuple[i]
                logging.debug(f'SQL metric {sql_metric.title} {metric_name} -> {metric_value}')
                measurement = Measurement(metric=metric_name, value=metric_value, column_name=column_name)
                test_variables[metric_name] = metric_value
                self._log_and_append_query_measurement(measurements, measurement)

            self._flush_measurements(measurements)

            sql_metric_test_results = self._execute_tests(sql_metric.tests, test_variables)
            self._flush_test_results(sql_metric_test_results)
        except Exception as e:
            self.scan_result.add_error(ScanError(f'Exception during sql metric query {resolved_sql}', e))
コード例 #3
0
    def _query_columns_metadata(self):
        sql = self.warehouse.dialect.sql_columns_metadata_query(self.scan_yml.table_name)
        column_tuples = self.warehouse.sql_fetchall(sql) if sql != '' else self.warehouse.dialect.sql_columns_metadata(
            self.scan_yml.table_name)
        self.queries_executed += 1
        self.column_metadatas = []
        for column_tuple in column_tuples:
            name = column_tuple[0]
            type = column_tuple[1]
            nullable = 'YES' == column_tuple[2].upper()
            self.column_metadatas.append(ColumnMetadata(name, type, nullable))
        logging.debug(str(len(self.column_metadatas)) + ' columns:')

        self.column_names: List[str] = [column_metadata.name for column_metadata in self.column_metadatas]
        self.scan_columns: dict = {}
        for column_metadata in self.column_metadatas:
            scan_column = ScanColumn(self, column_metadata)
            if scan_column.is_supported:
                logging.debug(f'  {scan_column.column_name} ({scan_column.column.type}) '
                              f'{"" if scan_column.column.nullable else "not null"}')
                self.scan_columns[column_metadata.name.lower()] = scan_column
            else:
                logging.debug(f'  {scan_column.column_name} ({scan_column.column.type}) -> unsupported, skipped!')

        schema_measurement_value = [column_metadata.to_json() for column_metadata in self.column_metadatas]
        schema_measurement = Measurement(Metric.SCHEMA, value=schema_measurement_value)
        self._log_measurement(schema_measurement)
        self._flush_measurements([schema_measurement])
コード例 #4
0
    def _run_sql_metric_default_and_run_tests(
            self,
            sql_metric: SqlMetricYml,
            resolved_sql: str,
            column_name_lower: Optional[str] = None):
        row_tuple, description = self.warehouse.sql_fetchone_description(
            resolved_sql)
        test_variables = self._get_test_variables(column_name_lower)

        measurements = []
        for i in range(len(row_tuple)):
            metric_name = sql_metric.metric_names[
                i] if sql_metric.metric_names is not None else description[i][0]
            metric_value = row_tuple[i]
            logging.debug(
                f'SQL metric {sql_metric.name} {metric_name} -> {metric_value}'
            )
            measurement = Measurement(metric=metric_name, value=metric_value)
            test_variables[metric_name] = metric_value
            self._log_and_append_query_measurement(measurements, measurement)

        self._flush_measurements(measurements)

        sql_metric_test_results = self._execute_tests(sql_metric.tests,
                                                      test_variables)
        self._flush_test_results(sql_metric_test_results)
コード例 #5
0
    def _query_columns_metadata(self):
        sql = self.warehouse.dialect.sql_columns_metadata_query(
            self.scan_yml.table_name)
        column_tuples = self.warehouse.sql_fetchall(sql)
        self.column_metadatas = []
        for column_tuple in column_tuples:
            name = column_tuple[0]
            type = column_tuple[1]
            nullable = 'YES' == column_tuple[2].upper()
            self.column_metadatas.append(ColumnMetadata(name, type, nullable))
        logging.debug(str(len(self.column_metadatas)) + ' columns:')
        for column in self.column_metadatas:
            logging.debug(
                f'  {column.name} {column.type} {"" if column.nullable else "not null"}'
            )

        self.column_names: List[str] = [
            column_metadata.name for column_metadata in self.column_metadatas
        ]
        self.scan_columns: dict = {
            column.name.lower(): ScanColumn(self, column)
            for column in self.column_metadatas
        }
        schema_measurement_value = [
            column_metadata.to_json()
            for column_metadata in self.column_metadatas
        ]
        schema_measurement = Measurement(Metric.SCHEMA,
                                         value=schema_measurement_value)
        self._log_measurement(schema_measurement)
        self._flush_measurements([schema_measurement])
コード例 #6
0
 def test_column_group_measurement(self):
     measurement_str = str(
         Measurement(metric=Metric.MIN,
                     column_name='AGE',
                     value=3.4,
                     group_values={'country': 'US'}))
     self.assertEqual(measurement_str, 'min(AGE){"country": "US"} = 3.4')
コード例 #7
0
    def _run_sql_metric_with_groups_and_run_tests(self,
                                                  sql_metric: SqlMetricYml,
                                                  resolved_sql: str,
                                                  scan_column: Optional[ScanColumn]):
        try:
            measurements = []
            test_results = []
            group_fields_lower = set(group_field.lower() for group_field in sql_metric.group_fields)

            rows, description = self.warehouse.sql_fetchall_description(resolved_sql)
            self.queries_executed += 1

            group_values_by_metric_name = {}
            for row in rows:
                group = {}
                metric_values = {}

                for i in range(len(row)):
                    metric_name = sql_metric.metric_names[i] if sql_metric.metric_names is not None else description[i][
                        0]
                    metric_value = row[i]
                    if metric_name.lower() in group_fields_lower:
                        group[metric_name] = metric_value
                    else:
                        metric_values[metric_name] = metric_value

                if not group:
                    logging.error(f'None of the declared group_fields were found in '
                                  f'result: {sql_metric.group_fields}. Skipping result.')
                else:
                    for metric_name in metric_values:
                        metric_value = metric_values[metric_name]
                        if metric_name not in group_values_by_metric_name:
                            group_values_by_metric_name[metric_name] = []
                        group_values = group_values_by_metric_name[metric_name]
                        group_values.append(GroupValue(group=group, value=metric_value))
                        logging.debug(f'SQL metric {sql_metric.title} {metric_name} {group} -> {metric_value}')

                    sql_metric_tests = sql_metric.tests
                    test_variables = self._get_test_variables(scan_column)
                    test_variables.update(metric_values)
                    sql_metric_test_results = self._execute_tests(sql_metric_tests, test_variables, group)
                    test_results.extend(sql_metric_test_results)

            column_name = scan_column.column_name if scan_column is not None else None
            for metric_name in group_values_by_metric_name:
                group_values = group_values_by_metric_name[metric_name]
                measurement = Measurement(metric=metric_name, group_values=group_values, column_name=column_name)
                self._log_and_append_query_measurement(measurements, measurement)

            self._flush_measurements(measurements)
            self._flush_test_results(test_results)
        except Exception as e:
            self.scan_result.add_error(ScanError(f'Exception during sql metric groups query {resolved_sql}', e))
コード例 #8
0
    def _run_sql_metric_with_groups_and_run_tests(
            self,
            sql_metric: SqlMetricYml,
            resolved_sql: str,
            column_name_lower: Optional[str] = None):
        measurements = []
        test_results = []
        group_fields_lower = set(group_field.lower()
                                 for group_field in sql_metric.group_fields)
        rows, description = self.warehouse.sql_fetchall_description(
            resolved_sql)
        group_values_by_metric_name = {}
        for row in rows:
            group = {}
            metric_values = {}
            for i in range(len(row)):
                metric_name = sql_metric.metric_names[
                    i] if sql_metric.metric_names is not None else description[
                        i][0]
                metric_value = row[i]
                if metric_name.lower() in group_fields_lower:
                    group[metric_name] = metric_value
                else:
                    metric_values[metric_name] = metric_value

            for metric_name in metric_values:
                metric_value = metric_values[metric_name]
                if metric_name not in group_values_by_metric_name:
                    group_values_by_metric_name[metric_name] = []
                group_values = group_values_by_metric_name[metric_name]
                group_values.append(GroupValue(group=group,
                                               value=metric_value))
                logging.debug(
                    f'SQL metric {sql_metric.name} {metric_name} {group} -> {metric_value}'
                )

            sql_metric_tests = sql_metric.tests
            test_variables = self._get_test_variables(column_name_lower)
            test_variables.update(metric_values)
            sql_metric_test_results = self._execute_tests(
                sql_metric_tests, test_variables, group)
            test_results.extend(sql_metric_test_results)

        for metric_name in group_values_by_metric_name:
            group_values = group_values_by_metric_name[metric_name]
            measurement = Measurement(metric=metric_name,
                                      group_values=group_values)
            self._log_and_append_query_measurement(measurements, measurement)

        self._flush_measurements(measurements)
        self._flush_test_results(test_results)
コード例 #9
0
 def test_column_group_measurement_empty_list(self):
     measurement = Measurement(metric=Metric.MIN,
                               column_name='AGE',
                               group_values=[])
     self.assertEqual('min(AGE): no groups', str(measurement))
     JsonHelper.to_json(JsonHelper.to_jsonnable(measurement.to_dict()))
コード例 #10
0
ファイル: scan.py プロジェクト: ashpreetbedi/soda-sql
    def _query_group_by_value(self):
        for column_name_lower, scan_column in self.scan_columns.items():
            measurements = []

            column_name = scan_column.column_name

            if scan_column.is_any_metric_enabled(
                    [Metric.DISTINCT, Metric.UNIQUENESS, Metric.UNIQUE_COUNT,
                     Metric.MINS, Metric.MAXS, Metric.FREQUENT_VALUES, Metric.DUPLICATE_COUNT]):

                group_by_cte = scan_column.get_group_by_cte()
                numeric_value_expr = scan_column.get_group_by_cte_numeric_value_expression()
                order_by_value_expr = scan_column.get_order_by_cte_value_expression(numeric_value_expr)

                if self.scan_yml.is_any_metric_enabled(
                        [Metric.DISTINCT, Metric.UNIQUENESS, Metric.UNIQUE_COUNT, Metric.DUPLICATE_COUNT],
                        column_name):

                    sql = (f'{group_by_cte} \n'
                           f'SELECT COUNT(*), \n'
                           f'       COUNT(CASE WHEN frequency = 1 THEN 1 END), \n'
                           f'       SUM(frequency) \n'
                           f'FROM group_by_value')

                    query_result_tuple = self.warehouse.sql_fetchone(sql)
                    self.queries_executed += 1

                    distinct_count = query_result_tuple[0]
                    unique_count = query_result_tuple[1]
                    valid_count = query_result_tuple[2] if query_result_tuple[2] else 0
                    duplicate_count = distinct_count - unique_count

                    self._log_and_append_query_measurement(
                        measurements, Measurement(Metric.DISTINCT, column_name, distinct_count))
                    self._log_and_append_query_measurement(
                        measurements, Measurement(Metric.UNIQUE_COUNT, column_name, unique_count))

                    derived_measurements = [Measurement(Metric.DUPLICATE_COUNT, column_name, duplicate_count)]
                    if valid_count > 1:
                        uniqueness = (distinct_count - 1) * 100 / (valid_count - 1)
                        derived_measurements.append(Measurement(Metric.UNIQUENESS, column_name, uniqueness))
                    self._log_and_append_derived_measurements(measurements, derived_measurements)

                if scan_column.is_metric_enabled(Metric.MINS) and order_by_value_expr:
                    sql = (f'{group_by_cte} \n'
                           f'SELECT value \n'
                           f'FROM group_by_value \n'
                           f'ORDER BY {order_by_value_expr} ASC \n'
                           f'LIMIT {scan_column.mins_maxs_limit} \n')

                    rows = self.warehouse.sql_fetchall(sql)
                    self.queries_executed += 1

                    mins = [row[0] for row in rows]
                    self._log_and_append_query_measurement(measurements, Measurement(Metric.MINS, column_name, mins))

                if self.scan_yml.is_metric_enabled(Metric.MAXS, column_name) and order_by_value_expr:

                    sql = (f'{group_by_cte} \n'
                           f'SELECT value \n'
                           f'FROM group_by_value \n'
                           f'ORDER BY {order_by_value_expr} DESC \n'
                           f'LIMIT {scan_column.mins_maxs_limit} \n')

                    rows = self.warehouse.sql_fetchall(sql)
                    self.queries_executed += 1

                    maxs = [row[0] for row in rows]
                    self._log_and_append_query_measurement(measurements, Measurement(Metric.MAXS, column_name, maxs))

                if self.scan_yml.is_metric_enabled(Metric.FREQUENT_VALUES, column_name) \
                        and (scan_column.is_number or scan_column.is_column_numeric_text_format):

                    frequent_values_limit = self.scan_yml.get_frequent_values_limit(column_name)
                    sql = (f'{group_by_cte} \n'
                           f'SELECT value, frequency \n'
                           f'FROM group_by_value \n'
                           f'ORDER BY frequency DESC \n'
                           f'LIMIT {frequent_values_limit} \n')

                    rows = self.warehouse.sql_fetchall(sql)
                    self.queries_executed += 1

                    frequent_values = [{'value': row[0], 'frequency': row[1]} for row in rows]
                    self._log_and_append_query_measurement(
                        measurements, Measurement(Metric.FREQUENT_VALUES, column_name, frequent_values))

            self._flush_measurements(measurements)
コード例 #11
0
 def test_row_count(self):
     measurement = Measurement(metric=Metric.ROW_COUNT, value=5)
     self.assertEqual('row_count = 5', str(measurement))
     JsonHelper.to_json(JsonHelper.to_jsonnable(measurement.to_dict()))
コード例 #12
0
 def test_column_measurement_list_value(self):
     measurement_str = str(
         Measurement(metric=Metric.MINS,
                     column_name='chars',
                     value=['a', 'b']))
     self.assertEqual(measurement_str, 'mins(chars) = a, b')
コード例 #13
0
 def test_column_measurement(self):
     measurement_str = str(
         Measurement(metric=Metric.MIN, column_name='AGE', value=3.4))
     self.assertEqual(measurement_str, 'min(AGE) = 3.4')
コード例 #14
0
 def test_row_count(self):
     measurement_str = str(Measurement(metric=Metric.ROW_COUNT, value=5))
     self.assertEqual(measurement_str, 'row_count = 5')
コード例 #15
0
ファイル: scan.py プロジェクト: udemy/soda-sql
    def _run_sql_metric_failed_rows(self,
                                    sql_metric: SqlMetricYml,
                                    resolved_sql: str,
                                    scan_column: Optional[ScanColumn] = None):

        try:
            if self.soda_server_client:

                logging.debug(
                    f'Sending failed rows for sql metric {sql_metric.name} to Soda Cloud'
                )
                with tempfile.TemporaryFile() as temp_file:
                    rows_stored, sample_columns = \
                        self.sampler.save_sample_to_local_file(resolved_sql, temp_file)

                    column_name = scan_column.column_name if scan_column else None
                    measurement = Measurement(metric=sql_metric.name,
                                              value=rows_stored,
                                              column_name=column_name)

                    measurements = []
                    self._log_and_append_query_measurement(
                        measurements, measurement)
                    self._flush_measurements(measurements)

                    test = sql_metric.tests[0]

                    test_variables = {sql_metric.name: rows_stored}

                    test_result = self.evaluate_test(test, test_variables)
                    self._flush_test_results([test_result])

                    if rows_stored > 0:
                        temp_file_size_in_bytes = temp_file.tell()
                        temp_file.seek(0)

                        file_path = self.sampler.create_file_path_failed_rows_sql_metric(
                            sql_metric)

                        file_id = self.soda_server_client.scan_upload(
                            self.scan_reference, file_path, temp_file,
                            temp_file_size_in_bytes)

                        self.soda_server_client.scan_file(
                            scan_reference=self.scan_reference,
                            sample_type='failedRowsSample',
                            stored=int(rows_stored),
                            total=int(rows_stored),
                            source_columns=sample_columns,
                            file_id=file_id,
                            column_name=sql_metric.column_name,
                            test_ids=[test.id],
                            sql_metric_name=sql_metric.name)
                        logging.debug(
                            f'Sent failed rows for sql metric ({rows_stored}/{rows_stored}) to Soda Cloud'
                        )
                    else:
                        logging.debug(
                            f'No failed rows for sql metric ({sql_metric.name})'
                        )
            else:
                failed_rows_tuples, description = self.warehouse.sql_fetchall_description(
                    sql=resolved_sql)
                if len(failed_rows_tuples) > 0:
                    table_text = self._table_to_text(failed_rows_tuples,
                                                     description)
                    logging.debug(
                        f'Failed rows for sql metric sql {sql_metric.name}:\n'
                        + table_text)
                else:
                    logging.debug(
                        f'No failed rows for sql metric sql {sql_metric.name}')
        except Exception as e:
            logging.exception(
                f'Could not perform sql metric failed rows \n{resolved_sql}',
                e)
            self.scan_result.add_error(
                ScanError(
                    f'Exception during sql metric failed rows query {resolved_sql}',
                    e))
コード例 #16
0
ファイル: scan.py プロジェクト: ashpreetbedi/soda-sql
    def _query_aggregations(self):
        # This measurements list is used to match measurements with the query field order.
        # After query execution, the value of the measurements will be extracted from the query result and
        # the measurements will be added with self.add_query(measurement)
        measurements: List[Measurement] = []

        fields: List[str] = []

        dialect = self.warehouse.dialect

        if self.scan_yml.is_metric_enabled(Metric.ROW_COUNT):
            fields.append(dialect.sql_expr_count_all())
            measurements.append(Measurement(Metric.ROW_COUNT))

        # maps db column names (lower) to missing and invalid metric indices in the measurements
        # eg { 'colname': {'missing': 2, 'invalid': 3}, ...}
        column_metric_indices = {}

        for column_name_lower, scan_column in self.scan_columns.items():
            metric_indices = {}
            column_metric_indices[column_name_lower] = metric_indices
            column_name = scan_column.column_name

            if scan_column.is_missing_enabled:
                metric_indices['non_missing'] = len(measurements)
                if scan_column.non_missing_condition:
                    fields.append(dialect.sql_expr_count_conditional(scan_column.non_missing_condition))
                else:
                    fields.append(dialect.sql_expr_count(scan_column.qualified_column_name))
                measurements.append(Measurement(Metric.VALUES_COUNT, column_name))

            if scan_column.is_valid_enabled:
                metric_indices['valid'] = len(measurements)
                if scan_column.non_missing_and_valid_condition:
                    fields.append(dialect.sql_expr_count_conditional(scan_column.non_missing_and_valid_condition))
                else:
                    fields.append(dialect.sql_expr_count(scan_column.qualified_column_name))
                measurements.append(Measurement(Metric.VALID_COUNT, column_name))

            if scan_column.is_text:
                length_expr = dialect.sql_expr_conditional(
                        scan_column.non_missing_and_valid_condition,
                        dialect.sql_expr_length(scan_column.qualified_column_name)) \
                    if scan_column.non_missing_and_valid_condition \
                    else dialect.sql_expr_length(scan_column.qualified_column_name)

                if self.scan_yml.is_metric_enabled(Metric.MIN_LENGTH, column_name):
                    fields.append(dialect.sql_expr_min(length_expr))
                    measurements.append(Measurement(Metric.MIN_LENGTH, column_name))

                if self.scan_yml.is_metric_enabled(Metric.MAX_LENGTH, column_name):
                    fields.append(dialect.sql_expr_max(length_expr))
                    measurements.append(Measurement(Metric.MAX_LENGTH, column_name))

            if scan_column.has_numeric_values:
                if scan_column.is_metric_enabled(Metric.MIN):
                    fields.append(dialect.sql_expr_min(scan_column.numeric_expr))
                    measurements.append(Measurement(Metric.MIN, column_name))

                if scan_column.is_metric_enabled(Metric.MAX):
                    fields.append(dialect.sql_expr_max(scan_column.numeric_expr))
                    measurements.append(Measurement(Metric.MAX, column_name))

                if scan_column.is_metric_enabled(Metric.AVG):
                    fields.append(dialect.sql_expr_avg(scan_column.numeric_expr))
                    measurements.append(Measurement(Metric.AVG, column_name))

                if scan_column.is_metric_enabled(Metric.SUM):
                    fields.append(dialect.sql_expr_sum(scan_column.numeric_expr))
                    measurements.append(Measurement(Metric.SUM, column_name))

                if scan_column.is_metric_enabled(Metric.VARIANCE):
                    fields.append(dialect.sql_expr_variance(scan_column.numeric_expr))
                    measurements.append(Measurement(Metric.VARIANCE, column_name))

                if scan_column.is_metric_enabled(Metric.STDDEV):
                    fields.append(dialect.sql_expr_stddev(scan_column.numeric_expr))
                    measurements.append(Measurement(Metric.STDDEV, column_name))

        if len(fields) > 0:
            sql = 'SELECT \n  ' + ',\n  '.join(fields) + ' \n' \
                  'FROM ' + self.qualified_table_name
            if self.table_sample_clause:
                sql += f'\n{self.table_sample_clause}'
            if self.filter_sql:
                sql += f'\nWHERE {self.filter_sql}'

            query_result_tuple = self.warehouse.sql_fetchone(sql)
            self.queries_executed += 1

            for i in range(0, len(measurements)):
                measurement = measurements[i]
                measurement.value = query_result_tuple[i]
                self._log_measurement(measurement)

            # Calculating derived measurements
            row_count_measurement = next((m for m in measurements if m.metric == Metric.ROW_COUNT), None)
            if row_count_measurement:
                row_count = row_count_measurement.value
                for column_name_lower, scan_column in self.scan_columns.items():
                    column_name = scan_column.column_name
                    metric_indices = column_metric_indices[column_name_lower]
                    non_missing_index = metric_indices.get('non_missing')
                    if non_missing_index is not None:
                        values_count = measurements[non_missing_index].value
                        missing_count = row_count - values_count
                        missing_percentage = missing_count * 100 / row_count if row_count > 0 else None
                        values_percentage = values_count * 100 / row_count if row_count > 0 else None

                        self._log_and_append_derived_measurements(measurements, [
                            Measurement(Metric.MISSING_PERCENTAGE, column_name, missing_percentage),
                            Measurement(Metric.MISSING_COUNT, column_name, missing_count),
                            Measurement(Metric.VALUES_PERCENTAGE, column_name, values_percentage)
                        ])

                        valid_index = metric_indices.get('valid')
                        if valid_index is not None:
                            valid_count = measurements[valid_index].value
                            invalid_count = row_count - missing_count - valid_count
                            invalid_percentage = invalid_count * 100 / row_count if row_count > 0 else None
                            valid_percentage = valid_count * 100 / row_count if row_count > 0 else None
                            self._log_and_append_derived_measurements(measurements, [
                                Measurement(Metric.INVALID_PERCENTAGE, column_name, invalid_percentage),
                                Measurement(Metric.INVALID_COUNT, column_name, invalid_count),
                                Measurement(Metric.VALID_PERCENTAGE, column_name, valid_percentage)
                            ])

        self._flush_measurements(measurements)
コード例 #17
0
 def test_column_measurement(self):
     measurement = Measurement(metric=Metric.MIN,
                               column_name='AGE',
                               value=3.4)
     self.assertEqual('min(AGE) = 3.4', str(measurement))
     JsonHelper.to_json(JsonHelper.to_jsonnable(measurement.to_dict()))
コード例 #18
0
ファイル: scan.py プロジェクト: ashpreetbedi/soda-sql
    def _query_histograms(self):
        measurements = []
        for column_name_lower, scan_column in self.scan_columns.items():
            column_name = scan_column.column_name

            if scan_column.is_metric_enabled(Metric.HISTOGRAM) and scan_column.numeric_expr:

                buckets: int = scan_column.get_histogram_buckets()

                min_value = scan_column.get_metric_value(Metric.MIN)
                max_value = scan_column.get_metric_value(Metric.MAX)

                if scan_column.has_numeric_values and min_value and max_value and min_value < max_value:
                    # Build the histogram query
                    min_value = floor(min_value * 1000) / 1000
                    max_value = ceil(max_value * 1000) / 1000
                    bucket_width = (max_value - min_value) / buckets

                    boundary = min_value
                    boundaries = [min_value]
                    for i in range(0, buckets):
                        boundary += bucket_width
                        boundaries.append(round(boundary, 3))

                    group_by_cte = scan_column.get_group_by_cte()
                    numeric_value_expr = scan_column.get_group_by_cte_numeric_value_expression()

                    field_clauses = []
                    for i in range(0, buckets):
                        lower_bound = '' if i == 0 else f'{boundaries[i]} <= {numeric_value_expr}'
                        upper_bound = '' if i == buckets - 1 else f'{numeric_value_expr} < {boundaries[i + 1]}'
                        optional_and = '' if lower_bound == '' or upper_bound == '' else ' and '
                        field_clauses.append(
                            f'SUM(CASE WHEN {lower_bound}{optional_and}{upper_bound} THEN frequency END)')

                    fields = ',\n  '.join(field_clauses)

                    sql = (f'{group_by_cte} \n'
                           f'SELECT \n'
                           f'  {fields} \n'
                           f'FROM group_by_value')

                    if self.filter_sql:
                        sql += f' \nWHERE {self.scan.filter_sql}'

                    row = self.warehouse.sql_fetchone(sql)
                    self.queries_executed += 1

                    # Process the histogram query
                    frequencies = []
                    for i in range(0, buckets):
                        frequency = row[i]
                        frequencies.append(0 if not frequency else int(frequency))
                    histogram = {
                        'boundaries': boundaries,
                        'frequencies': frequencies
                    }

                    self._log_and_append_query_measurement(
                        measurements, Measurement(Metric.HISTOGRAM, column_name, histogram))
        self._flush_measurements(measurements)
コード例 #19
0
 def test_column_measurement_list_value(self):
     measurement = Measurement(metric=Metric.MINS,
                               column_name='chars',
                               value=['a', 'b'])
     self.assertEqual("mins(chars) = ['a', 'b']", str(measurement))
     JsonHelper.to_json(JsonHelper.to_jsonnable(measurement.to_dict()))