标签：
转自：http://www.cnblogs.com/fxjwind/archive/2011/12/10/2283344.html
首先谈谈为什么需要NoSQL?
这儿看到一篇blog说的不错http://robbin.iteye.com/blog/524977, 摘录一下
首先是面对Web2.0网站, 出现的3高问题,
1、High performance - 对数据库高并发读写的需求
web2.0网站要根据用户个性化信息来实时生成动态页面和提供动态信息，所以基本上无法使用动态页面静态化技术，因此数据库并发负载非常高，往往要达到每秒上万次读写请求。关系数据库应付上万次SQL查询还勉强顶得住，但是应付上万次SQL写数据请求，硬盘IO就已经无法承受了。其实对于普通的 BBS网站，往往也存在对高并发写请求的需求，例如像JavaEye网站的实时统计在线用户状态，记录热门帖子的点击次数，投票计数等，因此这是一个相当普遍的需求。
2、Huge Storage - 对海量数据的高效率存储和访问的需求
类似Facebook，twitter，Friendfeed这样的SNS网站，每天用户产生海量的用户动态，以Friendfeed为例，一个月就达到了2.5亿条用户动态，对于关系数据库来说，在一张2.5亿条记录的表里面进行SQL查询，效率是极其低下乃至不可忍受的。再例如大型web网站的用户登录系统，例如腾讯，盛大，动辄数以亿计的帐号，关系数据库也很难应付。
3、High Scalability && High Availability- 对数据库的高可扩展性和高可用性的需求
在基于web的架构当中，数据库是最难进行横向扩展的，当一个应用系统的用户量和访问量与日俱增的时候，你的数据库却没有办法像web server和app server那样简单的通过添加更多的硬件和服务节点来扩展性能和负载能力。对于很多需要提供24小时不间断服务的网站来说，对数据库系统进行升级和扩展是非常痛苦的事情，往往需要停机维护和数据迁移，为什么数据库不能通过不断的添加服务器节点来实现扩展呢？
而对于web2.0网站来说，关系数据库的很多主要特性却往往无用武之地，例如：
1、数据库事务一致性需求
很多web实时系统并不要求严格的数据库事务，对读一致性的要求很低，有些场合对写一致性要求也不高。因此数据库事务管理成了数据库高负载下一个沉重的负担。
2、数据库的写实时性和读实时性需求
对关系数据库来说，插入一条数据之后立刻查询，是肯定可以读出来这条数据的，但是对于很多web应用来说，并不要求这么高的实时性，比方说我（JavaEye的robbin）发一条消息之后，过几秒乃至十几秒之后，我的订阅者才看到这条动态是完全可以接受的。
3、对复杂的SQL查询，特别是多表关联查询的需求
任何大数据量的web系统，都非常忌讳多个大表的关联查询，以及复杂的数据分析类型的复杂SQL报表查询，特别是SNS类型的网站，从需求以及产品设计角度，就避免了这种情况的产生。往往更多的只是单表的主键查询，以及单表的简单条件分页查询，SQL的功能被极大的弱化了。
Redis Cookbook
1 Introduction
1.1 Are your application and data a good fit for NoSQL?
When working on the web, chances are your data and data model keep changing with added functionality and business updates. Evolving the schema to support these changes in a relational database is a painful process, especially if you can’t really afford downtime—which people most often can’t these days, because applications are expected to run 24/7.
Examples of data that are a particularly good fit for nonrelation storage are transactional details, historical data, and server logs. These are normally highly dynamic, changing quite often, and their storage tends to grow quite quickly, also don’t typically feel “relational”.
ACID，指数据库事务正确执行的四个基本要素的缩写
原子性(Atomicity）
整个事务中的所有操作，要么全部完成，要么全部不完成，不可能停滞在中间某个环节。事务在执行过程中发生错误，会被回滚（Rollback）到事务开始前的状态，就像这个事务从来没有执行过一样。
一致性（Consistency）
在事务开始之前和事务结束以后，数据库的完整性约束没有被破坏。
隔离性（Isolation）
两个事务的执行是互不干扰的，一个事务不可能看到其他事务运行时，中间某一时刻的数据。
持久性（Durability）
在事务完成以后，该事务所对数据库所作的更改便持久的保存在数据库之中，并不会被回滚。
Performance can also be a key factor. NoSQL databases are generally faster, particularly for write operations, making them a good fit for applications that are write-heavy.
NoSQL databases generally don’t provide ACID or do it only partially.
Redis provides partial ACID compliance by design due to the fact that it is single threaded (which guarantees consistency and isolation), and full compliance if configured with appendfsync always, providing durability as well.
Once you’ve weighted all the options, picking between SQL (for stable, predictable, relational data) and NoSQL (for temporary, highly dynamic data) should be an easy task.
There are also big differences between NoSQL databases that you should account for.
MongoDB (a popular NoSQL database) is a feature-heavy document database that allows you to perform range queries, regular expression searches, indexing, and MapReduce.
Redis is extremely fast, making it perfectly suited for applications that are write-heavy, data that changes often, and data that naturally fits one of Redis’s data structures (for instance, analytics data).
1.2 Using Redis Data Types
Unlike most other NoSQL solutions and key-value storage engines, Redis includes several built-in data types, allowing developers to structure their data in meaningful semantic ways—with the added benefit of being able to perform data-type specific operations inside Redis, which is typically faster than processing the data externally.
Strings
The simplest data type in Redis is a string. Strings are also the typical (and frequently the sole) data type in other key-value storage engines. You can store strings of any kind, including binary data. You might, for example, want to cache image data for avatars in a social network. The only thing you need to keep in mind is that a specific value inside Redis shouldn’t go beyond 512MB of data.
Lists
Lists in Redis are ordered lists of binary safe strings, implemented on the idea of a linked list.
是用linked list实现的, 而不是用数组, 所以不适合按index去取, 比较适合实现queue和stack
Hashes
Much like traditional hashtables, hashes in Redis store several fields and their values inside a specific key.
象python里面的字典, 非常有用的结构
Sets and Sorted Sets
Sets in Redis are an unordered collection of binary-safe strings. Elements in a given set can have no duplicates. Sets allow you to perform typical set operations such as intersections and unions.
对于web应用, 如朋友圈...
1.3 Using Redis from Python with redis-py
easy_install redis
>>> import redis
>>> redis = redis.Redis(host=‘localhost‘, port=6379, db=0)
>>> redis.smembers(‘circle:jdoe:soccer‘)
set([‘users:toby‘, ‘users:adam‘, ‘users:apollo‘, ‘users:mike‘])
>>> redis.sadd(‘circle:jdoe:soccer‘, ‘users:fred‘)
True
In order to squeeze a bit more performance out of your Redis and Python setup, you may want to install the Python bindings for Hiredis, a C-based Redis client library developed by the Redis authors.
easy_install hiredis
redis-py will then automatically detect the Python bindings and use Hiredis to connect to the server and process responses—hopefully much faster than before.
2 Leveraging Redis
2.1 Using Redis as a Key/Value Store
Storing application usage counters
Let’s begin by storing something quite basic: counters. Imagine we run a business social network and want to track profile/page visit data. We could just add a column to whatever table is storing our page data in our RDBMS, but hopefully our traffic is high enough that updates to this column have trouble keeping up. We need something much faster to update and to query. So we’ll use Redis for this instead.
SET visits:1:totals 21389
SET visits:2:totals 1367894
INCR visits:635:totals
GET visits:635:totals
Storing object data in hashes
实现了类似python字典的功能
Let’s also assume we want to store a number of fields about our users, such as a full name, email address, phone number, and number of visits to our application. We’ll use Redis’s hash management commands—like HSET, HGET, and HINCRBY—to store this
information.
redis> hset users:jdoe name "John Doe"
(integer) 1
redis> hset users:jdoe email "jdoe@test.com"
(integer) 1
redis> hset users:jdoe phone "+1555313940"
(integer) 1
redis> hincrby users:jdoe visits 1
(integer) 1
redis> hget users:jdoe email
"jdoe@test.com"
redis> hgetall users:jdoe
1) "name"
2) "John Doe"
3) "email"
4) "jdoe@test.com"
5) "phone"
6) "+1555313940"
7) "visits"
8) "1"
Storing user “Circles” using sets
Let’s look at how we can use Redis’s support for sets to create functionality similar to the circles in the recently launched Google+. Sets are a natural fit for circles, because sets represent collections of data, and have native functionality to do interesting things like intersections and unions.
redis> sadd circle:jdoe:family users:anna
(integer) 1
redis> sadd circle:jdoe:family users:richard
(integer) 1
redis> sadd circle:jdoe:family users:mike
(integer) 1
redis> sadd circle:jdoe:family users:mike
(integer) 1
redis> sadd circle:jdoe:soccer users:mike
(integer) 1
redis> sadd circle:jdoe:soccer users:adam
(integer) 1
redis> sadd circle:jdoe:soccer users:toby
(integer) 1
redis> sadd circle:jdoe:soccer users:apollo
(integer) 1
redis> sinter circle:jdoe:family circle:jdoe:soccer
1) "users:mike"
redis> sunion circle:jdoe:family circle:jdoe:soccer
1) "users:anna"
2) "users:mike"
3) "users:apollo"
4) "users:adam"
5) "users:richard"
6) "users:toby"
2.2 Inspecting Your Data
While developing (or perhaps debugging) with Redis, you may find you need to take a look at your data. Even though it’s not as simple (or powerful) as MySQL’s SHOW TABLES; and SELECT * FROM table WHERE conditions; commands, there are ways of viewing data with Redis.
KEYS pattern
Lists all the keys in the current database that match the given pattern.
TYPE key-name
Tells the type of the key. Possible types are: string, list, hash, set, zset, and none.
MONITOR
Outputs the commands received by the Redis server in real time.
Keep in mind that every time you use the KEYS command, Redis has to scan all the keys in the database. Therefore, this can really slow down your server, so you probably shouldn’t use it as a normal operation. If you need a list of all your keys (or a subset) you might want to add those keys to a set and then query it. 慎用Keys, 因为需要遍历所有keys, overload server
2.3 Implementing OAuth on Top of Redis
http://zh.wikipedia.org/zh/OAuth
对于OAuth不是很熟悉, 其实是通过这个场景介绍了数据过期设置, 可以设置key的有效时间, 过期自动删除.
对于认证和授权是有时限的, 用Redis提供的这个特性很合适.
EXPIRE key seconds
Sets an expiration timeout on a key, after which it will be deleted. This can be used on any type of key (strings, hashes, lists, sets or sorted sets) and is one of the most powerful Redis features.
EXPIREAT key timestamp
Performs the same operation as EXPIRE, except you can specify a UNIX timestamp (seconds since midnight, January 1, 1970) instead of the number of elapsed seconds.
TTL key
Tells you the remaining time to live of a key with an expiration timeout.
PERSIST key
Removes the expiration timeout on the given key.
2.4 Using Redis’s Pub/Sub Functionality to Create a Chat System
通过聊天室场景来介绍Redis另一个非常有用的功能, 发布/订阅（publish/subscribe）功能类似于传统的消息路由功能，发布者发布消息，订阅者接收消息，沟通发布者和订阅者之间的桥梁是订阅的channel或者pattern. 这种设计模式实现了两者之间的松耦合.
Redis就是一种基于内存的服务器, 客户端可以通过简单的命令, 进行创建Chanel, subscribe, publish的操作, 有Redis来保存chanel状态并负责消息路由和发送. 这个功能和他提供的基本功能不一样, 算是副业.
PUBLISH
Publishes to a specific channel
SUBSCRIBE
Subscribes to a specific channel
UNSUBSCRIBE
Unsubscribes from a specific channel
PSUBSCRIBE, 批量订阅
Subscribes to channels that match a given pattern
PUNSUBSCRIBE
Unsubscribes from channels that match a given pattern
2.5 Implementing an Inverted-Index Text Search with Redis
用Redis实现一个内存倒排索引相当容易的, 因为它本身对数据结构的支持.
在生成索引时, 为每个word创建doc_id的集合
$redis.sadd("word:#{word}", doc_id)
在search的时候, 把所有搜索word的doc_id集合求交集
$redis.sinter(*terms.map{|t| "word:#{t}"})
相当简单.但是这章通过这个场景是为了介绍Redis的事务特性
document_ids = redis.multidoredis.zinterstore("temp_set", terms.map{|t| "word:#{t}"})
$redis.zrevrange("temp_set", 0, -1)
end.last
The MULTI and EXEC commands allow transactional behavior in Redis.
Using DISCARD inside a transaction will abort the transaction, discarding all the commands and return to the normal state.
上面那段代码是在search中用到, zinterstore会创建一个temp_set存放临时结果, 如果多个客户端调用search的时候, 这个temp_set就会冲突, race condition.
做法就是把这段代码加到Multi…last的事务中, EXEC会由Redis在Last时, 自动调用.
Redis也是基于异步系统的, 异步系统是单线程的, 所以通过这个方法可以解决这个问题.
2,6 Analytics and Time-Based Data
Storing analytics or other time-based data poses somewhat of a challenge for traditional storage systems (like an RDBMS). Perhaps you want to do rate limiting on incoming traffic (which requires fast and highly concurrent updates) or simply track visitors (or more complex metrics) to your website and plot them on a chart.
这是Redis一个非常典型和完美的应用, 实时数据统计, 对于关系数据库, 这种短时间内, 大量快速的更新, 和海量的查询, 明显是力不从心的.
Redis is ideally suited for storing this type of data, and for tracking events in particular.
A good and memory-efficient way of storing this data in Redis is to use hashes to store the counters, increment them using HINCRBY, and then fetch them using HGET and HMGET. Finding the top elements is also easily achieved using the SORT command.
2.7 Implementing a Job Queue with Redis
A typical use case for Redis has been a queue. Although this is owed mostly to Resque (a project started by Github after trying all the other Ruby job queueing solutions). In fact, there are several other implementations (Barbershop, qr, presque) and tutorials
(“Creating Processing Queues with Redis”). Nevertheless, it’s interesting in the context of this book to give an example implementation (inspired by existing ones).
这也是Redis一个典型的应用, Redis可以被看作是共享内存, 作为进程间的消息queue, 再合适不过...
3 Redis Administration and Maintenance
3.1 Configuring Persistence
One of the advantages of Redis over other key/value stores like memcached is its support for persistence.
(1)The default persistence model is snapshotting, which consists of saving the entire database to disk in the RDB format (basically a compressed database dump). This can be done periodically at set times, or every time a configurable number of keys changes.
You can manually trigger snapshotting with the SAVE and BGSAVE commands. SAVE performs the same operation as BGSAVE but does so in the foreground, thereby blocking your Redis server.
If you come to the conclusion that snapshotting is putting too much strain on your Redis servers you might want to consider using slaves for persistence (by commenting out all the save statements in your masters and enabling them only on the slaves), or using AOF instead.
(2)The alternative is using an Append Only File (AOF). This might be a better option if you have a large dataset or your data doesn’t change very frequently.
The Append Only File persistence mode keeps a log of the commands that change your dataset in a separate file. Like most writes on modern operating systems, any data logged to AOF is left in memory buffers and written to disk at intervals of a few secondsusing the system’s fsync call. You can configure how often the AOF gets synched to disk by putting appendfsync statements in your configuration file. Valid options are always, everysec, and no.
3.2 Starting a Redis Slave
Database slaves are useful for a number of reasons. You might need them to loadbalance your queries, keep hot standby servers, perform maintenance operations, or simply inspect your data.
Redis supports master-slave replication natively: you can have multiple slaves per master and slaves connecting to slaves. You can configure replication on the configuration file before starting a server or by connecting to a running server and using the SLAVEOFcommand.
3.3 Handling a Dataset Larger Than Memory
Often you might find yourself with a dataset that won’t fit in your available memory.
While you could try to get around that by adding more RAM or sharding (which in addition would allow you you to scale horizontally), it might not be feasible or practical to do so.
Redis has supported a feature called Virtual Memory (VM) since version 2.0. This allows you to have a dataset bigger than your available RAM by swapping rarely used values to disk and keeping all the keys and the frequently used values in memory.
However, this has one downside: before Redis reads or performs an operation on swapped values, they must be read into real memory.
任何东西都是balance, VM就是时间换空间, 所以使用VM必然带来tradeoff,
If you decide to use VM, you should be aware of its ideal use cases and the tradeoffs you’re making.
• The keys are always kept in memory. This means that if you have a big number of small keys, VM might not be the best option for you or you might have the change your data structure and use large strings, hashes, lists, or sets instead. 对于Redis, 因为它提供丰富的数据结果, 所以尽量使用数据结构来封装数据, 而不应该单纯将数据以kv的格式存放, 明显大量的key会降低Redis系统的效率.
• VM is ideal for some patterns of data access, not all. If you regularly query all your data, VM is probably not a good fit because your Redis server might end up blocking clients in order to fetch the values from disk. VM is ideally suited for situations when you have a reasonable amount of frequently accessed values that fit in memory.
• Doing a full dump of your Redis server will be extremely slow. In order to generate a snapshot, Redis needs to read all the values swapped to disk in order to write them to the RDB file (see “Configuring Persistence” on page 45). This generates a lot of I/O. Due to this, it might be better for you to use AOF as a persistence mode.
• VM also affects the speed of replication, because Redis masters need to perform a BGSAVE when a new slave connects.
3.4 Upgrading Redis
Our solution will involve starting a new Redis server in slave mode, switching over the clients to the slave and promoting the new server to the master role.
3.5 Backing up Redis
Our proposed solution is heavily dependent on your Redis persistence model:
• With the default persistence model (snapshotting), you’re best off using a snapshot as a backup.
• If you’re using only AOF, you’ll have to back up your log in order to be able to replay it on startup.
• If you’re running your Redis in VM mode, you might want to use an AOF log for the purpose of backups, as the use of snapshotting is not advised with VM.
3.6 Sharding Redis
Sharding is a horizontal partitioning tecnique often used with databases. It allows you to scale them by distributing your data across several database instances. Not only does this allow you to have a bigger dataset, as you can use more memory, it will also help if CPU usage is the problem, since you can distribute your instances through different servers (or servers with multiple CPUs).
In Redis’s case, sharding can be easily implemented in the client library or application.
Redis的一些资源
Redis, from the Ground Up
http://blog.mjrusso.com/2010/10/17/redis-from-the-ground-up.html
Redis tutorial
http://simonwillison.net/static/2010/redis-tutorial/
Redis几个认识误区
转一篇博客, 因为要翻*墙才能访问, 觉得写的不错.
http://timyang.net/data/redis-misunderstanding/
Saturday, Dec 4th, 2010 by Tim | Tags: key value store, redis
前几天微博发生了一起大的系统故障，很多技术的朋友都比较关心，其中的原因不会超出James Hamilton在On Designing and Deploying Internet-Scale Service(1)概括的那几个范围，James第一条经验“Design for failure”是所有互联网架构成功的一个关键。互联网系统的工程理论其实非常简单，James paper中内容几乎称不上理论，而是多条实践经验分享，每个公司对这些经验的理解及执行力决定了架构成败。
题外话说完，最近又研究了Redis。去年曾做过一个MemcacheDB, Tokyo Tyrant, Redis performance test，到目前为止，这个benchmark结果依然有效。这1年我们经历了很多眼花缭乱的key value存储产品的诱惑，从Cassandra的淡出(Twitter暂停在主业务使用)到HBase的兴起(Facebook新的邮箱业务选用 HBase(2))，当再回头再去看Redis，发现这个只有1万多行源代码的程序充满了神奇及大量未经挖掘的特性。Redis性能惊人，国内前十大网站的子产品估计用1台Redis就可以满足存储及Cache的需求。除了性能印象之外，业界其实普遍对Redis的认识存在一定误区。本文提出一些观点供大家探讨。
这个问题的结果影响了我们怎么用Redis。如果你认为Redis是一个key value store, 那可能会用它来代替MySQL；如果认为它是一个可以持久化的cache, 可能只是它保存一些频繁访问的临时数据。Redis是REmote DIctionary Server的缩写，在Redis在官方网站的的副标题是A persistent key-value database with built-in net interface written in ANSI-C for Posix systems，这个定义偏向key value store。还有一些看法则认为Redis是一个memory database，因为它的高性能都是基于内存操作的基础。另外一些人则认为Redis是一个data structure server，因为Redis支持复杂的数据特性，比如List, Set等。对Redis的作用的不同解读决定了你对Redis的使用方式。
互联网数据目前基本使用两种方式来存储，关系数据库或者key value。但是这些互联网业务本身并不属于这两种数据类型，比如用户在社会化平台中的关系，它是一个list，如果要用关系数据库存储就需要转换成一种多行记录的形式，这种形式存在很多冗余数据，每一行需要存储一些重复信息。如果用key value存储则修改和删除比较麻烦，需要将全部数据读出再写入。Redis在内存中设计了各种数据类型，让业务能够高速原子的访问这些数据结构，并且不需要关心持久存储的问题，从架构上解决了前面两种存储需要走一些弯路的问题。
很多开发者都认为Redis不可能比Memcached快，Memcached完全基于内存，而Redis具有持久化保存特性，即使是异步的，Redis也不可能比Memcached快。但是测试结果基本是Redis占绝对优势。一直在思考这个原因，目前想到的原因有这几方面。
Redis的数据全部放在内存带来了高速的性能，但是也带来一些不合理之处。比如一个中型网站有100万注册用户，如果这些资料要用Redis来存储，内存的容量必须能够容纳这100万用户。但是业务实际情况是100万用户只有5万活跃用户，1周来访问过1次的也只有15万用户，因此全部100万用户的数据都放在内存有不合理之处，RAM需要为冷数据买单。
这跟操作系统非常相似，操作系统所有应用访问的数据都在内存，但是如果物理内存容纳不下新的数据，操作系统会智能将部分长期没有访问的数据交换到磁盘，为新的应用留出空间。现代操作系统给应用提供的并不是物理内存，而是虚拟内存(Virtual Memory)的概念。
基于相同的考虑，Redis 2.0也增加了VM特性。让Redis数据容量突破了物理内存的限制。并实现了数据冷热分离。
Redis的VM依照之前的epoll实现思路依旧是自己实现。但是在前面操作系统的介绍提到OS也可以自动帮程序实现冷热数据分离，Redis只需要OS申请一块大内存，OS会自动将热数据放入物理内存，冷数据交换到硬盘，另外一个知名的“理解了现代操作系统(3)”的Varnish就是这样实现，也取得了非常成功的效果。
作者antirez在解释为什么要自己实现VM中提到几个原因(6)。主要OS的VM换入换出是基于Page概念，比如OS VM1个Page是4K, 4K中只要还有一个元素即使只有1个字节被访问，这个页也不会被SWAP, 换入也同样道理，读到一个字节可能会换入4K无用的内存。而Redis自己实现则可以达到控制换入的粒度。另外访问操作系统SWAP内存区域时block 进程，也是导致Redis要自己实现VM原因之一。
作为一个key value存在，很多开发者自然的使用set/get方式来使用Redis，实际上这并不是最优化的使用方法。尤其在未启用VM情况下，Redis全部数据需要放入内存，节约内存尤其重要。
假如一个key-value单元需要最小占用512字节，即使只存一个字节也占了512字节。这时候就有一个设计模式，可以把key复用，几个key-value放入一个key中，value再作为一个set存入，这样同样512字节就会存放10-100倍的容量。
这就是为了节约内存，建议使用hashset而不是set/get的方式来使用Redis，详细方法见参考文献(7)。
Redis有两种存储方式，默认是snapshot方式，实现方法是定时将内存的快照(snapshot)持久化到硬盘，这种方法缺点是持久化之后如果出现crash则会丢失一段数据。因此在完美主义者的推动下作者增加了aof方式。aof即append only mode，在写入内存数据的同时将操作命令保存到日志文件，在一个并发更改上万的系统中，命令日志是一个非常庞大的数据，管理维护成本非常高，恢复重建时间会非常长，这样导致失去aof高可用性本意。另外更重要的是Redis是一个内存数据结构模型，所有的优势都是建立在对内存复杂数据结构高效的原子操作上，这样就看出aof是一个非常不协调的部分。
其实aof目的主要是数据可靠性及高可用性，在Redis中有另外一种方法来达到目的：Replication。由于Redis的高性能，复制基本没有延迟。这样达到了防止单点故障及实现了高可用。
要想成功使用一种产品，我们需要深入了解它的特性。Redis性能突出，如果能够熟练的驾驭，对国内很多大型应用具有很大帮助。希望更多同行加入到Redis使用及代码研究行列。
标签：
原文地址：http://www.cnblogs.com/cxzdy/p/5128970.html
