Exemple #1
0
def main():
    # 1.获取Kafka对象
    kafka = Kafka()
    # 2.读取Kafka数据
    consumer = kafka.read_kafka(CONSUMER_TOPIC)
    # 3.获取HBase对象
    hbase = HBase()
    # 4.创建HBase表
    hbase.get_create_table(TARGET_TABLE_NAME, COLUMN_FAMILY)

    for message in consumer:
        try:
            data = json.loads(message.value)
            print(data)
        except Exception as e:
            print(e)
            continue
        # obj = data[TARGET_TABLE_NAME][0]
        # row_key = obj[ROW_KEY]
        # 5.写入HBase
        # hbase.put_row(TARGET_TABLE_NAME, COLUMN_FAMILY, row_key, dict(obj))
        # 6.写入Kafka
        # message = json.dumps({TARGET_TABLE_NAME: obj})
        # kafka.write_kafka(PRODUCER_TOPIC, message)
        kafka.update_offset(message)

    hbase.close()
Exemple #2
0
def main():
    # 1.获取Kafka对象
    kafka = Kafka(group_id="DP1")
    # 2.读取Kafka数据
    # consumer = kafka.read_kafka(CONSUMER_TOPIC)
    for i in range(100):
        print(i)
        kafka.write_kafka(CONSUMER_TOPIC, f"message_{i}")
    print("end")
Exemple #3
0
def write_kafka():
    topic = request.args.get("topic")
    message = request.args.get("message")
    if not topic or not message:
        return to_dict_msg(
            status=200,
            msg="write kafka failed, topic and message can not empty")
    try:
        kafka = Kafka()
        kafka.write_kafka(topic, json.loads(message))
        return to_dict_msg(status=200, msg="write kafka success")
    except Exception as e:
        print(e)
        return to_dict_msg(status=500)
Exemple #4
0
def main():
    # 1.获取Kafka对象
    kafka = Kafka()
    # 2.读取Kafka数据
    consumer = kafka.read_kafka(CONSUMER_TOPIC)
    # 3.获取HBase对象
    hbase = HBase()
    # 4.创建HBase表
    hbase.get_create_table(TARGET_TABLE_NAME, COLUMN_FAMILY)
    for message in consumer:
        data = json.loads(message.value)
        if data['table'] == SOURCE_TABLE_NAME:
            obj = data['data'][0]
            row_key = obj[ROW_KEY]
            # 5.写入HBase
            hbase.put_row(TARGET_TABLE_NAME, COLUMN_FAMILY, row_key, dict(obj))
            # 6.写入Kafka
            message = json.dumps({TARGET_TABLE_NAME: obj})
            kafka.write_kafka(PRODUCER_TOPIC, message)
def main():
    # 1.获取Kafka对象
    kafka = Kafka(group_id='DP3')
    print('1')
    # 2.读取Kafka数据
    consumer = kafka.read_kafka(CONSUMER_TOPIC)

    # 3.获取HBase对象
    hbase = HBase()
    # 4.创建HBase表
    hbase.get_create_table(SINGLE_TARGET_TABLE_NAME, COLUMN_FAMILY)
    hbase.get_create_table(STATS_TARGET_TABLE_NAME, COLUMN_FAMILY)

    # modelSuccess 改成 isSuccess。内容要转义  True->1 ,False->0
    for msg in consumer:
        try:
            data = json.loads(msg.value)
            # modelSuccess 改成 isSuccess
            data["isSuccess"] = '1' if data["modelSuccess"] else '0'
            del data["modelSuccess"]

            print('2')

            row_key_model_code = data[STATS_ROW_KEY_MODEL_CODE]
            row_key_timestamp = data[STATS_ROW_KEY_TIMESTAMP]
            row_key_subject_code = data[STATS_ROW_KEY_SUBJECT_CODE]
            log_time_cost = data["timeCost"]  # 响应时长

            row_key_model_code = row_key_model_code[::-1]

            row_key_single = data[SINGLE_ROW_KEY]
            # 模型编码5+用户标识符18+时间戳13=36位
            row_kwy_stats = row_key_model_code + row_key_subject_code + row_key_timestamp

            result = hbase.get_rows(STATS_TARGET_TABLE_NAME,
                                    row_prefix=row_key_model_code +
                                    row_key_subject_code)
            print('3')
            if result:
                # 表中已有记录
                print('4')
                tmp_data = result[len(result) - 1]
                # 更新callDate、callTimes、successTimes、sumTime
                tmp_data["callDate"], tmp_data["callTimes"], tmp_data["successTimes"], tmp_data["sumTime"] = \
                    row_key_timestamp, \
                    str(int(tmp_data["callTimes"]) + 1), \
                    str(int(tmp_data["successTimes"]) + 1) if int(data["isSuccess"]) else tmp_data["successTimes"], \
                    str(int(tmp_data["sumTime"]) + log_time_cost)
                print("tmp_data: ", tmp_data)
                hbase.put_row(STATS_TARGET_TABLE_NAME, COLUMN_FAMILY,
                              row_kwy_stats, tmp_data)
                print('5')
            else:
                # 表中没有记录,日志统计表数据初始化
                print('6')
                tmp_data = {
                    "callDate": row_key_timestamp,
                    "subjectCode": row_key_subject_code,
                    "subjectName": data["subjectName"],
                    "subjectType": data["subjectType"],
                    "modelCode": row_key_model_code,
                    "modelName": data["modelName"],
                    "callTimes": '0',
                    "successTimes": '0',
                    "sumTime": '0'
                }
                hbase.put_row(STATS_TARGET_TABLE_NAME, COLUMN_FAMILY,
                              row_kwy_stats, tmp_data)
                print('7')

        except Exception as e:
            print(e)
            continue
        # 5.写入HBase
        # STEP1: 全部数据写进hbase al:s003_evt_model_log表中
        hbase.put_row(SINGLE_TARGET_TABLE_NAME, COLUMN_FAMILY, row_key_single,
                      data)
        kafka.update_offset(msg)
        print('8')