zhen ma's Public Library

instead of deleting a message immediately after consumption, we can retain messages for a relative long period (say a week).

First the stream is partitioned on the brokers into a set of distinct partitions. The semantic meaning of these partitions is left up to the producer and the producer specifies which partition a message belongs to. Within a partition messages are stored in the order in which they arrive at the broker, and will be given out to consumers in that same order. This means that rather than store metadata for each message (marking it as consumed, say), we just need to store the "high water mark" for each combination of consumer, topic, and partition. Hence the total metadata required to summarize the state of the consumer is actually quite small. In Kafka we refer to this high-water mark as "the offset" for reasons that will become clear in the implementation section.

Kafka also maintains this state about what has been consumed to the client.

所以将redo放在SSD上，可以提高写的响应延迟，但是不必将数据文件放在SSD上，因为他们并不是必须被写入的。

选择简单但是不完美的技术解决问题，先做！然后再不断优化。如果不去尝试，我们永远也不知道下一步要做什么，总是停留在对技术方案本身优劣的讨论上，是没有意义的。

科学研究发现，儿童看很多电视会影响后天的智商和持续思维能力，因为电视屏幕平均每隔90秒就要换一次场景，儿童的大脑就会发展为适应这样的持续激励的右脑思维方式，而左脑得不到必要的训练。