{"id":5304,"date":"2021-06-08T23:22:16","date_gmt":"2021-06-08T15:22:16","guid":{"rendered":"https:\/\/damogame.cn\/wordpress\/?p=5304"},"modified":"2021-06-08T23:22:16","modified_gmt":"2021-06-08T15:22:16","slug":"%e8%b5%b0%e8%bf%91%e6%ba%90%e7%a0%81%ef%bc%9a%e7%a5%9e%e5%a5%87%e7%9a%84hyperloglog","status":"publish","type":"post","link":"https:\/\/i007.cc\/wordpress\/archives\/5304","title":{"rendered":"\u8d70\u8fd1\u6e90\u7801\uff1a\u795e\u5947\u7684HyperLogLog"},"content":{"rendered":"
\n
\u539f\u6587\u5730\u5740<\/a><\/div>\n
<\/div>\n<\/header>\n
\n
\n

HyperLogLog\u662fRedis\u7684\u9ad8\u7ea7\u6570\u636e\u7ed3\u6784\uff0c\u662f\u7edf\u8ba1\u57fa\u6570\u7684\u5229\u5668\u3002\u524d\u6587<\/a>\u6211\u4eec\u5df2\u7ecf\u4ecb\u7ecd\u8fc7HyperLogLog\u7684\u57fa\u672c\u7528\u6cd5\uff0c\u5982\u679c\u53ea\u6c42\u4f1a\u7528\uff0c\u53ea\u9700\u8981\u638c\u63e1HyperLogLog\u7684\u4e09\u4e2a\u547d\u4ee4\u5373\u53ef\uff0c\u5982\u679c\u60f3\u8981\u66f4\u8fdb\u4e00\u6b65\u4e86\u89e3HyperLogLog\u7684\u539f\u7406\u4ee5\u53ca\u6e90\u7801\u5b9e\u73b0\uff0c\u76f8\u4fe1\u8fd9\u7bc7\u6587\u7ae0\u4f1a\u7ed9\u4f60\u5e26\u6765\u4e00\u4e9b\u542f\u53d1\u3002<\/p>\n

\u57fa\u6570<\/h2>\n

\u5728\u6570\u5b66\u4e0a\uff0c\u57fa\u6570\u6216\u52bf\uff0c\u5373\u96c6\u5408\u4e2d\u5305\u542b\u7684\u5143\u7d20\u7684\u201c\u4e2a\u6570\u201d\uff08\u53c2\u89c1\u52bf\u7684\u6bd4\u8f83\uff09\uff0c\u662f\u65e5\u5e38\u4ea4\u6d41\u4e2d\u57fa\u6570\u7684\u6982\u5ff5\u5728\u6570\u5b66\u4e0a\u7684\u7cbe\u786e\u5316\uff08\u5e76\u4f7f\u4e4b\u4e0d\u518d\u53d7\u9650\u4e8e\u6709\u9650\u60c5\u5f62\uff09\u3002\u6709\u9650\u96c6\u5408\u7684\u57fa\u6570\uff0c\u5176\u610f\u4e49\u4e0e\u65e5\u5e38\u7528\u8bed\u4e2d\u7684\u201c\u57fa\u6570\u201d\u76f8\u540c\uff0c\u4f8b\u5982{{a,b,c}\u7684\u57fa\u6570\u662f3\u3002\u65e0\u9650\u96c6\u5408\u7684\u57fa\u6570\uff0c\u5176\u610f\u4e49\u5728\u4e8e\u6bd4\u8f83\u4e24\u4e2a\u96c6\u7684\u5927\u5c0f\uff0c\u4f8b\u5982\u6574\u6570\u96c6\u548c\u6709\u7406\u6570\u96c6\u7684\u57fa\u6570\u76f8\u540c\uff1b\u6574\u6570\u96c6\u7684\u57fa\u6570\u6bd4\u5b9e\u6570\u96c6\u7684\u5c0f\u3002<\/p><\/blockquote>\n

\u5728\u4ecb\u7ecdHyperLogLog\u7684\u539f\u7406\u4e4b\u524d\uff0c\u8bf7\u4f60\u5148\u6765\u601d\u8003\u4e00\u4e0b\uff0c\u5982\u679c\u8ba9\u4f60\u6765\u7edf\u8ba1\u57fa\u6570\uff0c\u4f60\u4f1a\u7528\u4ec0\u4e48\u65b9\u6cd5\u3002<\/p>\n

Set<\/h2>\n

\u719f\u6089Redis\u6570\u636e\u7ed3\u6784\u7684\u540c\u5b66\u4e00\u5b9a\u9996\u5148\u4f1a\u60f3\u5230Set\u8fd9\u4e2a\u7ed3\u6784\uff0c\u6211\u4eec\u53ea\u9700\u8981\u628a\u6570\u636e\u90fd\u5b58\u5165Set\uff0c\u7136\u540e\u7528scard\u547d\u4ee4\u5c31\u53ef\u4ee5\u5f97\u5230\u7ed3\u679c\uff0c\u8fd9\u662f\u4e00\u79cd\u601d\u8def\uff0c\u4f46\u662f\u5b58\u5728\u4e00\u5b9a\u7684\u95ee\u9898\u3002\u5982\u679c\u6570\u636e\u91cf\u975e\u5e38\u5927\uff0c\u90a3\u4e48\u5c06\u4f1a\u8017\u8d39\u5f88\u5927\u7684\u5185\u5b58\u7a7a\u95f4\uff0c\u5982\u679c\u8fd9\u4e9b\u6570\u636e\u4ec5\u4ec5\u662f\u7528\u6765\u7edf\u8ba1\u57fa\u6570\uff0c\u90a3\u4e48\u65e0\u7591\u662f\u9020\u6210\u4e86\u5de8\u5927\u7684\u6d6a\u8d39\uff0c\u56e0\u6b64\uff0c\u6211\u4eec\u9700\u8981\u627e\u5230\u4e00\u79cd\u5360\u7528\u5185\u5b58\u8f83\u5c0f\u7684\u65b9\u6cd5\u3002<\/p>\n

bitmap<\/h2>\n

bitmap\u540c\u6837\u662f\u4e00\u79cd\u53ef\u4ee5\u7edf\u8ba1\u57fa\u6570\u7684\u65b9\u6cd5\uff0c\u53ef\u4ee5\u7406\u89e3\u4e3a\u7528bit\u6570\u7ec4\u5b58\u50a8\u5143\u7d20\uff0c\u4f8b\u598201101001\uff0c\u8868\u793a\u7684\u662f[1,2,4,8]\uff0cbitmap\u4e2d1\u7684\u4e2a\u6570\u5c31\u662f\u57fa\u6570\u3002bitmap\u4e5f\u53ef\u4ee5\u8f7b\u677e\u5408\u5e76\u591a\u4e2a\u96c6\u5408\uff0c\u53ea\u9700\u8981\u5c06\u591a\u4e2a\u6570\u7ec4\u8fdb\u884c\u5f02\u6216\u64cd\u4f5c\u5c31\u53ef\u4ee5\u4e86\u3002bitmap\u76f8\u6bd4\u4e8eSet\u4e5f\u5927\u5927\u8282\u7701\u4e86\u5185\u5b58\uff0c\u6211\u4eec\u6765\u7c97\u7565\u8ba1\u7b97\u4e00\u4e0b\uff0c\u7edf\u8ba11\u4ebf\u4e2a\u6570\u636e\u7684\u57fa\u6570\uff0c\u9700\u8981\u7684\u5185\u5b58\u662f\uff1a100000000\/8\/1024\/1024 \u2248 12M\u3002<\/p>\n

\u867d\u7136bitmap\u5728\u8282\u7701\u7a7a\u95f4\u65b9\u9762\u5df2\u7ecf\u6709\u4e86\u4e0d\u9519\u7684\u8868\u73b0\uff0c\u4f46\u662f\u5982\u679c\u9700\u8981\u7edf\u8ba11000\u4e2a\u5bf9\u8c61\uff0c\u5c31\u9700\u8981\u5927\u7ea612G\u7684\u5185\u5b58\uff0c\u663e\u7136\u8fd9\u4e2a\u7ed3\u679c\u4ecd\u7136\u4e0d\u80fd\u4ee4\u6211\u4eec\u6ee1\u610f\u3002\u5728\u8fd9\u79cd\u60c5\u51b5\u4e0b\uff0cHyperLogLog\u5c06\u4f1a\u51fa\u6765\u62ef\u6551\u6211\u4eec\u3002<\/p>\n

HyperLogLog\u539f\u7406<\/h2>\n

HyperLogLog\u5b9e\u9645\u4e0a\u4e0d\u4f1a\u5b58\u50a8\u6bcf\u4e2a\u5143\u7d20\u7684\u503c\uff0c\u5b83\u4f7f\u7528\u7684\u662f\u6982\u7387\u7b97\u6cd5\uff0c\u901a\u8fc7\u5b58\u50a8\u5143\u7d20\u7684hash\u503c\u7684\u7b2c\u4e00\u4e2a1\u7684\u4f4d\u7f6e\uff0c\u6765\u8ba1\u7b97\u5143\u7d20\u6570\u91cf\u3002\u8fd9\u4e48\u8bf4\u4e0d\u592a\u5bb9\u6613\u7406\u89e3\uff0c\u5bb9\u6211\u5148\u642c\u51fa\u6765\u4e00\u4e2a\u6817\u5b50\u3002<\/p>\n

\u6709\u4e00\u5929Jack\u548c\u4e2b\u4e2b\u73a9\u629b\u786c\u5e01\u7684\u6e38\u620f\uff0c\u89c4\u5219\u662f\u4e2b\u4e2b\u8d1f\u8d23\u629b\u786c\u5e01\uff0c\u6bcf\u6b21\u629b\u5230\u6b63\u9762\u4e3a\u4e00\u56de\u5408\uff0c\u4e2b\u4e2b\u53ef\u4ee5\u81ea\u5df1\u51b3\u5b9a\u8fdb\u884c\u51e0\u4e2a\u56de\u5408\u3002\u6700\u540e\u9700\u8981\u544a\u8bc9Jack\u6700\u957f\u7684\u90a3\u4e2a\u56de\u5408\u629b\u4e86\u591a\u5c11\u6b21\uff0c\u518d\u7531Jack\u6765\u731c\u4e2b\u4e2b\u4e00\u5171\u8fdb\u884c\u4e86\u51e0\u4e2a\u56de\u5408\u3002Jack\u5fc3\u60f3\uff1a\u8fd9\u53ef\u4e0d\u597d\u731c\u554a\uff0c\u6211\u5f97\u7b97\u7b97\u6982\u7387\u4e86\u3002\u4e8e\u662f\u5728\u8111\u6d77\u4e2d\u7ed8\u5236\u8fd9\u6837\u4e00\u5f20\u56fe\u3002<\/p>\n

<\/figure>\n

k\u662f\u6bcf\u56de\u5408\u629b\u52301\u6240\u7528\u7684\u6b21\u6570\uff0c\u6211\u4eec\u5df2\u77e5\u7684\u662f\u6700\u5927\u7684k\u503c\uff0c\u53ef\u4ee5\u7528kmax\u8868\u793a\uff0c\u7531\u4e8e\u6bcf\u6b21\u629b\u786c\u5e01\u7684\u7ed3\u679c\u53ea\u67090\u548c1\u4e24\u79cd\u60c5\u51b5\uff0c\u56e0\u6b64\uff0ckmax\u5728\u4efb\u610f\u56de\u5408\u51fa\u73b0\u7684\u6982\u7387\u5373\u4e3a\u00a0\"[\u516c\u5f0f]\"\u00a0\uff0c\u56e0\u6b64\u53ef\u4ee5\u63a8\u6d4b\u00a0\"[\u516c\u5f0f]\"\u00a0\u3002\u6982\u7387\u5b66\u628a\u8fd9\u79cd\u95ee\u9898\u53eb\u505a\u4f2f\u52aa\u5229\u5b9e\u9a8c\u3002\u6b64\u65f6\u4e2b\u4e2b\u5df2\u7ecf\u5b8c\u6210\u4e86n\u4e2a\u56de\u5408\uff0c\u5e76\u4e14\u544a\u8bc9Jack\u6700\u957f\u7684\u4e00\u6b21\u629b\u4e863\u6b21\uff0cJack\u6b64\u65f6\u4e5f\u80f8\u6709\u6210\u7af9\uff0c\u9a6c\u4e0a\u8bf4\u51fa\u4ed6\u7684\u7b54\u68488\uff0c\u6700\u540e\u7684\u7ed3\u679c\u662f\uff1a\u4e2b\u4e2b\u53ea\u629b\u4e86\u4e00\u56de\u5408\uff0cJack\u8f93\u4e86\uff0c\u8981\u8d1f\u8d23\u5237\u7897\u4e00\u4e2a\u6708\u3002<\/p>\n

\u7ec8\u4e8e\uff0c\u6211\u4eec\u7684Philippe Flajolet<\/a>\u6559\u6388\u9047\u5230\u4e86Jack\u4e00\u6837\u7684\u95ee\u9898\uff0c\u4ed6\u51b3\u5fc3\u5438\u53d6Jack\u7684\u6559\u8bad\uff0c\u8981\u8ba9\u8fd9\u4e2a\u7b97\u6cd5\u66f4\u52a0\u51c6\u786e\uff0c\u4e8e\u662f\u5f15\u5165\u4e86\u6876\u7684\u6982\u5ff5\uff0c\u8ba1\u7b97m\u4e2a\u6876\u7684\u52a0\u6743\u5e73\u5747\u503c\uff0c\u8fd9\u6837\u5c31\u80fd\u5f97\u5230\u6bd4\u8f83\u51c6\u786e\u7684\u7b54\u6848\u4e86\uff08\u5b9e\u9645\u4e0a\u8fd8\u8981\u8fdb\u884c\u5176\u4ed6\u4fee\u6b63\uff09\u3002\u6700\u7ec8\u7684\u516c\u5f0f\u5982\u56fe<\/p>\n

<\/figure>\n

\u5176\u4e2dm\u662f\u6876\u7684\u6570\u91cf\uff0cconst\u662f\u4fee\u6b63\u5e38\u6570\uff0c\u5b83\u7684\u53d6\u503c\u4f1a\u6839\u636em\u800c\u53d8\u5316\u3002\u00a0\"[\u516c\u5f0f]\"<\/p>\n

\n
switch<\/span> (<\/span>p<\/span>)<\/span> {<\/span>\r\n   case<\/span> 4<\/span>:<\/span>\r\n       constant<\/span> =<\/span> 0<\/span>.<\/span>673<\/span> *<\/span> m<\/span> *<\/span> m<\/span>;<\/span>\r\n   case<\/span> 5<\/span>:<\/span>\r\n       constant<\/span> =<\/span> 0<\/span>.<\/span>697<\/span> *<\/span> m<\/span> *<\/span> m<\/span>;<\/span>\r\n   case<\/span> 6<\/span>:<\/span>\r\n       constant<\/span> =<\/span> 0<\/span>.<\/span>709<\/span> *<\/span> m<\/span> *<\/span> m<\/span>;<\/span>\r\n   default<\/span>:<\/span>\r\n       constant<\/span> =<\/span> (<\/span>0<\/span>.<\/span>7213<\/span> \/<\/span> (<\/span>1<\/span> +<\/span> 1<\/span>.<\/span>079<\/span> \/<\/span> m<\/span>))<\/span> *<\/span> m<\/span> *<\/span> m<\/span>;<\/span>\r\n}<\/span><\/code><\/pre>\n<\/div>\n
\u6211\u4eec\u56de\u5230Redis\uff0c\u5bf9\u4e8e\u4e00\u4e2a\u8f93\u5165\u7684\u5b57\u7b26\u4e32\uff0c\u9996\u5148\u5f97\u523064\u4f4d\u7684hash\u503c\uff0c\u7528\u524d14\u4f4d\u6765\u5b9a\u4f4d\u6876\u7684\u4f4d\u7f6e\uff08\u5171\u6709\u00a0\"[\u516c\u5f0f]\"\u00a0\uff0c\u537316384\u4e2a\u6876\uff09\u3002\u540e\u976250\u4f4d\u5373\u4e3a\u4f2f\u52aa\u5229\u8fc7\u7a0b\uff0c\u6bcf\u4e2a\u6876\u67096bit\uff0c\u8bb0\u5f55\u7b2c\u4e00\u6b21\u51fa\u73b01\u7684\u4f4d\u7f6ecount\uff0c\u5982\u679ccount>oldcount\uff0c\u5c31\u7528count\u66ff\u6362oldcount\u3002<\/p>\n
\u4e86\u89e3\u539f\u7406\u4e4b\u540e\uff0c\u6211\u4eec\u518d\u6765\u804a\u4e00\u4e0bHyperLogLog\u7684\u5b58\u50a8\u3002HyperLogLog\u7684\u5b58\u50a8\u7ed3\u6784\u5206\u4e3a\u5bc6\u96c6\u5b58\u50a8\u7ed3\u6784\u548c\u7a00\u758f\u5b58\u50a8\u7ed3\u6784\u4e24\u79cd\uff0c\u9ed8\u8ba4\u4e3a\u7a00\u758f\u5b58\u50a8\u7ed3\u6784\uff0c\u800c\u6211\u4eec\u5e38\u8bf4\u7684\u5360\u752812K\u5185\u5b58\u7684\u5219\u662f\u5bc6\u96c6\u5b58\u50a8\u7ed3\u6784\u3002<\/p>\n
\u5bc6\u96c6\u5b58\u50a8\u7ed3\u6784<\/h2>\n
\u5bc6\u96c6\u5b58\u50a8\u6bd4\u8f83\u7b80\u5355\uff0c\u5c31\u662f\u8fde\u7eed16384\u4e2a6bit\u7684\u4e32\u6210\u7684\u4f4d\u56fe\u3002\u7531\u4e8e\u6bcf\u4e2a\u6876\u662f6bit\uff0c\u56e0\u6b64\u5bf9\u6876\u7684\u5b9a\u4f4d\u8981\u9ebb\u70e6\u4e00\u4e9b\u3002<\/p>\n
\n
#define HLL_BITS 6 <\/span>\/* Enough to count up to 63 leading zeroes. *\/<\/span>\r\n#define HLL_REGISTER_MAX ((1<<HLL_BITS)-1)\r\n<\/span>\/* Store the value of the register at position 'regnum' into variable 'target'.\r\n<\/span> * 'p' is an array of unsigned bytes. *\/<\/span>\r\n#define HLL_DENSE_GET_REGISTER(target,p,regnum) do { \\\r\n<\/span>    uint8_t *_p = (uint8_t*) p; \\\r\n<\/span>    unsigned long _byte = regnum*HLL_BITS\/8; \\\r\n<\/span>    unsigned long _fb = regnum*HLL_BITS&7; \\\r\n<\/span>    unsigned long _fb8 = 8 - _fb; \\\r\n<\/span>    unsigned long b0 = _p[_byte]; \\\r\n<\/span>    unsigned long b1 = _p[_byte+1]; \\\r\n<\/span>    target = ((b0 >> _fb) | (b1 << _fb8)) & HLL_REGISTER_MAX; \\\r\n<\/span>} while(0)\r\n<\/span>\r\n\/* Set the value of the register at position 'regnum' to 'val'.\r\n<\/span> * 'p' is an array of unsigned bytes. *\/<\/span>\r\n#define HLL_DENSE_SET_REGISTER(p,regnum,val) do { \\\r\n<\/span>    uint8_t *_p = (uint8_t*) p; \\\r\n<\/span>    unsigned long _byte = regnum*HLL_BITS\/8; \\\r\n<\/span>    unsigned long _fb = regnum*HLL_BITS&7; \\\r\n<\/span>    unsigned long _fb8 = 8 - _fb; \\\r\n<\/span>    unsigned long _v = val; \\\r\n<\/span>    _p[_byte] &= ~(HLL_REGISTER_MAX << _fb); \\\r\n<\/span>    _p[_byte] |= _v << _fb; \\\r\n<\/span>    _p[_byte+1] &= ~(HLL_REGISTER_MAX >> _fb8); \\\r\n<\/span>    _p[_byte+1] |= _v >> _fb8; \\\r\n<\/span>} while(0)<\/span><\/code><\/pre>\n<\/div>\n
\u5982\u679c\u6211\u4eec\u8981\u5b9a\u4f4d\u7684\u6876\u7f16\u53f7\u4e3aregnum\uff0c\u5b83\u7684\u504f\u79fb\u5b57\u8282\u91cf\u4e3a(regnum * 6) \/ 8\uff0c\u8d77\u59cbbit\u504f\u79fb\u4e3a(regnum * 6) % 8\uff0c\u4f8b\u5982\uff0c\u6211\u4eec\u8981\u5b9a\u4f4d\u7f16\u53f7\u4e3a5\u7684\u6876\uff0c\u5b57\u8282\u504f\u79fb\u662f3\uff0c\u4f4d\u504f\u79fb\u4e5f\u662f6\uff0c\u4e5f\u5c31\u662f\u8bf4\uff0c\u4ece\u7b2c4\u4e2a\u5b57\u8282\u7684\u7b2c7\u4f4d\u5f00\u59cb\u662f\u7f16\u53f7\u4e3a3\u7684\u6876\u3002\u8fd9\u91cc\u9700\u8981\u6ce8\u610f\uff0c\u5b57\u8282\u5e8f\u548c\u6211\u4eec\u5e73\u65f6\u7684\u5b57\u8282\u5e8f\u76f8\u53cd\uff0c\u56e0\u6b64\u9700\u8981\u8fdb\u884c\u5012\u7f6e\u3002\u6211\u4eec\u7528\u4e00\u5f20\u56fe\u6765\u8bf4\u660eRedis\u662f\u5982\u4f55\u5b9a\u4f4d\u6876\u5e76\u4e14\u5f97\u5230\u5b58\u50a8\u7684\u503c\uff08\u5373HLL_DENSE_GET_REGISTER\u51fd\u6570\u7684\u89e3\u91ca\uff09\u3002<\/p>\n
<\/figure>\n
\u5bf9\u4e8e\u7f16\u53f7\u4e3a5\u7684\u6876\uff0c\u6211\u4eec\u5df2\u7ecf\u5f97\u5230\u4e86\u5b57\u8282\u504f\u79fb_byte\u548c\u4e3a\u504f\u79fb_fb\uff0cb0 >> _fb\u548cb1 << _fb8\u64cd\u4f5c\u662f\u5c06\u5b57\u8282\u5012\u7f6e\uff0c\u7136\u540e\u8fdb\u884c\u62fc\u63a5\uff0c\u5e76\u4e14\u4fdd\u7559\u6700\u540e6\u4f4d\u3002<\/p>\n
\u7a00\u758f\u5b58\u50a8\u7ed3\u6784<\/h2>\n
\u4f60\u4ee5\u4e3aRedis\u771f\u7684\u4f1a\u752816384\u4e2a6bit\u5b58\u50a8\u6bcf\u4e00\u4e2aHLL\u5bf9\u8c61\u5417\uff0c\u90a3\u5c31too naive\u4e86\uff0c\u867d\u7136\u5b83\u53ea\u5360\u7528\u4e8612K\u5185\u5b58\uff0c\u4f46\u662fRedis\u5bf9\u4e8e\u5185\u5b58\u7684\u8282\u7ea6\u5df2\u7ecf\u5230\u4e86\u4e27\u5fc3\u75c5\u72c2\u7684\u5730\u6b65\u4e86\u3002\u56e0\u6b64\uff0c\u5982\u679c\u6bd4\u8f83\u591a\u7684\u8ba1\u6570\u503c\u90fd\u662f0\uff0c\u90a3\u4e48\u5c31\u4f1a\u91c7\u7528\u7a00\u758f\u5b58\u50a8\u7684\u7ed3\u6784\u3002<\/p>\n
\u5bf9\u4e8e\u8fde\u7eed\u591a\u4e2a\u8ba1\u6570\u503c\u4e3a0\u7684\u6876\uff0cRedis\u4f7f\u7528\u7684\u5b58\u50a8\u65b9\u5f0f\u662f\uff1a00xxxxxx\uff0c\u524d\u7f00\u4e24\u4e2a0\uff0c\u540e\u97626\u4f4d\u7684\u503c\u52a01\u8868\u793a\u6709\u8fde\u7eed\u591a\u5c11\u4e2a\u6876\u7684\u8ba1\u6570\u503c\u4e3a0\uff0c\u7531\u4e8e6bit\u6700\u5927\u80fd\u8868\u793a64\u4e2a\u6876\uff0c\u6240\u4ee5Redis\u53c8\u8bbe\u8ba1\u4e86\u53e6\u4e00\u79cd\u8868\u793a\u65b9\u6cd5\uff1a01xxxxxx yyyyyyyy\uff0c\u8fd9\u6837\u540e\u976214bit\u5c31\u53ef\u4ee5\u8868\u793a16384\u4e2a\u6876\u4e86\uff0c\u800c\u4e00\u4e2a\u521d\u59cb\u72b6\u6001\u7684HyperLogLog\u5bf9\u8c61\u53ea\u9700\u8981\u75282\u4e2a\u5b57\u8282\u6765\u5b58\u50a8\u3002<\/p>\n
\u5982\u679c\u8fde\u7eed\u7684\u6876\u6570\u90fd\u4e0d\u662f0\uff0c\u90a3\u4e48Redis\u7684\u8868\u793a\u65b9\u5f0f\u4e3a1vvvvvxx\uff0c\u5373\u4e3a\u8fde\u7eed(xx+1)\u4e2a\u6876\u7684\u8ba1\u6570\u503c\u90fd\u662f(vvvvv+1)\u3002\u4f8b\u5982\uff0c10011110\u8868\u793a\u8fde\u7eed3\u4e2a8\u3002\u8fd9\u91cc\u4f7f\u75285bit\uff0c\u6700\u5927\u53ea\u80fd\u8868\u793a32\u3002\u56e0\u6b64\uff0c\u5f53\u67d0\u4e2a\u8ba1\u6570\u503c\u5927\u4e8e32\u65f6\uff0cRedis\u4f1a\u5c06\u8fd9\u4e2aHyperLogLog\u5bf9\u8c61\u8c03\u6574\u4e3a\u5bc6\u96c6\u5b58\u50a8\u3002<\/p>\n
Redis\u7528\u4e09\u6761\u6307\u4ee4\u6765\u8868\u8fbe\u7a00\u758f\u5b58\u50a8\u7684\u65b9\u5f0f\uff1a<\/p>\n
    \n
  1. ZERO:len \u5355\u4e2a\u5b57\u8282\u8868\u793a 00[len-1]\uff0c\u8fde\u7eed\u6700\u591a64\u4e2a\u96f6\u8ba1\u6570\u503c<\/li>\n
  2. VAL:value,len \u5355\u4e2a\u5b57\u8282\u8868\u793a 1[value-1][len-1]\uff0c\u8fde\u7eed len \u4e2a\u503c\u4e3a value \u7684\u8ba1\u6570\u503c<\/li>\n
  3. XZERO:len \u53cc\u5b57\u8282\u8868\u793a 01[len-1]\uff0c\u8fde\u7eed\u6700\u591a16384\u4e2a\u96f6\u8ba1\u6570\u503c<\/li>\n<\/ol>\n
    Redis\u4ece\u7a00\u758f\u5b58\u50a8\u8f6c\u6362\u5230\u5bc6\u96c6\u5b58\u50a8\u7684\u6761\u4ef6\u662f\uff1a<\/p>\n
      \n
    1. \u4efb\u610f\u4e00\u4e2a\u8ba1\u6570\u503c\u4ece 32 \u53d8\u6210 33\uff0c\u56e0\u4e3aVAL\u6307\u4ee4\u5df2\u7ecf\u65e0\u6cd5\u5bb9\u7eb3\uff0c\u5b83\u80fd\u8868\u793a\u7684\u8ba1\u6570\u503c\u6700\u5927\u4e3a 32<\/li>\n
    2. \u7a00\u758f\u5b58\u50a8\u5360\u7528\u7684\u603b\u5b57\u8282\u6570\u8d85\u8fc7 3000 \u5b57\u8282\uff0c\u8fd9\u4e2a\u9608\u503c\u53ef\u4ee5\u901a\u8fc7 hll_sparse_max_bytes \u53c2\u6570\u8fdb\u884c\u8c03\u6574\u3002<\/li>\n<\/ol>\n
      \u6e90\u7801\u89e3\u6790<\/h2>\n
      \u63a5\u4e0b\u6765\u901a\u8fc7\u6e90\u7801\u6765\u770b\u4e00\u4e0bpfadd\u548cpfcount\u4e24\u4e2a\u547d\u4ee4\u7684\u5177\u4f53\u6d41\u7a0b\u3002\u5728\u8fd9\u4e4b\u524d\u6211\u4eec\u9996\u5148\u8981\u4e86\u89e3\u7684\u662fHyperLogLog\u7684\u5934\u7ed3\u6784\u4f53\u548c\u521b\u5efa\u4e00\u4e2aHyperLogLog\u5bf9\u8c61\u7684\u6b65\u9aa4\u3002<\/p>\n
      HyperLogLog\u5934\u7ed3\u6784\u4f53<\/h2>\n
      \n
      struct<\/span> hllhdr<\/span> {<\/span>\r\n    char<\/span> magic<\/span>[<\/span>4<\/span>];<\/span>      \/* \"HYLL\" *\/<\/span>\r\n    uint8_t<\/span> encoding<\/span>;<\/span>   \/* HLL_DENSE or HLL_SPARSE. *\/<\/span>\r\n    uint8_t<\/span> notused<\/span>[<\/span>3<\/span>];<\/span> \/* Reserved for future use, must be zero. *\/<\/span>\r\n    uint8_t<\/span> card<\/span>[<\/span>8<\/span>];<\/span>    \/* Cached cardinality, little endian. *\/<\/span>\r\n    uint8_t<\/span> registers<\/span>[];<\/span> \/* Data bytes. *\/<\/span>\r\n};<\/span><\/code><\/pre>\n<\/div>\n
      \u521b\u5efaHyperLogLog\u5bf9\u8c61<\/h2>\n
      \n
      #define HLL_P 14 <\/span>\/* The greater is P, the smaller the error. *\/<\/span>\r\n#define HLL_REGISTERS (1<<HLL_P) <\/span>\/* With P=14, 16384 registers. *\/<\/span>\r\n#define HLL_SPARSE_XZERO_MAX_LEN 16384\r\n<\/span>\r\n\r\n#define HLL_SPARSE_XZERO_SET(p,len) do { \\\r\n<\/span>    int _l = (len)-1; \\\r\n<\/span>    *(p) = (_l>>8) | HLL_SPARSE_XZERO_BIT; \\\r\n<\/span>    *((p)+1) = (_l&0xff); \\\r\n<\/span>} while(0)\r\n<\/span>\r\n\/* Create an HLL object. We always create the HLL using sparse encoding.\r\n<\/span> * This will be upgraded to the dense representation as needed. *\/<\/span>\r\nrobj<\/span> *<\/span>createHLLObject<\/span>(<\/span>void<\/span>)<\/span> {<\/span>\r\n    robj<\/span> *<\/span>o<\/span>;<\/span>\r\n    struct<\/span> hllhdr<\/span> *<\/span>hdr<\/span>;<\/span>\r\n    sds<\/span> s<\/span>;<\/span>\r\n    uint8_t<\/span> *<\/span>p<\/span>;<\/span>\r\n    int<\/span> sparselen<\/span> =<\/span> HLL_HDR_SIZE<\/span> +<\/span>\r\n                    (((<\/span>HLL_REGISTERS<\/span>+<\/span>(<\/span>HLL_SPARSE_XZERO_MAX_LEN<\/span>-<\/span>1<\/span>))<\/span> \/<\/span>\r\n                     HLL_SPARSE_XZERO_MAX_LEN<\/span>)<\/span>*<\/span>2<\/span>);<\/span>\r\n    int<\/span> aux<\/span>;<\/span>\r\n\r\n    \/* Populate the sparse representation with as many XZERO opcodes as\r\n<\/span>     * needed to represent all the registers. *\/<\/span>\r\n    aux<\/span> =<\/span> HLL_REGISTERS<\/span>;<\/span>\r\n    s<\/span> =<\/span> sdsnewlen<\/span>(<\/span>NULL<\/span>,<\/span>sparselen<\/span>);<\/span>\r\n    p<\/span> =<\/span> (<\/span>uint8_t<\/span>*<\/span>)<\/span>s<\/span> +<\/span> HLL_HDR_SIZE<\/span>;<\/span>\r\n    while<\/span>(<\/span>aux<\/span>)<\/span> {<\/span>\r\n        int<\/span> xzero<\/span> =<\/span> HLL_SPARSE_XZERO_MAX_LEN<\/span>;<\/span>\r\n        if<\/span> (<\/span>xzero<\/span> ><\/span> aux<\/span>)<\/span> xzero<\/span> =<\/span> aux<\/span>;<\/span>\r\n        HLL_SPARSE_XZERO_SET<\/span>(<\/span>p<\/span>,<\/span>xzero<\/span>);<\/span>\r\n        p<\/span> +=<\/span> 2<\/span>;<\/span>\r\n        aux<\/span> -=<\/span> xzero<\/span>;<\/span>\r\n    }<\/span>\r\n    serverAssert<\/span>((<\/span>p<\/span>-<\/span>(<\/span>uint8_t<\/span>*<\/span>)<\/span>s<\/span>)<\/span> ==<\/span> sparselen<\/span>);<\/span>\r\n\r\n    \/* Create the actual object. *\/<\/span>\r\n    o<\/span> =<\/span> createObject<\/span>(<\/span>OBJ_STRING<\/span>,<\/span>s<\/span>);<\/span>\r\n    hdr<\/span> =<\/span> o<\/span>-><\/span>ptr<\/span>;<\/span>\r\n    memcpy<\/span>(<\/span>hdr<\/span>-><\/span>magic<\/span>,<\/span>\"HYLL\"<\/span>,<\/span>4<\/span>);<\/span>\r\n    hdr<\/span>-><\/span>encoding<\/span> =<\/span> HLL_SPARSE<\/span>;<\/span>\r\n    return<\/span> o<\/span>;<\/span>\r\n}<\/span><\/code><\/pre>\n<\/div>\n
      \u8fd9\u91ccsparselen=HLL_HDR_SIZE+2\uff0c\u56e0\u4e3a\u521d\u59cb\u5316\u65f6\u9ed8\u8ba4\u6240\u6709\u6876\u7684\u8ba1\u6570\u503c\u90fd\u662f0\u3002\u5176\u4ed6\u8fc7\u7a0b\u4e0d\u96be\u7406\u89e3\uff0c\u7528\u7684\u5b58\u50a8\u65b9\u5f0f\u662f\u6211\u4eec\u524d\u9762\u63d0\u5230\u8fc7\u7684\u7a00\u758f\u5b58\u50a8\uff0c\u521b\u5efa\u7684\u5bf9\u8c61\u5b9e\u8d28\u4e0a\u662f\u4e00\u4e2a\u5b57\u7b26\u4e32\u5bf9\u8c61\uff0c\u8fd9\u4e5f\u662f\u5b57\u7b26\u4e32\u547d\u4ee4\u53ef\u4ee5\u64cd\u4f5cHyperLogLog\u5bf9\u8c61\u7684\u539f\u56e0\u3002<\/p>\n
      PFADD\u547d\u4ee4<\/h2>\n
      \n
      \/* PFADD var ele ele ele ... ele => :0 or :1 *\/<\/span>\r\nvoid<\/span> pfaddCommand<\/span>(<\/span>client<\/span> *<\/span>c<\/span>)<\/span> {<\/span>\r\n    robj<\/span> *<\/span>o<\/span> =<\/span> lookupKeyWrite<\/span>(<\/span>c<\/span>-><\/span>db<\/span>,<\/span>c<\/span>-><\/span>argv<\/span>[<\/span>1<\/span>]);<\/span>\r\n    struct<\/span> hllhdr<\/span> *<\/span>hdr<\/span>;<\/span>\r\n    int<\/span> updated<\/span> =<\/span> 0<\/span>,<\/span> j<\/span>;<\/span>\r\n\r\n    if<\/span> (<\/span>o<\/span> ==<\/span> NULL<\/span>)<\/span> {<\/span>\r\n        \/* Create the key with a string value of the exact length to\r\n<\/span>         * hold our HLL data structure. sdsnewlen() when NULL is passed\r\n<\/span>         * is guaranteed to return bytes initialized to zero. *\/<\/span>\r\n        o<\/span> =<\/span> createHLLObject<\/span>();<\/span>\r\n        dbAdd<\/span>(<\/span>c<\/span>-><\/span>db<\/span>,<\/span>c<\/span>-><\/span>argv<\/span>[<\/span>1<\/span>],<\/span>o<\/span>);<\/span>\r\n        updated<\/span>++<\/span>;<\/span>\r\n    }<\/span> else<\/span> {<\/span>\r\n        if<\/span> (<\/span>isHLLObjectOrReply<\/span>(<\/span>c<\/span>,<\/span>o<\/span>)<\/span> !=<\/span> C_OK<\/span>)<\/span> return<\/span>;<\/span>\r\n        o<\/span> =<\/span> dbUnshareStringValue<\/span>(<\/span>c<\/span>-><\/span>db<\/span>,<\/span>c<\/span>-><\/span>argv<\/span>[<\/span>1<\/span>],<\/span>o<\/span>);<\/span>\r\n    }<\/span>\r\n    \/* Perform the low level ADD operation for every element. *\/<\/span>\r\n    for<\/span> (<\/span>j<\/span> =<\/span> 2<\/span>;<\/span> j<\/span> <<\/span> c<\/span>-><\/span>argc<\/span>;<\/span> j<\/span>++<\/span>)<\/span> {<\/span>\r\n        int<\/span> retval<\/span> =<\/span> hllAdd<\/span>(<\/span>o<\/span>,<\/span> (<\/span>unsigned<\/span> char<\/span>*<\/span>)<\/span>c<\/span>-><\/span>argv<\/span>[<\/span>j<\/span>]<\/span>-><\/span>ptr<\/span>,<\/span>\r\n                               sdslen<\/span>(<\/span>c<\/span>-><\/span>argv<\/span>[<\/span>j<\/span>]<\/span>-><\/span>ptr<\/span>));<\/span>\r\n        switch<\/span>(<\/span>retval<\/span>)<\/span> {<\/span>\r\n        case<\/span> 1<\/span>:<\/span>\r\n            updated<\/span>++<\/span>;<\/span>\r\n            break<\/span>;<\/span>\r\n        case<\/span> -<\/span>1<\/span>:<\/span>\r\n            addReplySds<\/span>(<\/span>c<\/span>,<\/span>sdsnew<\/span>(<\/span>invalid_hll_err<\/span>));<\/span>\r\n            return<\/span>;<\/span>\r\n        }<\/span>\r\n    }<\/span>\r\n    hdr<\/span> =<\/span> o<\/span>-><\/span>ptr<\/span>;<\/span>\r\n    if<\/span> (<\/span>updated<\/span>)<\/span> {<\/span>\r\n        signalModifiedKey<\/span>(<\/span>c<\/span>-><\/span>db<\/span>,<\/span>c<\/span>-><\/span>argv<\/span>[<\/span>1<\/span>]);<\/span>\r\n        notifyKeyspaceEvent<\/span>(<\/span>NOTIFY_STRING<\/span>,<\/span>\"pfadd\"<\/span>,<\/span>c<\/span>-><\/span>argv<\/span>[<\/span>1<\/span>],<\/span>c<\/span>-><\/span>db<\/span>-><\/span>id<\/span>);<\/span>\r\n        server<\/span>.<\/span>dirty<\/span>++<\/span>;<\/span>\r\n        HLL_INVALIDATE_CACHE<\/span>(<\/span>hdr<\/span>);<\/span>\r\n    }<\/span>\r\n    addReply<\/span>(<\/span>c<\/span>,<\/span> updated<\/span> ?<\/span> shared<\/span>.<\/span>cone<\/span> :<\/span> shared<\/span>.<\/span>czero<\/span>);<\/span>\r\n}<\/span><\/code><\/pre>\n<\/div>\n
      PFADD\u547d\u4ee4\u4f1a\u5148\u5224\u65adkey\u662f\u5426\u5b58\u5728\uff0c\u5982\u679c\u4e0d\u5b58\u5728\uff0c\u5219\u521b\u5efa\u4e00\u4e2a\u65b0\u7684HyperLogLog\u5bf9\u8c61\uff1b\u5982\u679c\u5b58\u5728\uff0c\u4f1a\u8c03\u7528isHLLObjectOrReply()\u51fd\u6570\u68c0\u67e5\u8fd9\u4e2a\u5bf9\u8c61\u662f\u4e0d\u662fHyperLogLog\u5bf9\u8c61\uff0c\u68c0\u67e5\u65b9\u6cd5\u4e3b\u8981\u662f\u68c0\u67e5\u9b54\u6570\u662f\u5426\u6b63\u786e\uff0c\u5b58\u50a8\u7ed3\u6784\u662f\u5426\u6b63\u786e\u4ee5\u53ca\u5934\u7ed3\u6784\u4f53\u7684\u957f\u5ea6\u662f\u5426\u6b63\u786e\u7b49\u3002<\/p>\n
      \u4e00\u5207\u5c31\u7eea\u540e\uff0c\u624d\u53ef\u4ee5\u8c03\u7528hllAdd()\u51fd\u6570\u6dfb\u52a0\u5143\u7d20\u3002hllAdd\u51fd\u6570\u5f88\u7b80\u5355\uff0c\u53ea\u662f\u6839\u636e\u5b58\u50a8\u7ed3\u6784\u5224\u65ad\u9700\u8981\u8c03\u7528hllDenseAdd()\u51fd\u6570\u8fd8\u662fhllSparseAdd()\u51fd\u6570\u3002<\/p>\n
      \u5bc6\u96c6\u5b58\u50a8\u7ed3\u6784\u53ea\u662f\u6bd4\u8f83\u65b0\u65e7\u8ba1\u6570\u503c\uff0c\u5982\u679c\u65b0\u8ba1\u6570\u503c\u5927\u4e8e\u5c31\u8ba1\u6570\u503c\uff0c\u5c31\u5c06\u5176\u66ff\u4ee3\u3002<\/p>\n
      \u800c\u7a00\u758f\u5b58\u50a8\u7ed3\u6784\u8981\u590d\u6742\u4e00\u4e9b\uff1a<\/p>\n
        \n
      1. \u5224\u65ad\u662f\u5426\u9700\u8981\u8c03\u6574\u4e3a\u5bc6\u96c6\u5b58\u50a8\u7ed3\u6784\uff0c\u5982\u679c\u4e0d\u9700\u8981\u5219\u7ee7\u7eed\u8fdb\u884c\uff0c\u5426\u5219\u5c31\u5148\u8c03\u6574\u4e3a\u5bc6\u96c6\u5b58\u50a8\u7ed3\u6784\uff0c\u7136\u540e\u6267\u884c\u6dfb\u52a0\u64cd\u4f5c<\/li>\n
      2. \u6211\u4eec\u9700\u8981\u5148\u5b9a\u4f4d\u8981\u4fee\u6539\u7684\u5b57\u8282\u6bb5\uff0c\u901a\u8fc7\u5faa\u73af\u8ba1\u7b97\u6bcf\u4e00\u6bb5\u8868\u793a\u7684\u6876\u7684\u8303\u56f4\u662f\u5426\u5305\u62ec\u8981\u4fee\u6539\u7684\u6876<\/li>\n
      3. \u5b9a\u4f4d\u5230\u6876\u540e\uff0c\u5982\u679c\u8fd9\u4e2a\u6876\u5df2\u7ecf\u662fVAL\uff0c\u5e76\u4e14\u8ba1\u6570\u503c\u5927\u4e8e\u5f53\u524d\u8981\u6dfb\u52a0\u7684\u8ba1\u6570\u503c\uff0c\u5219\u8fd4\u56de0\uff0c\u5982\u679c\u5c0f\u4e8e\u5f53\u524d\u8ba1\u6570\u503c\uff0c\u5c31\u8fdb\u884c\u66f4\u65b0<\/li>\n
      4. \u5982\u679c\u662fZERO\uff0c\u5e76\u4e14\u957f\u5ea6\u4e3a1\uff0c\u90a3\u4e48\u53ef\u4ee5\u76f4\u63a5\u628a\u5b83\u66ff\u6362\u4e3aVAL\uff0c\u5e76\u4e14\u8bbe\u7f6e\u8ba1\u6570\u503c<\/li>\n
      5. \u5982\u679c\u4e0d\u662f\u4e0a\u8ff0\u4e24\u79cd\u60c5\u51b5\uff0c\u5219\u9700\u8981\u5bf9\u73b0\u6709\u7684\u5b58\u50a8\u8fdb\u884c\u62c6\u5206<\/li>\n<\/ol>\n
        PFCOUNT\u547d\u4ee4<\/h2>\n
        \n
        \/* PFCOUNT var -> approximated cardinality of set. *\/<\/span>\r\nvoid<\/span> pfcountCommand<\/span>(<\/span>client<\/span> *<\/span>c<\/span>)<\/span> {<\/span>\r\n    robj<\/span> *<\/span>o<\/span>;<\/span>\r\n    struct<\/span> hllhdr<\/span> *<\/span>hdr<\/span>;<\/span>\r\n    uint64_t<\/span> card<\/span>;<\/span>\r\n\r\n    \/* Case 1: multi-key keys, cardinality of the union.\r\n<\/span>     *\r\n<\/span>     * When multiple keys are specified, PFCOUNT actually computes\r\n<\/span>     * the cardinality of the merge of the N HLLs specified. *\/<\/span>\r\n    if<\/span> (<\/span>c<\/span>-><\/span>argc<\/span> ><\/span> 2<\/span>)<\/span> {<\/span>\r\n        uint8_t<\/span> max<\/span>[<\/span>HLL_HDR_SIZE<\/span>+<\/span>HLL_REGISTERS<\/span>],<\/span> *<\/span>registers<\/span>;<\/span>\r\n        int<\/span> j<\/span>;<\/span>\r\n\r\n        \/* Compute an HLL with M[i] = MAX(M[i]_j). *\/<\/span>\r\n        memset<\/span>(<\/span>max<\/span>,<\/span>0<\/span>,<\/span>sizeof<\/span>(<\/span>max<\/span>));<\/span>\r\n        hdr<\/span> =<\/span> (<\/span>struct<\/span> hllhdr<\/span>*<\/span>)<\/span> max<\/span>;<\/span>\r\n        hdr<\/span>-><\/span>encoding<\/span> =<\/span> HLL_RAW<\/span>;<\/span> \/* Special internal-only encoding. *\/<\/span>\r\n        registers<\/span> =<\/span> max<\/span> +<\/span> HLL_HDR_SIZE<\/span>;<\/span>\r\n        for<\/span> (<\/span>j<\/span> =<\/span> 1<\/span>;<\/span> j<\/span> <<\/span> c<\/span>-><\/span>argc<\/span>;<\/span> j<\/span>++<\/span>)<\/span> {<\/span>\r\n            \/* Check type and size. *\/<\/span>\r\n            robj<\/span> *<\/span>o<\/span> =<\/span> lookupKeyRead<\/span>(<\/span>c<\/span>-><\/span>db<\/span>,<\/span>c<\/span>-><\/span>argv<\/span>[<\/span>j<\/span>]);<\/span>\r\n            if<\/span> (<\/span>o<\/span> ==<\/span> NULL<\/span>)<\/span> continue<\/span>;<\/span> \/* Assume empty HLL for non existing var.*\/<\/span>\r\n            if<\/span> (<\/span>isHLLObjectOrReply<\/span>(<\/span>c<\/span>,<\/span>o<\/span>)<\/span> !=<\/span> C_OK<\/span>)<\/span> return<\/span>;<\/span>\r\n\r\n            \/* Merge with this HLL with our 'max' HHL by setting max[i]\r\n<\/span>             * to MAX(max[i],hll[i]). *\/<\/span>\r\n            if<\/span> (<\/span>hllMerge<\/span>(<\/span>registers<\/span>,<\/span>o<\/span>)<\/span> ==<\/span> C_ERR<\/span>)<\/span> {<\/span>\r\n                addReplySds<\/span>(<\/span>c<\/span>,<\/span>sdsnew<\/span>(<\/span>invalid_hll_err<\/span>));<\/span>\r\n                return<\/span>;<\/span>\r\n            }<\/span>\r\n        }<\/span>\r\n\r\n        \/* Compute cardinality of the resulting set. *\/<\/span>\r\n        addReplyLongLong<\/span>(<\/span>c<\/span>,<\/span>hllCount<\/span>(<\/span>hdr<\/span>,<\/span>NULL<\/span>));<\/span>\r\n        return<\/span>;<\/span>\r\n    }<\/span>\r\n\r\n    \/* Case 2: cardinality of the single HLL.\r\n<\/span>     *\r\n<\/span>     * The user specified a single key. Either return the cached value\r\n<\/span>     * or compute one and update the cache. *\/<\/span>\r\n    o<\/span> =<\/span> lookupKeyWrite<\/span>(<\/span>c<\/span>-><\/span>db<\/span>,<\/span>c<\/span>-><\/span>argv<\/span>[<\/span>1<\/span>]);<\/span>\r\n    if<\/span> (<\/span>o<\/span> ==<\/span> NULL<\/span>)<\/span> {<\/span>\r\n        \/* No key? Cardinality is zero since no element was added, otherwise\r\n<\/span>         * we would have a key as HLLADD creates it as a side effect. *\/<\/span>\r\n        addReply<\/span>(<\/span>c<\/span>,<\/span>shared<\/span>.<\/span>czero<\/span>);<\/span>\r\n    }<\/span> else<\/span> {<\/span>\r\n        if<\/span> (<\/span>isHLLObjectOrReply<\/span>(<\/span>c<\/span>,<\/span>o<\/span>)<\/span> !=<\/span> C_OK<\/span>)<\/span> return<\/span>;<\/span>\r\n        o<\/span> =<\/span> dbUnshareStringValue<\/span>(<\/span>c<\/span>-><\/span>db<\/span>,<\/span>c<\/span>-><\/span>argv<\/span>[<\/span>1<\/span>],<\/span>o<\/span>);<\/span>\r\n\r\n        \/* Check if the cached cardinality is valid. *\/<\/span>\r\n        hdr<\/span> =<\/span> o<\/span>-><\/span>ptr<\/span>;<\/span>\r\n        if<\/span> (<\/span>HLL_VALID_CACHE<\/span>(<\/span>hdr<\/span>))<\/span> {<\/span>\r\n            \/* Just return the cached value. *\/<\/span>\r\n            card<\/span> =<\/span> (<\/span>uint64_t<\/span>)<\/span>hdr<\/span>-><\/span>card<\/span>[<\/span>0<\/span>];<\/span>\r\n            card<\/span> |=<\/span> (<\/span>uint64_t<\/span>)<\/span>hdr<\/span>-><\/span>card<\/span>[<\/span>1<\/span>]<\/span> <<<\/span> 8<\/span>;<\/span>\r\n            card<\/span> |=<\/span> (<\/span>uint64_t<\/span>)<\/span>hdr<\/span>-><\/span>card<\/span>[<\/span>2<\/span>]<\/span> <<<\/span> 16<\/span>;<\/span>\r\n            card<\/span> |=<\/span> (<\/span>uint64_t<\/span>)<\/span>hdr<\/span>-><\/span>card<\/span>[<\/span>3<\/span>]<\/span> <<<\/span> 24<\/span>;<\/span>\r\n            card<\/span> |=<\/span> (<\/span>uint64_t<\/span>)<\/span>hdr<\/span>-><\/span>card<\/span>[<\/span>4<\/span>]<\/span> <<<\/span> 32<\/span>;<\/span>\r\n            card<\/span> |=<\/span> (<\/span>uint64_t<\/span>)<\/span>hdr<\/span>-><\/span>card<\/span>[<\/span>5<\/span>]<\/span> <<<\/span> 40<\/span>;<\/span>\r\n            card<\/span> |=<\/span> (<\/span>uint64_t<\/span>)<\/span>hdr<\/span>-><\/span>card<\/span>[<\/span>6<\/span>]<\/span> <<<\/span> 48<\/span>;<\/span>\r\n            card<\/span> |=<\/span> (<\/span>uint64_t<\/span>)<\/span>hdr<\/span>-><\/span>card<\/span>[<\/span>7<\/span>]<\/span> <<<\/span> 56<\/span>;<\/span>\r\n        }<\/span> else<\/span> {<\/span>\r\n            int<\/span> invalid<\/span> =<\/span> 0<\/span>;<\/span>\r\n            \/* Recompute it and update the cached value. *\/<\/span>\r\n            card<\/span> =<\/span> hllCount<\/span>(<\/span>hdr<\/span>,<\/span>&<\/span>invalid<\/span>);<\/span>\r\n            if<\/span> (<\/span>invalid<\/span>)<\/span> {<\/span>\r\n                addReplySds<\/span>(<\/span>c<\/span>,<\/span>sdsnew<\/span>(<\/span>invalid_hll_err<\/span>));<\/span>\r\n                return<\/span>;<\/span>\r\n            }<\/span>\r\n            hdr<\/span>-><\/span>card<\/span>[<\/span>0<\/span>]<\/span> =<\/span> card<\/span> &<\/span> 0xff<\/span>;<\/span>\r\n            hdr<\/span>-><\/span>card<\/span>[<\/span>1<\/span>]<\/span> =<\/span> (<\/span>card<\/span> >><\/span> 8<\/span>)<\/span> &<\/span> 0xff<\/span>;<\/span>\r\n            hdr<\/span>-><\/span>card<\/span>[<\/span>2<\/span>]<\/span> =<\/span> (<\/span>card<\/span> >><\/span> 16<\/span>)<\/span> &<\/span> 0xff<\/span>;<\/span>\r\n            hdr<\/span>-><\/span>card<\/span>[<\/span>3<\/span>]<\/span> =<\/span> (<\/span>card<\/span> >><\/span> 24<\/span>)<\/span> &<\/span> 0xff<\/span>;<\/span>\r\n            hdr<\/span>-><\/span>card<\/span>[<\/span>4<\/span>]<\/span> =<\/span> (<\/span>card<\/span> >><\/span> 32<\/span>)<\/span> &<\/span> 0xff<\/span>;<\/span>\r\n            hdr<\/span>-><\/span>card<\/span>[<\/span>5<\/span>]<\/span> =<\/span> (<\/span>card<\/span> >><\/span> 40<\/span>)<\/span> &<\/span> 0xff<\/span>;<\/span>\r\n            hdr<\/span>-><\/span>card<\/span>[<\/span>6<\/span>]<\/span> =<\/span> (<\/span>card<\/span> >><\/span> 48<\/span>)<\/span> &<\/span> 0xff<\/span>;<\/span>\r\n            hdr<\/span>-><\/span>card<\/span>[<\/span>7<\/span>]<\/span> =<\/span> (<\/span>card<\/span> >><\/span> 56<\/span>)<\/span> &<\/span> 0xff<\/span>;<\/span>\r\n            \/* This is not considered a read-only command even if the\r\n<\/span>             * data structure is not modified, since the cached value\r\n<\/span>             * may be modified and given that the HLL is a Redis string\r\n<\/span>             * we need to propagate the change. *\/<\/span>\r\n            signalModifiedKey<\/span>(<\/span>c<\/span>-><\/span>db<\/span>,<\/span>c<\/span>-><\/span>argv<\/span>[<\/span>1<\/span>]);<\/span>\r\n            server<\/span>.<\/span>dirty<\/span>++<\/span>;<\/span>\r\n        }<\/span>\r\n        addReplyLongLong<\/span>(<\/span>c<\/span>,<\/span>card<\/span>);<\/span>\r\n    }<\/span>\r\n}<\/span><\/code><\/pre>\n<\/div>\n
        \u5982\u679c\u8981\u8ba1\u7b97\u591a\u4e2aHyperLogLog\u7684\u57fa\u6570\uff0c\u5219\u9700\u8981\u5c06\u591a\u4e2aHyperLogLog\u5bf9\u8c61\u5408\u5e76\uff0c\u8fd9\u91cc\u5408\u5e76\u65b9\u6cd5\u662f\u5c06\u6240\u6709\u7684HyperLogLog\u5bf9\u8c61\u5408\u5e76\u5230\u4e00\u4e2a\u540d\u4e3amax\u7684\u5bf9\u8c61\u4e2d\uff0cmax\u91c7\u7528\u7684\u662f\u5bc6\u96c6\u5b58\u50a8\u7ed3\u6784\uff0c\u5982\u679c\u88ab\u5408\u5e76\u7684\u5bf9\u8c61\u4e5f\u662f\u5bc6\u96c6\u5b58\u50a8\u7ed3\u6784\uff0c\u5219\u5faa\u73af\u6bd4\u8f83\u6bcf\u4e00\u4e2a\u8ba1\u6570\u503c\uff0c\u5c06\u5927\u7684\u90a3\u4e2a\u5b58\u5165max\u3002\u5982\u679c\u88ab\u5408\u5e76\u7684\u662f\u7a00\u758f\u5b58\u50a8\uff0c\u5219\u53ea\u9700\u8981\u6bd4\u8f83VAL\u5373\u53ef\u3002<\/p>\n
        \u5982\u679c\u8ba1\u7b97\u5355\u4e2aHyperLogLog\u5bf9\u8c61\u7684\u57fa\u6570\uff0c\u5219\u5148\u5224\u65ad\u5bf9\u8c61\u5934\u7ed3\u6784\u4f53\u4e2d\u7684\u57fa\u6570\u7f13\u5b58\u662f\u5426\u6709\u6548\uff0c\u5982\u679c\u6709\u6548\uff0c\u53ef\u76f4\u63a5\u8fd4\u56de\u3002\u5982\u679c\u5df2\u7ecf\u5931\u6548\uff0c\u5219\u9700\u8981\u91cd\u65b0\u8ba1\u7b97\u57fa\u6570\uff0c\u5e76\u4fee\u6539\u539f\u6709\u7f13\u5b58\uff0c\u8fd9\u4e5f\u662fPFCOUNT\u547d\u4ee4\u4e0d\u88ab\u5f53\u505a\u53ea\u8bfb\u547d\u4ee4\u7684\u539f\u56e0\u3002<\/p>\n
        \u7ed3\u8bed<\/h2>\n
        \u6700\u540e\uff0c\u7ed9\u5927\u5bb6\u63a8\u8350\u4e00\u4e2a\u5e2e\u52a9\u7406\u89e3HyperLogLog\u539f\u7406\u7684\u5de5\u5177\uff1ahttp:\/\/<\/span>content.research.neustar.biz<\/span>\/blog\/hll.html<\/span><\/a>\uff0c\u6709\u5174\u8da3\u7684\u8bdd\u53ef\u4ee5\u53bb\u5b66\u4e60\u4e00\u4e0b\u3002<\/p>\n
        <\/figure>\n
        \u53c2\u8003\u9605\u8bfb<\/i><\/p>\n
        Redis new data structure: the HyperLogLog<\/a><\/p>\n
        \u63a2\u7d22HyperLogLog\u7b97\u6cd5\uff08\u542bJava\u5b9e\u73b0\uff09<\/a><\/p>\n
        Redis \u6df1\u5ea6\u5386\u9669\uff1a\u6838\u5fc3\u539f\u7406\u4e0e\u5e94\u7528\u5b9e\u8df5<\/a><\/p>\n<\/div>\n<\/div>\n
        \u53d1\u5e03\u4e8e 2019-03-06<\/div>\n","protected":false},"excerpt":{"rendered":"
        \u539f\u6587\u5730\u5740 HyperLogLog\u662fRed<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"colormag_page_container_layout":"default_layout","colormag_page_sidebar_layout":"default_layout","footnotes":""},"categories":[],"tags":[119],"class_list":["post-5304","post","type-post","status-publish","format-standard","hentry","tag-03-"],"_links":{"self":[{"href":"https:\/\/i007.cc\/wordpress\/wp-json\/wp\/v2\/posts\/5304","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/i007.cc\/wordpress\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/i007.cc\/wordpress\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/i007.cc\/wordpress\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/i007.cc\/wordpress\/wp-json\/wp\/v2\/comments?post=5304"}],"version-history":[{"count":0,"href":"https:\/\/i007.cc\/wordpress\/wp-json\/wp\/v2\/posts\/5304\/revisions"}],"wp:attachment":[{"href":"https:\/\/i007.cc\/wordpress\/wp-json\/wp\/v2\/media?parent=5304"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/i007.cc\/wordpress\/wp-json\/wp\/v2\/categories?post=5304"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/i007.cc\/wordpress\/wp-json\/wp\/v2\/tags?post=5304"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}