数据库键空间

Redis 是一个键值对(key-value pair)数据库服务器,服务器中的每个数据库都由一个 redis.h/redisDb 结构表示,其中,redisDb 结构的 dict 字典保存了数据库中的所有键值对,我们将这个字典称为键空间(key space):

  1. typedef struct redisDb {
  2.  
  3. // ...
  4.  
  5. // 数据库键空间,保存着数据库中的所有键值对
  6. dict *dict;
  7.  
  8. // ...
  9.  
  10. } redisDb;

键空间和用户所见的数据库是直接对应的:

  • 键空间的键也就是数据库的键,每个键都是一个字符串对象。
  • 键空间的值也就是数据库的值,每个值可以是字符串对象、列表对象、哈希表对象、集合对象和有序集合对象在内的任意一种 Redis 对象。

举个例子,如果我们在空白的数据库中执行以下命令:

  1. redis> SET message "hello world"
  2. OK
  3.  
  4. redis> RPUSH alphabet "a" "b" "c"
  5. (integer) 3
  6.  
  7. redis> HSET book name "Redis in Action"
  8. (integer) 1
  9.  
  10. redis> HSET book author "Josiah L. Carlson"
  11. (integer) 1
  12.  
  13. redis> HSET book publisher "Manning"
  14. (integer) 1

那么在这些命令执行之后,数据库的键空间将会是图 IMAGE_DB_EXAMPLE 所展示的样子:

  • alphabet 是一个列表键,键的名字是一个包含字符串 "alphabet" 的字符串对象,键的值则是一个包含三个元素的列表对象。
  • book 是一个哈希表键,键的名字是一个包含字符串 "book" 的字符串对象,键的值则是一个包含三个键值对的哈希表对象。
  • message 是一个字符串键,键的名字是一个包含字符串 "message" 的字符串对象,键的值则是一个包含字符串 "hello world" 的字符串对象。

digraph { label = "\n图 IMAGE_DB_EXAMPLE 数据库键空间例子"; rankdir = LR; node [shape = record]; // redisDb [label = "redisDb | ... | <dict> dict | ..."]; dict [label = "<dict> dict | <alphabet> StringObject \n \"alphabet\" | <book> StringObject \n \"book\" | <message> StringObject \n \"message\""]; subgraph cluster_alphabet { a [label = " StringObject \n \"a\" "]; b [label = " StringObject \n \"b\" "]; c [label = " StringObject \n \"c\" "]; a -> b -> c; label = "ListObject"; } //alphabet [label = " ListObject | { StringObject \n \"a\" | \"b\" | \"c\" }"]; book [label = " HashObject | <name> StringObject \n \"name\" | <author> StringObject \n \"author\" | <publisher> StringObject \n \"publisher\""]; //name [label = " StringObject \n \"Redis in Action\""]; name [label = " StringObject \n \"Redis in Action\""]; author [label = " StringObject \n \"Josiah L. Carlson\""]; publisher [label = " StringObject \n \"Manning\""]; message [label = " StringObject \n \"hello world\""]; // redisDb:dict -> dict:dict; dict:alphabet -> a; dict:book -> book:head; dict:message -> message; book:name -> name; book:publisher -> publisher; book:author -> author;}

因为数据库的键空间是一个字典,所以所有针对数据库的操作 ——比如添加一个键值对到数据库,或者从数据库中删除一个键值对,又或者在数据库中获取某个键值对,等等,实际上都是通过对键空间字典进行操作来实现的,以下几个小节将分别介绍数据库的添加、删除、更新、取值等操作的实现原理。

添加新键

添加一个新键值对到数据库,实际上就是将一个新键值对添加到键空间字典里面,其中键为字符串对象,而值则为任意一种类型的 Redis 对象。

举个例子,如果键空间当前的状态如图 IMAGE_DB_EXAMPLE 所示,那么在执行以下命令之后:

  1. redis> SET date "2013.12.1"
  2. OK

键空间将添加一个新的键值对,这个新键值对的键是一个包含字符串 "date" 的字符串对象,而键值对的值则是一个包含字符串 "2013.12.1" 的字符串对象,如图 IMAGE_DB_AFTER_ADD_NEW_KEY 所示。

digraph { label = "\n图 IMAGE_DB_AFTER_ADD_NEW_KEY 添加 date 键之后的键空间"; rankdir = LR; node [shape = record]; // redisDb [label = "redisDb | ... | <dict> dict | ..."]; dict [label = "<dict> dict | <alphabet> StringObject \n \"alphabet\" | <book> StringObject \n \"book\" | <message> StringObject \n \"message\" | <date> StringObject \n \"date\""]; subgraph cluster_alphabet { a [label = " StringObject \n \"a\" "]; b [label = " StringObject \n \"b\" "]; c [label = " StringObject \n \"c\" "]; a -> b -> c; label = "ListObject"; } book [label = " HashObject | <name> StringObject \n \"name\" | <author> StringObject \n \"author\" | <publisher> StringObject \n \"publisher\""]; name [label = " StringObject \n \"Redis in Action\""]; author [label = " StringObject \n \"Josiah L. Carlson\""]; publisher [label = " StringObject \n \"Manning\""]; message [label = " StringObject \n \"hello world\""]; date [label = " StringObject \n \"2013.12.1\""]; // redisDb:dict -> dict:dict; dict:alphabet -> a; dict:book -> book:head; dict:message -> message; book:name -> name; book:publisher -> publisher; book:author -> author; dict:date -> date; // node [shape = plaintext] newadd [label = "新添加"] newadd -> dict:date [style = dashed]}

删除键

删除数据库中的一个键,实际上就是在键空间里面删除键所对应的键值对对象。

举个例子,如果键空间当前的状态如图 IMAGE_DB_EXAMPLE 所示,那么在执行以下命令之后:

  1. redis> DEL book
  2. (integer) 1

book 以及它的值将从键空间中被删除,如图 IMAGE_DB_AFTER_DEL 所示。

digraph { label = "\n图 IMAGE_DB_AFTER_DEL 删除 book 键之后的键空间"; rankdir = LR; node [shape = record]; // redisDb [label = "redisDb | ... | <dict> dict | ..."]; dict [label = "<dict> dict | <alphabet> StringObject \n \"alphabet\" | <message> StringObject \n \"message\""]; subgraph cluster_alphabet { a [label = " StringObject \n \"a\" "]; b [label = " StringObject \n \"b\" "]; c [label = " StringObject \n \"c\" "]; a -> b -> c; label = "ListObject"; } message [label = " StringObject \n \"hello world\""]; // redisDb:dict -> dict:dict; dict:alphabet -> a; dict:message -> message;}

更新键

对一个数据库键进行更新,实际上就是对键空间里面键所对应的值对象进行更新,根据值对象的类型不同,更新的具体方法也会有所不同。

举个例子,如果键空间当前的状态如图 IMAGE_DB_EXAMPLE 所示,那么在执行以下命令之后:

  1. redis> SET message "blah blah"
  2. OK

message 的值对象将从之前包含 "hello world" 字符串更新为包含 "blah blah" 字符串,如图 IMAGE_DB_UPDATE_CAUSE_SET 所示。

digraph { label = "\n图 IMAGE_DB_UPDATE_CAUSE_SET 使用 SET 命令更新 message 键"; rankdir = LR; node [shape = record]; // redisDb [label = "redisDb | ... | <dict> dict | ..."]; dict [label = "<dict> dict | <alphabet> StringObject \n \"alphabet\" | <book> StringObject \n \"book\" | <message> StringObject \n \"message\""]; subgraph cluster_alphabet { a [label = " StringObject \n \"a\" "]; b [label = " StringObject \n \"b\" "]; c [label = " StringObject \n \"c\" "]; a -> b -> c; label = "ListObject"; } book [label = " HashObject | <name> StringObject \n \"name\" | <author> StringObject \n \"author\" | <publisher> StringObject \n \"publisher\""]; name [label = " StringObject \n \"Redis in Action\""]; author [label = " StringObject \n \"Josiah L. Carlson\""]; publisher [label = " StringObject \n \"Manning\""]; message [label = " StringObject \n \"blah blah\""]; // redisDb:dict -> dict:dict; dict:alphabet -> a; dict:book -> book:head; dict:message -> message; book:name -> name; book:publisher -> publisher; book:author -> author; // node [shape = plaintext] update [label = "更新值对象"] update -> message [style = dashed]}

再举个例子,如果我们继续执行以下命令:

  1. redis> HSET book page 320
  2. (integer) 1

那么键空间中 book 键的值对象(一个哈希对象)将被更新,新的键值对 page320 会被添加到值对象里面,如图 IMAGE_UPDATE_BY_HSET 所示。

digraph { label = "\n图 IMAGE_UPDATE_BY_HSET 使用 HSET 更新 book 键"; rankdir = LR; node [shape = record]; // redisDb [label = "redisDb | ... | <dict> dict | ..."]; dict [label = "<dict> dict | <alphabet> StringObject \n \"alphabet\" | <book> StringObject \n \"book\" | <message> StringObject \n \"message\" "]; subgraph cluster_alphabet { a [label = " StringObject \n \"a\" "]; b [label = " StringObject \n \"b\" "]; c [label = " StringObject \n \"c\" "]; a -> b -> c; label = "ListObject"; } book [label = " HashObject | <name> StringObject \n \"name\" | <author> StringObject \n \"author\" | <publisher> StringObject \n \"publisher\" | <page> StringObject \n \"page\" "]; name [label = " StringObject \n \"Redis in Action\""]; author [label = " StringObject \n \"Josiah L. Carlson\""]; publisher [label = " StringObject \n \"Manning\""]; page [label = " StringObject \n 320"]; message [label = " StringObject \n \"blah blah\""]; // redisDb:dict -> dict:dict; dict:alphabet -> a; dict:book -> book:head; dict:message -> message; book:name -> name; book:publisher -> publisher; book:author -> author; book:page -> page; // node [shape = plaintext] update [label = "新添加"] update -> book:page [style = dashed]}

对键取值

对一个数据库键进行取值,实际上就是在键空间中取出键所对应的值对象,根据值对象的类型不同,具体的取值方法也会有所不同。

举个例子,如果键空间当前的状态如图 IMAGE_DB_EXAMPLE 所示,那么当执行以下命令时:

  1. redis> GET message
  2. "hello world"

GET 命令将首先在键空间中查找键 message ,找到键之后接着取得该键所对应的字符串对象值,之后再返回值对象所包含的字符串 "hello world" ,取值过程如图 IMAGE_FETCH_VALUE_VIA_GET 所示。

digraph { label = "\n图 IMAGE_FETCH_VALUE_VIA_GET 使用 GET 命令取值的过程"; rankdir = LR; node [shape = record]; // redisDb [label = "redisDb | ... | <dict> dict | ..."]; dict [label = "<dict> dict | <alphabet> StringObject \n \"alphabet\" | <book> StringObject \n \"book\" | <message> StringObject \n \"message\""]; subgraph cluster_alphabet { a [label = " StringObject \n \"a\" "]; b [label = " StringObject \n \"b\" "]; c [label = " StringObject \n \"c\" "]; a -> b -> c; label = "ListObject"; } book [label = " HashObject | <name> StringObject \n \"name\" | <author> StringObject \n \"author\" | <publisher> StringObject \n \"publisher\""]; name [label = " StringObject \n \"Redis in Action\""]; author [label = " StringObject \n \"Josiah L. Carlson\""]; publisher [label = " StringObject \n \"Manning\""]; message [label = " StringObject \n \"hello world\""]; get [label = "GET", shape = plaintext]; // redisDb:dict -> dict:dict; dict:alphabet -> a; dict:book -> book:head; dict:message -> message:head [label = "2)取值", style = dashed]; book:name -> name; book:publisher -> publisher; book:author -> author; get -> dict:message [label = "1)查找键", style = dashed];}

再举一个例子,当执行以下命令时:

  1. redis> LRANGE alphabet 0 -1
  2. 1) "a"
  3. 2) "b"
  4. 3) "c"

LRANGE 命令将首先在键空间中查找键 alphabet ,找到键之后接着取得该键所对应的列表对象值,之后再返回列表对象中包含的三个字符串对象的值,取值过程如图 IMAGE_FETCH_VALUE_VIA_LRANGE 所示。

digraph { label = "\n图 IMAGE_FETCH_VALUE_VIA_LRANGE 使用 LRANGE 命令取值的过程"; rankdir = LR; node [shape = record]; // redisDb [label = "redisDb | ... | <dict> dict | ..."]; dict [label = "<dict> dict | <alphabet> StringObject \n \"alphabet\" | <book> StringObject \n \"book\" | <message> StringObject \n \"message\""]; subgraph cluster_alphabet { a [label = " StringObject \n \"a\" "]; b [label = " StringObject \n \"b\" "]; c [label = " StringObject \n \"c\" "]; a -> b -> c [style = dashed]; label = "ListObject"; } book [label = " HashObject | <name> StringObject \n \"name\" | <author> StringObject \n \"author\" | <publisher> StringObject \n \"publisher\""]; name [label = " StringObject \n \"Redis in Action\""]; author [label = " StringObject \n \"Josiah L. Carlson\""]; publisher [label = " StringObject \n \"Manning\""]; message [label = " StringObject \n \"hello world\""]; lrange [label = "LRANGE", shape = plaintext]; // redisDb:dict -> dict:dict; dict:alphabet -> a [label = "2)取值", style = dashed]; dict:book -> book:head; dict:message -> message:head; book:name -> name:head; book:publisher -> publisher:head; book:author -> author:head; lrange -> dict:alphabet [label = "1)查找键", style = dashed];}

其他键空间操作

除了上面列出的添加、删除、更新、取值操作之外,还有很多针对数据库本身的 Redis 命令,也是通过对键空间进行处理来完成的。

比如说,用于清空整个数据库的 FLUSHDB 命令,就是通过删除键空间中的所有键值对来实现的。

又比如说,用于随机返回数据库中某个键的 RANDOMKEY 命令,就是通过在键空间中随机返回一个键来实现的。

另外,用于返回数据库键数量的 DBSIZE 命令,就是通过返回键空间中包含键值对的数量来实现的。

类似的命令还有 EXISTS 、 RENAME 、 KEYS ,等等,这些命令都是通过对键空间进行操作来实现的。

读写键空间时的维护操作

当使用 Redis 命令对数据库进行读写时,服务器不仅会对键空间执行指定的读写操作,还会执行一些额外的维护操作,其中包括:

  • 在读取一个键之后(读操作和写操作都要对键进行读取),服务器会根据键是否存在,以此来更新服务器的键空间命中(hit)次数或键空间不命中(miss)次数,这两个值可以在 INFO stats 命令的 keyspace_hits 属性和 keyspace_misses 属性中查看。
  • 在读取一个键之后,服务器会更新键的 LRU (最后一次使用)时间,这个值可以用于计算键的闲置时间,使用命令 OBJECT idletime 命令可以查看键 key 的闲置时间。
  • 如果服务器在读取一个键时,发现该键已经过期,那么服务器会先删除这个过期键,然后才执行余下的其他操作,本章稍后对过期键的讨论会详细说明这一点。
  • 如果有客户端使用 WATCH 命令监视了某个键,那么服务器在对被监视的键进行修改之后,会将这个键标记为脏(dirty),从而让事务程序注意到这个键已经被修改过,《事务》一章会详细说明这一点。
  • 服务器每次修改一个键之后,都会对脏(dirty)键计数器的值增一,这个计数器会触发服务器的持久化以及复制操作执行,《RDB 持久化》、《AOF 持久化》和《复制》这三章都会说到这一点。
  • 如果服务器开启了数据库通知功能,那么在对键进行修改之后,服务器将按配置发送相应的数据库通知,本章稍后讨论数据库通知功能的实现时会详细说明这一点。