给一个2层嵌套的数据

{"metric":"smsSendSucessCout1","nested":{"clientId":"client-id","number":20},"time":1537075089042,"value":"1.0"}

这个其实只要定义好schema就可以了，直接上源码

/**
 * 实体对象
 * @author liu.zhiqiang
 * @version $Id: AthenaxKafkaDTO.java, v 0.1 2018-09-12 18:14 liu.zhiqiang Exp $$
 */
@Data
public class AthenaxKafkaDTO {
    private String                  topic;
    private String                  brokerAddress;
    private String                  groupId;
    private List<AthenaxJSONSchema> schemas;

    // org.apache.flink.formats.json.JsonRowDeserializationSchema.deserialize(JsonRowDeserializationSchema.java:97)
    private TypeInformation<?> getTypeInfo(AthenaxJSONSchema schema) throws TableNotExistException {
        String type = schema.getType();
        if (type.toLowerCase().equals("string")) {
            return BasicTypeInfo.STRING_TYPE_INFO;
        } else if (type.toLowerCase().equals("boolean")) {
            return BasicTypeInfo.BOOLEAN_TYPE_INFO;
        } else if (type.toLowerCase().equals("byte")) {
            return BasicTypeInfo.BYTE_TYPE_INFO;
        } else if (type.toLowerCase().equals("short")) {
            return BasicTypeInfo.SHORT_TYPE_INFO;
        } else if (type.toLowerCase().equals("int")) {
            return BasicTypeInfo.INT_TYPE_INFO;
        } else if (type.toLowerCase().equals("long")) {
            return BasicTypeInfo.LONG_TYPE_INFO;
        } else if (type.toLowerCase().equals("float")) {
            return BasicTypeInfo.FLOAT_TYPE_INFO;
        } else if (type.toLowerCase().equals("double")) {
            return BasicTypeInfo.DOUBLE_TYPE_INFO;
        } else if (type.toLowerCase().equals("char")) {
            return BasicTypeInfo.CHAR_TYPE_INFO;
        } else if (type.toLowerCase().equals("void")) {
            return BasicTypeInfo.VOID_TYPE_INFO;
        } else if (type.toLowerCase().equals("biginteger")) {
            return BasicTypeInfo.BIG_INT_TYPE_INFO;
        } else if (type.toLowerCase().equals("bigdecimal")) {
            return BasicTypeInfo.BIG_DEC_TYPE_INFO;
        } else if (type.toLowerCase().equals("date")) {
            return SqlTimeTypeInfo.DATE;
        } else if (type.toLowerCase().equals("timestamp")) {
            return SqlTimeTypeInfo.TIMESTAMP;
        } else if (type.toLowerCase().equals("time")) {
            return SqlTimeTypeInfo.TIME;
        } else if (type.toLowerCase().equals("row")) {
            //目前复杂的支持row,其它的待扩展
            List<AthenaxJSONSchema> schemaList = schema.getSchema();
            List<String> nameList = new ArrayList<String>();
            List<TypeInformation<?>> typeInformationList = new ArrayList<TypeInformation<?>>();
            //有数据就继续处理
            if (null != schemaList) {
                for (AthenaxJSONSchema childSchema : schemaList) {
                    nameList.add(childSchema.getKey());
                    typeInformationList.add(getTypeInfo(childSchema));
                }
            }
            //构造返回结果
            String[] nameArray = new String[nameList.size()];
            nameList.toArray(nameArray);
            TypeInformation<?>[] typeInformationArray = new TypeInformation<?>[typeInformationList
                .size()];
            typeInformationList.toArray(typeInformationArray);
            //return new RowTypeInfo();
            return Types.ROW(nameArray, typeInformationArray);
        } else {
            String errorMsg = "fail to find TypeInformation for type[" + type + "]";
            LoggerUtil.error(errorMsg);
            throw new TableNotExistException(type, type);
        }
    }

    // TableSchema SCHEMA = new TableSchema(new String[] { "id", "proctime" },
    // new TypeInformation<?>[] { BasicTypeInfo.INT_TYPE_INFO,
    // SqlTimeTypeInfo.TIMESTAMP });
    public TableSchema getTableSchema() throws TableNotExistException {
        String[] keys = new String[schemas.size()];
        TypeInformation<?>[] types = new TypeInformation<?>[schemas.size()];
        int index = 0;
        for (AthenaxJSONSchema schema : schemas) {
            keys[index] = schema.getKey();
            types[index++] = getTypeInfo(schema);
        }
        return new TableSchema(keys, types);
    }

    public String getTimeField() {
        //目前只支持第1层定义,以后再优化
        for (AthenaxJSONSchema schema : schemas) {
            if (1 == schema.getTime()) {
                return schema.getKey();
            }
        }
        return null;
    }

}

然后sql定义这么写

SELECT SUM(nested.number) as nestedNumber,hundredFunction(SUM(CAST(`value` AS DOUBLE))) as `sum`,COUNT(`value`) as `count`,AVG(CAST(`value` AS DOUBLE)) as `avg`,MAX(CAST(`value` AS DOUBLE)) as `max`,MIN(CAST(`value` AS DOUBLE)) as `min`,MAX(`time`)  as `time` FROM input.tumble_topic_input_15 WHERE metric IS NOT NULL AND `value` IS NOT NULL and `time` IS NOT NULL GROUP BY metric,TUMBLE(`time`, INTERVAL '3' SECOND)

schema定义如下：

"schemas": [{"key": "metric","type": "string"},{"key": "time","type": "timestamp","time": 1},{"key":"value","type": "string"},{"key": "nested","type": "row","schema":[{"key": "clientId","type": "string"},{"key": "number","type": "int"}]}]

测试通过！