石建 | Fat Mind

meta学习 - 前辈kafka


http://incubator.apache.org/kafka/design.html

1.Why we built this
    asd(activity stream data)数据是任何网站的一部分,反映网站使用情况,如:那些内容被搜索、展示。通常,此部分数据被以log方式记录在文件,然后定期的整合和分析。od(operation data)是关于机器性能数据,和其它不同途径整合的操作数据。
    在近几年,asd和od变成一个网站重要的一部分,更复杂的基础设施是必须的。
     数据特点:
        a、大吞吐量的不变的ad,对实时计算是一个挑战,会很容易超过10倍or100倍。
 
        b、传统的记录log方式是respectable and scalable方式去支持离线处理,但是延迟太高。
    Kafka is intended to be a single queuing platform that can support both offline and online use cases.

2.Major Design Elements

There is a small number of major design decisions that make Kafka different from most other messaging systems:

  1. Kafka is designed for persistent messages as the common case;消息持久
  2. Throughput rather than features are the primary design constraint;吞吐量是第一要求
  3. State about what has been consumed is maintained as part of the consumer not the server;状态由客户端维护
  4. Kafka is explicitly distributed. It is assumed that producers, brokers, and consumers are all spread over multiple machines;必须是分布式
3.Basics
    Messages are the fundamental unit of communication
    Messages are
 published to a topic by a producer which means they are physically sent to a server acting as a broker,消息被生产者发布到一个topic,意味着物理的发送消息到broker;
    多个consumer订阅一个topic,则此topic的每个消息都会被分发到每个consumer;
    kafka是分布式:producer、broker、consumer,均可以由集群的多台机器组成,相互协作 a logic group;
    属于同一个consumer group的每一个consumer process,每个消息能准确的由其中的一个process消费;A more common case in our own usage is that we have multiple logical consumer groups, each consisting of a cluster of consuming machines that act as a logical whole.
    kafka不管一个topic有多少个consumer,其消息仅会存储一份。

4.Message Persistence and Caching

4.1 Don't fear the filesystem !
    kafka完全依赖文件系统去存储和cache消息;
    大家通常对磁盘的直觉是'很慢',则使人们对持久化结构,是否能提供有竞争力的性能表示怀疑;实际上,磁盘到底有多慢或多块,完全取决于如何使用磁盘,a properly designed disk structure can often be as fast as the network.
    http://baike.baidu.com/view/969385.htm raid-5 
    http://www.china001.com/show_hdr.php?xname=PPDDMV0&dname=66IP341&xpos=172 磁盘种类
    磁盘顺序读写的性能非常高, linear writes on a 6 7200rpm SATA RAID-5 array is about 300MB/sec;These linear reads and writes are the most predictable of all usage patterns, and hence the one detected and optimized best by the operating system using read-ahead and write-behind techniques。顺序读写是最可预见的模式,因此操作系统通过read-head和write-behind技术去优化。
    现代操作系统,用mem作为disk的cache;Any modern OS will happily divert all free memory to disk caching with little performance penalty when the memory is reclaimed. All disk reads and writes will go through this unified cache. 
    Jvm:a、对象的内存开销是非常大的,通常是数据存储的2倍;b、当heap数据增大时,gc代价越来越大;
    As a result of these factors using the filesystem and relying on pagecache is superior to maintaining an in-memory cache or other structure。依赖文件系统和pagecache是优于mem cahce或其它结构的。
    数据压缩,Doing so will result in a cache of up to 28-30GB on a 32GB machine without GC penalties. 
    This suggests a design which is very simple: maintain as much as possible in-memory and flush to the filesystem only when necessary. 尽可能的维持在内存中,仅当必须时写回到文件系统.
    当数据被立即写回到持久化的文件,而未调用flush,其意味着数据仅被写入到os pagecahe,在后续某个时间由os flush。Then we add a configuration driven flush policy to allow the user of the system to control how often data is flushed to the physical disk (every N messages or every M seconds) to put a bound on the amount of data "at risk" in the event of a hard crash. 提供flush策略。

4.2 
Constant Time Suffices
    
The persistent data structure used in messaging systems metadata is often a BTree. BTrees are the most versatile data structure available, and make it possible to support a wide variety of transactional and non-transactional semantics in the messaging system.
    Disk seeks come at 10 ms a pop, and each disk can do only one seek at a time so parallelism is limited. Hence even a handful of disk seeks leads to very high overhead. 
    Furthermore BTrees require a very sophisticated page or row locking implementation to avoid locking the entire tree on each operation.
The implementation must pay a fairly high price for row-locking or else effectively serialize all reads.
    持久化消息的元数据通常是BTree结构,但磁盘结构,其代价太大。原因:寻道、避免锁整棵树。
    
Intuitively a persistent queue could be built on simple reads and appends to files as is commonly the case with logging solutions.
    持久化队列可以构建在读和append to 文件。所以不支持BTree的一些语义,但其好处是:O(1)消耗,无锁读写。
    
the performance is completely decoupled from the data size--one server can now take full advantage of a number of cheap, low-rotational speed 1+TB SATA drives. 
Though they have poor seek performance, these drives often have comparable performance for large reads and writes at 1/3 the price and 3x the capacity.

4.3 Maximizing Efficiency
    Furthermore we assume each message published is read at least once (and often multiple times), hence we optimize for consumption rather than production. 更进一步,我们假设被发布的消息至少会读一次,因此优化consumer优先于producer。
    
There are two common causes of inefficiency :
        two many network requests, (
 APIs are built around a "message set" abstraction,
This allows network requests to group messages together and amortize the overhead of the network roundtrip rather than sending a single message at a time.) 仅提供批量操作api,则每次网络开销是平分在一组消息,而不是单个消息。
    and excessive byte copying.(
The message log maintained by the broker is itself just a directory of message sets that have been written to disk.
Maintaining this common format allows optimization of the most important operation : network transfer of persistent log chunks.
    To understand the impact of sendfile, it is important to understand the common data path for transfer of data from file to socket:
  1. The operating system reads data from the disk into pagecache in kernel space
  2. The application reads the data from kernel space into a user-space buffer
  3. The application writes the data back into kernel space into a socket buffer
  4. The operating system copies the data from the socket buffer to the NIC buffer where it is sent over the network
    利用os提供的zero-copy,
only the final copy to the NIC buffer is needed.

4.4 End-to-end Batch Compression
    In many cases the bottleneck is actually not CPU but network. This is particularly true for a data pipeline that needs to send messages across data centers.
Efficient compression requires compressing multiple messages together rather than compressing each message individually. 
Ideally this would be possible in an end-to-end fashion — that is, data would be compressed prior to sending by the producer and remain compressed on the server, only being decompressed by the eventual consumers. 
    
A batch of messages can be clumped together compressed and sent to the server in this form. This batch of messages will be delivered all to the same consumer and will remain in compressed form until it arrives there.
    理解:kafka 
producer api 提供批量压缩,broker不对此批消息做任何操作,且以压缩的方式,一起被发送到consumer。

4.5 Consumer state
    Keeping track of what has been consumed is one of the key things a messaging system must provide. 
State tracking requires updating a persistent entity and potentially causes random accesses. 
    
Most messaging systems keep metadata about what messages have been consumed on the broker. That is, as a message is handed out to a consumer, the broker records that fact locally. 大部分消息系统,存储是否被消费的元信息在broker。则是说,一个消息被分发到一个consumer,broker记录。
    问题:当consumer消费失败后,会导致消息丢失;改进:每次consumer消费后,给broker ack,若broker在超时时间未收到ack,则重发此消息。
    问题:1.当消费成功,但未ack时,会导致消费2次  2.
 now the broker must keep multiple states about every single message  3.当broker是多台机器时,则状态之间需要同步

4.5.1 Message delivery semantics
    
So clearly there are multiple possible message delivery guarantees that could be provided : at most once 、at least once、exactly once。
    
This problem is heavily studied, and is a variation of the "transaction commit" problem. Algorithms that provide exactly once semantics exist, two- or three-phase commits and Paxos variants being examples, but they come with some drawbacks. They typically require multiple round trips and may have poor guarantees of liveness (they can halt indefinitely). 
    消费分发语义,是 ‘事务提交’ 问题的变种。算法提供 exactly onece 语义,两阶段 or 三阶段提交,paxos 均是例子,但它们存在缺点。典型的问题是要求多次round trip,且
poor guarantees of liveness。
    
Kafka does two unusual things with respect to metadata. 
First the stream is partitioned on the brokers into a set of distinct partitions. 
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.  
    
4.5.2 
Consumer state
    In Kafka, the consumers are responsible for maintaining state information (offset) on what has been consumed. 
Typically, the Kafka consumer library writes their state data to zookeeper.
    
This solves a distributed consensus problem, by removing the distributed part!
    
There is a side benefit of this decision. A consumer can deliberately rewind back to an old offset and re-consume data.

4.5.3 Push vs. pull
    
A related question is whether consumers should pull data from brokers or brokers should push data to the subscriber.
There are pros and cons to both approaches.
    However a push-based system has difficulty dealing with diverse consumers as the broker controls the rate at which data is transferred. push目标是consumer能在最大速率去消费,可不幸的是,当consume速率小于生产速率时,the consumer tends to be overwhelmed。
    
A pull-based system has the nicer property that the consumer simply falls behind and catches up when it can. This can be mitigated with some kind of backoff protocol by which the consumer can indicate it is overwhelmed, but getting the rate of transfer to fully utilize (but never over-utilize) the consumer is trickier than it seems. Previous attempts at building systems in this fashion led us to go with a more traditional pull model.  不存在push问题,且也保证充分利用consumer能力。

5. Distribution
    Kafka is built to be run across a cluster of machines as the common case. There is no central "master" node. Brokers are peers to each other and can be added and removed at anytime without any manual configuration changes. Similarly, producers and consumers can be started dynamically at any time. Each broker registers some metadata (e.g., available topics) in Zookeeper. Producers and consumers can use Zookeeper to discover topics and to co-ordinate the production and consumption. The details of producers and consumers will be described below.

6. Producer

6.1 Automatic producer load balancing
    Kafka supports client-side load balancing for message producers or use of a dedicated load balancer to balance TCP connections. 
 
    The advantage of using a level-4 load balancer is that each producer only needs a single TCP connection, and no connection to zookeeper is needed. 
The disadvantage is that the balancing is done at the TCP connection level, and hence it may not be well balanced (if some producers produce many more messages then others, evenly dividing up the connections per broker may not result in evenly dividing up the messages per broker).
    
Client-side zookeeper-based load balancing solves some of these problems. It allows the producer to dynamically discover new brokers, and balance load on a per-request basis. It allows the producer to partition data according to some key instead of randomly.

    The working of the zookeeper-based load balancing is described below. Zookeeper watchers are registered on the following events—

  • a new broker comes up
  • a broker goes down
  • a new topic is registered
  • a broker gets registered for an existing topic

    Internally, the producer maintains an elastic pool of connections to the brokers, one per broker. This pool is kept updated to establish/maintain connections to all the live brokers, through the zookeeper watcher callbacks. When a producer request for a particular topic comes in, a broker partition is picked by the partitioner (see section on semantic partitioning). The available producer connection is used from the pool to send the data to the selected broker partition.
    producer通过zk,管理与broker的连接。当一个请求,根据partition rule 计算分区,从连接池选择对应的connection,发送数据。

6.2 Asynchronous send

    Asynchronous non-blocking operations are fundamental to scaling messaging systems.
    
This allows buffering of produce requests in a in-memory queue and batch sends that are triggered by a time interval or a pre-configured batch size. 

6.3 Semantic partitioning
    
The producer has the capability to be able to semantically map messages to the available kafka nodes and partitions. 
This allows partitioning the stream of messages with some semantic partition function based on some key in the message to spread them over broker machines. 


posted on 2013-07-06 14:57 石建 | Fat Mind 阅读(1743) 评论(0)  编辑  收藏 所属分类: 一点理解


只有注册用户登录后才能发表评论。


网站导航:
 

导航

<2013年7月>
30123456
78910111213
14151617181920
21222324252627
28293031123
45678910

统计

常用链接

留言簿

随笔分类

随笔档案

搜索

最新评论

What 、How、Why,从细节中寻找不断的成长点