1. dict的特点

　　字典dict采用hash表作为底层的存储结构。

　　1. hash表的长度保持为2的N次方，最大长度为LONG_MAX。

　　2. hash表采用链式法来解决hash值冲突。

　　3. dict数据结构中保存了两个hash表指针，用于实现rehash的过程。

　　4. 为了防止大数据量情况下rehash过程过分耗时，dict采用渐进式rehash，将rehash的过程分散到每一个增删改查的操作中，从而分摊耗时。

2. dict的定义

 1 //hash表元素
 2 typedef struct dictEntry {
 3     void *key;
 4     union {
 5         void *val;
 6         uint64_t u64;
 7         int64_t s64;
 8         double d;
 9     } v;
10     struct dictEntry *next;     // 链接指针
11 } dictEntry;
12
13 /*
14  * 自定义的函数，用于实现深度赋值，删除和自定义比较等
15  */
16 typedef struct dictType {
17     unsigned int (*hashFunction)(const void *key);
18     void *(*keyDup)(void *privdata, const void *key);
19     void *(*valDup)(void *privdata, const void *obj);
20     int (*keyCompare)(void *privdata, const void *key1, const void *key2);
21     void (*keyDestructor)(void *privdata, void *key);
22     void (*valDestructor)(void *privdata, void *obj);
23 } dictType;
24
25 /* This is our hash table structure. Every dictionary has two of this as we
26  * implement incremental rehashing, for the old to the new table. */
27 typedef struct dictht {
28     dictEntry **table;
29     unsigned long size;             // table长度
30     unsigned long sizemask;         // 长度mask，值为size-1
31     unsigned long used;             // 当前保存了多少个元素
32 } dictht;
33
34 // hash表，其中有两个dictht结构，用于实现rehashing
35 typedef struct dict {
36     dictType *type;         // 自定义操作
37     void *privdata;
38     dictht ht[2];
39     long rehashidx; /* rehashing not in progress if rehashidx == -1 */
40     unsigned long iterators; /* number of iterators currently running */
41 } dict;

　　dict结构图

3. dict创建

 1 /* Create a new hash table */
 2 dict *dictCreate(dictType *type,
 3         void *privDataPtr)
 4 {
 5     dict *d = zmalloc(sizeof(*d));
 6
 7     _dictInit(d,type,privDataPtr);
 8     return d;
 9 }
10
11 /* Initialize the hash table */
12 int _dictInit(dict *d, dictType *type,
13         void *privDataPtr)
14 {
15     _dictReset(&d->ht[0]);
16     _dictReset(&d->ht[1]);
17     d->type = type;
18     d->privdata = privDataPtr;
19     d->rehashidx = -1;
20     d->iterators = 0;
21     return DICT_OK;
22 }

4. dict插入新元素

　　字典增加新元素使用函数dictAdd，内部首先调用dictAddRaw根据key在合适的位置插入新的Entry，然后再设置value。

 1 /* Add an element to the target hash table */
 2 int dictAdd(dict *d, void *key, void *val)
 3 {
 4     // 增加新的key
 5     dictEntry *entry = dictAddRaw(d,key);
 6
 7     if (!entry) return DICT_ERR;
 8     // 设置value
 9     dictSetVal(d, entry, val);
10     return DICT_OK;
11 }

　 dictAddRaw的流程总共分为4步，其中比较重要的是step2和step3，step2用于获取key在hash表的下标，如果正在rehash过程中，则这个下标是ht[1]中的，否则就是在ht[0]中，相应的step3选择相应的ht来插入Entry

 1 dictEntry *dictAddRaw(dict *d, void *key)
 2 {
 3     int index;
 4     dictEntry *entry;
 5     dictht *ht;
 6
 7     // step1 : 如果当前正在rehash，则执行一次rehash
 8     if (dictIsRehashing(d)) _dictRehashStep(d);
 9
10     /* Get the index of the new element, or -1 if
11      * the element already exists. */
12     // step2 : 获取key在hash表的下标
13     if ((index = _dictKeyIndex(d, key)) == -1)
14         return NULL;
15
16     /* Allocate the memory and store the new entry.
17      * Insert the element in top, with the assumption that in a database
18      * system it is more likely that recently added entries are accessed
19      * more frequently. */
20     // step3 : 如果当前正在渐进式hash的过程中，则使用ht[1]，将新元素加到队列头
21     ht = dictIsRehashing(d) ? &d->ht[1] : &d->ht[0];
22     entry = zmalloc(sizeof(*entry));
23     entry->next = ht->table[index];
24     ht->table[index] = entry;
25     ht->used++;
26
27     /* Set the hash entry fields. */
28     // step4 : 设置key，这里要决定是否需要为key重新分配内存
29     dictSetKey(d, entry, key);
30     return entry;
31 }

　　_dictKeyIndex用来返回key所在的hash表的下标，它首先会检查是否需要扩展hash表(执行Expand操作)。_dictKeyIndex是一个内部函数，不应该被外部直接调用。

 1 /* Returns the index of a free slot that can be populated with
 2  * a hash entry for the given ‘key‘.
 3  * If the key already exists, -1 is returned.
 4  *
 5  * Note that if we are in the process of rehashing the hash table, the
 6  * index is always returned in the context of the second (new) hash table. */
 7 /**
 8  *  获取key在字典的下标，如果key已经存在，则返回-1
 9  */
10 static int _dictKeyIndex(dict *d, const void *key)
11 {
12     unsigned int h, idx, table;
13     dictEntry *he;
14
15     /* Expand the hash table if needed */
16     // step1 : 检查是否需要扩展hash表，并开始执行扩展操作
17     if (_dictExpandIfNeeded(d) == DICT_ERR)
18         return -1;
19     /* Compute the key hash value */
20     h = dictHashKey(d, key);
21     // step2 : 获取index
22     for (table = 0; table <= 1; table++) {
23         idx = h & d->ht[table].sizemask;
24         /* Search if this slot does not already contain the given key */
25         he = d->ht[table].table[idx];
26         while(he) {
27             if (dictCompareKeys(d, key, he->key))
28                 return -1;
29             he = he->next;
30         }
31         // 如果字典没有执行rehash操作，则直接返回ht[0]中的下标
32         if (!dictIsRehashing(d)) break;
33     }
34     return idx;
35 }

　　_dictExpandIfNeeded 会判断当前是否需要扩展字典dict并在需要的时候执行扩展操作，它也是一个内部函数，红色的代码标明了扩展的条件：

　　1.  如果dict_can_resize被设置并且 used/size >= 1，则扩展dict，并且扩展的长度为used * 2；

　　2.  如果 used / size > dict_force_resize_ratio，则强制扩展操作 (dict_force_resize_ratio值为5)。

 1 /* Expand the hash table if needed */
 2 /**
 3  *  检查当前是否需要扩展hash表
 4  */
 5 static int _dictExpandIfNeeded(dict *d)
 6 {
 7     /* Incremental rehashing already in progress. Return. */
 8     // step1 : 检查当前是否已经在进行渐进式rehash，如果是则返回OK
 9     if (dictIsRehashing(d)) return DICT_OK;
10
11     /* If the hash table is empty expand it to the initial size. */
12     // step2 : 如果当前hash表是空的，则直接扩展为初始大小
13     if (d->ht[0].size == 0) return dictExpand(d, DICT_HT_INITIAL_SIZE);
14
15     /* If we reached the 1:1 ratio, and we are allowed to resize the hash
16      * table (global setting) or we should avoid it but the ratio between
17      * elements/buckets is over the "safe" threshold, we resize doubling
18      * the number of buckets. */
19     if (d->ht[0].used >= d->ht[0].size &&
20         (dict_can_resize ||
21          d->ht[0].used/d->ht[0].size > dict_force_resize_ratio))
22     {
23         return dictExpand(d, d->ht[0].used*2);
24     }
25     return DICT_OK;
26 }

　　dictExpand函数执行扩展操作，如果是扩展空的hash表，则直接扩展，如果ht[0]不为空，则开启渐进式rehash过程

 1 /* Expand or create the hash table */
 2 int dictExpand(dict *d, unsigned long size)
 3 {
 4     dictht n; /* the new hash table */
 5     // step1 : 计算真实长度，真实长度为第一个大于等于size的2的次方
 6     unsigned long realsize = _dictNextPower(size);
 7
 8     /* the size is invalid if it is smaller than the number of
 9      * elements already inside the hash table */
10     if (dictIsRehashing(d) || d->ht[0].used > size)
11         return DICT_ERR;
12
13     /* Rehashing to the same table size is not useful. */
14     if (realsize == d->ht[0].size) return DICT_ERR;
15
16     /* Allocate the new hash table and initialize all pointers to NULL */
17     // step2 分配新的hash表结构并初始化
18     n.size = realsize;
19     n.sizemask = realsize-1;
20     //分配桶(bucket)数组
21     n.table = zcalloc(realsize*sizeof(dictEntry*));
22     n.used = 0;
23
24     /* Is this the first initialization? If so it‘s not really a rehashing
25      * we just set the first hash table so that it can accept keys. */
26     // step3 : 判断是初始化ht[0]还是执行rehash过程
27     if (d->ht[0].table == NULL) {
28         d->ht[0] = n;
29         return DICT_OK;
30     }
31
32     /* Prepare a second hash table for incremental rehashing */
33     d->ht[1] = n;
34     d->rehashidx = 0;
35     return DICT_OK;
36 }

5. 删除元素

　　从dict中删除元素，注意如果在rehash的过程中，需要判断是否从ht[1]中删除

 1 /* Search and remove an element */
 2 static int dictGenericDelete(dict *d, const void *key, int nofree)
 3 {
 4     unsigned int h, idx;
 5     dictEntry *he, *prevHe;
 6     int table;
 7
 8     if (d->ht[0].size == 0) return DICT_ERR; /* d->ht[0].table is NULL */
 9     if (dictIsRehashing(d)) _dictRehashStep(d);
10     h = dictHashKey(d, key);
11
12     for (table = 0; table <= 1; table++) {
13         // step1 : 获取key在hash表中的index
14         idx = h & d->ht[table].sizemask;
15         he = d->ht[table].table[idx];
16         prevHe = NULL;
17         while(he) {
18             // step2 : 比较key值，如果相同，则从链表中移除，然后释放
19             if (dictCompareKeys(d, key, he->key)) {
20                 /* Unlink the element from the list */
21                 if (prevHe)
22                     prevHe->next = he->next;
23                 else
24                     d->ht[table].table[idx] = he->next;
25                 if (!nofree) {
26                     dictFreeKey(d, he);
27                     dictFreeVal(d, he);
28                 }
29                 zfree(he);
30                 d->ht[table].used--;
31                 return DICT_OK;
32             }
33             prevHe = he;
34             he = he->next;
35         }
36         if (!dictIsRehashing(d)) break;
37     }
38     return DICT_ERR; /* not found */
39 }
40
41 int dictDelete(dict *ht, const void *key) {
42     return dictGenericDelete(ht,key,0);
43 }
44
45 int dictDeleteNoFree(dict *ht, const void *key) {
46     return dictGenericDelete(ht,key,1);
47 }

6. rehash过程

　　前面说到dict使用渐进式rehash方式，就是将rehash的过程平摊到每一个增删查改的操作中，在这些操作中，都可以看到这样两条语句： if (dictIsRehashing(d)) _dictRehashStep(d);

 1 /* This function performs just a step of rehashing, and only if there are
 2  * no safe iterators bound to our hash table. When we have iterators in the
 3  * middle of a rehashing we can‘t mess with the two hash tables otherwise
 4  * some element can be missed or duplicated.
 5  *
 6  * This function is called by common lookup or update operations in the
 7  * dictionary so that the hash table automatically migrates from H1 to H2
 8  * while it is actively used. */
 9 static void _dictRehashStep(dict *d) {
10     if (d->iterators == 0) dictRehash(d,1);
11 }

　　dictRehash是真正执行rehash过程的函数

 1 int dictRehash(dict *d, int n) {
 2     int empty_visits = n*10; /* Max number of empty buckets to visit. */
 3     if (!dictIsRehashing(d)) return 0;
 4
 5     while(n-- && d->ht[0].used != 0) {
 6         dictEntry *de, *nextde;
 7
 8         /* Note that rehashidx can‘t overflow as we are sure there are more
 9          * elements because ht[0].used != 0 */
10         assert(d->ht[0].size > (unsigned long)d->rehashidx);
11         // step1 : 跳过连续为空的index，最多访问empty_visits个空index
12         while(d->ht[0].table[d->rehashidx] == NULL) {
13             d->rehashidx++;
14             if (--empty_visits == 0) return 1;
15         }
16         // step2 : 找到一个不为空的index，然后将该bucked里面所有的key移动到新的hash表中
17         de = d->ht[0].table[d->rehashidx];
18         /* Move all the keys in this bucket from the old to the new hash HT */
19         while(de) {
20             unsigned int h;
21
22             nextde = de->next;
23             /* Get the index in the new hash table */
24             // 计算下标
25             h = dictHashKey(d, de->key) & d->ht[1].sizemask;
26             de->next = d->ht[1].table[h];
27             d->ht[1].table[h] = de;
28             d->ht[0].used--;
29             d->ht[1].used++;
30             de = nextde;
31         }
32         // step3 : 将原来的bucket指针置为NULL
33         d->ht[0].table[d->rehashidx] = NULL;
34         d->rehashidx++;
35     }
36
37     // step3 : 如果已经移动完了所有的，则释放ht[0]
38     /* Check if we already rehashed the whole table... */
39     if (d->ht[0].used == 0) {
40         zfree(d->ht[0].table);
41         d->ht[0] = d->ht[1];
42         _dictReset(&d->ht[1]);
43         d->rehashidx = -1;
44         return 0;
45     }
46
47     /* More to rehash... */
48     return 1;
49 }

7. hash函数

　　dict中使用的hash函数有3种：

　　1.   Thomas Wang‘s 32bit整数hash函数。

　　2.   MurmurHash2 hash函数

　　3.   大小写敏感的hash函数，基于djb hash。

 1 /* -------------------------- hash functions -------------------------------- */
 2
 3 /* Thomas Wang‘s 32 bit Mix Function */
 4 unsigned int dictIntHashFunction(unsigned int key)
 5 {
 6     key += ~(key << 15);
 7     key ^=  (key >> 10);
 8     key +=  (key << 3);
 9     key ^=  (key >> 6);
10     key += ~(key << 11);
11     key ^=  (key >> 16);
12     return key;
13 }
14
15 static uint32_t dict_hash_function_seed = 5381;
16
17 void dictSetHashFunctionSeed(uint32_t seed) {
18     dict_hash_function_seed = seed;
19 }
20
21 uint32_t dictGetHashFunctionSeed(void) {
22     return dict_hash_function_seed;
23 }
24
25 /* MurmurHash2, by Austin Appleby
26  * Note - This code makes a few assumptions about how your machine behaves -
27  * 1. We can read a 4-byte value from any address without crashing
28  * 2. sizeof(int) == 4
29  *
30  * And it has a few limitations -
31  *
32  * 1. It will not work incrementally.
33  * 2. It will not produce the same results on little-endian and big-endian
34  *    machines.
35  */
36 unsigned int dictGenHashFunction(const void *key, int len) {
37     /* ‘m‘ and ‘r‘ are mixing constants generated offline.
38      They‘re not really ‘magic‘, they just happen to work well.  */
39     uint32_t seed = dict_hash_function_seed;
40     const uint32_t m = 0x5bd1e995;
41     const int r = 24;
42
43     /* Initialize the hash to a ‘random‘ value */
44     uint32_t h = seed ^ len;
45
46     /* Mix 4 bytes at a time into the hash */
47     const unsigned char *data = (const unsigned char *)key;
48
49     while(len >= 4) {
50         uint32_t k = *(uint32_t*)data;
51
52         k *= m;
53         k ^= k >> r;
54         k *= m;
55
56         h *= m;
57         h ^= k;
58
59         data += 4;
60         len -= 4;
61     }
62
63     /* Handle the last few bytes of the input array  */
64     switch(len) {
65     case 3: h ^= data[2] << 16;
66     case 2: h ^= data[1] << 8;
67     case 1: h ^= data[0]; h *= m;
68     };
69
70     /* Do a few final mixes of the hash to ensure the last few
71      * bytes are well-incorporated. */
72     h ^= h >> 13;
73     h *= m;
74     h ^= h >> 15;
75
76     return (unsigned int)h;
77 }
78
79 /* And a case insensitive hash function (based on djb hash) */
80 unsigned int dictGenCaseHashFunction(const unsigned char *buf, int len) {
81     unsigned int hash = (unsigned int)dict_hash_function_seed;
82
83     while (len--)
84         hash = ((hash << 5) + hash) + (tolower(*buf++)); /* hash * 33 + c */
85     return hash;
86 }