kafka 按时间戳消费

步骤

• 获取当前 topic 的分区列表
• 利用 offsets_for_times() + 时间戳查找给定分区的偏移量，如：找到开始时间的偏移量
• 循环每个分区，设置偏移量
• 根据 end_offset 或结束时间退出

import json

from kafka import KafkaConsumer, TopicPartition

TOPIC = ""
BROKER = ""
GROUP_ID = ""

consumer = KafkaConsumer(bootstrap_servers=BROKER,
                         group_id=GROUP_ID,
                         api_version=(1, 1, 0),
                         max_poll_records=100,  # 每次最大消费数量
                         connections_max_idle_ms=900000,
                         auto_offset_reset='earliest',
                         consumer_timeout_ms=1000,
                         enable_auto_commit=True,  # 每过一段时间自动提交所有已消费的消息（在迭代时提交）
                         auto_commit_interval_ms=5000,
                         sasl_mechanism="",
                         sasl_plain_username="",
                         sasl_plain_password="")


def main():
    start_time, end_time = 1666097095932, 1666098095932

    partition_list = consumer.partitions_for_topic(TOPIC)  # 获取 topic 分区 id 集合，当前 topic存在多少个分区
    assigned_topics = [TopicPartition(topic=TOPIC, partition=index) for index in partition_list]

    consumer.assign(assigned_topics)  # 手动将 TopicPartitions 列表分配给次消费者

    partitions = consumer.assignment()  # 获取当前分配给该消费者的 TopicPartitions

    partition_start_timestamp = {part: start_time for part in partitions}
    partition_end_timestamp = {part: end_time for part in partitions}

    # 根据时间戳查找给定分区的偏移量，这里分别查了 start_time 和 end_time {TopicPartition: OffsetAndTimestamp}
    mapping = consumer.offsets_for_times(partition_start_timestamp)
    # end_offsets = consumer.end_offsets(list(partition_end_timestamp.keys()))
    end_offsets = consumer.offsets_for_times(partition_end_timestamp)

    # 按分区消费 {"TopicPartition(topic="", partition=5): OffsetAndTimestamp(offset=26616345, timestamp=1663560125950)}
    for partition, ts in mapping.items():
        # end_offset = end_offsets.get(partition)
        consumer.seek(partition, ts[0])  # 每个分区设置偏移量
        for msg in consumer:
            value = json.loads(msg.value.decode('utf-8'))

            # 方法一：按时间戳来退出
            if msg.timestamp >= end_time:
                consumer.close()
                break

            # # 方法二：使用 end_offset 来退出
            # if msg.offset == end_offset - 1:
            #     consumer.close()
            #     break


if __name__ == '__main__':
    main()

参考：get results from kafka for a specific period of time