blob: c672ecc8a5d106177476e871c7a0623bca2ba3d5 [file] [log] [blame]
<!DOCTYPE HTML>
<html lang="en">
<head>
<!-- Generated by javadoc (17) -->
<title>Source code</title>
<meta name="viewport" content="width=device-width, initial-scale=1">
<meta name="description" content="source: package: org.apache.hadoop.hbase.util, class: RegionSplitCalculator">
<meta name="generator" content="javadoc/SourceToHTMLConverter">
<link rel="stylesheet" type="text/css" href="../../../../../../stylesheet.css" title="Style">
</head>
<body class="source-page">
<main role="main">
<div class="source-container">
<pre><span class="source-line-no">001</span><span id="line-1">/*</span>
<span class="source-line-no">002</span><span id="line-2"> * Licensed to the Apache Software Foundation (ASF) under one</span>
<span class="source-line-no">003</span><span id="line-3"> * or more contributor license agreements. See the NOTICE file</span>
<span class="source-line-no">004</span><span id="line-4"> * distributed with this work for additional information</span>
<span class="source-line-no">005</span><span id="line-5"> * regarding copyright ownership. The ASF licenses this file</span>
<span class="source-line-no">006</span><span id="line-6"> * to you under the Apache License, Version 2.0 (the</span>
<span class="source-line-no">007</span><span id="line-7"> * "License"); you may not use this file except in compliance</span>
<span class="source-line-no">008</span><span id="line-8"> * with the License. You may obtain a copy of the License at</span>
<span class="source-line-no">009</span><span id="line-9"> *</span>
<span class="source-line-no">010</span><span id="line-10"> * http://www.apache.org/licenses/LICENSE-2.0</span>
<span class="source-line-no">011</span><span id="line-11"> *</span>
<span class="source-line-no">012</span><span id="line-12"> * Unless required by applicable law or agreed to in writing, software</span>
<span class="source-line-no">013</span><span id="line-13"> * distributed under the License is distributed on an "AS IS" BASIS,</span>
<span class="source-line-no">014</span><span id="line-14"> * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.</span>
<span class="source-line-no">015</span><span id="line-15"> * See the License for the specific language governing permissions and</span>
<span class="source-line-no">016</span><span id="line-16"> * limitations under the License.</span>
<span class="source-line-no">017</span><span id="line-17"> */</span>
<span class="source-line-no">018</span><span id="line-18">package org.apache.hadoop.hbase.util;</span>
<span class="source-line-no">019</span><span id="line-19"></span>
<span class="source-line-no">020</span><span id="line-20">import java.util.ArrayList;</span>
<span class="source-line-no">021</span><span id="line-21">import java.util.Collection;</span>
<span class="source-line-no">022</span><span id="line-22">import java.util.Comparator;</span>
<span class="source-line-no">023</span><span id="line-23">import java.util.List;</span>
<span class="source-line-no">024</span><span id="line-24">import java.util.Map.Entry;</span>
<span class="source-line-no">025</span><span id="line-25">import java.util.TreeMap;</span>
<span class="source-line-no">026</span><span id="line-26">import java.util.TreeSet;</span>
<span class="source-line-no">027</span><span id="line-27">import org.apache.hadoop.hbase.util.Bytes.ByteArrayComparator;</span>
<span class="source-line-no">028</span><span id="line-28">import org.apache.yetus.audience.InterfaceAudience;</span>
<span class="source-line-no">029</span><span id="line-29">import org.slf4j.Logger;</span>
<span class="source-line-no">030</span><span id="line-30">import org.slf4j.LoggerFactory;</span>
<span class="source-line-no">031</span><span id="line-31"></span>
<span class="source-line-no">032</span><span id="line-32">import org.apache.hbase.thirdparty.com.google.common.collect.ArrayListMultimap;</span>
<span class="source-line-no">033</span><span id="line-33">import org.apache.hbase.thirdparty.com.google.common.collect.Multimap;</span>
<span class="source-line-no">034</span><span id="line-34">import org.apache.hbase.thirdparty.com.google.common.collect.TreeMultimap;</span>
<span class="source-line-no">035</span><span id="line-35"></span>
<span class="source-line-no">036</span><span id="line-36">/**</span>
<span class="source-line-no">037</span><span id="line-37"> * This is a generic region split calculator. It requires Ranges that provide start, end, and a</span>
<span class="source-line-no">038</span><span id="line-38"> * comparator. It works in two phases -- the first adds ranges and rejects backwards ranges. Then</span>
<span class="source-line-no">039</span><span id="line-39"> * one calls calcRegions to generate the multimap that has a start split key as a key and possibly</span>
<span class="source-line-no">040</span><span id="line-40"> * multiple Ranges as members. To traverse, one normally would get the split set, and iterate</span>
<span class="source-line-no">041</span><span id="line-41"> * through the calcRegions. Normal regions would have only one entry, holes would have zero, and any</span>
<span class="source-line-no">042</span><span id="line-42"> * overlaps would have multiple entries. The interface is a bit cumbersome currently but is exposed</span>
<span class="source-line-no">043</span><span id="line-43"> * this way so that clients can choose how to iterate through the region splits.</span>
<span class="source-line-no">044</span><span id="line-44"> * @param &lt;R&gt;</span>
<span class="source-line-no">045</span><span id="line-45"> */</span>
<span class="source-line-no">046</span><span id="line-46">@InterfaceAudience.Private</span>
<span class="source-line-no">047</span><span id="line-47">public class RegionSplitCalculator&lt;R extends KeyRange&gt; {</span>
<span class="source-line-no">048</span><span id="line-48"> private static final Logger LOG = LoggerFactory.getLogger(RegionSplitCalculator.class);</span>
<span class="source-line-no">049</span><span id="line-49"></span>
<span class="source-line-no">050</span><span id="line-50"> private final Comparator&lt;R&gt; rangeCmp;</span>
<span class="source-line-no">051</span><span id="line-51"> /**</span>
<span class="source-line-no">052</span><span id="line-52"> * This contains a sorted set of all the possible split points Invariant: once populated this has</span>
<span class="source-line-no">053</span><span id="line-53"> * 0 entries if empty or at most n+1 values where n == number of added ranges.</span>
<span class="source-line-no">054</span><span id="line-54"> */</span>
<span class="source-line-no">055</span><span id="line-55"> private final TreeSet&lt;byte[]&gt; splits = new TreeSet&lt;&gt;(BYTES_COMPARATOR);</span>
<span class="source-line-no">056</span><span id="line-56"></span>
<span class="source-line-no">057</span><span id="line-57"> /**</span>
<span class="source-line-no">058</span><span id="line-58"> * This is a map from start key to regions with the same start key. Invariant: This always have n</span>
<span class="source-line-no">059</span><span id="line-59"> * values in total</span>
<span class="source-line-no">060</span><span id="line-60"> */</span>
<span class="source-line-no">061</span><span id="line-61"> private final Multimap&lt;byte[], R&gt; starts = ArrayListMultimap.create();</span>
<span class="source-line-no">062</span><span id="line-62"></span>
<span class="source-line-no">063</span><span id="line-63"> /**</span>
<span class="source-line-no">064</span><span id="line-64"> * SPECIAL CASE</span>
<span class="source-line-no">065</span><span id="line-65"> */</span>
<span class="source-line-no">066</span><span id="line-66"> private final static byte[] ENDKEY = null;</span>
<span class="source-line-no">067</span><span id="line-67"></span>
<span class="source-line-no">068</span><span id="line-68"> public RegionSplitCalculator(Comparator&lt;R&gt; cmp) {</span>
<span class="source-line-no">069</span><span id="line-69"> rangeCmp = cmp;</span>
<span class="source-line-no">070</span><span id="line-70"> }</span>
<span class="source-line-no">071</span><span id="line-71"></span>
<span class="source-line-no">072</span><span id="line-72"> public final static Comparator&lt;byte[]&gt; BYTES_COMPARATOR = new ByteArrayComparator() {</span>
<span class="source-line-no">073</span><span id="line-73"> @Override</span>
<span class="source-line-no">074</span><span id="line-74"> public int compare(byte[] l, byte[] r) {</span>
<span class="source-line-no">075</span><span id="line-75"> if (l == null &amp;&amp; r == null) return 0;</span>
<span class="source-line-no">076</span><span id="line-76"> if (l == null) return 1;</span>
<span class="source-line-no">077</span><span id="line-77"> if (r == null) return -1;</span>
<span class="source-line-no">078</span><span id="line-78"> return super.compare(l, r);</span>
<span class="source-line-no">079</span><span id="line-79"> }</span>
<span class="source-line-no">080</span><span id="line-80"> };</span>
<span class="source-line-no">081</span><span id="line-81"></span>
<span class="source-line-no">082</span><span id="line-82"> /**</span>
<span class="source-line-no">083</span><span id="line-83"> * SPECIAL CASE wrapper for empty end key</span>
<span class="source-line-no">084</span><span id="line-84"> * @return ENDKEY if end key is empty, else normal endkey.</span>
<span class="source-line-no">085</span><span id="line-85"> */</span>
<span class="source-line-no">086</span><span id="line-86"> private static &lt;R extends KeyRange&gt; byte[] specialEndKey(R range) {</span>
<span class="source-line-no">087</span><span id="line-87"> byte[] end = range.getEndKey();</span>
<span class="source-line-no">088</span><span id="line-88"> if (end.length == 0) {</span>
<span class="source-line-no">089</span><span id="line-89"> return ENDKEY;</span>
<span class="source-line-no">090</span><span id="line-90"> }</span>
<span class="source-line-no">091</span><span id="line-91"> return end;</span>
<span class="source-line-no">092</span><span id="line-92"> }</span>
<span class="source-line-no">093</span><span id="line-93"></span>
<span class="source-line-no">094</span><span id="line-94"> /**</span>
<span class="source-line-no">095</span><span id="line-95"> * Adds an edge to the split calculator</span>
<span class="source-line-no">096</span><span id="line-96"> * @return true if is included, false if backwards/invalid</span>
<span class="source-line-no">097</span><span id="line-97"> */</span>
<span class="source-line-no">098</span><span id="line-98"> public boolean add(R range) {</span>
<span class="source-line-no">099</span><span id="line-99"> byte[] start = range.getStartKey();</span>
<span class="source-line-no">100</span><span id="line-100"> byte[] end = specialEndKey(range);</span>
<span class="source-line-no">101</span><span id="line-101"></span>
<span class="source-line-no">102</span><span id="line-102"> // No need to use Arrays.equals because ENDKEY is null</span>
<span class="source-line-no">103</span><span id="line-103"> if (end != ENDKEY &amp;&amp; Bytes.compareTo(start, end) &gt; 0) {</span>
<span class="source-line-no">104</span><span id="line-104"> // don't allow backwards edges</span>
<span class="source-line-no">105</span><span id="line-105"> LOG.debug(</span>
<span class="source-line-no">106</span><span id="line-106"> "attempted to add backwards edge: " + Bytes.toString(start) + " " + Bytes.toString(end));</span>
<span class="source-line-no">107</span><span id="line-107"> return false;</span>
<span class="source-line-no">108</span><span id="line-108"> }</span>
<span class="source-line-no">109</span><span id="line-109"></span>
<span class="source-line-no">110</span><span id="line-110"> splits.add(start);</span>
<span class="source-line-no">111</span><span id="line-111"> splits.add(end);</span>
<span class="source-line-no">112</span><span id="line-112"> starts.put(start, range);</span>
<span class="source-line-no">113</span><span id="line-113"> return true;</span>
<span class="source-line-no">114</span><span id="line-114"> }</span>
<span class="source-line-no">115</span><span id="line-115"></span>
<span class="source-line-no">116</span><span id="line-116"> /**</span>
<span class="source-line-no">117</span><span id="line-117"> * Generates a coverage multimap from split key to Regions that start with the split key.</span>
<span class="source-line-no">118</span><span id="line-118"> * @return coverage multimap</span>
<span class="source-line-no">119</span><span id="line-119"> */</span>
<span class="source-line-no">120</span><span id="line-120"> public Multimap&lt;byte[], R&gt; calcCoverage() {</span>
<span class="source-line-no">121</span><span id="line-121"> // This needs to be sorted to force the use of the comparator on the values,</span>
<span class="source-line-no">122</span><span id="line-122"> // otherwise byte array comparison isn't used</span>
<span class="source-line-no">123</span><span id="line-123"> Multimap&lt;byte[], R&gt; regions = TreeMultimap.create(BYTES_COMPARATOR, rangeCmp);</span>
<span class="source-line-no">124</span><span id="line-124"></span>
<span class="source-line-no">125</span><span id="line-125"> // march through all splits from the start points</span>
<span class="source-line-no">126</span><span id="line-126"> for (Entry&lt;byte[], Collection&lt;R&gt;&gt; start : starts.asMap().entrySet()) {</span>
<span class="source-line-no">127</span><span id="line-127"> byte[] key = start.getKey();</span>
<span class="source-line-no">128</span><span id="line-128"> for (R r : start.getValue()) {</span>
<span class="source-line-no">129</span><span id="line-129"> regions.put(key, r);</span>
<span class="source-line-no">130</span><span id="line-130"></span>
<span class="source-line-no">131</span><span id="line-131"> for (byte[] coveredSplit : splits.subSet(r.getStartKey(), specialEndKey(r))) {</span>
<span class="source-line-no">132</span><span id="line-132"> regions.put(coveredSplit, r);</span>
<span class="source-line-no">133</span><span id="line-133"> }</span>
<span class="source-line-no">134</span><span id="line-134"> }</span>
<span class="source-line-no">135</span><span id="line-135"> }</span>
<span class="source-line-no">136</span><span id="line-136"> return regions;</span>
<span class="source-line-no">137</span><span id="line-137"> }</span>
<span class="source-line-no">138</span><span id="line-138"></span>
<span class="source-line-no">139</span><span id="line-139"> public TreeSet&lt;byte[]&gt; getSplits() {</span>
<span class="source-line-no">140</span><span id="line-140"> return splits;</span>
<span class="source-line-no">141</span><span id="line-141"> }</span>
<span class="source-line-no">142</span><span id="line-142"></span>
<span class="source-line-no">143</span><span id="line-143"> public Multimap&lt;byte[], R&gt; getStarts() {</span>
<span class="source-line-no">144</span><span id="line-144"> return starts;</span>
<span class="source-line-no">145</span><span id="line-145"> }</span>
<span class="source-line-no">146</span><span id="line-146"></span>
<span class="source-line-no">147</span><span id="line-147"> /**</span>
<span class="source-line-no">148</span><span id="line-148"> * Find specified number of top ranges in a big overlap group. It could return less if there are</span>
<span class="source-line-no">149</span><span id="line-149"> * not that many top ranges. Once these top ranges are excluded, the big overlap group will be</span>
<span class="source-line-no">150</span><span id="line-150"> * broken into ranges with no overlapping, or smaller overlapped groups, and most likely some</span>
<span class="source-line-no">151</span><span id="line-151"> * holes.</span>
<span class="source-line-no">152</span><span id="line-152"> * @param bigOverlap a list of ranges that overlap with each other</span>
<span class="source-line-no">153</span><span id="line-153"> * @param count the max number of ranges to find</span>
<span class="source-line-no">154</span><span id="line-154"> * @return a list of ranges that overlap with most others</span>
<span class="source-line-no">155</span><span id="line-155"> */</span>
<span class="source-line-no">156</span><span id="line-156"> public static &lt;R extends KeyRange&gt; List&lt;R&gt; findBigRanges(Collection&lt;R&gt; bigOverlap, int count) {</span>
<span class="source-line-no">157</span><span id="line-157"> List&lt;R&gt; bigRanges = new ArrayList&lt;&gt;();</span>
<span class="source-line-no">158</span><span id="line-158"></span>
<span class="source-line-no">159</span><span id="line-159"> // The key is the count of overlaps,</span>
<span class="source-line-no">160</span><span id="line-160"> // The value is a list of ranges that have that many overlaps</span>
<span class="source-line-no">161</span><span id="line-161"> TreeMap&lt;Integer, List&lt;R&gt;&gt; overlapRangeMap = new TreeMap&lt;&gt;();</span>
<span class="source-line-no">162</span><span id="line-162"> for (R r : bigOverlap) {</span>
<span class="source-line-no">163</span><span id="line-163"> // Calculates the # of overlaps for each region</span>
<span class="source-line-no">164</span><span id="line-164"> // and populates rangeOverlapMap</span>
<span class="source-line-no">165</span><span id="line-165"> byte[] startKey = r.getStartKey();</span>
<span class="source-line-no">166</span><span id="line-166"> byte[] endKey = specialEndKey(r);</span>
<span class="source-line-no">167</span><span id="line-167"></span>
<span class="source-line-no">168</span><span id="line-168"> int overlappedRegions = 0;</span>
<span class="source-line-no">169</span><span id="line-169"> for (R rr : bigOverlap) {</span>
<span class="source-line-no">170</span><span id="line-170"> byte[] start = rr.getStartKey();</span>
<span class="source-line-no">171</span><span id="line-171"> byte[] end = specialEndKey(rr);</span>
<span class="source-line-no">172</span><span id="line-172"></span>
<span class="source-line-no">173</span><span id="line-173"> if (</span>
<span class="source-line-no">174</span><span id="line-174"> BYTES_COMPARATOR.compare(startKey, end) &lt; 0 &amp;&amp; BYTES_COMPARATOR.compare(endKey, start) &gt; 0</span>
<span class="source-line-no">175</span><span id="line-175"> ) {</span>
<span class="source-line-no">176</span><span id="line-176"> overlappedRegions++;</span>
<span class="source-line-no">177</span><span id="line-177"> }</span>
<span class="source-line-no">178</span><span id="line-178"> }</span>
<span class="source-line-no">179</span><span id="line-179"></span>
<span class="source-line-no">180</span><span id="line-180"> // One region always overlaps with itself,</span>
<span class="source-line-no">181</span><span id="line-181"> // so overlappedRegions should be more than 1</span>
<span class="source-line-no">182</span><span id="line-182"> // for actual overlaps.</span>
<span class="source-line-no">183</span><span id="line-183"> if (overlappedRegions &gt; 1) {</span>
<span class="source-line-no">184</span><span id="line-184"> Integer key = Integer.valueOf(overlappedRegions);</span>
<span class="source-line-no">185</span><span id="line-185"> List&lt;R&gt; ranges = overlapRangeMap.get(key);</span>
<span class="source-line-no">186</span><span id="line-186"> if (ranges == null) {</span>
<span class="source-line-no">187</span><span id="line-187"> ranges = new ArrayList&lt;&gt;();</span>
<span class="source-line-no">188</span><span id="line-188"> overlapRangeMap.put(key, ranges);</span>
<span class="source-line-no">189</span><span id="line-189"> }</span>
<span class="source-line-no">190</span><span id="line-190"> ranges.add(r);</span>
<span class="source-line-no">191</span><span id="line-191"> }</span>
<span class="source-line-no">192</span><span id="line-192"> }</span>
<span class="source-line-no">193</span><span id="line-193"> int toBeAdded = count;</span>
<span class="source-line-no">194</span><span id="line-194"> for (Integer key : overlapRangeMap.descendingKeySet()) {</span>
<span class="source-line-no">195</span><span id="line-195"> List&lt;R&gt; chunk = overlapRangeMap.get(key);</span>
<span class="source-line-no">196</span><span id="line-196"> int chunkSize = chunk.size();</span>
<span class="source-line-no">197</span><span id="line-197"> if (chunkSize &lt;= toBeAdded) {</span>
<span class="source-line-no">198</span><span id="line-198"> bigRanges.addAll(chunk);</span>
<span class="source-line-no">199</span><span id="line-199"> toBeAdded -= chunkSize;</span>
<span class="source-line-no">200</span><span id="line-200"> if (toBeAdded &gt; 0) continue;</span>
<span class="source-line-no">201</span><span id="line-201"> } else {</span>
<span class="source-line-no">202</span><span id="line-202"> // Try to use the middle chunk in case the overlapping is</span>
<span class="source-line-no">203</span><span id="line-203"> // chained, for example: [a, c), [b, e), [d, g), [f h)...</span>
<span class="source-line-no">204</span><span id="line-204"> // In such a case, sideline the middle chunk will break</span>
<span class="source-line-no">205</span><span id="line-205"> // the group efficiently.</span>
<span class="source-line-no">206</span><span id="line-206"> int start = (chunkSize - toBeAdded) / 2;</span>
<span class="source-line-no">207</span><span id="line-207"> int end = start + toBeAdded;</span>
<span class="source-line-no">208</span><span id="line-208"> for (int i = start; i &lt; end; i++) {</span>
<span class="source-line-no">209</span><span id="line-209"> bigRanges.add(chunk.get(i));</span>
<span class="source-line-no">210</span><span id="line-210"> }</span>
<span class="source-line-no">211</span><span id="line-211"> }</span>
<span class="source-line-no">212</span><span id="line-212"> break;</span>
<span class="source-line-no">213</span><span id="line-213"> }</span>
<span class="source-line-no">214</span><span id="line-214"> return bigRanges;</span>
<span class="source-line-no">215</span><span id="line-215"> }</span>
<span class="source-line-no">216</span><span id="line-216">}</span>
</pre>
</div>
</main>
</body>
</html>