{"id":3562,"date":"2021-12-16T17:51:43","date_gmt":"2021-12-16T09:51:43","guid":{"rendered":"https:\/\/nj.transwarp.cn:8180\/?p=3562"},"modified":"2021-12-16T17:51:43","modified_gmt":"2021-12-16T09:51:43","slug":"hdfs-balancer%e7%ae%80%e5%8d%95%e8%af%b4%e6%98%8e","status":"publish","type":"post","link":"https:\/\/kbwp.transwarp.cn\/?p=3562","title":{"rendered":"HDFS balancer\u7b80\u5355\u8bf4\u660e"},"content":{"rendered":"

\u6982\u8981\u8bf4\u660e<\/h3>\n
\n

\u7ebf\u4e0a\u957f\u65f6\u95f4\u8fd0\u884c\u7684\u5927\u89c4\u6a21Hadoop\u96c6\u7fa4\uff0c\u5404\u4e2adatanode\u8282\u70b9\u78c1\u76d8\u7a7a\u95f4\u4f7f\u7528\u7387\u7ecf\u5e38\u4f1a\u51fa\u73b0\u5206\u5e03\u4e0d\u5747\u8861\u7684\u60c5\u51b5\uff0c\u5c24\u5176\u5728\u65b0\u589e\u548c\u4e0b\u67b6\u8282\u70b9\u3001\u6216\u8005\u4eba\u4e3a\u5e72\u9884\u526f\u672c\u6570\u91cf\u7684\u65f6\u5019\u3002\u8282\u70b9\u7a7a\u95f4\u4f7f\u7528\u7387\u4e0d\u5747\u5300\u4f1a\u5bfc\u81f4\u8ba1\u7b97\u5f15\u64ce\u9891\u7e41\u5728\u8de8\u8282\u70b9\u62f7\u8d1d\u6570\u636e(A\u8282\u70b9\u4e0a\u8fd0\u884c\u7684task\u6240\u9700\u6570\u636e\u5728\u5176\u5b83\u8282\u70b9\u4e0a)\uff0c\u5f15\u8d77\u4e0d\u5fc5\u8981\u7684\u8017\u65f6\u548c\u5e26\u5bbd\u3002<\/p>\n

\u8be6\u7ec6\u8bf4\u660e<\/h3>\n
\n

\u5f53\u90e8\u5206\u8282\u70b9\u7a7a\u95f4\u4f7f\u7528\u7387\u5f88\u9ad8\u4f46\u672a\u6ee1(90%\u5de6\u53f3)\u65f6\uff0c\u5206\u914d\u5728\u8be5\u8282\u70b9\u4e0a\u7684task\u4f1a\u5b58\u5728\u4efb\u52a1\u5931\u8d25\u7684\u98ce\u9669\u3002\u56e0\u6b64\uff0c\u5f15\u5165balance\u7b56\u7565\u4f7f\u96c6\u7fa4\u4e2d\u7684\u8282\u70b9\u7a7a\u95f4\u4f7f\u7528\u7387\u5747\u5300\u5206\u5e03\u5fc5\u4e0d\u53ef\u5c11\u3002<\/p>\n

\"file\"<\/p>\n

Rebalance\u7a0b\u5e8f\u4f5c\u4e3a\u4e00\u4e2a\u72ec\u7acb\u7684\u8fdb\u7a0b\u4e0ename node\u8fdb\u884c\u5206\u5f00\u6267\u884c\u3002<\/p>\n

1 Rebalance Server\u4eceName Node\u4e2d\u83b7\u53d6\u6240\u6709\u7684Data Node\u60c5\u51b5\uff1a\u6bcf\u4e00\u4e2aData Node\u78c1\u76d8\u4f7f\u7528\u60c5\u51b5\u3002<\/p>\n

2 Rebalance Server\u8ba1\u7b97\u54ea\u4e9b\u673a\u5668\u9700\u8981\u5c06\u6570\u636e\u79fb\u52a8\uff0c\u54ea\u4e9b\u673a\u5668\u53ef\u4ee5\u63a5\u53d7\u79fb\u52a8\u7684\u6570\u636e\u3002\u5e76\u4e14\u4eceName Node\u4e2d\u83b7\u53d6\u9700\u8981\u79fb\u52a8\u7684\u6570\u636e\u5206\u5e03\u60c5\u51b5\u3002<\/p>\n

3 Rebalance Server\u8ba1\u7b97\u51fa\u6765\u53ef\u4ee5\u5c06\u54ea\u4e00\u53f0\u673a\u5668\u7684block\u79fb\u52a8\u5230\u53e6\u4e00\u53f0\u673a\u5668\u4e2d\u53bb\u3002<\/p>\n

4,5,6 \u9700\u8981\u79fb\u52a8block\u7684\u673a\u5668\u5c06\u6570\u636e\u79fb\u52a8\u7684\u76ee\u7684\u673a\u5668\u4e0a\u53bb\uff0c\u540c\u65f6\u5220\u9664\u81ea\u5df1\u673a\u5668\u4e0a\u7684block\u6570\u636e\u3002<\/p>\n

7 Rebalance Server\u83b7\u53d6\u5230\u672c\u6b21\u6570\u636e\u79fb\u52a8\u7684\u6267\u884c\u7ed3\u679c\uff0c\u5e76\u7ee7\u7eed\u6267\u884c\u8fd9\u4e2a\u8fc7\u7a0b\uff0c\u4e00\u76f4\u6ca1\u6709\u6570\u636e\u53ef\u4ee5\u79fb\u52a8\u6216\u8005HDFS\u96c6\u7fa4\u4ee5\u53ca\u8fbe\u5230\u4e86\u5e73\u8861\u7684\u6807\u51c6\u4e3a\u6b62\u3002<\/p>\n

1.Balance\u547d\u4ee4\u8be6\u89e3<\/h4>\n
Usage: hdfs balancer\n    [-policy ]  the balancing policy: datanode or blockpool\n    [-threshold ]    Percentage of disk capacity\n    [-exclude [-f  | ]]  Excludes the specified datanodes.\n    [-include [-f  | ]]  Includes only the specified datanodes.\n    [-idleiterations ]  Number of consecutive idle iterations (-1 for Infinite) before exit.\n    [-runDuringUpgrade] Whether to run the balancer during an ongoing HDFS upgrade.This is usually not desired since it will not affect used space on over-utilized machines.\n\nGeneric options supported are\n-conf      specify an application configuration file\n-D             use value for given property\n-fs       specify a namenode\n-jt     specify a ResourceManager\n-files     specify comma separated files to be copied to the map reduce cluster\n-libjars     specify comma separated jar files to include in the classpath.\n-archives     specify comma separated archives to be unarchived on the compute machines.\n\nThe general command line syntax is\nbin\/hadoop command [genericOptions] [commandOptions]<\/code><\/pre>\n
\u53c2\u6570\u4e3e\u4f8b\u8bf4\u660e\uff1a<\/p>\n
hdfs balancer\n-threshold  10                    \\\\\u96c6\u7fa4\u5e73\u8861\u7684\u6761\u4ef6\uff0cdatanode\u95f4\u78c1\u76d8\u4f7f\u7528\u7387\u76f8\u5dee\u9608\u503c\uff0c\u533a\u95f4\u9009\u62e9\uff1a0~100\n-policy datanode                  \\\\\u9ed8\u8ba4\u4e3adatanode\uff0cdatanode\u7ea7\u522b\u7684\u5e73\u8861\u7b56\u7565\n-exclude  -f  \/tmp\/ip1.txt        \\\\\u9ed8\u8ba4\u4e3a\u7a7a\uff0c\u6307\u5b9a\u8be5\u90e8\u5206ip\u4e0d\u53c2\u4e0ebalance\uff0c -f\uff1a\u6307\u5b9a\u8f93\u5165\u4e3a\u6587\u4ef6\n-include  -f  \/tmp\/ip2.txt        \\\\\u9ed8\u8ba4\u4e3a\u7a7a\uff0c\u53ea\u5141\u8bb8\u8be5\u90e8\u5206ip\u53c2\u4e0ebalance\uff0c-f\uff1a\u6307\u5b9a\u8f93\u5165\u4e3a\u6587\u4ef6\n -idleiterations  5               \\\\\u8fed\u4ee3\u6b21\u6570\uff0c\u9ed8\u8ba4\u4e3a 5<\/code><\/pre>\n
2.Balance\u9000\u51fa\u6761\u4ef6<\/h4>\n
    \n
  • The cluster is balanced;<\/li>\n
  • No block can be moved;<\/li>\n
  • No block has been moved for specified consecutive iterations (5 by default);<\/li>\n
  • An IOException occurs while communicating with the namenode;<\/li>\n
  • Another balancer is running.<\/li>\n<\/ul>\n
    3.\u7edf\u8ba1\u53c2\u4e0ebalancer\u7684datanode<\/h4>\n
      \n
    • 1.\u5ffd\u7565\u9000\u5f79\u8282\u70b9<\/li>\n
    • 2.\u5ffd\u7565\u6b63\u5728\u9000\u5f79\u7684\u8282\u70b9<\/li>\n
    • 3.\u5ffd\u7565-exclude\u91cc\u9762\u5305\u542b\u7684\u8282\u70b9<\/li>\n
    • 4.\u5ffd\u7565\u4e0d\u5728-include\u91cc\u9762\u7684\u8282\u70b9<\/li>\n<\/ul>\n
      4.\u76f8\u5173\u53c2\u6570\u548c\u9608\u503c\u7684\u8ba1\u7b97\u89c4\u5219<\/h4>\n
      \u96c6\u7fa4\u5e73\u5747\u4f7f\u7528\u7387(\u8ba1\u7b97\u516c\u5f0f)\uff1aaverage = totalUsedSpaces * 100 \/ totalCapacities
      \ntotalUsedSpaces\uff1a\u5404datanode\u5df2\u4f7f\u7528\u7a7a\u95f4\uff08dfsUsed\uff0c\u4e0d\u5305\u542bnon dfsUsed\uff09\u76f8\u52a0\uff1b
      \ntotalCapacities\uff1a\u5404datanode\u603b\u7a7a\u95f4\uff08DataNode\u914d\u7f6e\u7684\u670d\u52a1\u5668\u78c1\u76d8\u76ee\u5f55\uff09\u76f8\u52a0\uff1b<\/p>\n
      \u5355\u4e2adatanode\u4f7f\u7528\u7387\uff1autilization = dfsUsed * 100.0 \/ \u8be5\u8282\u70b9capacity\uff1b
      \ndfsUsed\uff1a\u5f53\u524ddatanode dfs\uff08dfsUsed\uff0c\u4e0d\u5305\u542bnon dfsUsed\uff09\u5df2\u4f7f\u7528\u7a7a\u95f4\uff1b
      \ncapacity\uff1a\u5f53\u524ddatanode\uff08DataNode\u914d\u7f6e\u7684\u670d\u52a1\u5668\u78c1\u76d8\u76ee\u5f55\uff09\u603b\u7a7a\u95f4\uff1b<\/p>\n
      \u5355\u4e2adatanode\u4f7f\u7528\u7387\u4e0e\u96c6\u7fa4\u5e73\u5747\u4f7f\u7528\u7387\u5dee\u503c\uff1autilizationDiff = utilization – average
      \n\u5355\u4e2adatanode utilizationDiff\u4e0e\u9608\u503c\u7684\u5dee\u503c: thresholdDiff = |utilizationDiff| – threshold<\/p>\n
      \u67d0\u4e2adatanode\u9700\u8981\u8fc1\u79fb\u6216\u8005\u53ef\u4ee5\u8fc1\u5165\u7684\u7a7a\u95f4\uff1amaxSize2Move = |utilizationDiff| * capacity<\/p>\n
      \u53ef\u4ee5\u8fc1\u5165\u7684\u7a7a\u95f4\u8ba1\u7b97\uff1aMath.min(remaining, maxSizeToMove)
      \n\u9700\u8981\u8fc1\u79fb\u7684\u7a7a\u95f4\u8ba1\u7b97\uff1aMath.min(max, maxSizeToMove)
      \nremaining: datanode\u8282\u70b9\u5269\u4f59\u7a7a\u95f4
      \nmax: \u9ed8\u8ba4\u5355\u4e2adatanode\u5355\u6b21balance\u8fed\u4ee3\u53ef\u4ee5\u8fc1\u79fb\u7684\u6700\u5927\u7a7a\u95f4\u9650\u5236\uff0c\u7f3a\u770110G)
      \n\u9ed8\u8ba4\u8fed\u4ee3\u6b21\u6570\u4e3a5\uff0c\u5373\u8fd0\u884c\u4e00\u6b21balance\u811a\u672c\uff0c\u5355\u4e2adatanode\u53ef\u4ee5\u6700\u5927\u8fc1\u79fb\u7684\u7a7a\u95f4\u4e3a\uff1a5*10G = 50G<\/p>\n
      overUtilized\uff1autilizationDiff > 0 && thresholdDiff > 0        <\u4f7f\u7528\u7387\u8d85\u8fc7\u5e73\u5747\u503c\uff0c\u4e14\u5dee\u503c\u5927\u4e8e\u9608\u503c>
      \naboveAvgUtilized\uff1autilizationDiff > 0 && thresholdDiff <= 0   <\u4f7f\u7528\u7387\u8d85\u8fc7\u5e73\u5747\u503c\uff0c\u4e14\u5dee\u503c\u5c0f\u4e8e\u7b49\u4e8e\u9608\u503c>
      \nbelowAvgUtilized\uff1autilizationDiff < 0 && thresholdDiff <= 0   <\u4f7f\u7528\u7387\u4f4e\u4e8e\u5e73\u5747\u503c\uff0c\u4e14\u5dee\u503c\u5c0f\u4e8e\u7b49\u4e8e\u9608\u503c>
      \nunderUtilized\uff1autilizationDiff > 0 && thresholdDiff > 0       <\u4f7f\u7528\u7387\u4f4e\u4e8e\u5e73\u5747\u503c\uff0c\u4e14\u5dee\u503c\u5927\u4e8e\u7b49\u4e8e\u9608\u503c><\/p>\n
      \u6570\u636e\u8fc1\u79fb\u914d\u5bf9(\u539f\u5219\uff1a1. \u4f18\u5148\u4e3a\u540c\u673a\u67b6\uff0c\u5176\u6b21\u4e3a\u5176\u5b83\u673a\u67b6; 2. \u4e00\u5bf9\u591a\u914d\u5bf9)\uff1a
      \n\u7b2c\u4e00\u6b65[Source -> Target]\uff1aeach overUtilized datanode => one or more underUtilized datanodes
      \n\u7b2c\u4e8c\u6b65[Source -> Target]\uff1amatch each remaining overutilized datanode => one or more belowAvgUtilized datanodes
      \n\u7b2c\u4e09\u6b65[Target -> Source]\uff1aeach remaining underutilized datanode (step 1\u672a\u548coverUtilized\u5339\u914d\u8fc7) => one or more aboveAvgUtilized datanodes<\/p>\n
      \u64cd\u4f5c\u6b65\u9aa4<\/h3>\n
      \n
      1. \u8bbe\u7f6ebalancer\u7684\u4f20\u8f93\u5e26\u5bbd<\/h4>\n
      \u5f00\u542f\u5b89\u5168\u9700\u8981\u5148\u8fdb\u884ckinit hdfs\u8ba4\u8bc1\uff1a<\/h5>\n
      $ kinit hdfs\n$ hadoop dfsadmin -setBalancerBandwidth xxxx<\/code><\/pre>\n
      \u672a\u5f00\u542f\u5b89\u5168\u9700\u8981\u5148export\uff1a<\/h5>\n
      $ export HADOOP_USER_NAME=hdfs\n$ hdfs dfsadmin -setBalancerBandwidth xxxx<\/code><\/pre>\n
      xxx\u4ee3\u8868\u8bbe\u7f6e\u7684\u5e26\u5bbd\u503c\uff0c\u9ed8\u8ba4\u4e3a1048576\uff0c\u53731MB\/s\uff0c\u8fd9\u4e2a\u9700\u8981\u6839\u636e\u81ea\u5df1\u7684\u673a\u5668\u4ee5\u53ca\u7f51\u7edc\u914d\u7f6e\u6765\u8fdb\u884c\u8bbe\u7f6e<\/p>\n
      2. \u9875\u9762\u4e0a\u5148\u786e\u8ba4\u4e0b\u5f53\u524ddatanode\u7684\u7a7a\u95f4\u4f7f\u7528\u60c5\u51b5\uff1a<\/h4>\n
      \"file\"
      \n\u6b64\u65f6\u53ef\u4ee5\u770b\u5230\uff0c\u65b0\u52a0\u7684datanode\u4f7f\u7528\u7387\u57fa\u672c\u4e3a0<\/p>\n
      3. \u6267\u884cbalancer\uff1a<\/h4>\n
      $ hdfs balancer -threshold 1<\/code><\/pre>\n
      \u5982\u679c\u5f53\u524d\u96c6\u7fa4\u6709\u82e5\u5e72datanode\u7a7a\u95f4\u5feb\u6ee1\uff0c\u5efa\u8bae\u5c06threshold\u8bbe\u7f6e\u4e3a\u6bd4\u8f83\u5927\u7684\u503c\uff0c\u8fd9\u6837\u4f1a\u4f18\u5148\u4ece\u4f7f\u7528\u7387\u6bd4\u8f83\u9ad8\u7684\u8282\u70b9\u5f80\u65b0\u8282\u70b9\u5e73\u8861\u3002\u5982\u679c\u662f\u5e73\u65f6\u6b63\u5e38\u65f6\u5019\u7684\u5efa\u8bae\u8bbe\u7f6e\u4e3a10%\u5de6\u53f3\u3002<\/p>\n
      4. balance\u5b8c\u6210\u540e\u9a8c\u8bc1\u7ed3\u679c<\/h4>\n
      \"file\"
      \n\u6b64\u65f6\u53ef\u4ee5\u770b\u5230\uff0c\u5404\u4e2a\u8282\u70b9\u57fa\u672c\u5747\u8861\uff0c\u5982\u679c\u5bf9\u5747\u8861\u7ed3\u679c\u4e0d\u6ee1\u610f\uff0c\u53ef\u4ee5\u5c1d\u8bd5\u518d\u6b21\u6267\u884cbalance\u547d\u4ee4<\/p>\n
      FAQ<\/h3>\n
      \n
      1. \u5982\u679c\u96c6\u7fa4\u672c\u8eab\u541e\u5410\u91cf\u5f88\u5927\u6216\u8005\u6570\u636e\u91cf\u5f88\u5927\uff0c\u53ef\u80fd\u5747\u8861\u6548\u679c\u4e0d\u662f\u5f88\u597d\uff0c\u9700\u8981\u591a\u6267\u884c\u51e0\u6b21<\/h5>\n
      2. \u5982\u679c\u9700\u8981balance\u7684\u6570\u636e\u91cf\u975e\u5e38\u5927\uff0c\u5e73\u8861\u901f\u5ea6\u8fdc\u8fdc\u8fbe\u4e0d\u5230\u8981\u6c42<\/h5>\n
      \u53ef\u4ee5\u5c1d\u8bd5\u91c7\u53d6\u9000\u5f79\u5bf9\u5e94\u7684datanode\uff0c\u7136\u540e\u683c\u5f0f\u5316\u5bf9\u5e94\u7684\u6570\u636e\u5b58\u50a8\u78c1\u76d8\uff0c\u7136\u540e\u91cd\u65b0\u6dfb\u52a0\u8282\u70b9\u7684\u65b9\u6cd5\u3002\u5982\u679c\u91c7\u53d6\u6b64\u79cd\u65b9\u6cd5\uff0c\u7406\u8bba\u4e0a\u4e3a\u4e86\u6570\u636e\u5b89\u5168\uff0c\u540c\u65f6\u9000\u5f79\u7684\u8282\u70b9\u6570\u4e0d\u80fd\u8d85\u8fc7\u9ed8\u8ba4\u7684\u526f\u672c\u6570\uff08\u9ed8\u8ba4\u662f3\uff09\uff0c\u5373\u540c\u65f6\u9000\u5f79\u7684\u6700\u5927\u8282\u70b9\u6570\u4e3a2\u3002<\/p>\n
      3. \u5efa\u8bae\u5728\u96c6\u7fa4\u4e1a\u52a1\u4f4e\u5cf0\u671f\u6267\u884cbalance\u64cd\u4f5c\uff0c\u5e76\u4e14\u8bbe\u7f6e\u7684\u5e26\u5bbd\u4e0d\u5b9c\u8fc7\u5927\uff0c\u9632\u6b62\u5f71\u54cd\u5230\u96c6\u7fa4\u6b63\u5e38\u4efb\u52a1\u3002<\/h5>\n","protected":false},"excerpt":{"rendered":"
      \u6982\u8981\u8bf4\u660e \u7ebf\u4e0a\u957f\u65f6\u95f4\u8fd0\u884c\u7684\u5927\u89c4\u6a21Hadoop\u96c6\u7fa4\uff0c\u5404\u4e2adatanode\u8282\u70b9\u78c1\u76d8\u7a7a\u95f4\u4f7f\u7528\u7387\u7ecf\u5e38\u4f1a\u51fa\u73b0\u5206\u5e03\u4e0d\u5747\u8861\u7684 ..<\/p>\n
      <\/div>\n
      Read more<\/a><\/p>\n","protected":false},"author":27,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-3562","post","type-post","status-publish","format-standard","hentry","category-uncategorized"],"acf":[],"_links":{"self":[{"href":"https:\/\/kbwp.transwarp.cn\/index.php?rest_route=\/wp\/v2\/posts\/3562","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/kbwp.transwarp.cn\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/kbwp.transwarp.cn\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/kbwp.transwarp.cn\/index.php?rest_route=\/wp\/v2\/users\/27"}],"replies":[{"embeddable":true,"href":"https:\/\/kbwp.transwarp.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=3562"}],"version-history":[{"count":2,"href":"https:\/\/kbwp.transwarp.cn\/index.php?rest_route=\/wp\/v2\/posts\/3562\/revisions"}],"predecessor-version":[{"id":3955,"href":"https:\/\/kbwp.transwarp.cn\/index.php?rest_route=\/wp\/v2\/posts\/3562\/revisions\/3955"}],"wp:attachment":[{"href":"https:\/\/kbwp.transwarp.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=3562"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/kbwp.transwarp.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=3562"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/kbwp.transwarp.cn\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=3562"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}