http://git-wip-us.apache.org/repos/asf/hbase-site/blob/f17356a7/devapidocs/src-html/org/apache/hadoop/hbase/util/RegionSplitter.UniformSplit.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/util/RegionSplitter.UniformSplit.html b/devapidocs/src-html/org/apache/hadoop/hbase/util/RegionSplitter.UniformSplit.html index 89b0f0c..df6f4e7 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/util/RegionSplitter.UniformSplit.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/util/RegionSplitter.UniformSplit.html @@ -28,1103 +28,1096 @@ <span class="sourceLineNo">020</span><a name="line.20"></a> <span class="sourceLineNo">021</span>import java.io.IOException;<a name="line.21"></a> <span class="sourceLineNo">022</span>import java.math.BigInteger;<a name="line.22"></a> -<span class="sourceLineNo">023</span>import java.util.Arrays;<a name="line.23"></a> -<span class="sourceLineNo">024</span>import java.util.Collection;<a name="line.24"></a> -<span class="sourceLineNo">025</span>import java.util.Collections;<a name="line.25"></a> -<span class="sourceLineNo">026</span>import java.util.Comparator;<a name="line.26"></a> -<span class="sourceLineNo">027</span>import java.util.LinkedList;<a name="line.27"></a> -<span class="sourceLineNo">028</span>import java.util.List;<a name="line.28"></a> +<span class="sourceLineNo">023</span><a name="line.23"></a> +<span class="sourceLineNo">024</span>import java.util.Arrays;<a name="line.24"></a> +<span class="sourceLineNo">025</span>import java.util.Collection;<a name="line.25"></a> +<span class="sourceLineNo">026</span>import java.util.LinkedList;<a name="line.26"></a> +<span class="sourceLineNo">027</span>import java.util.List;<a name="line.27"></a> +<span class="sourceLineNo">028</span>import java.util.Map;<a name="line.28"></a> <span class="sourceLineNo">029</span>import java.util.Set;<a name="line.29"></a> <span class="sourceLineNo">030</span>import java.util.TreeMap;<a name="line.30"></a> -<span class="sourceLineNo">031</span><a name="line.31"></a> -<span class="sourceLineNo">032</span>import org.apache.commons.cli.CommandLine;<a name="line.32"></a> -<span class="sourceLineNo">033</span>import org.apache.commons.cli.GnuParser;<a name="line.33"></a> -<span class="sourceLineNo">034</span>import org.apache.commons.cli.HelpFormatter;<a name="line.34"></a> -<span class="sourceLineNo">035</span>import org.apache.commons.cli.OptionBuilder;<a name="line.35"></a> -<span class="sourceLineNo">036</span>import org.apache.commons.cli.Options;<a name="line.36"></a> -<span class="sourceLineNo">037</span>import org.apache.commons.cli.ParseException;<a name="line.37"></a> -<span class="sourceLineNo">038</span>import org.apache.commons.lang.ArrayUtils;<a name="line.38"></a> -<span class="sourceLineNo">039</span>import org.apache.commons.lang.StringUtils;<a name="line.39"></a> -<span class="sourceLineNo">040</span>import org.apache.commons.logging.Log;<a name="line.40"></a> -<span class="sourceLineNo">041</span>import org.apache.commons.logging.LogFactory;<a name="line.41"></a> -<span class="sourceLineNo">042</span>import org.apache.hadoop.conf.Configuration;<a name="line.42"></a> -<span class="sourceLineNo">043</span>import org.apache.hadoop.fs.FSDataInputStream;<a name="line.43"></a> -<span class="sourceLineNo">044</span>import org.apache.hadoop.fs.FSDataOutputStream;<a name="line.44"></a> -<span class="sourceLineNo">045</span>import org.apache.hadoop.fs.FileSystem;<a name="line.45"></a> -<span class="sourceLineNo">046</span>import org.apache.hadoop.fs.Path;<a name="line.46"></a> -<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.ClusterStatus;<a name="line.47"></a> -<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.HBaseConfiguration;<a name="line.48"></a> -<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.HColumnDescriptor;<a name="line.49"></a> -<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.HRegionInfo;<a name="line.50"></a> -<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.HRegionLocation;<a name="line.51"></a> -<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.HTableDescriptor;<a name="line.52"></a> -<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.MetaTableAccessor;<a name="line.53"></a> -<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.ServerName;<a name="line.54"></a> -<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.TableName;<a name="line.55"></a> -<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.classification.InterfaceAudience;<a name="line.56"></a> -<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.client.Admin;<a name="line.57"></a> -<span class="sourceLineNo">058</span>import org.apache.hadoop.hbase.client.ClusterConnection;<a name="line.58"></a> -<span class="sourceLineNo">059</span>import org.apache.hadoop.hbase.client.Connection;<a name="line.59"></a> -<span class="sourceLineNo">060</span>import org.apache.hadoop.hbase.client.ConnectionFactory;<a name="line.60"></a> -<span class="sourceLineNo">061</span>import org.apache.hadoop.hbase.client.NoServerForRegionException;<a name="line.61"></a> -<span class="sourceLineNo">062</span>import org.apache.hadoop.hbase.client.RegionLocator;<a name="line.62"></a> -<span class="sourceLineNo">063</span>import org.apache.hadoop.hbase.client.Table;<a name="line.63"></a> -<span class="sourceLineNo">064</span>import org.apache.hadoop.hbase.regionserver.HRegionFileSystem;<a name="line.64"></a> -<span class="sourceLineNo">065</span><a name="line.65"></a> -<span class="sourceLineNo">066</span>import com.google.common.base.Preconditions;<a name="line.66"></a> -<span class="sourceLineNo">067</span>import com.google.common.collect.Lists;<a name="line.67"></a> -<span class="sourceLineNo">068</span>import com.google.common.collect.Maps;<a name="line.68"></a> -<span class="sourceLineNo">069</span>import com.google.common.collect.Sets;<a name="line.69"></a> -<span class="sourceLineNo">070</span><a name="line.70"></a> -<span class="sourceLineNo">071</span>/**<a name="line.71"></a> -<span class="sourceLineNo">072</span> * The {@link RegionSplitter} class provides several utilities to help in the<a name="line.72"></a> -<span class="sourceLineNo">073</span> * administration lifecycle for developers who choose to manually split regions<a name="line.73"></a> -<span class="sourceLineNo">074</span> * instead of having HBase handle that automatically. The most useful utilities<a name="line.74"></a> -<span class="sourceLineNo">075</span> * are:<a name="line.75"></a> -<span class="sourceLineNo">076</span> * <p><a name="line.76"></a> -<span class="sourceLineNo">077</span> * <ul><a name="line.77"></a> -<span class="sourceLineNo">078</span> * <li>Create a table with a specified number of pre-split regions<a name="line.78"></a> -<span class="sourceLineNo">079</span> * <li>Execute a rolling split of all regions on an existing table<a name="line.79"></a> -<span class="sourceLineNo">080</span> * </ul><a name="line.80"></a> -<span class="sourceLineNo">081</span> * <p><a name="line.81"></a> -<span class="sourceLineNo">082</span> * Both operations can be safely done on a live server.<a name="line.82"></a> -<span class="sourceLineNo">083</span> * <p><a name="line.83"></a> -<span class="sourceLineNo">084</span> * <b>Question:</b> How do I turn off automatic splitting? <br><a name="line.84"></a> -<span class="sourceLineNo">085</span> * <b>Answer:</b> Automatic splitting is determined by the configuration value<a name="line.85"></a> -<span class="sourceLineNo">086</span> * <i>HConstants.HREGION_MAX_FILESIZE</i>. It is not recommended that you set this<a name="line.86"></a> -<span class="sourceLineNo">087</span> * to Long.MAX_VALUE in case you forget about manual splits. A suggested setting<a name="line.87"></a> -<span class="sourceLineNo">088</span> * is 100GB, which would result in &gt; 1hr major compactions if reached.<a name="line.88"></a> -<span class="sourceLineNo">089</span> * <p><a name="line.89"></a> -<span class="sourceLineNo">090</span> * <b>Question:</b> Why did the original authors decide to manually split? <br><a name="line.90"></a> -<span class="sourceLineNo">091</span> * <b>Answer:</b> Specific workload characteristics of our use case allowed us<a name="line.91"></a> -<span class="sourceLineNo">092</span> * to benefit from a manual split system.<a name="line.92"></a> -<span class="sourceLineNo">093</span> * <p><a name="line.93"></a> -<span class="sourceLineNo">094</span> * <ul><a name="line.94"></a> -<span class="sourceLineNo">095</span> * <li>Data (~1k) that would grow instead of being replaced<a name="line.95"></a> -<span class="sourceLineNo">096</span> * <li>Data growth was roughly uniform across all regions<a name="line.96"></a> -<span class="sourceLineNo">097</span> * <li>OLTP workload. Data loss is a big deal.<a name="line.97"></a> -<span class="sourceLineNo">098</span> * </ul><a name="line.98"></a> -<span class="sourceLineNo">099</span> * <p><a name="line.99"></a> -<span class="sourceLineNo">100</span> * <b>Question:</b> Why is manual splitting good for this workload? <br><a name="line.100"></a> -<span class="sourceLineNo">101</span> * <b>Answer:</b> Although automated splitting is not a bad option, there are<a name="line.101"></a> -<span class="sourceLineNo">102</span> * benefits to manual splitting.<a name="line.102"></a> -<span class="sourceLineNo">103</span> * <p><a name="line.103"></a> -<span class="sourceLineNo">104</span> * <ul><a name="line.104"></a> -<span class="sourceLineNo">105</span> * <li>With growing amounts of data, splits will continually be needed. Since<a name="line.105"></a> -<span class="sourceLineNo">106</span> * you always know exactly what regions you have, long-term debugging and<a name="line.106"></a> -<span class="sourceLineNo">107</span> * profiling is much easier with manual splits. It is hard to trace the logs to<a name="line.107"></a> -<span class="sourceLineNo">108</span> * understand region level problems if it keeps splitting and getting renamed.<a name="line.108"></a> -<span class="sourceLineNo">109</span> * <li>Data offlining bugs + unknown number of split regions == oh crap! If an<a name="line.109"></a> -<span class="sourceLineNo">110</span> * WAL or StoreFile was mistakenly unprocessed by HBase due to a weird bug and<a name="line.110"></a> -<span class="sourceLineNo">111</span> * you notice it a day or so later, you can be assured that the regions<a name="line.111"></a> -<span class="sourceLineNo">112</span> * specified in these files are the same as the current regions and you have<a name="line.112"></a> -<span class="sourceLineNo">113</span> * less headaches trying to restore/replay your data.<a name="line.113"></a> -<span class="sourceLineNo">114</span> * <li>You can finely tune your compaction algorithm. With roughly uniform data<a name="line.114"></a> -<span class="sourceLineNo">115</span> * growth, it's easy to cause split / compaction storms as the regions all<a name="line.115"></a> -<span class="sourceLineNo">116</span> * roughly hit the same data size at the same time. With manual splits, you can<a name="line.116"></a> -<span class="sourceLineNo">117</span> * let staggered, time-based major compactions spread out your network IO load.<a name="line.117"></a> -<span class="sourceLineNo">118</span> * </ul><a name="line.118"></a> -<span class="sourceLineNo">119</span> * <p><a name="line.119"></a> -<span class="sourceLineNo">120</span> * <b>Question:</b> What's the optimal number of pre-split regions to create? <br><a name="line.120"></a> -<span class="sourceLineNo">121</span> * <b>Answer:</b> Mileage will vary depending upon your application.<a name="line.121"></a> -<span class="sourceLineNo">122</span> * <p><a name="line.122"></a> -<span class="sourceLineNo">123</span> * The short answer for our application is that we started with 10 pre-split<a name="line.123"></a> -<span class="sourceLineNo">124</span> * regions / server and watched our data growth over time. It's better to err on<a name="line.124"></a> -<span class="sourceLineNo">125</span> * the side of too little regions and rolling split later.<a name="line.125"></a> -<span class="sourceLineNo">126</span> * <p><a name="line.126"></a> -<span class="sourceLineNo">127</span> * The more complicated answer is that this depends upon the largest storefile<a name="line.127"></a> -<span class="sourceLineNo">128</span> * in your region. With a growing data size, this will get larger over time. You<a name="line.128"></a> -<span class="sourceLineNo">129</span> * want the largest region to be just big enough that the<a name="line.129"></a> -<span class="sourceLineNo">130</span> * {@link org.apache.hadoop.hbase.regionserver.HStore} compact<a name="line.130"></a> -<span class="sourceLineNo">131</span> * selection algorithm only compacts it due to a timed major. If you don't, your<a name="line.131"></a> -<span class="sourceLineNo">132</span> * cluster can be prone to compaction storms as the algorithm decides to run<a name="line.132"></a> -<span class="sourceLineNo">133</span> * major compactions on a large series of regions all at once. Note that<a name="line.133"></a> -<span class="sourceLineNo">134</span> * compaction storms are due to the uniform data growth, not the manual split<a name="line.134"></a> -<span class="sourceLineNo">135</span> * decision.<a name="line.135"></a> -<span class="sourceLineNo">136</span> * <p><a name="line.136"></a> -<span class="sourceLineNo">137</span> * If you pre-split your regions too thin, you can increase the major compaction<a name="line.137"></a> -<span class="sourceLineNo">138</span> * interval by configuring HConstants.MAJOR_COMPACTION_PERIOD. If your data size<a name="line.138"></a> -<span class="sourceLineNo">139</span> * grows too large, use this script to perform a network IO safe rolling split<a name="line.139"></a> -<span class="sourceLineNo">140</span> * of all regions.<a name="line.140"></a> -<span class="sourceLineNo">141</span> */<a name="line.141"></a> -<span class="sourceLineNo">142</span>@InterfaceAudience.Private<a name="line.142"></a> -<span class="sourceLineNo">143</span>public class RegionSplitter {<a name="line.143"></a> -<span class="sourceLineNo">144</span> private static final Log LOG = LogFactory.getLog(RegionSplitter.class);<a name="line.144"></a> -<span class="sourceLineNo">145</span><a name="line.145"></a> -<span class="sourceLineNo">146</span> /**<a name="line.146"></a> -<span class="sourceLineNo">147</span> * A generic interface for the RegionSplitter code to use for all it's<a name="line.147"></a> -<span class="sourceLineNo">148</span> * functionality. Note that the original authors of this code use<a name="line.148"></a> -<span class="sourceLineNo">149</span> * {@link HexStringSplit} to partition their table and set it as default, but<a name="line.149"></a> -<span class="sourceLineNo">150</span> * provided this for your custom algorithm. To use, create a new derived class<a name="line.150"></a> -<span class="sourceLineNo">151</span> * from this interface and call {@link RegionSplitter#createPresplitTable} or<a name="line.151"></a> -<span class="sourceLineNo">152</span> * RegionSplitter#rollingSplit(TableName, SplitAlgorithm, Configuration) with the<a name="line.152"></a> -<span class="sourceLineNo">153</span> * argument splitClassName giving the name of your class.<a name="line.153"></a> -<span class="sourceLineNo">154</span> */<a name="line.154"></a> -<span class="sourceLineNo">155</span> public interface SplitAlgorithm {<a name="line.155"></a> -<span class="sourceLineNo">156</span> /**<a name="line.156"></a> -<span class="sourceLineNo">157</span> * Split a pre-existing region into 2 regions.<a name="line.157"></a> -<span class="sourceLineNo">158</span> *<a name="line.158"></a> -<span class="sourceLineNo">159</span> * @param start<a name="line.159"></a> -<span class="sourceLineNo">160</span> * first row (inclusive)<a name="line.160"></a> -<span class="sourceLineNo">161</span> * @param end<a name="line.161"></a> -<span class="sourceLineNo">162</span> * last row (exclusive)<a name="line.162"></a> -<span class="sourceLineNo">163</span> * @return the split row to use<a name="line.163"></a> -<span class="sourceLineNo">164</span> */<a name="line.164"></a> -<span class="sourceLineNo">165</span> byte[] split(byte[] start, byte[] end);<a name="line.165"></a> -<span class="sourceLineNo">166</span><a name="line.166"></a> -<span class="sourceLineNo">167</span> /**<a name="line.167"></a> -<span class="sourceLineNo">168</span> * Split an entire table.<a name="line.168"></a> -<span class="sourceLineNo">169</span> *<a name="line.169"></a> -<span class="sourceLineNo">170</span> * @param numRegions<a name="line.170"></a> -<span class="sourceLineNo">171</span> * number of regions to split the table into<a name="line.171"></a> -<span class="sourceLineNo">172</span> *<a name="line.172"></a> -<span class="sourceLineNo">173</span> * @throws RuntimeException<a name="line.173"></a> -<span class="sourceLineNo">174</span> * user input is validated at this time. may throw a runtime<a name="line.174"></a> -<span class="sourceLineNo">175</span> * exception in response to a parse failure<a name="line.175"></a> -<span class="sourceLineNo">176</span> * @return array of split keys for the initial regions of the table. The<a name="line.176"></a> -<span class="sourceLineNo">177</span> * length of the returned array should be numRegions-1.<a name="line.177"></a> -<span class="sourceLineNo">178</span> */<a name="line.178"></a> -<span class="sourceLineNo">179</span> byte[][] split(int numRegions);<a name="line.179"></a> -<span class="sourceLineNo">180</span><a name="line.180"></a> -<span class="sourceLineNo">181</span> /**<a name="line.181"></a> -<span class="sourceLineNo">182</span> * In HBase, the first row is represented by an empty byte array. This might<a name="line.182"></a> -<span class="sourceLineNo">183</span> * cause problems with your split algorithm or row printing. All your APIs<a name="line.183"></a> -<span class="sourceLineNo">184</span> * will be passed firstRow() instead of empty array.<a name="line.184"></a> -<span class="sourceLineNo">185</span> *<a name="line.185"></a> -<span class="sourceLineNo">186</span> * @return your representation of your first row<a name="line.186"></a> -<span class="sourceLineNo">187</span> */<a name="line.187"></a> -<span class="sourceLineNo">188</span> byte[] firstRow();<a name="line.188"></a> -<span class="sourceLineNo">189</span><a name="line.189"></a> -<span class="sourceLineNo">190</span> /**<a name="line.190"></a> -<span class="sourceLineNo">191</span> * In HBase, the last row is represented by an empty byte array. This might<a name="line.191"></a> -<span class="sourceLineNo">192</span> * cause problems with your split algorithm or row printing. All your APIs<a name="line.192"></a> -<span class="sourceLineNo">193</span> * will be passed firstRow() instead of empty array.<a name="line.193"></a> -<span class="sourceLineNo">194</span> *<a name="line.194"></a> -<span class="sourceLineNo">195</span> * @return your representation of your last row<a name="line.195"></a> -<span class="sourceLineNo">196</span> */<a name="line.196"></a> -<span class="sourceLineNo">197</span> byte[] lastRow();<a name="line.197"></a> -<span class="sourceLineNo">198</span><a name="line.198"></a> -<span class="sourceLineNo">199</span> /**<a name="line.199"></a> -<span class="sourceLineNo">200</span> * In HBase, the last row is represented by an empty byte array. Set this<a name="line.200"></a> -<span class="sourceLineNo">201</span> * value to help the split code understand how to evenly divide the first<a name="line.201"></a> -<span class="sourceLineNo">202</span> * region.<a name="line.202"></a> -<span class="sourceLineNo">203</span> *<a name="line.203"></a> -<span class="sourceLineNo">204</span> * @param userInput<a name="line.204"></a> -<span class="sourceLineNo">205</span> * raw user input (may throw RuntimeException on parse failure)<a name="line.205"></a> -<span class="sourceLineNo">206</span> */<a name="line.206"></a> -<span class="sourceLineNo">207</span> void setFirstRow(String userInput);<a name="line.207"></a> -<span class="sourceLineNo">208</span><a name="line.208"></a> -<span class="sourceLineNo">209</span> /**<a name="line.209"></a> -<span class="sourceLineNo">210</span> * In HBase, the last row is represented by an empty byte array. Set this<a name="line.210"></a> -<span class="sourceLineNo">211</span> * value to help the split code understand how to evenly divide the last<a name="line.211"></a> -<span class="sourceLineNo">212</span> * region. Note that this last row is inclusive for all rows sharing the<a name="line.212"></a> -<span class="sourceLineNo">213</span> * same prefix.<a name="line.213"></a> -<span class="sourceLineNo">214</span> *<a name="line.214"></a> -<span class="sourceLineNo">215</span> * @param userInput<a name="line.215"></a> -<span class="sourceLineNo">216</span> * raw user input (may throw RuntimeException on parse failure)<a name="line.216"></a> -<span class="sourceLineNo">217</span> */<a name="line.217"></a> -<span class="sourceLineNo">218</span> void setLastRow(String userInput);<a name="line.218"></a> -<span class="sourceLineNo">219</span><a name="line.219"></a> -<span class="sourceLineNo">220</span> /**<a name="line.220"></a> -<span class="sourceLineNo">221</span> * @param input<a name="line.221"></a> -<span class="sourceLineNo">222</span> * user or file input for row<a name="line.222"></a> -<span class="sourceLineNo">223</span> * @return byte array representation of this row for HBase<a name="line.223"></a> -<span class="sourceLineNo">224</span> */<a name="line.224"></a> -<span class="sourceLineNo">225</span> byte[] strToRow(String input);<a name="line.225"></a> -<span class="sourceLineNo">226</span><a name="line.226"></a> -<span class="sourceLineNo">227</span> /**<a name="line.227"></a> -<span class="sourceLineNo">228</span> * @param row<a name="line.228"></a> -<span class="sourceLineNo">229</span> * byte array representing a row in HBase<a name="line.229"></a> -<span class="sourceLineNo">230</span> * @return String to use for debug &amp; file printing<a name="line.230"></a> -<span class="sourceLineNo">231</span> */<a name="line.231"></a> -<span class="sourceLineNo">232</span> String rowToStr(byte[] row);<a name="line.232"></a> -<span class="sourceLineNo">233</span><a name="line.233"></a> -<span class="sourceLineNo">234</span> /**<a name="line.234"></a> -<span class="sourceLineNo">235</span> * @return the separator character to use when storing / printing the row<a name="line.235"></a> -<span class="sourceLineNo">236</span> */<a name="line.236"></a> -<span class="sourceLineNo">237</span> String separator();<a name="line.237"></a> -<span class="sourceLineNo">238</span><a name="line.238"></a> -<span class="sourceLineNo">239</span> /**<a name="line.239"></a> -<span class="sourceLineNo">240</span> * Set the first row<a name="line.240"></a> -<span class="sourceLineNo">241</span> * @param userInput byte array of the row key.<a name="line.241"></a> -<span class="sourceLineNo">242</span> */<a name="line.242"></a> -<span class="sourceLineNo">243</span> void setFirstRow(byte[] userInput);<a name="line.243"></a> -<span class="sourceLineNo">244</span><a name="line.244"></a> -<span class="sourceLineNo">245</span> /**<a name="line.245"></a> -<span class="sourceLineNo">246</span> * Set the last row<a name="line.246"></a> -<span class="sourceLineNo">247</span> * @param userInput byte array of the row key.<a name="line.247"></a> -<span class="sourceLineNo">248</span> */<a name="line.248"></a> -<span class="sourceLineNo">249</span> void setLastRow(byte[] userInput);<a name="line.249"></a> -<span class="sourceLineNo">250</span> }<a name="line.250"></a> -<span class="sourceLineNo">251</span><a name="line.251"></a> -<span class="sourceLineNo">252</span> /**<a name="line.252"></a> -<span class="sourceLineNo">253</span> * The main function for the RegionSplitter application. Common uses:<a name="line.253"></a> -<span class="sourceLineNo">254</span> * <p><a name="line.254"></a> -<span class="sourceLineNo">255</span> * <ul><a name="line.255"></a> -<span class="sourceLineNo">256</span> * <li>create a table named 'myTable' with 60 pre-split regions containing 2<a name="line.256"></a> -<span class="sourceLineNo">257</span> * column families 'test' &amp; 'rs', assuming the keys are hex-encoded ASCII:<a name="line.257"></a> -<span class="sourceLineNo">258</span> * <ul><a name="line.258"></a> -<span class="sourceLineNo">259</span> * <li>bin/hbase org.apache.hadoop.hbase.util.RegionSplitter -c 60 -f test:rs<a name="line.259"></a> -<span class="sourceLineNo">260</span> * myTable HexStringSplit<a name="line.260"></a> -<span class="sourceLineNo">261</span> * </ul><a name="line.261"></a> -<span class="sourceLineNo">262</span> * <li>perform a rolling split of 'myTable' (i.e. 60 =&gt; 120 regions), # 2<a name="line.262"></a> -<span class="sourceLineNo">263</span> * outstanding splits at a time, assuming keys are uniformly distributed<a name="line.263"></a> -<span class="sourceLineNo">264</span> * bytes:<a name="line.264"></a> -<span class="sourceLineNo">265</span> * <ul><a name="line.265"></a> -<span class="sourceLineNo">266</span> * <li>bin/hbase org.apache.hadoop.hbase.util.RegionSplitter -r -o 2 myTable<a name="line.266"></a> -<span class="sourceLineNo">267</span> * UniformSplit<a name="line.267"></a> +<span class="sourceLineNo">031</span>import org.apache.commons.cli.CommandLine;<a name="line.31"></a> +<span class="sourceLineNo">032</span>import org.apache.commons.cli.GnuParser;<a name="line.32"></a> +<span class="sourceLineNo">033</span>import org.apache.commons.cli.HelpFormatter;<a name="line.33"></a> +<span class="sourceLineNo">034</span>import org.apache.commons.cli.OptionBuilder;<a name="line.34"></a> +<span class="sourceLineNo">035</span>import org.apache.commons.cli.Options;<a name="line.35"></a> +<span class="sourceLineNo">036</span>import org.apache.commons.cli.ParseException;<a name="line.36"></a> +<span class="sourceLineNo">037</span>import org.apache.commons.lang.ArrayUtils;<a name="line.37"></a> +<span class="sourceLineNo">038</span>import org.apache.commons.lang.StringUtils;<a name="line.38"></a> +<span class="sourceLineNo">039</span>import org.apache.commons.logging.Log;<a name="line.39"></a> +<span class="sourceLineNo">040</span>import org.apache.commons.logging.LogFactory;<a name="line.40"></a> +<span class="sourceLineNo">041</span>import org.apache.hadoop.conf.Configuration;<a name="line.41"></a> +<span class="sourceLineNo">042</span>import org.apache.hadoop.fs.FSDataInputStream;<a name="line.42"></a> +<span class="sourceLineNo">043</span>import org.apache.hadoop.fs.FSDataOutputStream;<a name="line.43"></a> +<span class="sourceLineNo">044</span>import org.apache.hadoop.fs.FileSystem;<a name="line.44"></a> +<span class="sourceLineNo">045</span>import org.apache.hadoop.fs.Path;<a name="line.45"></a> +<span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.ClusterStatus;<a name="line.46"></a> +<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.HBaseConfiguration;<a name="line.47"></a> +<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.HColumnDescriptor;<a name="line.48"></a> +<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.HRegionInfo;<a name="line.49"></a> +<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.HRegionLocation;<a name="line.50"></a> +<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.HTableDescriptor;<a name="line.51"></a> +<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.MetaTableAccessor;<a name="line.52"></a> +<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.ServerName;<a name="line.53"></a> +<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.TableName;<a name="line.54"></a> +<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.classification.InterfaceAudience;<a name="line.55"></a> +<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.client.Admin;<a name="line.56"></a> +<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.client.ClusterConnection;<a name="line.57"></a> +<span class="sourceLineNo">058</span>import org.apache.hadoop.hbase.client.Connection;<a name="line.58"></a> +<span class="sourceLineNo">059</span>import org.apache.hadoop.hbase.client.ConnectionFactory;<a name="line.59"></a> +<span class="sourceLineNo">060</span>import org.apache.hadoop.hbase.client.NoServerForRegionException;<a name="line.60"></a> +<span class="sourceLineNo">061</span>import org.apache.hadoop.hbase.client.RegionLocator;<a name="line.61"></a> +<span class="sourceLineNo">062</span>import org.apache.hadoop.hbase.client.Table;<a name="line.62"></a> +<span class="sourceLineNo">063</span>import org.apache.hadoop.hbase.regionserver.HRegionFileSystem;<a name="line.63"></a> +<span class="sourceLineNo">064</span><a name="line.64"></a> +<span class="sourceLineNo">065</span>import com.google.common.base.Preconditions;<a name="line.65"></a> +<span class="sourceLineNo">066</span>import com.google.common.collect.Lists;<a name="line.66"></a> +<span class="sourceLineNo">067</span>import com.google.common.collect.Maps;<a name="line.67"></a> +<span class="sourceLineNo">068</span>import com.google.common.collect.Sets;<a name="line.68"></a> +<span class="sourceLineNo">069</span><a name="line.69"></a> +<span class="sourceLineNo">070</span>/**<a name="line.70"></a> +<span class="sourceLineNo">071</span> * The {@link RegionSplitter} class provides several utilities to help in the<a name="line.71"></a> +<span class="sourceLineNo">072</span> * administration lifecycle for developers who choose to manually split regions<a name="line.72"></a> +<span class="sourceLineNo">073</span> * instead of having HBase handle that automatically. The most useful utilities<a name="line.73"></a> +<span class="sourceLineNo">074</span> * are:<a name="line.74"></a> +<span class="sourceLineNo">075</span> * <p><a name="line.75"></a> +<span class="sourceLineNo">076</span> * <ul><a name="line.76"></a> +<span class="sourceLineNo">077</span> * <li>Create a table with a specified number of pre-split regions<a name="line.77"></a> +<span class="sourceLineNo">078</span> * <li>Execute a rolling split of all regions on an existing table<a name="line.78"></a> +<span class="sourceLineNo">079</span> * </ul><a name="line.79"></a> +<span class="sourceLineNo">080</span> * <p><a name="line.80"></a> +<span class="sourceLineNo">081</span> * Both operations can be safely done on a live server.<a name="line.81"></a> +<span class="sourceLineNo">082</span> * <p><a name="line.82"></a> +<span class="sourceLineNo">083</span> * <b>Question:</b> How do I turn off automatic splitting? <br><a name="line.83"></a> +<span class="sourceLineNo">084</span> * <b>Answer:</b> Automatic splitting is determined by the configuration value<a name="line.84"></a> +<span class="sourceLineNo">085</span> * <i>HConstants.HREGION_MAX_FILESIZE</i>. It is not recommended that you set this<a name="line.85"></a> +<span class="sourceLineNo">086</span> * to Long.MAX_VALUE in case you forget about manual splits. A suggested setting<a name="line.86"></a> +<span class="sourceLineNo">087</span> * is 100GB, which would result in &gt; 1hr major compactions if reached.<a name="line.87"></a> +<span class="sourceLineNo">088</span> * <p><a name="line.88"></a> +<span class="sourceLineNo">089</span> * <b>Question:</b> Why did the original authors decide to manually split? <br><a name="line.89"></a> +<span class="sourceLineNo">090</span> * <b>Answer:</b> Specific workload characteristics of our use case allowed us<a name="line.90"></a> +<span class="sourceLineNo">091</span> * to benefit from a manual split system.<a name="line.91"></a> +<span class="sourceLineNo">092</span> * <p><a name="line.92"></a> +<span class="sourceLineNo">093</span> * <ul><a name="line.93"></a> +<span class="sourceLineNo">094</span> * <li>Data (~1k) that would grow instead of being replaced<a name="line.94"></a> +<span class="sourceLineNo">095</span> * <li>Data growth was roughly uniform across all regions<a name="line.95"></a> +<span class="sourceLineNo">096</span> * <li>OLTP workload. Data loss is a big deal.<a name="line.96"></a> +<span class="sourceLineNo">097</span> * </ul><a name="line.97"></a> +<span class="sourceLineNo">098</span> * <p><a name="line.98"></a> +<span class="sourceLineNo">099</span> * <b>Question:</b> Why is manual splitting good for this workload? <br><a name="line.99"></a> +<span class="sourceLineNo">100</span> * <b>Answer:</b> Although automated splitting is not a bad option, there are<a name="line.100"></a> +<span class="sourceLineNo">101</span> * benefits to manual splitting.<a name="line.101"></a> +<span class="sourceLineNo">102</span> * <p><a name="line.102"></a> +<span class="sourceLineNo">103</span> * <ul><a name="line.103"></a> +<span class="sourceLineNo">104</span> * <li>With growing amounts of data, splits will continually be needed. Since<a name="line.104"></a> +<span class="sourceLineNo">105</span> * you always know exactly what regions you have, long-term debugging and<a name="line.105"></a> +<span class="sourceLineNo">106</span> * profiling is much easier with manual splits. It is hard to trace the logs to<a name="line.106"></a> +<span class="sourceLineNo">107</span> * understand region level problems if it keeps splitting and getting renamed.<a name="line.107"></a> +<span class="sourceLineNo">108</span> * <li>Data offlining bugs + unknown number of split regions == oh crap! If an<a name="line.108"></a> +<span class="sourceLineNo">109</span> * WAL or StoreFile was mistakenly unprocessed by HBase due to a weird bug and<a name="line.109"></a> +<span class="sourceLineNo">110</span> * you notice it a day or so later, you can be assured that the regions<a name="line.110"></a> +<span class="sourceLineNo">111</span> * specified in these files are the same as the current regions and you have<a name="line.111"></a> +<span class="sourceLineNo">112</span> * less headaches trying to restore/replay your data.<a name="line.112"></a> +<span class="sourceLineNo">113</span> * <li>You can finely tune your compaction algorithm. With roughly uniform data<a name="line.113"></a> +<span class="sourceLineNo">114</span> * growth, it's easy to cause split / compaction storms as the regions all<a name="line.114"></a> +<span class="sourceLineNo">115</span> * roughly hit the same data size at the same time. With manual splits, you can<a name="line.115"></a> +<span class="sourceLineNo">116</span> * let staggered, time-based major compactions spread out your network IO load.<a name="line.116"></a> +<span class="sourceLineNo">117</span> * </ul><a name="line.117"></a> +<span class="sourceLineNo">118</span> * <p><a name="line.118"></a> +<span class="sourceLineNo">119</span> * <b>Question:</b> What's the optimal number of pre-split regions to create? <br><a name="line.119"></a> +<span class="sourceLineNo">120</span> * <b>Answer:</b> Mileage will vary depending upon your application.<a name="line.120"></a> +<span class="sourceLineNo">121</span> * <p><a name="line.121"></a> +<span class="sourceLineNo">122</span> * The short answer for our application is that we started with 10 pre-split<a name="line.122"></a> +<span class="sourceLineNo">123</span> * regions / server and watched our data growth over time. It's better to err on<a name="line.123"></a> +<span class="sourceLineNo">124</span> * the side of too little regions and rolling split later.<a name="line.124"></a> +<span class="sourceLineNo">125</span> * <p><a name="line.125"></a> +<span class="sourceLineNo">126</span> * The more complicated answer is that this depends upon the largest storefile<a name="line.126"></a> +<span class="sourceLineNo">127</span> * in your region. With a growing data size, this will get larger over time. You<a name="line.127"></a> +<span class="sourceLineNo">128</span> * want the largest region to be just big enough that the<a name="line.128"></a> +<span class="sourceLineNo">129</span> * {@link org.apache.hadoop.hbase.regionserver.HStore} compact<a name="line.129"></a> +<span class="sourceLineNo">130</span> * selection algorithm only compacts it due to a timed major. If you don't, your<a name="line.130"></a> +<span class="sourceLineNo">131</span> * cluster can be prone to compaction storms as the algorithm decides to run<a name="line.131"></a> +<span class="sourceLineNo">132</span> * major compactions on a large series of regions all at once. Note that<a name="line.132"></a> +<span class="sourceLineNo">133</span> * compaction storms are due to the uniform data growth, not the manual split<a name="line.133"></a> +<span class="sourceLineNo">134</span> * decision.<a name="line.134"></a> +<span class="sourceLineNo">135</span> * <p><a name="line.135"></a> +<span class="sourceLineNo">136</span> * If you pre-split your regions too thin, you can increase the major compaction<a name="line.136"></a> +<span class="sourceLineNo">137</span> * interval by configuring HConstants.MAJOR_COMPACTION_PERIOD. If your data size<a name="line.137"></a> +<span class="sourceLineNo">138</span> * grows too large, use this script to perform a network IO safe rolling split<a name="line.138"></a> +<span class="sourceLineNo">139</span> * of all regions.<a name="line.139"></a> +<span class="sourceLineNo">140</span> */<a name="line.140"></a> +<span class="sourceLineNo">141</span>@InterfaceAudience.Private<a name="line.141"></a> +<span class="sourceLineNo">142</span>public class RegionSplitter {<a name="line.142"></a> +<span class="sourceLineNo">143</span> private static final Log LOG = LogFactory.getLog(RegionSplitter.class);<a name="line.143"></a> +<span class="sourceLineNo">144</span><a name="line.144"></a> +<span class="sourceLineNo">145</span> /**<a name="line.145"></a> +<span class="sourceLineNo">146</span> * A generic interface for the RegionSplitter code to use for all it's<a name="line.146"></a> +<span class="sourceLineNo">147</span> * functionality. Note that the original authors of this code use<a name="line.147"></a> +<span class="sourceLineNo">148</span> * {@link HexStringSplit} to partition their table and set it as default, but<a name="line.148"></a> +<span class="sourceLineNo">149</span> * provided this for your custom algorithm. To use, create a new derived class<a name="line.149"></a> +<span class="sourceLineNo">150</span> * from this interface and call {@link RegionSplitter#createPresplitTable} or<a name="line.150"></a> +<span class="sourceLineNo">151</span> * RegionSplitter#rollingSplit(TableName, SplitAlgorithm, Configuration) with the<a name="line.151"></a> +<span class="sourceLineNo">152</span> * argument splitClassName giving the name of your class.<a name="line.152"></a> +<span class="sourceLineNo">153</span> */<a name="line.153"></a> +<span class="sourceLineNo">154</span> public interface SplitAlgorithm {<a name="line.154"></a> +<span class="sourceLineNo">155</span> /**<a name="line.155"></a> +<span class="sourceLineNo">156</span> * Split a pre-existing region into 2 regions.<a name="line.156"></a> +<span class="sourceLineNo">157</span> *<a name="line.157"></a> +<span class="sourceLineNo">158</span> * @param start<a name="line.158"></a> +<span class="sourceLineNo">159</span> * first row (inclusive)<a name="line.159"></a> +<span class="sourceLineNo">160</span> * @param end<a name="line.160"></a> +<span class="sourceLineNo">161</span> * last row (exclusive)<a name="line.161"></a> +<span class="sourceLineNo">162</span> * @return the split row to use<a name="line.162"></a> +<span class="sourceLineNo">163</span> */<a name="line.163"></a> +<span class="sourceLineNo">164</span> byte[] split(byte[] start, byte[] end);<a name="line.164"></a> +<span class="sourceLineNo">165</span><a name="line.165"></a> +<span class="sourceLineNo">166</span> /**<a name="line.166"></a> +<span class="sourceLineNo">167</span> * Split an entire table.<a name="line.167"></a> +<span class="sourceLineNo">168</span> *<a name="line.168"></a> +<span class="sourceLineNo">169</span> * @param numRegions<a name="line.169"></a> +<span class="sourceLineNo">170</span> * number of regions to split the table into<a name="line.170"></a> +<span class="sourceLineNo">171</span> *<a name="line.171"></a> +<span class="sourceLineNo">172</span> * @throws RuntimeException<a name="line.172"></a> +<span class="sourceLineNo">173</span> * user input is validated at this time. may throw a runtime<a name="line.173"></a> +<span class="sourceLineNo">174</span> * exception in response to a parse failure<a name="line.174"></a> +<span class="sourceLineNo">175</span> * @return array of split keys for the initial regions of the table. The<a name="line.175"></a> +<span class="sourceLineNo">176</span> * length of the returned array should be numRegions-1.<a name="line.176"></a> +<span class="sourceLineNo">177</span> */<a name="line.177"></a> +<span class="sourceLineNo">178</span> byte[][] split(int numRegions);<a name="line.178"></a> +<span class="sourceLineNo">179</span><a name="line.179"></a> +<span class="sourceLineNo">180</span> /**<a name="line.180"></a> +<span class="sourceLineNo">181</span> * In HBase, the first row is represented by an empty byte array. This might<a name="line.181"></a> +<span class="sourceLineNo">182</span> * cause problems with your split algorithm or row printing. All your APIs<a name="line.182"></a> +<span class="sourceLineNo">183</span> * will be passed firstRow() instead of empty array.<a name="line.183"></a> +<span class="sourceLineNo">184</span> *<a name="line.184"></a> +<span class="sourceLineNo">185</span> * @return your representation of your first row<a name="line.185"></a> +<span class="sourceLineNo">186</span> */<a name="line.186"></a> +<span class="sourceLineNo">187</span> byte[] firstRow();<a name="line.187"></a> +<span class="sourceLineNo">188</span><a name="line.188"></a> +<span class="sourceLineNo">189</span> /**<a name="line.189"></a> +<span class="sourceLineNo">190</span> * In HBase, the last row is represented by an empty byte array. This might<a name="line.190"></a> +<span class="sourceLineNo">191</span> * cause problems with your split algorithm or row printing. All your APIs<a name="line.191"></a> +<span class="sourceLineNo">192</span> * will be passed firstRow() instead of empty array.<a name="line.192"></a> +<span class="sourceLineNo">193</span> *<a name="line.193"></a> +<span class="sourceLineNo">194</span> * @return your representation of your last row<a name="line.194"></a> +<span class="sourceLineNo">195</span> */<a name="line.195"></a> +<span class="sourceLineNo">196</span> byte[] lastRow();<a name="line.196"></a> +<span class="sourceLineNo">197</span><a name="line.197"></a> +<span class="sourceLineNo">198</span> /**<a name="line.198"></a> +<span class="sourceLineNo">199</span> * In HBase, the last row is represented by an empty byte array. Set this<a name="line.199"></a> +<span class="sourceLineNo">200</span> * value to help the split code understand how to evenly divide the first<a name="line.200"></a> +<span class="sourceLineNo">201</span> * region.<a name="line.201"></a> +<span class="sourceLineNo">202</span> *<a name="line.202"></a> +<span class="sourceLineNo">203</span> * @param userInput<a name="line.203"></a> +<span class="sourceLineNo">204</span> * raw user input (may throw RuntimeException on parse failure)<a name="line.204"></a> +<span class="sourceLineNo">205</span> */<a name="line.205"></a> +<span class="sourceLineNo">206</span> void setFirstRow(String userInput);<a name="line.206"></a> +<span class="sourceLineNo">207</span><a name="line.207"></a> +<span class="sourceLineNo">208</span> /**<a name="line.208"></a> +<span class="sourceLineNo">209</span> * In HBase, the last row is represented by an empty byte array. Set this<a name="line.209"></a> +<span class="sourceLineNo">210</span> * value to help the split code understand how to evenly divide the last<a name="line.210"></a> +<span class="sourceLineNo">211</span> * region. Note that this last row is inclusive for all rows sharing the<a name="line.211"></a> +<span class="sourceLineNo">212</span> * same prefix.<a name="line.212"></a> +<span class="sourceLineNo">213</span> *<a name="line.213"></a> +<span class="sourceLineNo">214</span> * @param userInput<a name="line.214"></a> +<span class="sourceLineNo">215</span> * raw user input (may throw RuntimeException on parse failure)<a name="line.215"></a> +<span class="sourceLineNo">216</span> */<a name="line.216"></a> +<span class="sourceLineNo">217</span> void setLastRow(String userInput);<a name="line.217"></a> +<span class="sourceLineNo">218</span><a name="line.218"></a> +<span class="sourceLineNo">219</span> /**<a name="line.219"></a> +<span class="sourceLineNo">220</span> * @param input<a name="line.220"></a> +<span class="sourceLineNo">221</span> * user or file input for row<a name="line.221"></a> +<span class="sourceLineNo">222</span> * @return byte array representation of this row for HBase<a name="line.222"></a> +<span class="sourceLineNo">223</span> */<a name="line.223"></a> +<span class="sourceLineNo">224</span> byte[] strToRow(String input);<a name="line.224"></a> +<span class="sourceLineNo">225</span><a name="line.225"></a> +<span class="sourceLineNo">226</span> /**<a name="line.226"></a> +<span class="sourceLineNo">227</span> * @param row<a name="line.227"></a> +<span class="sourceLineNo">228</span> * byte array representing a row in HBase<a name="line.228"></a> +<span class="sourceLineNo">229</span> * @return String to use for debug &amp; file printing<a name="line.229"></a> +<span class="sourceLineNo">230</span> */<a name="line.230"></a> +<span class="sourceLineNo">231</span> String rowToStr(byte[] row);<a name="line.231"></a> +<span class="sourceLineNo">232</span><a name="line.232"></a> +<span class="sourceLineNo">233</span> /**<a name="line.233"></a> +<span class="sourceLineNo">234</span> * @return the separator character to use when storing / printing the row<a name="line.234"></a> +<span class="sourceLineNo">235</span> */<a name="line.235"></a> +<span class="sourceLineNo">236</span> String separator();<a name="line.236"></a> +<span class="sourceLineNo">237</span><a name="line.237"></a> +<span class="sourceLineNo">238</span> /**<a name="line.238"></a> +<span class="sourceLineNo">239</span> * Set the first row<a name="line.239"></a> +<span class="sourceLineNo">240</span> * @param userInput byte array of the row key.<a name="line.240"></a> +<span class="sourceLineNo">241</span> */<a name="line.241"></a> +<span class="sourceLineNo">242</span> void setFirstRow(byte[] userInput);<a name="line.242"></a> +<span class="sourceLineNo">243</span><a name="line.243"></a> +<span class="sourceLineNo">244</span> /**<a name="line.244"></a> +<span class="sourceLineNo">245</span> * Set the last row<a name="line.245"></a> +<span class="sourceLineNo">246</span> * @param userInput byte array of the row key.<a name="line.246"></a> +<span class="sourceLineNo">247</span> */<a name="line.247"></a> +<span class="sourceLineNo">248</span> void setLastRow(byte[] userInput);<a name="line.248"></a> +<span class="sourceLineNo">249</span> }<a name="line.249"></a> +<span class="sourceLineNo">250</span><a name="line.250"></a> +<span class="sourceLineNo">251</span> /**<a name="line.251"></a> +<span class="sourceLineNo">252</span> * The main function for the RegionSplitter application. Common uses:<a name="line.252"></a> +<span class="sourceLineNo">253</span> * <p><a name="line.253"></a> +<span class="sourceLineNo">254</span> * <ul><a name="line.254"></a> +<span class="sourceLineNo">255</span> * <li>create a table named 'myTable' with 60 pre-split regions containing 2<a name="line.255"></a> +<span class="sourceLineNo">256</span> * column families 'test' &amp; 'rs', assuming the keys are hex-encoded ASCII:<a name="line.256"></a> +<span class="sourceLineNo">257</span> * <ul><a name="line.257"></a> +<span class="sourceLineNo">258</span> * <li>bin/hbase org.apache.hadoop.hbase.util.RegionSplitter -c 60 -f test:rs<a name="line.258"></a> +<span class="sourceLineNo">259</span> * myTable HexStringSplit<a name="line.259"></a> +<span class="sourceLineNo">260</span> * </ul><a name="line.260"></a> +<span class="sourceLineNo">261</span> * <li>perform a rolling split of 'myTable' (i.e. 60 =&gt; 120 regions), # 2<a name="line.261"></a> +<span class="sourceLineNo">262</span> * outstanding splits at a time, assuming keys are uniformly distributed<a name="line.262"></a> +<span class="sourceLineNo">263</span> * bytes:<a name="line.263"></a> +<span class="sourceLineNo">264</span> * <ul><a name="line.264"></a> +<span class="sourceLineNo">265</span> * <li>bin/hbase org.apache.hadoop.hbase.util.RegionSplitter -r -o 2 myTable<a name="line.265"></a> +<span class="sourceLineNo">266</span> * UniformSplit<a name="line.266"></a> +<span class="sourceLineNo">267</span> * </ul><a name="line.267"></a> <span class="sourceLineNo">268</span> * </ul><a name="line.268"></a> -<span class="sourceLineNo">269</span> * </ul><a name="line.269"></a> -<span class="sourceLineNo">270</span> *<a name="line.270"></a> -<span class="sourceLineNo">271</span> * There are two SplitAlgorithms built into RegionSplitter, HexStringSplit<a name="line.271"></a> -<span class="sourceLineNo">272</span> * and UniformSplit. These are different strategies for choosing region<a name="line.272"></a> -<span class="sourceLineNo">273</span> * boundaries. See their source code for details.<a name="line.273"></a> -<span class="sourceLineNo">274</span> *<a name="line.274"></a> -<span class="sourceLineNo">275</span> * @param args<a name="line.275"></a> -<span class="sourceLineNo">276</span> * Usage: RegionSplitter &lt;TABLE&gt; &lt;SPLITALGORITHM&gt;<a name="line.276"></a> -<span class="sourceLineNo">277</span> * &lt;-c &lt;# regions&gt; -f &lt;family:family:...&gt; | -r<a name="line.277"></a> -<span class="sourceLineNo">278</span> * [-o &lt;# outstanding splits&gt;]&gt;<a name="line.278"></a> -<span class="sourceLineNo">279</span> * [-D &lt;conf.param=value&gt;]<a name="line.279"></a> -<span class="sourceLineNo">280</span> * @throws IOException<a name="line.280"></a> -<span class="sourceLineNo">281</span> * HBase IO problem<a name="line.281"></a> -<span class="sourceLineNo">282</span> * @throws InterruptedException<a name="line.282"></a> -<span class="sourceLineNo">283</span> * user requested exit<a name="line.283"></a> -<span class="sourceLineNo">284</span> * @throws ParseException<a name="line.284"></a> -<span class="sourceLineNo">285</span> * problem parsing user input<a name="line.285"></a> -<span class="sourceLineNo">286</span> */<a name="line.286"></a> -<span class="sourceLineNo">287</span> @SuppressWarnings("static-access")<a name="line.287"></a> -<span class="sourceLineNo">288</span> public static void main(String[] args) throws IOException,<a name="line.288"></a> -<span class="sourceLineNo">289</span> InterruptedException, ParseException {<a name="line.289"></a> -<span class="sourceLineNo">290</span> Configuration conf = HBaseConfiguration.create();<a name="line.290"></a> -<span class="sourceLineNo">291</span><a name="line.291"></a> -<span class="sourceLineNo">292</span> // parse user input<a name="line.292"></a> -<span class="sourceLineNo">293</span> Options opt = new Options();<a name="line.293"></a> -<span class="sourceLineNo">294</span> opt.addOption(OptionBuilder.withArgName("property=value").hasArg()<a name="line.294"></a> -<span class="sourceLineNo">295</span> .withDescription("Override HBase Configuration Settings").create("D"));<a name="line.295"></a> -<span class="sourceLineNo">296</span> opt.addOption(OptionBuilder.withArgName("region count").hasArg()<a name="line.296"></a> -<span class="sourceLineNo">297</span> .withDescription(<a name="line.297"></a> -<span class="sourceLineNo">298</span> "Create a new table with a pre-split number of regions")<a name="line.298"></a> -<span class="sourceLineNo">299</span> .create("c"));<a name="line.299"></a> -<span class="sourceLineNo">300</span> opt.addOption(OptionBuilder.withArgName("family:family:...").hasArg()<a name="line.300"></a> -<span class="sourceLineNo">301</span> .withDescription(<a name="line.301"></a> -<span class="sourceLineNo">302</span> "Column Families to create with new table. Required with -c")<a name="line.302"></a> -<span class="sourceLineNo">303</span> .create("f"));<a name="line.303"></a> -<span class="sourceLineNo">304</span> opt.addOption("h", false, "Print this usage help");<a name="line.304"></a> -<span class="sourceLineNo">305</span> opt.addOption("r", false, "Perform a rolling split of an existing region");<a name="line.305"></a> -<span class="sourceLineNo">306</span> opt.addOption(OptionBuilder.withArgName("count").hasArg().withDescription(<a name="line.306"></a> -<span class="sourceLineNo">307</span> "Max outstanding splits that have unfinished major compactions")<a name="line.307"></a> -<span class="sourceLineNo">308</span> .create("o"));<a name="line.308"></a> -<span class="sourceLineNo">309</span> opt.addOption(null, "firstrow", true,<a name="line.309"></a> -<span class="sourceLineNo">310</span> "First Row in Table for Split Algorithm");<a name="line.310"></a> -<span class="sourceLineNo">311</span> opt.addOption(null, "lastrow", true,<a name="line.311"></a> -<span class="sourceLineNo">312</span> "Last Row in Table for Split Algorithm");<a name="line.312"></a> -<span class="sourceLineNo">313</span> opt.addOption(null, "risky", false,<a name="line.313"></a> -<span class="sourceLineNo">314</span> "Skip verification steps to complete quickly."<a name="line.314"></a> -<span class="sourceLineNo">315</span> + "STRONGLY DISCOURAGED for production systems. ");<a name="line.315"></a> -<span class="sourceLineNo">316</span> CommandLine cmd = new GnuParser().parse(opt, args);<a name="line.316"></a> -<span class="sourceLineNo">317</span><a name="line.317"></a> -<span class="sourceLineNo">318</span> if (cmd.hasOption("D")) {<a name="line.318"></a> -<span class="sourceLineNo">319</span> for (String confOpt : cmd.getOptionValues("D")) {<a name="line.319"></a> -<span class="sourceLineNo">320</span> String[] kv = confOpt.split("=", 2);<a name="line.320"></a> -<span class="sourceLineNo">321</span> if (kv.length == 2) {<a name="line.321"></a> -<span class="sourceLineNo">322</span> conf.set(kv[0], kv[1]);<a name="line.322"></a> -<span class="sourceLineNo">323</span> LOG.debug("-D configuration override: " + kv[0] + "=" + kv[1]);<a name="line.323"></a> -<span class="sourceLineNo">324</span> } else {<a name="line.324"></a> -<span class="sourceLineNo">325</span> throw new ParseException("-D option format invalid: " + confOpt);<a name="line.325"></a> -<span class="sourceLineNo">326</span> }<a name="line.326"></a> -<span class="sourceLineNo">327</span> }<a name="line.327"></a> -<span class="sourceLineNo">328</span> }<a name="line.328"></a> -<span class="sourceLineNo">329</span><a name="line.329"></a> -<span class="sourceLineNo">330</span> if (cmd.hasOption("risky")) {<a name="line.330"></a> -<span class="sourceLineNo">331</span> conf.setBoolean("split.verify", false);<a name="line.331"></a> -<span class="sourceLineNo">332</span> }<a name="line.332"></a> -<span class="sourceLineNo">333</span><a name="line.333"></a> -<span class="sourceLineNo">334</span> boolean createTable = cmd.hasOption("c") && cmd.hasOption("f");<a name="line.334"></a> -<span class="sourceLineNo">335</span> boolean rollingSplit = cmd.hasOption("r");<a name="line.335"></a> -<span class="sourceLineNo">336</span> boolean oneOperOnly = createTable ^ rollingSplit;<a name="line.336"></a> -<span class="sourceLineNo">337</span><a name="line.337"></a> -<span class="sourceLineNo">338</span> if (2 != cmd.getArgList().size() || !oneOperOnly || cmd.hasOption("h")) {<a name="line.338"></a> -<span class="sourceLineNo">339</span> new HelpFormatter().printHelp("RegionSplitter <TABLE> <SPLITALGORITHM>\n"+<a name="line.339"></a> -<span class="sourceLineNo">340</span> "SPLITALGORITHM is a java class name of a class implementing " +<a name="line.340"></a> -<span class="sourceLineNo">341</span> "SplitAlgorithm, or one of the special strings HexStringSplit " +<a name="line.341"></a> -<span class="sourceLineNo">342</span> "or UniformSplit, which are built-in split algorithms. " +<a name="line.342"></a> -<span class="sourceLineNo">343</span> "HexStringSplit treats keys as hexadecimal ASCII, and " +<a name="line.343"></a> -<span class="sourceLineNo">344</span> "UniformSplit treats keys as arbitrary bytes.", opt);<a name="line.344"></a> -<span class="sourceLineNo">345</span> return;<a name="line.345"></a> -<span class="sourceLineNo">346</span> }<a name="line.346"></a> -<span class="sourceLineNo">347</span> TableName tableName = TableName.valueOf(cmd.getArgs()[0]);<a name="line.347"></a> -<span class="sourceLineNo">348</span> String splitClass = cmd.getArgs()[1];<a name="line.348"></a> -<span class="sourceLineNo">349</span> SplitAlgorithm splitAlgo = newSplitAlgoInstance(conf, splitClass);<a name="line.349"></a> -<span class="sourceLineNo">350</span><a name="line.350"></a> -<span class="sourceLineNo">351</span> if (cmd.hasOption("firstrow")) {<a name="line.351"></a> -<span class="sourceLineNo">352</span> splitAlgo.setFirstRow(cmd.getOptionValue("firstrow"));<a name="line.352"></a> -<span class="sourceLineNo">353</span> }<a name="line.353"></a> -<span class="sourceLineNo">354</span> if (cmd.hasOption("lastrow")) {<a name="line.354"></a> -<span class="sourceLineNo">355</span> splitAlgo.setLastRow(cmd.getOptionValue("lastrow"));<a name="line.355"></a> -<span class="sourceLineNo">356</span> }<a name="line.356"></a> -<span class="sourceLineNo">357</span><a name="line.357"></a> -<span class="sourceLineNo">358</span> if (createTable) {<a name="line.358"></a> -<span class="sourceLineNo">359</span> conf.set("split.count", cmd.getOptionValue("c"));<a name="line.359"></a> -<span class="sourceLineNo">360</span> createPresplitTable(tableName, splitAlgo, cmd.getOptionValue("f").split(":"), conf);<a name="line.360"></a> -<span class="sourceLineNo">361</span> }<a name="line.361"></a> -<span class="sourceLineNo">362</span><a name="line.362"></a> -<span class="sourceLineNo">363</span> if (rollingSplit) {<a name="line.363"></a> -<span class="sourceLineNo">364</span> if (cmd.hasOption("o")) {<a name="line.364"></a> -<span class="sourceLineNo">365</span> conf.set("split.outstanding", cmd.getOptionValue("o"));<a name="line.365"></a> -<span class="sourceLineNo">366</span> }<a name="line.366"></a> -<span class="sourceLineNo">367</span> rollingSplit(tableName, splitAlgo, conf);<a name="line.367"></a> -<span class="sourceLineNo">368</span> }<a name="line.368"></a> -<span class="sourceLineNo">369</span> }<a name="line.369"></a> -<span class="sourceLineNo">370</span><a name="line.370"></a> -<span class="sourceLineNo">371</span> static void createPresplitTable(TableName tableName, SplitAlgorithm splitAlgo,<a name="line.371"></a> -<span class="sourceLineNo">372</span> String[] columnFamilies, Configuration conf)<a name="line.372"></a> -<span class="sourceLineNo">373</span> throws IOException, InterruptedException {<a name="line.373"></a> -<span class="sourceLineNo">374</span> final int splitCount = conf.getInt("split.count", 0);<a name="line.374"></a> -<span class="sourceLineNo">375</span> Preconditions.checkArgument(splitCount > 1, "Split count must be > 1");<a name="line.375"></a> -<span class="sourceLineNo">376</span><a name="line.376"></a> -<span class="sourceLineNo">377</span> Preconditions.checkArgument(columnFamilies.length > 0,<a name="line.377"></a> -<span class="sourceLineNo">378</span> "Must specify at least one column family. ");<a name="line.378"></a> -<span class="sourceLineNo">379</span> LOG.debug("Creating table " + tableName + " with " + columnFamilies.length<a name="line.379"></a> -<span class="sourceLineNo">380</span> + " column families. Presplitting to " + splitCount + " regions");<a name="line.380"></a> -<span class="sourceLineNo">381</span><a name="line.381"></a> -<span class="sourceLineNo">382</span> HTableDescriptor desc = new HTableDescriptor(tableName);<a name="line.382"></a> -<span class="sourceLineNo">383</span> for (String cf : columnFamilies) {<a name="line.383"></a> -<span class="sourceLineNo">384</span> desc.addFamily(new HColumnDescriptor(Bytes.toBytes(cf)));<a name="line.384"></a> -<span class="sourceLineNo">385</span> }<a name="line.385"></a> -<span class="sourceLineNo">386</span> try (Connection connection = ConnectionFactory.createConnection(conf)) {<a name="line.386"></a> -<span class="sourceLineNo">387</span> Admin admin = connection.getAdmin();<a name="line.387"></a> -<span class="sourceLineNo">388</span> try {<a name="line.388"></a> -<span class="sourceLineNo">389</span> Preconditions.checkArgument(!admin.tableExists(tableName),<a name="line.389"></a> -<span class="sourceLineNo">390</span> "Table already exists: " + tableName);<a name="line.390"></a> -<span class="sourceLineNo">391</span> admin.createTable(desc, splitAlgo.split(splitCount));<a name="line.391"></a> -<span class="sourceLineNo">392</span> } finally {<a name="line.392"></a> -<span class="sourceLineNo">393</span> admin.close();<a name="line.393"></a> -<span class="sourceLineNo">394</span> }<a name="line.394"></a> -<span class="sourceLineNo">395</span> LOG.debug("Table created! Waiting for regions to show online in META...");<a name="line.395"></a> -<span class="sourceLineNo">396</span> if (!conf.getBoolean("split.verify", true)) {<a name="line.396"></a> -<span class="sourceLineNo">397</span> // NOTE: createTable is synchronous on the table, but not on the regions<a name="line.397"></a> -<span class="sourceLineNo">398</span> int onlineRegions = 0;<a name="line.398"></a> -<span class="sourceLineNo">399</span> while (onlineRegions < splitCount) {<a name="line.399"></a> -<span class="sourceLineNo">400</span> onlineRegions = MetaTableAccessor.getRegionCount(connection, tableName);<a name="line.400"></a> -<span class="sourceLineNo">401</span> LOG.debug(onlineRegions + " of " + splitCount + " regions online...");<a name="line.401"></a> -<span class="sourceLineNo">402</span> if (onlineRegions < splitCount) {<a name="line.402"></a> -<span class="sourceLineNo">403</span> Thread.sleep(10 * 1000); // sleep<a name="line.403"></a> -<span class="sourceLineNo">404</span> }<a name="line.404"></a> -<span class="sourceLineNo">405</span> }<a name="line.405"></a> -<span class="sourceLineNo">406</span> }<a name="line.406"></a> -<span class="sourceLineNo">407</span> LOG.debug("Finished creating table with " + splitCount + " regions");<a name="line.407"></a> -<span class="sourceLineNo">408</span> }<a name="line.408"></a> -<span class="sourceLineNo">409</span> }<a name="line.409"></a> -<span class="sourceLineNo">410</span><a name="line.410"></a> -<span class="sourceLineNo">411</span> /**<a name="line.411"></a> -<span class="sourceLineNo">412</span> * Alternative getCurrentNrHRS which is no longer available.<a name="line.412"></a> -<span class="sourceLineNo">413</span> * @param connection<a name="line.413"></a> -<span class="sourceLineNo">414</span> * @return Rough count of regionservers out on cluster.<a name="line.414"></a> -<span class="sourceLineNo">415</span> * @throws IOException <a name="line.415"></a> -<span class="sourceLineNo">416</span> */<a name="line.416"></a> -<span class="sourceLineNo">417</span> private static int getRegionServerCount(final Connection connection) throws IOException {<a name="line.417"></a> -<span class="sourceLineNo">418</span> try (Admin admin = connection.getAdmin()) {<a name="line.418"></a> -<span class="sourceLineNo">419</span> ClusterStatus status = admin.getClusterStatus();<a name="line.419"></a> -<span class="sourceLineNo">420</span> Collection<ServerName> servers = status.getServers();<a name="line.420"></a> -<span class="sourceLineNo">421</span> return servers == null || servers.isEmpty()? 0: servers.size();<a name="line.421"></a> -<span class="sourceLineNo">422</span> }<a name="line.422"></a> -<span class="sourceLineNo">423</span> }<a name="line.423"></a> -<span class="sourceLineNo">424</span><a name="line.424"></a> -<span class="sourceLineNo">425</span> private static byte [] readFile(final FileSystem fs, final Path path) throws IOException {<a name="line.425"></a> -<span class="sourceLineNo">426</span> FSDataInputStream tmpIn = fs.open(path);<a name="line.426"></a> -<span class="sourceLineNo">427</span> try {<a name="line.427"></a> -<span class="sourceLineNo">428</span> byte [] rawData = new byte[tmpIn.available()];<a name="line.428"></a> -<span class="sourceLineNo">429</span> tmpIn.readFully(rawData);<a name="line.429"></a> -<span class="sourceLineNo">430</span> return rawData;<a name="line.430"></a> -<span class="sourceLineNo">431</span> } finally {<a name="line.431"></a> -<span class="sourceLineNo">432</span> tmpIn.close();<a name="line.432"></a> -<span class="sourceLineNo">433</span> }<a name="line.433"></a> -<span class="sourceLineNo">434</span> }<a name="line.434"></a> -<span class="sourceLineNo">435</span><a name="line.435"></a> -<span class="sourceLineNo">436</span> static void rollingSplit(TableName tableName, SplitAlgorithm splitAlgo, Configuration conf)<a name="line.436"></a> -<span class="sourceLineNo">437</span> throws IOException, InterruptedException {<a name="line.437"></a> -<span class="sourceLineNo">438</span> final int minOS = conf.getInt("split.outstanding", 2);<a name="line.438"></a> -<span class="sourceLineNo">439</span> try (Connection connection = ConnectionFactory.createConnection(conf)) {<a name="line.439"></a> -<span class="sourceLineNo">440</span> // Max outstanding splits. default == 50% of servers<a name="line.440"></a> -<span class="sourceLineNo">441</span> final int MAX_OUTSTANDING = Math.max(getRegionServerCount(connection) / 2, minOS);<a name="line.441"></a> -<span class="sourceLineNo">442</span><a name="line.442"></a> -<span class="sourceLineNo">443</span> Path hbDir = FSUtils.getRootDir(conf);<a name="line.443"></a> -<span class="sourceLineNo">444</span> Path tableDir = FSUtils.getTableDir(hbDir, tableName);<a name="line.444"></a> -<span class="sourceLineNo">445</span> Path splitFile = new Path(tableDir, "_balancedSplit");<a name="line.445"></a> -<span class="sourceLineNo">446</span> FileSystem fs = FileSystem.get(conf);<a name="line.446"></a> -<span class="sourceLineNo">447</span><a name="line.447"></a> -<span class="sourceLineNo">448</span> // Get a list of daughter regions to create<a name="line.448"></a> -<span class="sourceLineNo">449</span> LinkedList<Pair<byte[], byte[]>> tmpRegionSet = null;<a name="line.449"></a> -<span class="sourceLineNo">450</span> try (Table table = connection.getTable(tableName)) {<a name="line.450"></a> -<span class="sourceLineNo">451</span> tmpRegionSet = getSplits(connection, tableName, splitAlgo);<a name="line.451"></a> -<span class="sourceLineNo">452</span> }<a name="line.452"></a> -<span class="sourceLineNo">453</span> LinkedList<Pair<byte[], byte[]>> outstanding = Lists.newLinkedList();<a name="line.453"></a> -<span class="sourceLineNo">454</span> int splitCount = 0;<a name="line.454"></a> -<span class="sourceLineNo">455</span> final int origCount = tmpRegionSet.size();<a name="line.455"></a> -<span class="sourceLineNo">456</span><a name="line.456"></a> -<span class="sourceLineNo">457</span> // all splits must compact & we have 1 compact thread, so 2 split<a name="line.457"></a> -<span class="sourceLineNo">458</span> // requests to the same RS can stall the outstanding split queue.<a name="line.458"></a> -<span class="sourceLineNo">459</span> // To fix, group the regions into an RS pool and round-robin through it<a name="line.459"></a> -<span class="sourceLineNo">460</span> LOG.debug("Bucketing regions by regionserver...");<a name="line.460"></a> -<span class="sourceLineNo">461</span> TreeMap<String, LinkedList<Pair<byte[], byte[]>>> daughterRegions =<a name="line.461"></a> -<span class="sourceLineNo">462</span> Maps.newTreeMap();<a name="line.462"></a> -<span class="sourceLineNo">463</span> // Get a regionLocator. Need it in below.<a name="line.463"></a> -<span class="sourceLineNo">464</span> try (RegionLocator regionLocator = connection.getRegionLocator(tableName)) {<a name="line.464"></a> -<span class="sourceLineNo">465</span> for (Pair<byte[], byte[]> dr : tmpRegionSet) {<a name="line.465"></a> -<span class="sourceLineNo">466</span> String rsLocation = regionLocator.getRegionLocation(dr.getSecond()).getHostnamePort();<a name="line.466"></a> -<span class="sourceLineNo">467</span> if (!daughterRegions.containsKey(rsLocation)) {<a name="line.467"></a> -<span class="sourceLineNo">468</span> LinkedList<Pair<byte[], byte[]>> entry = Lists.newLinkedList();<a name="line.468"></a> -<span class="sourceLineNo">469</span> daughterRegions.put(rsLocation, entry);<a name="line.469"></a> -<span class="sourceLineNo">470</span> }<a name="line.470"></a> -<span class="sourceLineNo">471</span> daughterRegions.get(rsLocation).add(dr);<a name="line.471"></a> -<span class="sourceLineNo">472</span> }<a name="line.472"></a> -<span class="sourceLineNo">473</span> LOG.debug("Done with bucketing. Split time!");<a name="line.473"></a> -<span class="sourceLineNo">474</span> long startTime = System.currentTimeMillis();<a name="line.474"></a> -<span class="sourceLineNo">475</span><a name="line.475"></a> -<span class="sourceLineNo">476</span> // Open the split file and modify it as splits finish<a name="line.476"></a> -<span class="sourceLineNo">477</span> byte[] rawData = readFile(fs, splitFile);<a name="line.477"></a> -<span class="sourceLineNo">478</span><a name="line.478"></a> -<span class="sourceLineNo">479</span> FSDataOutputStream splitOut = fs.create(splitFile);<a name="line.479"></a> -<span class="sourceLineNo">480</span> try {<a name="line.480"></a> -<span class="sourceLineNo">481</span> splitOut.write(rawData);<a name="line.481"></a> -<span class="sourceLineNo">482</span><a name="line.482"></a> -<span class="sourceLineNo">483</span> try {<a name="line.483"></a> -<span class="sourceLineNo">484</span> // *** split code ***<a name="line.484"></a> -<span class="sourceLineNo">485</span> while (!daughterRegions.isEmpty()) {<a name="line.485"></a> -<span class="sourceLineNo">486</span> LOG.debug(daughterRegions.size() + " RS have regions to splt.");<a name="line.486"></a> -<span class="sourceLineNo">487</span><a name="line.487"></a> -<span class="sourceLineNo">488</span> // Get ServerName to region count mapping<a name="line.488"></a> -<span class="sourceLineNo">489</span> final TreeMap<ServerName, Integer> rsSizes = Maps.newTreeMap();<a name="line.489"></a> -<span class="sourceLineNo">490</span> List<HRegionLocation> hrls = regionLocator.getAllRegionLocations();<a name="line.490"></a> -<span class="sourceLineNo">491</span> for (HRegionLocation hrl: hrls) {<a name="line.491"></a> -<span class="sourceLineNo">492</span> ServerName sn = hrl.getServerName();<a name="line.492"></a> -<span class="sourceLineNo">493</span> if (rsSizes.containsKey(sn)) {<a name="line.493"></a> -<span class="sourceLineNo">494</span> rsSizes.put(sn, rsSizes.get(sn) + 1);<a name="line.494"></a> -<span class="sourceLineNo">495</span> } else {<a name="line.495"></a> -<span class="sourceLineNo">496</span> rsSizes.put(sn, 1);<a name="line.496"></a> -<span class="sourceLineNo">497</span> }<a name="line.497"></a> -<span class="sourceLineNo">498</span> }<a name="line.498"></a> -<span class="sourceLineNo">499</span><a name="line.499"></a> -<span class="sourceLineNo">500</span> // Sort the ServerNames by the number of regions they have<a name="line.500"></a> -<span class="sourceLineNo">501</span> List<String> serversLeft = Lists.newArrayList(daughterRegions .keySet());<a name="line.501"></a> -<span class="sourceLineNo">502</span> Collections.sort(serversLeft, new Comparator<String>() {<a name="line.502"></a> -<span class="sourceLineNo">503</span> public int compare(String o1, String o2) {<a name="line.503"></a> -<span class="sourceLineNo">504</span> return rsSizes.get(o1).compareTo(rsSizes.get(o2));<a name="line.504"></a> -<span class="sourceLineNo">505</span> }<a name="line.505"></a> -<span class="sourceLineNo">506</span> });<a name="line.506"></a> -<span class="sourceLineNo">507</span><a name="line.507"></a> -<span class="sourceLineNo">508</span> // Round-robin through the ServerName list. Choose the lightest-loaded servers<a name="line.508"></a> -<span class="sourceLineNo">509</span> // first to keep the master from load-balancing regions as we split.<a name="line.509"></a> -<span class="sourceLineNo">510</span> for (String rsLoc : serversLeft) {<a name="line.510"></a> -<span class="sourceLineNo">511</span> Pair<byte[], byte[]> dr = null;<a name="line.511"></a> -<span class="sourceLineNo">512</span><a name="line.512"></a> -<span class="sourceLineNo">513</span> // Find a region in the ServerName list that hasn't been moved<a name="line.513"></a> -<span class="sourceLineNo">514</span> LOG.debug("Finding a region on " + rsLoc);<a name="line.514"></a> -<span class="sourceLineNo">515</span> LinkedList<Pair<byte[], byte[]>> regionList = daughterRegions.get(rsLoc);<a name="line.515"></a> -<span class="sourceLineNo">516</span> while (!regionList.isEmpty()) {<a name="line.516"></a> -<span class="sourceLineNo">517</span> dr = regionList.pop();<a name="line.517"></a> -<span class="sourceLineNo">518</span><a name="line.518"></a> -<span class="sourceLineNo">519</span> // get current region info<a name="line.519"></a> -<span class="sourceLineNo">520</span> byte[] split = dr.getSecond();<a name="line.520"></a> -<span class="sourceLineNo">521</span> HRegionLocation regionLoc = regionLocator.getRegionLocation(split);<a name="line.521"></a> -<span class="sourceLineNo">522</span><a name="line.522"></a> -<span class="sourceLineNo">523</span> // if this region moved locations<a name="line.523"></a> -<span class="sourceLineNo">524</span> String newRs = regionLoc.getHostnamePort();<a name="line.524"></a> -<span class="sourceLineNo">525</span> if (newRs.compareTo(rsLoc) != 0) {<a name="line.525"></a> -<span class="sourceLineNo">526</span> LOG.debug("Region with " + splitAlgo.rowToStr(split)<a name="line.526"></a> -<span class="sourceLineNo">527</span> + " moved to " + newRs + ". Relocating...");<a name="line.527"></a> -<span class="sourceLineNo">528</span> // relocate it, don't use it right now<a name="line.528"></a> -<span class="sourceLineNo">529</span> if (!daughterRegions.containsKey(newRs)) {<a name="line.529"></a> -<span class="sourceLineNo">530</span> LinkedList<Pair<byte[], byte[]>> entry = Lists.newLinkedList();<a name="line.530"></a> -<span class="sourceLineNo">531</span> daughterRegions.put(newRs, entry);<a name="line.531"></a> -<span class="sourceLineNo">532</span> }<a name="line.532"></a> -<span class="sourceLineNo">533</span> daughterRegions.get(newRs).add(dr);<a name="line.533"></a> -<span class="sourceLineNo">534</span> dr = null;<a name="line.534"></a> -<span class="sourceLineNo">535</span> continue;<a name="line.535"></a> -<span class="sourceLineNo">536</span> }<a name="line.536"></a> -<span class="sourceLineNo">537</span><a name="line.537"></a> -<span class="sourceLineNo">538</span> // make sure this region wasn't already split<a name="line.538"></a> -<span class="sourceLineNo">539</span> byte[] sk = regionLoc.getRegionInfo().getStartKey();<a name="line.539"></a> -<span class="sourceLineNo">540</span> if (sk.length != 0) {<a name="line.540"></a> -<span class="sourceLineNo">541</span> if (Bytes.equals(split, sk)) {<a name="line.541"></a> -<span class="sourceLineNo">542</span> LOG.debug("Region already split on "<a name="line.542"></a> -<span class="sourceLineNo">543</span> + splitAlgo.rowToStr(split) + ". Skipping this region...");<a name="line.543"></a> -<span class="sourceLineNo">544</span> ++splitCount;<a name="line.544"></a> -<span class="sourceLineNo">545</span> dr = null;<a name="line.545"></a> -<span class="sourceLineNo">546</span> continue;<a name="line.546"></a> -<span class="sourceLineNo">547</span> }<a name="line.547"></a> -<span class="sourceLineNo">548</span> byte[] start = dr.getFirst();<a name="line.548"></a> -<span class="sourceLineNo">549</span> Preconditions.checkArgument(Bytes.equals(start, sk), splitAlgo<a name="line.549"></a> -<span class="sourceLineNo">550</span> .rowToStr(start) + " != " + splitAlgo.rowToStr(sk));<a name="line.550"></a> -<span class="sourceLineNo">551</span> }<a name="line.551"></a> -<span class="sourceLineNo">552</span><a name="line.552"></a> -<span class="sourceLineNo">553</span> // passed all checks! found a good region<a name="line.553"></a> -<span class="sourceLineNo">554</span> break;<a name="line.554"></a> -<span class="sourceLineNo">555</span> }<a name="line.555"></a> -<span class="sourceLineNo">556</span> if (regionList.isEmpty()) {<a name="line.556"></a> -<span class="sourceLineNo">557</span> daughterRegions.remove(rsLoc);<a name="line.557"></a> -<span class="sourceLineNo">558</span> }<a name="line.558"></a> -<span class="sourceLineNo">559</span> if (dr == null)<a name="line.559"></a> -<span class="sourceLineNo">560</span> continue;<a name="line.560"></a> +<span class="sourceLineNo">269</span> *<a name="line.269"></a> +<span class="sourceLineNo">270</span> * There are two SplitAlgorithms built into RegionSplitter, HexStringSplit<a name="line.270"></a> +<span class="sourceLineNo">271</span> * and UniformSplit. These are different strategies for choosing region<a name="line.271"></a> +<span class="sourceLineNo">272</span> * boundaries. See their source code for details.<a name="line.272"></a> +<span class="sourceLineNo">273</span> *<a name="line.273"></a> +<span class="sourceLineNo">274</span> * @param args<a name="line.274"></a> +<span class="sourceLineNo">275</span> * Usage: RegionSplitter &lt;TABLE&gt; &lt;SPLITALGORITHM&gt;<a name="line.275"></a> +<span class="sourceLineNo">276</span> * &lt;-c &lt;# regions&gt; -f &lt;family:family:...&gt; | -r<a name="line.276"></a> +<span class="sourceLineNo">277</span> * [-o &lt;# outstanding splits&gt;]&gt;<a name="line.277"></a> +<span class="sourceLineNo">278</span> * [-D &lt;conf.param=value&gt;]<a name="line.278"></a> +<span class="sourceLineNo">279</span> * @throws IOException<a name="line.279"></a> +<span class="sourceLineNo">280</span> * HBase IO problem<a name="line.280"></a> +<span class="sourceLineNo">281</span> * @throws InterruptedException<a name="line.281"></a> +<span class="sourceLineNo">282</span> * user requested exit<a name="line.282"></a> +<span class="sourceLineNo">283</span> * @throws ParseException<a name="line.283"></a> +<span class="sourceLineNo">284</span> * problem parsing user input<a name="line.284"></a> +<span class="sourceLineNo">285</span> */<a name="line.285"></a> +<span class="sourceLineNo">286</span> @SuppressWarnings("static-access")<a name="line.286"></a> +<span class="sourceLineNo">287</span> public static void main(String[] args) throws IOException,<a name="line.287"></a> +<span class="sourceLineNo">288</span> InterruptedException, ParseException {<a name="line.288"></a> +<span class="sourceLineNo">289</span> Configuration conf = HBaseConfiguration.create();<a name="line.289"></a> +<span class="sourceLineNo">290</span><a name="line.290"></a> +<span class="sourceLineNo">291</span> // parse user input<a name="line.291"></a> +<span class="sourceLineNo">292</span> Options opt = new Options();<a name="line.292"></a> +<span class="sourceLineNo">293</span> opt.addOption(OptionBuilder.withArgName("property=value").hasArg()<a name="line.293"></a> +<span class="sourceLineNo">294</span> .withDescription("Override HBase Configuration Settings").create("D"));<a name="line.294"></a> +<span class="sourceLineNo">295</span> opt.addOption(OptionBuilder.withArgName("region count").hasArg()<a name="line.295"></a> +<span class="sourceLineNo">296</span> .withDescription(<a name="line.296"></a> +<span class="sourceLineNo">297</span> "Create a new table with a pre-split number of regions")<a name="line.297"></a> +<span class="sourceLineNo">298</span> .create("c"));<a name="line.298"></a> +<span class="sourceLineNo">299</span> opt.addOption(OptionBuilder.withArgName("family:family:...").hasArg()<a name="line.299"></a> +<span class="sourceLineNo">300</span> .withDescription(<a name="line.300"></a> +<span class="sourceLineNo">301</span> "Column Families to create with new table. Required with -c")<a name="line.301"></a> +<span class="sourceLineNo">302</span> .create("f"));<a name="line.302"></a> +<span class="sourceLineNo">303</span> opt.addOption("h", false, "Print this usage help");<a name="line.303"></a> +<span class="sourceLineNo">304</span> opt.addOption("r", false, "Perform a rolling split of an existing region");<a name="line.304"></a> +<span class="sourceLineNo">305</span> opt.addOption(OptionBuilder.withArgName("count").hasArg().withDescription(<a name="line.305"></a> +<span class="sourceLineNo">306</span> "Max outstanding splits that have unfinished major compactions")<a name="line.306"></a> +<span class="sourceLineNo">307</span> .create("o"));<a name="line.307"></a> +<span class="sourceLineNo">308</span> opt.addOption(null, "firstrow", true,<a name="line.308"></a> +<span class="sourceLineNo">309</span> "First Row in Table for Split Algorithm");<a name="line.309"></a> +<span class="sourceLineNo">310</span> opt.addOption(null, "lastrow", true,<a name="line.310"></a> +<span class="sourceLineNo">311</span> "Last Row in Table for Split Algorithm");<a name="line.311"></a> +<span class="sourceLineNo">312</span> opt.addOption(null, "risky", false,<a name="line.312"></a> +<span class="sourceLineNo">313</span> "Skip verification steps to complete quickly."<a name="line.313"></a> +<span class="sourceLineNo">314</span> + "STRONGLY DISCOURAGED for production systems. ");<a name="line.314"></a> +<span class="sourceLineNo">315</span> CommandLine cmd = new GnuParser().parse(opt, args);<a name="line.315"></a> +<span class="sourceLineNo">316</span><a name="line.316"></a> +<span class="sourceLineNo">317</span> if (cmd.hasOption("D")) {<a name="line.317"></a> +<span class="sourceLineNo">318</span> for (String confOpt : cmd.getOptionValues("D")) {<a name="line.318"></a> +<span class="sourceLineNo">319</span> String[] kv = confOpt.split("=", 2);<a name="line.319"></a> +<span class="sourceLineNo">320</span> if (kv.length == 2) {<a name="line.320"></a> +<span class="sourceLineNo">321</span> conf.set(kv[0], kv[1]);<a name="line.321"></a> +<span class="sourceLineNo">322</span> LOG.debug("-D configuration override: " + kv[0] + "=" + kv[1]);<a name="line.322"></a> +<span class="sourceLineNo">323</span> } else {<a name="line.323"></a> +<span class="sourceLineNo">324</span> throw new ParseException("-D option format invalid: " + confOpt);<a name="line.324"></a> +<span class="sourceLineNo">325</span> }<a name="line.325"></a> +<span class="sourceLineNo">326</span> }<a name="line.326"></a> +<span class="sourceLineNo">327</span> }<a name="line.327"></a> +<span class="sourceLineNo">328</span><a name="line.328"></a> +<span class="sourceLineNo">329</span> if (cmd.hasOption("risky")) {<a name="line.329"></a> +<span class="sourceLineNo">330</span> conf.setBoolean("split.verify", false);<a name="line.330"></a> +<span class="sourceLineNo">331</span> }<a name="line.331"></a> +<span class="sourceLineNo">332</span><a name="line.332"></a> +<span class="sourceLineNo">333</span> boolean createTable = cmd.hasOption("c") && cmd.hasOption("f");<a name="line.333"></a> +<span class="sourceLineNo">334</span> boolean rollingSplit = cmd.hasOption("r");<a name="line.334"></a> +<span class="sourceLineNo">335</span> boolean oneOperOnly = createTable ^ rollingSplit;<a name="line.335"></a> +<span class="sourceLineNo">336</span><a name="line.336"></a> +<span class="sourceLineNo">337</span> if (2 != cmd.getArgList().size() || !oneOperOnly || cmd.hasOption("h")) {<a name="line.337"></a> +<span class="sourceLineNo">338</span> new HelpFormatter().printHelp("RegionSplitter <TABLE> <SPLITALGORITHM>\n"+<a name="line.338"></a> +<span class="sourceLineNo">339</span> "SPLITALGORITHM is a java class name of a class implementing " +<a name="line.339"></a> +<span class="sourceLineNo">340</span> "SplitAlgorithm, or one of the special strings HexStringSplit " +<a name="line.340"></a> +<span class="sourceLineNo">341</span> "or UniformSplit, which are built-in split algorithms. " +<a name="line.341"></a> +<span class="sourceLineNo">342</span> "HexStringSplit treats keys as hexadeci
<TRUNCATED>
