hbase-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From mi...@apache.org
Subject [09/15] hbase-site git commit: Published site at 8eedc967515a4d9133068962fe029160d24e6f95.
Date Fri, 19 Feb 2016 19:26:59 GMT
http://git-wip-us.apache.org/repos/asf/hbase-site/blob/f564c619/devapidocs/src-html/org/apache/hadoop/hbase/master/AssignmentManager.html
----------------------------------------------------------------------
diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/master/AssignmentManager.html b/devapidocs/src-html/org/apache/hadoop/hbase/master/AssignmentManager.html
index addaeb1..96c4e73 100644
--- a/devapidocs/src-html/org/apache/hadoop/hbase/master/AssignmentManager.html
+++ b/devapidocs/src-html/org/apache/hadoop/hbase/master/AssignmentManager.html
@@ -26,2906 +26,2948 @@
 <span class="sourceLineNo">018</span> */<a name="line.18"></a>
 <span class="sourceLineNo">019</span>package org.apache.hadoop.hbase.master;<a name="line.19"></a>
 <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.util.ArrayList;<a name="line.22"></a>
-<span class="sourceLineNo">023</span>import java.util.Collection;<a name="line.23"></a>
-<span class="sourceLineNo">024</span>import java.util.Collections;<a name="line.24"></a>
-<span class="sourceLineNo">025</span>import java.util.HashMap;<a name="line.25"></a>
-<span class="sourceLineNo">026</span>import java.util.HashSet;<a name="line.26"></a>
-<span class="sourceLineNo">027</span>import java.util.Iterator;<a name="line.27"></a>
-<span class="sourceLineNo">028</span>import java.util.List;<a name="line.28"></a>
-<span class="sourceLineNo">029</span>import java.util.Map;<a name="line.29"></a>
-<span class="sourceLineNo">030</span>import java.util.NavigableMap;<a name="line.30"></a>
-<span class="sourceLineNo">031</span>import java.util.Random;<a name="line.31"></a>
-<span class="sourceLineNo">032</span>import java.util.Set;<a name="line.32"></a>
-<span class="sourceLineNo">033</span>import java.util.TreeMap;<a name="line.33"></a>
-<span class="sourceLineNo">034</span>import java.util.concurrent.Callable;<a name="line.34"></a>
-<span class="sourceLineNo">035</span>import java.util.concurrent.ConcurrentHashMap;<a name="line.35"></a>
-<span class="sourceLineNo">036</span>import java.util.concurrent.CopyOnWriteArrayList;<a name="line.36"></a>
-<span class="sourceLineNo">037</span>import java.util.concurrent.TimeUnit;<a name="line.37"></a>
-<span class="sourceLineNo">038</span>import java.util.concurrent.atomic.AtomicBoolean;<a name="line.38"></a>
-<span class="sourceLineNo">039</span>import java.util.concurrent.atomic.AtomicInteger;<a name="line.39"></a>
-<span class="sourceLineNo">040</span>import java.util.concurrent.locks.Lock;<a name="line.40"></a>
-<span class="sourceLineNo">041</span>import java.util.concurrent.locks.ReentrantLock;<a name="line.41"></a>
-<span class="sourceLineNo">042</span><a name="line.42"></a>
-<span class="sourceLineNo">043</span>import org.apache.commons.logging.Log;<a name="line.43"></a>
-<span class="sourceLineNo">044</span>import org.apache.commons.logging.LogFactory;<a name="line.44"></a>
-<span class="sourceLineNo">045</span>import org.apache.hadoop.hbase.classification.InterfaceAudience;<a name="line.45"></a>
-<span class="sourceLineNo">046</span>import org.apache.hadoop.conf.Configuration;<a name="line.46"></a>
-<span class="sourceLineNo">047</span>import org.apache.hadoop.fs.FileSystem;<a name="line.47"></a>
-<span class="sourceLineNo">048</span>import org.apache.hadoop.fs.Path;<a name="line.48"></a>
-<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.CoordinatedStateException;<a name="line.49"></a>
-<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.HBaseIOException;<a name="line.50"></a>
-<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.HConstants;<a name="line.51"></a>
-<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.HRegionInfo;<a name="line.52"></a>
-<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.HRegionLocation;<a name="line.53"></a>
-<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.HTableDescriptor;<a name="line.54"></a>
-<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.MetaTableAccessor;<a name="line.55"></a>
-<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.NotServingRegionException;<a name="line.56"></a>
-<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.RegionLocations;<a name="line.57"></a>
-<span class="sourceLineNo">058</span>import org.apache.hadoop.hbase.RegionStateListener;<a name="line.58"></a>
-<span class="sourceLineNo">059</span>import org.apache.hadoop.hbase.ServerName;<a name="line.59"></a>
-<span class="sourceLineNo">060</span>import org.apache.hadoop.hbase.TableName;<a name="line.60"></a>
-<span class="sourceLineNo">061</span>import org.apache.hadoop.hbase.TableNotFoundException;<a name="line.61"></a>
-<span class="sourceLineNo">062</span>import org.apache.hadoop.hbase.client.RegionReplicaUtil;<a name="line.62"></a>
-<span class="sourceLineNo">063</span>import org.apache.hadoop.hbase.client.Result;<a name="line.63"></a>
-<span class="sourceLineNo">064</span>import org.apache.hadoop.hbase.client.TableState;<a name="line.64"></a>
-<span class="sourceLineNo">065</span>import org.apache.hadoop.hbase.executor.EventHandler;<a name="line.65"></a>
-<span class="sourceLineNo">066</span>import org.apache.hadoop.hbase.executor.EventType;<a name="line.66"></a>
-<span class="sourceLineNo">067</span>import org.apache.hadoop.hbase.executor.ExecutorService;<a name="line.67"></a>
-<span class="sourceLineNo">068</span>import org.apache.hadoop.hbase.ipc.FailedServerException;<a name="line.68"></a>
-<span class="sourceLineNo">069</span>import org.apache.hadoop.hbase.ipc.RpcClient;<a name="line.69"></a>
-<span class="sourceLineNo">070</span>import org.apache.hadoop.hbase.ipc.ServerNotRunningYetException;<a name="line.70"></a>
-<span class="sourceLineNo">071</span>import org.apache.hadoop.hbase.master.RegionState.State;<a name="line.71"></a>
-<span class="sourceLineNo">072</span>import org.apache.hadoop.hbase.master.balancer.FavoredNodeAssignmentHelper;<a name="line.72"></a>
-<span class="sourceLineNo">073</span>import org.apache.hadoop.hbase.master.balancer.FavoredNodeLoadBalancer;<a name="line.73"></a>
-<span class="sourceLineNo">074</span>import org.apache.hadoop.hbase.master.handler.DisableTableHandler;<a name="line.74"></a>
-<span class="sourceLineNo">075</span>import org.apache.hadoop.hbase.master.handler.EnableTableHandler;<a name="line.75"></a>
-<span class="sourceLineNo">076</span>import org.apache.hadoop.hbase.master.normalizer.NormalizationPlan.PlanType;<a name="line.76"></a>
-<span class="sourceLineNo">077</span>import org.apache.hadoop.hbase.protobuf.generated.RegionServerStatusProtos.RegionStateTransition;<a name="line.77"></a>
-<span class="sourceLineNo">078</span>import org.apache.hadoop.hbase.protobuf.generated.RegionServerStatusProtos.RegionStateTransition.TransitionCode;<a name="line.78"></a>
-<span class="sourceLineNo">079</span>import org.apache.hadoop.hbase.quotas.QuotaExceededException;<a name="line.79"></a>
-<span class="sourceLineNo">080</span>import org.apache.hadoop.hbase.regionserver.RegionOpeningState;<a name="line.80"></a>
-<span class="sourceLineNo">081</span>import org.apache.hadoop.hbase.regionserver.RegionServerAbortedException;<a name="line.81"></a>
-<span class="sourceLineNo">082</span>import org.apache.hadoop.hbase.regionserver.RegionServerStoppedException;<a name="line.82"></a>
-<span class="sourceLineNo">083</span>import org.apache.hadoop.hbase.wal.DefaultWALProvider;<a name="line.83"></a>
-<span class="sourceLineNo">084</span>import org.apache.hadoop.hbase.util.EnvironmentEdgeManager;<a name="line.84"></a>
-<span class="sourceLineNo">085</span>import org.apache.hadoop.hbase.util.FSUtils;<a name="line.85"></a>
-<span class="sourceLineNo">086</span>import org.apache.hadoop.hbase.util.KeyLocker;<a name="line.86"></a>
-<span class="sourceLineNo">087</span>import org.apache.hadoop.hbase.util.Pair;<a name="line.87"></a>
-<span class="sourceLineNo">088</span>import org.apache.hadoop.hbase.util.PairOfSameType;<a name="line.88"></a>
-<span class="sourceLineNo">089</span>import org.apache.hadoop.hbase.util.Threads;<a name="line.89"></a>
-<span class="sourceLineNo">090</span>import org.apache.hadoop.hbase.zookeeper.MetaTableLocator;<a name="line.90"></a>
-<span class="sourceLineNo">091</span>import org.apache.hadoop.ipc.RemoteException;<a name="line.91"></a>
-<span class="sourceLineNo">092</span>import org.apache.hadoop.util.StringUtils;<a name="line.92"></a>
-<span class="sourceLineNo">093</span>import org.apache.zookeeper.KeeperException;<a name="line.93"></a>
-<span class="sourceLineNo">094</span><a name="line.94"></a>
-<span class="sourceLineNo">095</span>import com.google.common.annotations.VisibleForTesting;<a name="line.95"></a>
-<span class="sourceLineNo">096</span><a name="line.96"></a>
-<span class="sourceLineNo">097</span>/**<a name="line.97"></a>
-<span class="sourceLineNo">098</span> * Manages and performs region assignment.<a name="line.98"></a>
-<span class="sourceLineNo">099</span> * Related communications with regionserver are all done over RPC.<a name="line.99"></a>
-<span class="sourceLineNo">100</span> */<a name="line.100"></a>
-<span class="sourceLineNo">101</span>@InterfaceAudience.Private<a name="line.101"></a>
-<span class="sourceLineNo">102</span>public class AssignmentManager {<a name="line.102"></a>
-<span class="sourceLineNo">103</span>  private static final Log LOG = LogFactory.getLog(AssignmentManager.class);<a name="line.103"></a>
-<span class="sourceLineNo">104</span><a name="line.104"></a>
-<span class="sourceLineNo">105</span>  protected final MasterServices server;<a name="line.105"></a>
-<span class="sourceLineNo">106</span><a name="line.106"></a>
-<span class="sourceLineNo">107</span>  private ServerManager serverManager;<a name="line.107"></a>
-<span class="sourceLineNo">108</span><a name="line.108"></a>
-<span class="sourceLineNo">109</span>  private boolean shouldAssignRegionsWithFavoredNodes;<a name="line.109"></a>
-<span class="sourceLineNo">110</span><a name="line.110"></a>
-<span class="sourceLineNo">111</span>  private LoadBalancer balancer;<a name="line.111"></a>
-<span class="sourceLineNo">112</span><a name="line.112"></a>
-<span class="sourceLineNo">113</span>  private final MetricsAssignmentManager metricsAssignmentManager;<a name="line.113"></a>
-<span class="sourceLineNo">114</span><a name="line.114"></a>
-<span class="sourceLineNo">115</span>  private final TableLockManager tableLockManager;<a name="line.115"></a>
-<span class="sourceLineNo">116</span><a name="line.116"></a>
-<span class="sourceLineNo">117</span>  private AtomicInteger numRegionsOpened = new AtomicInteger(0);<a name="line.117"></a>
-<span class="sourceLineNo">118</span><a name="line.118"></a>
-<span class="sourceLineNo">119</span>  final private KeyLocker&lt;String&gt; locker = new KeyLocker&lt;String&gt;();<a name="line.119"></a>
-<span class="sourceLineNo">120</span><a name="line.120"></a>
-<span class="sourceLineNo">121</span>  Set&lt;HRegionInfo&gt; replicasToClose = Collections.synchronizedSet(new HashSet&lt;HRegionInfo&gt;());<a name="line.121"></a>
-<span class="sourceLineNo">122</span><a name="line.122"></a>
-<span class="sourceLineNo">123</span>  /**<a name="line.123"></a>
-<span class="sourceLineNo">124</span>   * Map of regions to reopen after the schema of a table is changed. Key -<a name="line.124"></a>
-<span class="sourceLineNo">125</span>   * encoded region name, value - HRegionInfo<a name="line.125"></a>
-<span class="sourceLineNo">126</span>   */<a name="line.126"></a>
-<span class="sourceLineNo">127</span>  private final Map &lt;String, HRegionInfo&gt; regionsToReopen;<a name="line.127"></a>
-<span class="sourceLineNo">128</span><a name="line.128"></a>
-<span class="sourceLineNo">129</span>  /*<a name="line.129"></a>
-<span class="sourceLineNo">130</span>   * Maximum times we recurse an assignment/unassignment.<a name="line.130"></a>
-<span class="sourceLineNo">131</span>   * See below in {@link #assign()} and {@link #unassign()}.<a name="line.131"></a>
-<span class="sourceLineNo">132</span>   */<a name="line.132"></a>
-<span class="sourceLineNo">133</span>  private final int maximumAttempts;<a name="line.133"></a>
-<span class="sourceLineNo">134</span><a name="line.134"></a>
-<span class="sourceLineNo">135</span>  /**<a name="line.135"></a>
-<span class="sourceLineNo">136</span>   * The sleep time for which the assignment will wait before retrying in case of<a name="line.136"></a>
-<span class="sourceLineNo">137</span>   * hbase:meta assignment failure due to lack of availability of region plan or bad region plan<a name="line.137"></a>
-<span class="sourceLineNo">138</span>   */<a name="line.138"></a>
-<span class="sourceLineNo">139</span>  private final long sleepTimeBeforeRetryingMetaAssignment;<a name="line.139"></a>
-<span class="sourceLineNo">140</span><a name="line.140"></a>
-<span class="sourceLineNo">141</span>  /** Plans for region movement. Key is the encoded version of a region name*/<a name="line.141"></a>
-<span class="sourceLineNo">142</span>  // TODO: When do plans get cleaned out?  Ever? In server open and in server<a name="line.142"></a>
-<span class="sourceLineNo">143</span>  // shutdown processing -- St.Ack<a name="line.143"></a>
-<span class="sourceLineNo">144</span>  // All access to this Map must be synchronized.<a name="line.144"></a>
-<span class="sourceLineNo">145</span>  final NavigableMap&lt;String, RegionPlan&gt; regionPlans =<a name="line.145"></a>
-<span class="sourceLineNo">146</span>    new TreeMap&lt;String, RegionPlan&gt;();<a name="line.146"></a>
-<span class="sourceLineNo">147</span><a name="line.147"></a>
-<span class="sourceLineNo">148</span>  private final TableStateManager tableStateManager;<a name="line.148"></a>
-<span class="sourceLineNo">149</span><a name="line.149"></a>
-<span class="sourceLineNo">150</span>  private final ExecutorService executorService;<a name="line.150"></a>
-<span class="sourceLineNo">151</span><a name="line.151"></a>
-<span class="sourceLineNo">152</span>  // Thread pool executor service. TODO, consolidate with executorService?<a name="line.152"></a>
-<span class="sourceLineNo">153</span>  private java.util.concurrent.ExecutorService threadPoolExecutorService;<a name="line.153"></a>
-<span class="sourceLineNo">154</span><a name="line.154"></a>
-<span class="sourceLineNo">155</span>  private final RegionStates regionStates;<a name="line.155"></a>
-<span class="sourceLineNo">156</span><a name="line.156"></a>
-<span class="sourceLineNo">157</span>  // The threshold to use bulk assigning. Using bulk assignment<a name="line.157"></a>
-<span class="sourceLineNo">158</span>  // only if assigning at least this many regions to at least this<a name="line.158"></a>
-<span class="sourceLineNo">159</span>  // many servers. If assigning fewer regions to fewer servers,<a name="line.159"></a>
-<span class="sourceLineNo">160</span>  // bulk assigning may be not as efficient.<a name="line.160"></a>
-<span class="sourceLineNo">161</span>  private final int bulkAssignThresholdRegions;<a name="line.161"></a>
-<span class="sourceLineNo">162</span>  private final int bulkAssignThresholdServers;<a name="line.162"></a>
-<span class="sourceLineNo">163</span>  private final int bulkPerRegionOpenTimeGuesstimate;<a name="line.163"></a>
-<span class="sourceLineNo">164</span><a name="line.164"></a>
-<span class="sourceLineNo">165</span>  // Should bulk assignment wait till all regions are assigned,<a name="line.165"></a>
-<span class="sourceLineNo">166</span>  // or it is timed out?  This is useful to measure bulk assignment<a name="line.166"></a>
-<span class="sourceLineNo">167</span>  // performance, but not needed in most use cases.<a name="line.167"></a>
-<span class="sourceLineNo">168</span>  private final boolean bulkAssignWaitTillAllAssigned;<a name="line.168"></a>
-<span class="sourceLineNo">169</span><a name="line.169"></a>
-<span class="sourceLineNo">170</span>  /**<a name="line.170"></a>
-<span class="sourceLineNo">171</span>   * Indicator that AssignmentManager has recovered the region states so<a name="line.171"></a>
-<span class="sourceLineNo">172</span>   * that ServerShutdownHandler can be fully enabled and re-assign regions<a name="line.172"></a>
-<span class="sourceLineNo">173</span>   * of dead servers. So that when re-assignment happens, AssignmentManager<a name="line.173"></a>
-<span class="sourceLineNo">174</span>   * has proper region states.<a name="line.174"></a>
-<span class="sourceLineNo">175</span>   *<a name="line.175"></a>
-<span class="sourceLineNo">176</span>   * Protected to ease testing.<a name="line.176"></a>
-<span class="sourceLineNo">177</span>   */<a name="line.177"></a>
-<span class="sourceLineNo">178</span>  protected final AtomicBoolean failoverCleanupDone = new AtomicBoolean(false);<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>   * A map to track the count a region fails to open in a row.<a name="line.181"></a>
-<span class="sourceLineNo">182</span>   * So that we don't try to open a region forever if the failure is<a name="line.182"></a>
-<span class="sourceLineNo">183</span>   * unrecoverable.  We don't put this information in region states<a name="line.183"></a>
-<span class="sourceLineNo">184</span>   * because we don't expect this to happen frequently; we don't<a name="line.184"></a>
-<span class="sourceLineNo">185</span>   * want to copy this information over during each state transition either.<a name="line.185"></a>
-<span class="sourceLineNo">186</span>   */<a name="line.186"></a>
-<span class="sourceLineNo">187</span>  private final ConcurrentHashMap&lt;String, AtomicInteger&gt;<a name="line.187"></a>
-<span class="sourceLineNo">188</span>    failedOpenTracker = new ConcurrentHashMap&lt;String, AtomicInteger&gt;();<a name="line.188"></a>
-<span class="sourceLineNo">189</span><a name="line.189"></a>
-<span class="sourceLineNo">190</span>  // In case not using ZK for region assignment, region states<a name="line.190"></a>
-<span class="sourceLineNo">191</span>  // are persisted in meta with a state store<a name="line.191"></a>
-<span class="sourceLineNo">192</span>  private final RegionStateStore regionStateStore;<a name="line.192"></a>
-<span class="sourceLineNo">193</span><a name="line.193"></a>
-<span class="sourceLineNo">194</span>  /**<a name="line.194"></a>
-<span class="sourceLineNo">195</span>   * For testing only!  Set to true to skip handling of split.<a name="line.195"></a>
-<span class="sourceLineNo">196</span>   */<a name="line.196"></a>
-<span class="sourceLineNo">197</span>  @edu.umd.cs.findbugs.annotations.SuppressWarnings(value="MS_SHOULD_BE_FINAL")<a name="line.197"></a>
-<span class="sourceLineNo">198</span>  public static boolean TEST_SKIP_SPLIT_HANDLING = false;<a name="line.198"></a>
-<span class="sourceLineNo">199</span><a name="line.199"></a>
-<span class="sourceLineNo">200</span>  /** Listeners that are called on assignment events. */<a name="line.200"></a>
-<span class="sourceLineNo">201</span>  private List&lt;AssignmentListener&gt; listeners = new CopyOnWriteArrayList&lt;AssignmentListener&gt;();<a name="line.201"></a>
-<span class="sourceLineNo">202</span><a name="line.202"></a>
-<span class="sourceLineNo">203</span>  private RegionStateListener regionStateListener;<a name="line.203"></a>
-<span class="sourceLineNo">204</span><a name="line.204"></a>
-<span class="sourceLineNo">205</span>  /**<a name="line.205"></a>
-<span class="sourceLineNo">206</span>   * Constructs a new assignment manager.<a name="line.206"></a>
-<span class="sourceLineNo">207</span>   *<a name="line.207"></a>
-<span class="sourceLineNo">208</span>   * @param server instance of HMaster this AM running inside<a name="line.208"></a>
-<span class="sourceLineNo">209</span>   * @param serverManager serverManager for associated HMaster<a name="line.209"></a>
-<span class="sourceLineNo">210</span>   * @param balancer implementation of {@link LoadBalancer}<a name="line.210"></a>
-<span class="sourceLineNo">211</span>   * @param service Executor service<a name="line.211"></a>
-<span class="sourceLineNo">212</span>   * @param metricsMaster metrics manager<a name="line.212"></a>
-<span class="sourceLineNo">213</span>   * @param tableLockManager TableLock manager<a name="line.213"></a>
-<span class="sourceLineNo">214</span>   * @throws IOException<a name="line.214"></a>
-<span class="sourceLineNo">215</span>   */<a name="line.215"></a>
-<span class="sourceLineNo">216</span>  public AssignmentManager(MasterServices server, ServerManager serverManager,<a name="line.216"></a>
-<span class="sourceLineNo">217</span>      final LoadBalancer balancer,<a name="line.217"></a>
-<span class="sourceLineNo">218</span>      final ExecutorService service, MetricsMaster metricsMaster,<a name="line.218"></a>
-<span class="sourceLineNo">219</span>      final TableLockManager tableLockManager,<a name="line.219"></a>
-<span class="sourceLineNo">220</span>      final TableStateManager tableStateManager)<a name="line.220"></a>
-<span class="sourceLineNo">221</span>          throws IOException {<a name="line.221"></a>
-<span class="sourceLineNo">222</span>    this.server = server;<a name="line.222"></a>
-<span class="sourceLineNo">223</span>    this.serverManager = serverManager;<a name="line.223"></a>
-<span class="sourceLineNo">224</span>    this.executorService = service;<a name="line.224"></a>
-<span class="sourceLineNo">225</span>    this.regionStateStore = new RegionStateStore(server);<a name="line.225"></a>
-<span class="sourceLineNo">226</span>    this.regionsToReopen = Collections.synchronizedMap<a name="line.226"></a>
-<span class="sourceLineNo">227</span>                           (new HashMap&lt;String, HRegionInfo&gt; ());<a name="line.227"></a>
-<span class="sourceLineNo">228</span>    Configuration conf = server.getConfiguration();<a name="line.228"></a>
-<span class="sourceLineNo">229</span>    // Only read favored nodes if using the favored nodes load balancer.<a name="line.229"></a>
-<span class="sourceLineNo">230</span>    this.shouldAssignRegionsWithFavoredNodes = conf.getClass(<a name="line.230"></a>
-<span class="sourceLineNo">231</span>           HConstants.HBASE_MASTER_LOADBALANCER_CLASS, Object.class).equals(<a name="line.231"></a>
-<span class="sourceLineNo">232</span>           FavoredNodeLoadBalancer.class);<a name="line.232"></a>
-<span class="sourceLineNo">233</span><a name="line.233"></a>
-<span class="sourceLineNo">234</span>    this.tableStateManager = tableStateManager;<a name="line.234"></a>
-<span class="sourceLineNo">235</span><a name="line.235"></a>
-<span class="sourceLineNo">236</span>    // This is the max attempts, not retries, so it should be at least 1.<a name="line.236"></a>
-<span class="sourceLineNo">237</span>    this.maximumAttempts = Math.max(1,<a name="line.237"></a>
-<span class="sourceLineNo">238</span>      this.server.getConfiguration().getInt("hbase.assignment.maximum.attempts", 10));<a name="line.238"></a>
-<span class="sourceLineNo">239</span>    this.sleepTimeBeforeRetryingMetaAssignment = this.server.getConfiguration().getLong(<a name="line.239"></a>
-<span class="sourceLineNo">240</span>        "hbase.meta.assignment.retry.sleeptime", 1000l);<a name="line.240"></a>
-<span class="sourceLineNo">241</span>    this.balancer = balancer;<a name="line.241"></a>
-<span class="sourceLineNo">242</span>    int maxThreads = conf.getInt("hbase.assignment.threads.max", 30);<a name="line.242"></a>
-<span class="sourceLineNo">243</span>    this.threadPoolExecutorService = Threads.getBoundedCachedThreadPool(<a name="line.243"></a>
-<span class="sourceLineNo">244</span>      maxThreads, 60L, TimeUnit.SECONDS, Threads.newDaemonThreadFactory("AM."));<a name="line.244"></a>
-<span class="sourceLineNo">245</span>    this.regionStates = new RegionStates(<a name="line.245"></a>
-<span class="sourceLineNo">246</span>      server, tableStateManager, serverManager, regionStateStore);<a name="line.246"></a>
-<span class="sourceLineNo">247</span><a name="line.247"></a>
-<span class="sourceLineNo">248</span>    this.bulkAssignWaitTillAllAssigned =<a name="line.248"></a>
-<span class="sourceLineNo">249</span>      conf.getBoolean("hbase.bulk.assignment.waittillallassigned", false);<a name="line.249"></a>
-<span class="sourceLineNo">250</span>    this.bulkAssignThresholdRegions = conf.getInt("hbase.bulk.assignment.threshold.regions", 7);<a name="line.250"></a>
-<span class="sourceLineNo">251</span>    this.bulkAssignThresholdServers = conf.getInt("hbase.bulk.assignment.threshold.servers", 3);<a name="line.251"></a>
-<span class="sourceLineNo">252</span>    this.bulkPerRegionOpenTimeGuesstimate =<a name="line.252"></a>
-<span class="sourceLineNo">253</span>      conf.getInt("hbase.bulk.assignment.perregion.open.time", 10000);<a name="line.253"></a>
-<span class="sourceLineNo">254</span><a name="line.254"></a>
-<span class="sourceLineNo">255</span>    this.metricsAssignmentManager = new MetricsAssignmentManager();<a name="line.255"></a>
-<span class="sourceLineNo">256</span>    this.tableLockManager = tableLockManager;<a name="line.256"></a>
-<span class="sourceLineNo">257</span>  }<a name="line.257"></a>
-<span class="sourceLineNo">258</span><a name="line.258"></a>
-<span class="sourceLineNo">259</span>  /**<a name="line.259"></a>
-<span class="sourceLineNo">260</span>   * Add the listener to the notification list.<a name="line.260"></a>
-<span class="sourceLineNo">261</span>   * @param listener The AssignmentListener to register<a name="line.261"></a>
-<span class="sourceLineNo">262</span>   */<a name="line.262"></a>
-<span class="sourceLineNo">263</span>  public void registerListener(final AssignmentListener listener) {<a name="line.263"></a>
-<span class="sourceLineNo">264</span>    this.listeners.add(listener);<a name="line.264"></a>
-<span class="sourceLineNo">265</span>  }<a name="line.265"></a>
-<span class="sourceLineNo">266</span><a name="line.266"></a>
-<span class="sourceLineNo">267</span>  /**<a name="line.267"></a>
-<span class="sourceLineNo">268</span>   * Remove the listener from the notification list.<a name="line.268"></a>
-<span class="sourceLineNo">269</span>   * @param listener The AssignmentListener to unregister<a name="line.269"></a>
-<span class="sourceLineNo">270</span>   */<a name="line.270"></a>
-<span class="sourceLineNo">271</span>  public boolean unregisterListener(final AssignmentListener listener) {<a name="line.271"></a>
-<span class="sourceLineNo">272</span>    return this.listeners.remove(listener);<a name="line.272"></a>
-<span class="sourceLineNo">273</span>  }<a name="line.273"></a>
-<span class="sourceLineNo">274</span><a name="line.274"></a>
-<span class="sourceLineNo">275</span>  /**<a name="line.275"></a>
-<span class="sourceLineNo">276</span>   * @return Instance of ZKTableStateManager.<a name="line.276"></a>
-<span class="sourceLineNo">277</span>   */<a name="line.277"></a>
-<span class="sourceLineNo">278</span>  public TableStateManager getTableStateManager() {<a name="line.278"></a>
-<span class="sourceLineNo">279</span>    // These are 'expensive' to make involving trip to zk ensemble so allow<a name="line.279"></a>
-<span class="sourceLineNo">280</span>    // sharing.<a name="line.280"></a>
-<span class="sourceLineNo">281</span>    return this.tableStateManager;<a name="line.281"></a>
-<span class="sourceLineNo">282</span>  }<a name="line.282"></a>
-<span class="sourceLineNo">283</span><a name="line.283"></a>
-<span class="sourceLineNo">284</span>  /**<a name="line.284"></a>
-<span class="sourceLineNo">285</span>   * This SHOULD not be public. It is public now<a name="line.285"></a>
-<span class="sourceLineNo">286</span>   * because of some unit tests.<a name="line.286"></a>
-<span class="sourceLineNo">287</span>   *<a name="line.287"></a>
-<span class="sourceLineNo">288</span>   * TODO: make it package private and keep RegionStates in the master package<a name="line.288"></a>
-<span class="sourceLineNo">289</span>   */<a name="line.289"></a>
-<span class="sourceLineNo">290</span>  public RegionStates getRegionStates() {<a name="line.290"></a>
-<span class="sourceLineNo">291</span>    return regionStates;<a name="line.291"></a>
-<span class="sourceLineNo">292</span>  }<a name="line.292"></a>
-<span class="sourceLineNo">293</span><a name="line.293"></a>
-<span class="sourceLineNo">294</span>  /**<a name="line.294"></a>
-<span class="sourceLineNo">295</span>   * Used in some tests to mock up region state in meta<a name="line.295"></a>
-<span class="sourceLineNo">296</span>   */<a name="line.296"></a>
-<span class="sourceLineNo">297</span>  @VisibleForTesting<a name="line.297"></a>
-<span class="sourceLineNo">298</span>  RegionStateStore getRegionStateStore() {<a name="line.298"></a>
-<span class="sourceLineNo">299</span>    return regionStateStore;<a name="line.299"></a>
-<span class="sourceLineNo">300</span>  }<a name="line.300"></a>
-<span class="sourceLineNo">301</span><a name="line.301"></a>
-<span class="sourceLineNo">302</span>  public RegionPlan getRegionReopenPlan(HRegionInfo hri) {<a name="line.302"></a>
-<span class="sourceLineNo">303</span>    return new RegionPlan(hri, null, regionStates.getRegionServerOfRegion(hri));<a name="line.303"></a>
-<span class="sourceLineNo">304</span>  }<a name="line.304"></a>
-<span class="sourceLineNo">305</span><a name="line.305"></a>
-<span class="sourceLineNo">306</span>  /**<a name="line.306"></a>
-<span class="sourceLineNo">307</span>   * Add a regionPlan for the specified region.<a name="line.307"></a>
-<span class="sourceLineNo">308</span>   * @param encodedName<a name="line.308"></a>
-<span class="sourceLineNo">309</span>   * @param plan<a name="line.309"></a>
-<span class="sourceLineNo">310</span>   */<a name="line.310"></a>
-<span class="sourceLineNo">311</span>  public void addPlan(String encodedName, RegionPlan plan) {<a name="line.311"></a>
-<span class="sourceLineNo">312</span>    synchronized (regionPlans) {<a name="line.312"></a>
-<span class="sourceLineNo">313</span>      regionPlans.put(encodedName, plan);<a name="line.313"></a>
-<span class="sourceLineNo">314</span>    }<a name="line.314"></a>
-<span class="sourceLineNo">315</span>  }<a name="line.315"></a>
-<span class="sourceLineNo">316</span><a name="line.316"></a>
-<span class="sourceLineNo">317</span>  /**<a name="line.317"></a>
-<span class="sourceLineNo">318</span>   * Add a map of region plans.<a name="line.318"></a>
-<span class="sourceLineNo">319</span>   */<a name="line.319"></a>
-<span class="sourceLineNo">320</span>  public void addPlans(Map&lt;String, RegionPlan&gt; plans) {<a name="line.320"></a>
-<span class="sourceLineNo">321</span>    synchronized (regionPlans) {<a name="line.321"></a>
-<span class="sourceLineNo">322</span>      regionPlans.putAll(plans);<a name="line.322"></a>
-<span class="sourceLineNo">323</span>    }<a name="line.323"></a>
-<span class="sourceLineNo">324</span>  }<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>   * Set the list of regions that will be reopened<a name="line.327"></a>
-<span class="sourceLineNo">328</span>   * because of an update in table schema<a name="line.328"></a>
-<span class="sourceLineNo">329</span>   *<a name="line.329"></a>
-<span class="sourceLineNo">330</span>   * @param regions<a name="line.330"></a>
-<span class="sourceLineNo">331</span>   *          list of regions that should be tracked for reopen<a name="line.331"></a>
-<span class="sourceLineNo">332</span>   */<a name="line.332"></a>
-<span class="sourceLineNo">333</span>  public void setRegionsToReopen(List &lt;HRegionInfo&gt; regions) {<a name="line.333"></a>
-<span class="sourceLineNo">334</span>    for(HRegionInfo hri : regions) {<a name="line.334"></a>
-<span class="sourceLineNo">335</span>      regionsToReopen.put(hri.getEncodedName(), hri);<a name="line.335"></a>
-<span class="sourceLineNo">336</span>    }<a name="line.336"></a>
-<span class="sourceLineNo">337</span>  }<a name="line.337"></a>
-<span class="sourceLineNo">338</span><a name="line.338"></a>
-<span class="sourceLineNo">339</span>  /**<a name="line.339"></a>
-<span class="sourceLineNo">340</span>   * Used by the client to identify if all regions have the schema updates<a name="line.340"></a>
-<span class="sourceLineNo">341</span>   *<a name="line.341"></a>
-<span class="sourceLineNo">342</span>   * @param tableName<a name="line.342"></a>
-<span class="sourceLineNo">343</span>   * @return Pair indicating the status of the alter command<a name="line.343"></a>
-<span class="sourceLineNo">344</span>   * @throws IOException<a name="line.344"></a>
-<span class="sourceLineNo">345</span>   */<a name="line.345"></a>
-<span class="sourceLineNo">346</span>  public Pair&lt;Integer, Integer&gt; getReopenStatus(TableName tableName)<a name="line.346"></a>
-<span class="sourceLineNo">347</span>      throws IOException {<a name="line.347"></a>
-<span class="sourceLineNo">348</span>    List&lt;HRegionInfo&gt; hris;<a name="line.348"></a>
-<span class="sourceLineNo">349</span>    if (TableName.META_TABLE_NAME.equals(tableName)) {<a name="line.349"></a>
-<span class="sourceLineNo">350</span>      hris = new MetaTableLocator().getMetaRegions(server.getZooKeeper());<a name="line.350"></a>
-<span class="sourceLineNo">351</span>    } else {<a name="line.351"></a>
-<span class="sourceLineNo">352</span>      hris = MetaTableAccessor.getTableRegions(server.getConnection(), tableName, true);<a name="line.352"></a>
-<span class="sourceLineNo">353</span>    }<a name="line.353"></a>
-<span class="sourceLineNo">354</span><a name="line.354"></a>
-<span class="sourceLineNo">355</span>    Integer pending = 0;<a name="line.355"></a>
-<span class="sourceLineNo">356</span>    for (HRegionInfo hri : hris) {<a name="line.356"></a>
-<span class="sourceLineNo">357</span>      String name = hri.getEncodedName();<a name="line.357"></a>
-<span class="sourceLineNo">358</span>      // no lock concurrent access ok: sequential consistency respected.<a name="line.358"></a>
-<span class="sourceLineNo">359</span>      if (regionsToReopen.containsKey(name)<a name="line.359"></a>
-<span class="sourceLineNo">360</span>          || regionStates.isRegionInTransition(name)) {<a name="line.360"></a>
-<span class="sourceLineNo">361</span>        pending++;<a name="line.361"></a>
-<span class="sourceLineNo">362</span>      }<a name="line.362"></a>
-<span class="sourceLineNo">363</span>    }<a name="line.363"></a>
-<span class="sourceLineNo">364</span>    return new Pair&lt;Integer, Integer&gt;(pending, hris.size());<a name="line.364"></a>
-<span class="sourceLineNo">365</span>  }<a name="line.365"></a>
-<span class="sourceLineNo">366</span><a name="line.366"></a>
-<span class="sourceLineNo">367</span>  /**<a name="line.367"></a>
-<span class="sourceLineNo">368</span>   * Used by ServerShutdownHandler to make sure AssignmentManager has completed<a name="line.368"></a>
-<span class="sourceLineNo">369</span>   * the failover cleanup before re-assigning regions of dead servers. So that<a name="line.369"></a>
-<span class="sourceLineNo">370</span>   * when re-assignment happens, AssignmentManager has proper region states.<a name="line.370"></a>
-<span class="sourceLineNo">371</span>   */<a name="line.371"></a>
-<span class="sourceLineNo">372</span>  public boolean isFailoverCleanupDone() {<a name="line.372"></a>
-<span class="sourceLineNo">373</span>    return failoverCleanupDone.get();<a name="line.373"></a>
-<span class="sourceLineNo">374</span>  }<a name="line.374"></a>
-<span class="sourceLineNo">375</span><a name="line.375"></a>
-<span class="sourceLineNo">376</span>  /**<a name="line.376"></a>
-<span class="sourceLineNo">377</span>   * To avoid racing with AM, external entities may need to lock a region,<a name="line.377"></a>
-<span class="sourceLineNo">378</span>   * for example, when SSH checks what regions to skip re-assigning.<a name="line.378"></a>
-<span class="sourceLineNo">379</span>   */<a name="line.379"></a>
-<span class="sourceLineNo">380</span>  public Lock acquireRegionLock(final String encodedName) {<a name="line.380"></a>
-<span class="sourceLineNo">381</span>    return locker.acquireLock(encodedName);<a name="line.381"></a>
-<span class="sourceLineNo">382</span>  }<a name="line.382"></a>
-<span class="sourceLineNo">383</span><a name="line.383"></a>
-<span class="sourceLineNo">384</span>  /**<a name="line.384"></a>
-<span class="sourceLineNo">385</span>   * Now, failover cleanup is completed. Notify server manager to<a name="line.385"></a>
-<span class="sourceLineNo">386</span>   * process queued up dead servers processing, if any.<a name="line.386"></a>
-<span class="sourceLineNo">387</span>   */<a name="line.387"></a>
-<span class="sourceLineNo">388</span>  void failoverCleanupDone() {<a name="line.388"></a>
-<span class="sourceLineNo">389</span>    failoverCleanupDone.set(true);<a name="line.389"></a>
-<span class="sourceLineNo">390</span>    serverManager.processQueuedDeadServers();<a name="line.390"></a>
-<span class="sourceLineNo">391</span>  }<a name="line.391"></a>
-<span class="sourceLineNo">392</span><a name="line.392"></a>
-<span class="sourceLineNo">393</span>  /**<a name="line.393"></a>
-<span class="sourceLineNo">394</span>   * Called on startup.<a name="line.394"></a>
-<span class="sourceLineNo">395</span>   * Figures whether a fresh cluster start of we are joining extant running cluster.<a name="line.395"></a>
-<span class="sourceLineNo">396</span>   * @throws IOException<a name="line.396"></a>
-<span class="sourceLineNo">397</span>   * @throws KeeperException<a name="line.397"></a>
-<span class="sourceLineNo">398</span>   * @throws InterruptedException<a name="line.398"></a>
-<span class="sourceLineNo">399</span>   * @throws CoordinatedStateException<a name="line.399"></a>
-<span class="sourceLineNo">400</span>   */<a name="line.400"></a>
-<span class="sourceLineNo">401</span>  void joinCluster()<a name="line.401"></a>
-<span class="sourceLineNo">402</span>  throws IOException, KeeperException, InterruptedException, CoordinatedStateException {<a name="line.402"></a>
-<span class="sourceLineNo">403</span>    long startTime = System.currentTimeMillis();<a name="line.403"></a>
-<span class="sourceLineNo">404</span>    // Concurrency note: In the below the accesses on regionsInTransition are<a name="line.404"></a>
-<span class="sourceLineNo">405</span>    // outside of a synchronization block where usually all accesses to RIT are<a name="line.405"></a>
-<span class="sourceLineNo">406</span>    // synchronized.  The presumption is that in this case it is safe since this<a name="line.406"></a>
-<span class="sourceLineNo">407</span>    // method is being played by a single thread on startup.<a name="line.407"></a>
-<span class="sourceLineNo">408</span><a name="line.408"></a>
-<span class="sourceLineNo">409</span>    // TODO: Regions that have a null location and are not in regionsInTransitions<a name="line.409"></a>
-<span class="sourceLineNo">410</span>    // need to be handled.<a name="line.410"></a>
-<span class="sourceLineNo">411</span><a name="line.411"></a>
-<span class="sourceLineNo">412</span>    // Scan hbase:meta to build list of existing regions, servers, and assignment<a name="line.412"></a>
-<span class="sourceLineNo">413</span>    // Returns servers who have not checked in (assumed dead) that some regions<a name="line.413"></a>
-<span class="sourceLineNo">414</span>    // were assigned to (according to the meta)<a name="line.414"></a>
-<span class="sourceLineNo">415</span>    Set&lt;ServerName&gt; deadServers = rebuildUserRegions();<a name="line.415"></a>
-<span class="sourceLineNo">416</span><a name="line.416"></a>
-<span class="sourceLineNo">417</span>    // This method will assign all user regions if a clean server startup or<a name="line.417"></a>
-<span class="sourceLineNo">418</span>    // it will reconstruct master state and cleanup any leftovers from previous master process.<a name="line.418"></a>
-<span class="sourceLineNo">419</span>    boolean failover = processDeadServersAndRegionsInTransition(deadServers);<a name="line.419"></a>
-<span class="sourceLineNo">420</span><a name="line.420"></a>
-<span class="sourceLineNo">421</span>    recoverTableInDisablingState();<a name="line.421"></a>
-<span class="sourceLineNo">422</span>    recoverTableInEnablingState();<a name="line.422"></a>
-<span class="sourceLineNo">423</span>    LOG.info("Joined the cluster in " + (System.currentTimeMillis()<a name="line.423"></a>
-<span class="sourceLineNo">424</span>      - startTime) + "ms, failover=" + failover);<a name="line.424"></a>
-<span class="sourceLineNo">425</span>  }<a name="line.425"></a>
-<span class="sourceLineNo">426</span><a name="line.426"></a>
-<span class="sourceLineNo">427</span>  /**<a name="line.427"></a>
-<span class="sourceLineNo">428</span>   * Process all regions that are in transition in zookeeper and also<a name="line.428"></a>
-<span class="sourceLineNo">429</span>   * processes the list of dead servers.<a name="line.429"></a>
-<span class="sourceLineNo">430</span>   * Used by master joining an cluster.  If we figure this is a clean cluster<a name="line.430"></a>
-<span class="sourceLineNo">431</span>   * startup, will assign all user regions.<a name="line.431"></a>
-<span class="sourceLineNo">432</span>   * @param deadServers Set of servers that are offline probably legitimately that were carrying<a name="line.432"></a>
-<span class="sourceLineNo">433</span>   * regions according to a scan of hbase:meta. Can be null.<a name="line.433"></a>
-<span class="sourceLineNo">434</span>   * @throws IOException<a name="line.434"></a>
-<span class="sourceLineNo">435</span>   * @throws InterruptedException<a name="line.435"></a>
-<span class="sourceLineNo">436</span>   */<a name="line.436"></a>
-<span class="sourceLineNo">437</span>  boolean processDeadServersAndRegionsInTransition(final Set&lt;ServerName&gt; deadServers)<a name="line.437"></a>
-<span class="sourceLineNo">438</span>  throws KeeperException, IOException, InterruptedException, CoordinatedStateException {<a name="line.438"></a>
-<span class="sourceLineNo">439</span>    // TODO Needed? List&lt;String&gt; nodes = ZKUtil.listChildrenNoWatch(watcher, watcher.assignmentZNode);<a name="line.439"></a>
-<span class="sourceLineNo">440</span>    boolean failover = !serverManager.getDeadServers().isEmpty();<a name="line.440"></a>
-<span class="sourceLineNo">441</span>    if (failover) {<a name="line.441"></a>
-<span class="sourceLineNo">442</span>      // This may not be a failover actually, especially if meta is on this master.<a name="line.442"></a>
-<span class="sourceLineNo">443</span>      if (LOG.isDebugEnabled()) {<a name="line.443"></a>
-<span class="sourceLineNo">444</span>        LOG.debug("Found dead servers out on cluster " + serverManager.getDeadServers());<a name="line.444"></a>
-<span class="sourceLineNo">445</span>      }<a name="line.445"></a>
-<span class="sourceLineNo">446</span>    } else {<a name="line.446"></a>
-<span class="sourceLineNo">447</span>      // If any one region except meta is assigned, it's a failover.<a name="line.447"></a>
-<span class="sourceLineNo">448</span>      Set&lt;ServerName&gt; onlineServers = serverManager.getOnlineServers().keySet();<a name="line.448"></a>
-<span class="sourceLineNo">449</span>      for (Map.Entry&lt;HRegionInfo, ServerName&gt; en:<a name="line.449"></a>
-<span class="sourceLineNo">450</span>          regionStates.getRegionAssignments().entrySet()) {<a name="line.450"></a>
-<span class="sourceLineNo">451</span>        HRegionInfo hri = en.getKey();<a name="line.451"></a>
-<span class="sourceLineNo">452</span>        if (!hri.isMetaTable()<a name="line.452"></a>
-<span class="sourceLineNo">453</span>            &amp;&amp; onlineServers.contains(en.getValue())) {<a name="line.453"></a>
-<span class="sourceLineNo">454</span>          LOG.debug("Found " + hri + " out on cluster");<a name="line.454"></a>
-<span class="sourceLineNo">455</span>          failover = true;<a name="line.455"></a>
-<span class="sourceLineNo">456</span>          break;<a name="line.456"></a>
-<span class="sourceLineNo">457</span>        }<a name="line.457"></a>
-<span class="sourceLineNo">458</span>      }<a name="line.458"></a>
-<span class="sourceLineNo">459</span>      if (!failover) {<a name="line.459"></a>
-<span class="sourceLineNo">460</span>        // If any region except meta is in transition on a live server, it's a failover.<a name="line.460"></a>
-<span class="sourceLineNo">461</span>        Map&lt;String, RegionState&gt; regionsInTransition = regionStates.getRegionsInTransition();<a name="line.461"></a>
-<span class="sourceLineNo">462</span>        if (!regionsInTransition.isEmpty()) {<a name="line.462"></a>
-<span class="sourceLineNo">463</span>          for (RegionState regionState: regionsInTransition.values()) {<a name="line.463"></a>
-<span class="sourceLineNo">464</span>            ServerName serverName = regionState.getServerName();<a name="line.464"></a>
-<span class="sourceLineNo">465</span>            if (!regionState.getRegion().isMetaRegion()<a name="line.465"></a>
-<span class="sourceLineNo">466</span>                &amp;&amp; serverName != null &amp;&amp; onlineServers.contains(serverName)) {<a name="line.466"></a>
-<span class="sourceLineNo">467</span>              LOG.debug("Found " + regionState + " in RITs");<a name="line.467"></a>
-<span class="sourceLineNo">468</span>              failover = true;<a name="line.468"></a>
-<span class="sourceLineNo">469</span>              break;<a name="line.469"></a>
-<span class="sourceLineNo">470</span>            }<a name="line.470"></a>
-<span class="sourceLineNo">471</span>          }<a name="line.471"></a>
-<span class="sourceLineNo">472</span>        }<a name="line.472"></a>
-<span class="sourceLineNo">473</span>      }<a name="line.473"></a>
-<span class="sourceLineNo">474</span>    }<a name="line.474"></a>
-<span class="sourceLineNo">475</span>    if (!failover) {<a name="line.475"></a>
-<span class="sourceLineNo">476</span>      // If we get here, we have a full cluster restart. It is a failover only<a name="line.476"></a>
-<span class="sourceLineNo">477</span>      // if there are some WALs are not split yet. For meta WALs, they should have<a name="line.477"></a>
-<span class="sourceLineNo">478</span>      // been split already, if any. We can walk through those queued dead servers,<a name="line.478"></a>
-<span class="sourceLineNo">479</span>      // if they don't have any WALs, this restart should be considered as a clean one<a name="line.479"></a>
-<span class="sourceLineNo">480</span>      Set&lt;ServerName&gt; queuedDeadServers = serverManager.getRequeuedDeadServers().keySet();<a name="line.480"></a>
-<span class="sourceLineNo">481</span>      if (!queuedDeadServers.isEmpty()) {<a name="line.481"></a>
-<span class="sourceLineNo">482</span>        Configuration conf = server.getConfiguration();<a name="line.482"></a>
-<span class="sourceLineNo">483</span>        Path rootdir = FSUtils.getRootDir(conf);<a name="line.483"></a>
-<span class="sourceLineNo">484</span>        FileSystem fs = rootdir.getFileSystem(conf);<a name="line.484"></a>
-<span class="sourceLineNo">485</span>        for (ServerName serverName: queuedDeadServers) {<a name="line.485"></a>
-<span class="sourceLineNo">486</span>          // In the case of a clean exit, the shutdown handler would have presplit any WALs and<a name="line.486"></a>
-<span class="sourceLineNo">487</span>          // removed empty directories.<a name="line.487"></a>
-<span class="sourceLineNo">488</span>          Path logDir = new Path(rootdir,<a name="line.488"></a>
-<span class="sourceLineNo">489</span>              DefaultWALProvider.getWALDirectoryName(serverName.toString()));<a name="line.489"></a>
-<span class="sourceLineNo">490</span>          Path splitDir = logDir.suffix(DefaultWALProvider.SPLITTING_EXT);<a name="line.490"></a>
-<span class="sourceLineNo">491</span>          if (fs.exists(logDir) || fs.exists(splitDir)) {<a name="line.491"></a>
-<span class="sourceLineNo">492</span>            LOG.debug("Found queued dead server " + serverName);<a name="line.492"></a>
-<span class="sourceLineNo">493</span>            failover = true;<a name="line.493"></a>
-<span class="sourceLineNo">494</span>            break;<a name="line.494"></a>
-<span class="sourceLineNo">495</span>          }<a name="line.495"></a>
-<span class="sourceLineNo">496</span>        }<a name="line.496"></a>
-<span class="sourceLineNo">497</span>        if (!failover) {<a name="line.497"></a>
-<span class="sourceLineNo">498</span>          // We figured that it's not a failover, so no need to<a name="line.498"></a>
-<span class="sourceLineNo">499</span>          // work on these re-queued dead servers any more.<a name="line.499"></a>
-<span class="sourceLineNo">500</span>          LOG.info("AM figured that it's not a failover and cleaned up "<a name="line.500"></a>
-<span class="sourceLineNo">501</span>            + queuedDeadServers.size() + " queued dead servers");<a name="line.501"></a>
-<span class="sourceLineNo">502</span>          serverManager.removeRequeuedDeadServers();<a name="line.502"></a>
-<span class="sourceLineNo">503</span>        }<a name="line.503"></a>
-<span class="sourceLineNo">504</span>      }<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>    Set&lt;TableName&gt; disabledOrDisablingOrEnabling = null;<a name="line.507"></a>
-<span class="sourceLineNo">508</span>    Map&lt;HRegionInfo, ServerName&gt; allRegions = null;<a name="line.508"></a>
-<span class="sourceLineNo">509</span><a name="line.509"></a>
-<span class="sourceLineNo">510</span>    if (!failover) {<a name="line.510"></a>
-<span class="sourceLineNo">511</span>      disabledOrDisablingOrEnabling = tableStateManager.getTablesInStates(<a name="line.511"></a>
-<span class="sourceLineNo">512</span>        TableState.State.DISABLED, TableState.State.DISABLING,<a name="line.512"></a>
-<span class="sourceLineNo">513</span>        TableState.State.ENABLING);<a name="line.513"></a>
-<span class="sourceLineNo">514</span><a name="line.514"></a>
-<span class="sourceLineNo">515</span>      // Clean re/start, mark all user regions closed before reassignment<a name="line.515"></a>
-<span class="sourceLineNo">516</span>      allRegions = regionStates.closeAllUserRegions(<a name="line.516"></a>
-<span class="sourceLineNo">517</span>        disabledOrDisablingOrEnabling);<a name="line.517"></a>
+<span class="sourceLineNo">021</span>import com.google.common.annotations.VisibleForTesting;<a name="line.21"></a>
+<span class="sourceLineNo">022</span><a name="line.22"></a>
+<span class="sourceLineNo">023</span>import java.io.IOException;<a name="line.23"></a>
+<span class="sourceLineNo">024</span>import java.util.ArrayList;<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.Collections;<a name="line.26"></a>
+<span class="sourceLineNo">027</span>import java.util.HashMap;<a name="line.27"></a>
+<span class="sourceLineNo">028</span>import java.util.HashSet;<a name="line.28"></a>
+<span class="sourceLineNo">029</span>import java.util.Iterator;<a name="line.29"></a>
+<span class="sourceLineNo">030</span>import java.util.List;<a name="line.30"></a>
+<span class="sourceLineNo">031</span>import java.util.Map;<a name="line.31"></a>
+<span class="sourceLineNo">032</span>import java.util.NavigableMap;<a name="line.32"></a>
+<span class="sourceLineNo">033</span>import java.util.Random;<a name="line.33"></a>
+<span class="sourceLineNo">034</span>import java.util.Set;<a name="line.34"></a>
+<span class="sourceLineNo">035</span>import java.util.TreeMap;<a name="line.35"></a>
+<span class="sourceLineNo">036</span>import java.util.concurrent.Callable;<a name="line.36"></a>
+<span class="sourceLineNo">037</span>import java.util.concurrent.ConcurrentHashMap;<a name="line.37"></a>
+<span class="sourceLineNo">038</span>import java.util.concurrent.CopyOnWriteArrayList;<a name="line.38"></a>
+<span class="sourceLineNo">039</span>import java.util.concurrent.TimeUnit;<a name="line.39"></a>
+<span class="sourceLineNo">040</span>import java.util.concurrent.atomic.AtomicBoolean;<a name="line.40"></a>
+<span class="sourceLineNo">041</span>import java.util.concurrent.atomic.AtomicInteger;<a name="line.41"></a>
+<span class="sourceLineNo">042</span>import java.util.concurrent.locks.Lock;<a name="line.42"></a>
+<span class="sourceLineNo">043</span>import java.util.concurrent.locks.ReentrantLock;<a name="line.43"></a>
+<span class="sourceLineNo">044</span><a name="line.44"></a>
+<span class="sourceLineNo">045</span>import org.apache.commons.logging.Log;<a name="line.45"></a>
+<span class="sourceLineNo">046</span>import org.apache.commons.logging.LogFactory;<a name="line.46"></a>
+<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.classification.InterfaceAudience;<a name="line.47"></a>
+<span class="sourceLineNo">048</span>import org.apache.hadoop.conf.Configuration;<a name="line.48"></a>
+<span class="sourceLineNo">049</span>import org.apache.hadoop.fs.FileStatus;<a name="line.49"></a>
+<span class="sourceLineNo">050</span>import org.apache.hadoop.fs.FileSystem;<a name="line.50"></a>
+<span class="sourceLineNo">051</span>import org.apache.hadoop.fs.Path;<a name="line.51"></a>
+<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.CoordinatedStateException;<a name="line.52"></a>
+<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.HBaseIOException;<a name="line.53"></a>
+<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.HConstants;<a name="line.54"></a>
+<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.HRegionInfo;<a name="line.55"></a>
+<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.HRegionLocation;<a name="line.56"></a>
+<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.HTableDescriptor;<a name="line.57"></a>
+<span class="sourceLineNo">058</span>import org.apache.hadoop.hbase.MetaTableAccessor;<a name="line.58"></a>
+<span class="sourceLineNo">059</span>import org.apache.hadoop.hbase.NotServingRegionException;<a name="line.59"></a>
+<span class="sourceLineNo">060</span>import org.apache.hadoop.hbase.RegionLocations;<a name="line.60"></a>
+<span class="sourceLineNo">061</span>import org.apache.hadoop.hbase.RegionStateListener;<a name="line.61"></a>
+<span class="sourceLineNo">062</span>import org.apache.hadoop.hbase.ServerName;<a name="line.62"></a>
+<span class="sourceLineNo">063</span>import org.apache.hadoop.hbase.TableName;<a name="line.63"></a>
+<span class="sourceLineNo">064</span>import org.apache.hadoop.hbase.TableNotFoundException;<a name="line.64"></a>
+<span class="sourceLineNo">065</span>import org.apache.hadoop.hbase.client.RegionReplicaUtil;<a name="line.65"></a>
+<span class="sourceLineNo">066</span>import org.apache.hadoop.hbase.client.Result;<a name="line.66"></a>
+<span class="sourceLineNo">067</span>import org.apache.hadoop.hbase.client.TableState;<a name="line.67"></a>
+<span class="sourceLineNo">068</span>import org.apache.hadoop.hbase.executor.EventHandler;<a name="line.68"></a>
+<span class="sourceLineNo">069</span>import org.apache.hadoop.hbase.executor.EventType;<a name="line.69"></a>
+<span class="sourceLineNo">070</span>import org.apache.hadoop.hbase.executor.ExecutorService;<a name="line.70"></a>
+<span class="sourceLineNo">071</span>import org.apache.hadoop.hbase.ipc.FailedServerException;<a name="line.71"></a>
+<span class="sourceLineNo">072</span>import org.apache.hadoop.hbase.ipc.RpcClient;<a name="line.72"></a>
+<span class="sourceLineNo">073</span>import org.apache.hadoop.hbase.ipc.ServerNotRunningYetException;<a name="line.73"></a>
+<span class="sourceLineNo">074</span>import org.apache.hadoop.hbase.master.RegionState.State;<a name="line.74"></a>
+<span class="sourceLineNo">075</span>import org.apache.hadoop.hbase.master.balancer.FavoredNodeAssignmentHelper;<a name="line.75"></a>
+<span class="sourceLineNo">076</span>import org.apache.hadoop.hbase.master.balancer.FavoredNodeLoadBalancer;<a name="line.76"></a>
+<span class="sourceLineNo">077</span>import org.apache.hadoop.hbase.master.handler.DisableTableHandler;<a name="line.77"></a>
+<span class="sourceLineNo">078</span>import org.apache.hadoop.hbase.master.handler.EnableTableHandler;<a name="line.78"></a>
+<span class="sourceLineNo">079</span>import org.apache.hadoop.hbase.master.normalizer.NormalizationPlan.PlanType;<a name="line.79"></a>
+<span class="sourceLineNo">080</span>import org.apache.hadoop.hbase.protobuf.generated.RegionServerStatusProtos.RegionStateTransition;<a name="line.80"></a>
+<span class="sourceLineNo">081</span>import org.apache.hadoop.hbase.protobuf.generated.RegionServerStatusProtos.RegionStateTransition.TransitionCode;<a name="line.81"></a>
+<span class="sourceLineNo">082</span>import org.apache.hadoop.hbase.quotas.QuotaExceededException;<a name="line.82"></a>
+<span class="sourceLineNo">083</span>import org.apache.hadoop.hbase.regionserver.RegionOpeningState;<a name="line.83"></a>
+<span class="sourceLineNo">084</span>import org.apache.hadoop.hbase.regionserver.RegionServerAbortedException;<a name="line.84"></a>
+<span class="sourceLineNo">085</span>import org.apache.hadoop.hbase.regionserver.RegionServerStoppedException;<a name="line.85"></a>
+<span class="sourceLineNo">086</span>import org.apache.hadoop.hbase.wal.DefaultWALProvider;<a name="line.86"></a>
+<span class="sourceLineNo">087</span>import org.apache.hadoop.hbase.util.EnvironmentEdgeManager;<a name="line.87"></a>
+<span class="sourceLineNo">088</span>import org.apache.hadoop.hbase.util.FSUtils;<a name="line.88"></a>
+<span class="sourceLineNo">089</span>import org.apache.hadoop.hbase.util.KeyLocker;<a name="line.89"></a>
+<span class="sourceLineNo">090</span>import org.apache.hadoop.hbase.util.Pair;<a name="line.90"></a>
+<span class="sourceLineNo">091</span>import org.apache.hadoop.hbase.util.PairOfSameType;<a name="line.91"></a>
+<span class="sourceLineNo">092</span>import org.apache.hadoop.hbase.util.Threads;<a name="line.92"></a>
+<span class="sourceLineNo">093</span>import org.apache.hadoop.hbase.zookeeper.MetaTableLocator;<a name="line.93"></a>
+<span class="sourceLineNo">094</span>import org.apache.hadoop.ipc.RemoteException;<a name="line.94"></a>
+<span class="sourceLineNo">095</span>import org.apache.hadoop.util.StringUtils;<a name="line.95"></a>
+<span class="sourceLineNo">096</span>import org.apache.zookeeper.KeeperException;<a name="line.96"></a>
+<span class="sourceLineNo">097</span><a name="line.97"></a>
+<span class="sourceLineNo">098</span>/**<a name="line.98"></a>
+<span class="sourceLineNo">099</span> * Manages and performs region assignment.<a name="line.99"></a>
+<span class="sourceLineNo">100</span> * Related communications with regionserver are all done over RPC.<a name="line.100"></a>
+<span class="sourceLineNo">101</span> */<a name="line.101"></a>
+<span class="sourceLineNo">102</span>@InterfaceAudience.Private<a name="line.102"></a>
+<span class="sourceLineNo">103</span>public class AssignmentManager {<a name="line.103"></a>
+<span class="sourceLineNo">104</span>  private static final Log LOG = LogFactory.getLog(AssignmentManager.class);<a name="line.104"></a>
+<span class="sourceLineNo">105</span><a name="line.105"></a>
+<span class="sourceLineNo">106</span>  protected final MasterServices server;<a name="line.106"></a>
+<span class="sourceLineNo">107</span><a name="line.107"></a>
+<span class="sourceLineNo">108</span>  private ServerManager serverManager;<a name="line.108"></a>
+<span class="sourceLineNo">109</span><a name="line.109"></a>
+<span class="sourceLineNo">110</span>  private boolean shouldAssignRegionsWithFavoredNodes;<a name="line.110"></a>
+<span class="sourceLineNo">111</span><a name="line.111"></a>
+<span class="sourceLineNo">112</span>  private LoadBalancer balancer;<a name="line.112"></a>
+<span class="sourceLineNo">113</span><a name="line.113"></a>
+<span class="sourceLineNo">114</span>  private final MetricsAssignmentManager metricsAssignmentManager;<a name="line.114"></a>
+<span class="sourceLineNo">115</span><a name="line.115"></a>
+<span class="sourceLineNo">116</span>  private final TableLockManager tableLockManager;<a name="line.116"></a>
+<span class="sourceLineNo">117</span><a name="line.117"></a>
+<span class="sourceLineNo">118</span>  private AtomicInteger numRegionsOpened = new AtomicInteger(0);<a name="line.118"></a>
+<span class="sourceLineNo">119</span><a name="line.119"></a>
+<span class="sourceLineNo">120</span>  final private KeyLocker&lt;String&gt; locker = new KeyLocker&lt;String&gt;();<a name="line.120"></a>
+<span class="sourceLineNo">121</span><a name="line.121"></a>
+<span class="sourceLineNo">122</span>  Set&lt;HRegionInfo&gt; replicasToClose = Collections.synchronizedSet(new HashSet&lt;HRegionInfo&gt;());<a name="line.122"></a>
+<span class="sourceLineNo">123</span><a name="line.123"></a>
+<span class="sourceLineNo">124</span>  /**<a name="line.124"></a>
+<span class="sourceLineNo">125</span>   * Map of regions to reopen after the schema of a table is changed. Key -<a name="line.125"></a>
+<span class="sourceLineNo">126</span>   * encoded region name, value - HRegionInfo<a name="line.126"></a>
+<span class="sourceLineNo">127</span>   */<a name="line.127"></a>
+<span class="sourceLineNo">128</span>  private final Map &lt;String, HRegionInfo&gt; regionsToReopen;<a name="line.128"></a>
+<span class="sourceLineNo">129</span><a name="line.129"></a>
+<span class="sourceLineNo">130</span>  /*<a name="line.130"></a>
+<span class="sourceLineNo">131</span>   * Maximum times we recurse an assignment/unassignment.<a name="line.131"></a>
+<span class="sourceLineNo">132</span>   * See below in {@link #assign()} and {@link #unassign()}.<a name="line.132"></a>
+<span class="sourceLineNo">133</span>   */<a name="line.133"></a>
+<span class="sourceLineNo">134</span>  private final int maximumAttempts;<a name="line.134"></a>
+<span class="sourceLineNo">135</span><a name="line.135"></a>
+<span class="sourceLineNo">136</span>  /**<a name="line.136"></a>
+<span class="sourceLineNo">137</span>   * The sleep time for which the assignment will wait before retrying in case of<a name="line.137"></a>
+<span class="sourceLineNo">138</span>   * hbase:meta assignment failure due to lack of availability of region plan or bad region plan<a name="line.138"></a>
+<span class="sourceLineNo">139</span>   */<a name="line.139"></a>
+<span class="sourceLineNo">140</span>  private final long sleepTimeBeforeRetryingMetaAssignment;<a name="line.140"></a>
+<span class="sourceLineNo">141</span><a name="line.141"></a>
+<span class="sourceLineNo">142</span>  /** Plans for region movement. Key is the encoded version of a region name*/<a name="line.142"></a>
+<span class="sourceLineNo">143</span>  // TODO: When do plans get cleaned out?  Ever? In server open and in server<a name="line.143"></a>
+<span class="sourceLineNo">144</span>  // shutdown processing -- St.Ack<a name="line.144"></a>
+<span class="sourceLineNo">145</span>  // All access to this Map must be synchronized.<a name="line.145"></a>
+<span class="sourceLineNo">146</span>  final NavigableMap&lt;String, RegionPlan&gt; regionPlans =<a name="line.146"></a>
+<span class="sourceLineNo">147</span>    new TreeMap&lt;String, RegionPlan&gt;();<a name="line.147"></a>
+<span class="sourceLineNo">148</span><a name="line.148"></a>
+<span class="sourceLineNo">149</span>  private final TableStateManager tableStateManager;<a name="line.149"></a>
+<span class="sourceLineNo">150</span><a name="line.150"></a>
+<span class="sourceLineNo">151</span>  private final ExecutorService executorService;<a name="line.151"></a>
+<span class="sourceLineNo">152</span><a name="line.152"></a>
+<span class="sourceLineNo">153</span>  // Thread pool executor service. TODO, consolidate with executorService?<a name="line.153"></a>
+<span class="sourceLineNo">154</span>  private java.util.concurrent.ExecutorService threadPoolExecutorService;<a name="line.154"></a>
+<span class="sourceLineNo">155</span><a name="line.155"></a>
+<span class="sourceLineNo">156</span>  private final RegionStates regionStates;<a name="line.156"></a>
+<span class="sourceLineNo">157</span><a name="line.157"></a>
+<span class="sourceLineNo">158</span>  // The threshold to use bulk assigning. Using bulk assignment<a name="line.158"></a>
+<span class="sourceLineNo">159</span>  // only if assigning at least this many regions to at least this<a name="line.159"></a>
+<span class="sourceLineNo">160</span>  // many servers. If assigning fewer regions to fewer servers,<a name="line.160"></a>
+<span class="sourceLineNo">161</span>  // bulk assigning may be not as efficient.<a name="line.161"></a>
+<span class="sourceLineNo">162</span>  private final int bulkAssignThresholdRegions;<a name="line.162"></a>
+<span class="sourceLineNo">163</span>  private final int bulkAssignThresholdServers;<a name="line.163"></a>
+<span class="sourceLineNo">164</span>  private final int bulkPerRegionOpenTimeGuesstimate;<a name="line.164"></a>
+<span class="sourceLineNo">165</span><a name="line.165"></a>
+<span class="sourceLineNo">166</span>  // Should bulk assignment wait till all regions are assigned,<a name="line.166"></a>
+<span class="sourceLineNo">167</span>  // or it is timed out?  This is useful to measure bulk assignment<a name="line.167"></a>
+<span class="sourceLineNo">168</span>  // performance, but not needed in most use cases.<a name="line.168"></a>
+<span class="sourceLineNo">169</span>  private final boolean bulkAssignWaitTillAllAssigned;<a name="line.169"></a>
+<span class="sourceLineNo">170</span><a name="line.170"></a>
+<span class="sourceLineNo">171</span>  /**<a name="line.171"></a>
+<span class="sourceLineNo">172</span>   * Indicator that AssignmentManager has recovered the region states so<a name="line.172"></a>
+<span class="sourceLineNo">173</span>   * that ServerShutdownHandler can be fully enabled and re-assign regions<a name="line.173"></a>
+<span class="sourceLineNo">174</span>   * of dead servers. So that when re-assignment happens, AssignmentManager<a name="line.174"></a>
+<span class="sourceLineNo">175</span>   * has proper region states.<a name="line.175"></a>
+<span class="sourceLineNo">176</span>   *<a name="line.176"></a>
+<span class="sourceLineNo">177</span>   * Protected to ease testing.<a name="line.177"></a>
+<span class="sourceLineNo">178</span>   */<a name="line.178"></a>
+<span class="sourceLineNo">179</span>  protected final AtomicBoolean failoverCleanupDone = new AtomicBoolean(false);<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>   * A map to track the count a region fails to open in a row.<a name="line.182"></a>
+<span class="sourceLineNo">183</span>   * So that we don't try to open a region forever if the failure is<a name="line.183"></a>
+<span class="sourceLineNo">184</span>   * unrecoverable.  We don't put this information in region states<a name="line.184"></a>
+<span class="sourceLineNo">185</span>   * because we don't expect this to happen frequently; we don't<a name="line.185"></a>
+<span class="sourceLineNo">186</span>   * want to copy this information over during each state transition either.<a name="line.186"></a>
+<span class="sourceLineNo">187</span>   */<a name="line.187"></a>
+<span class="sourceLineNo">188</span>  private final ConcurrentHashMap&lt;String, AtomicInteger&gt;<a name="line.188"></a>
+<span class="sourceLineNo">189</span>    failedOpenTracker = new ConcurrentHashMap&lt;String, AtomicInteger&gt;();<a name="line.189"></a>
+<span class="sourceLineNo">190</span><a name="line.190"></a>
+<span class="sourceLineNo">191</span>  // In case not using ZK for region assignment, region states<a name="line.191"></a>
+<span class="sourceLineNo">192</span>  // are persisted in meta with a state store<a name="line.192"></a>
+<span class="sourceLineNo">193</span>  private final RegionStateStore regionStateStore;<a name="line.193"></a>
+<span class="sourceLineNo">194</span><a name="line.194"></a>
+<span class="sourceLineNo">195</span>  /**<a name="line.195"></a>
+<span class="sourceLineNo">196</span>   * For testing only!  Set to true to skip handling of split.<a name="line.196"></a>
+<span class="sourceLineNo">197</span>   */<a name="line.197"></a>
+<span class="sourceLineNo">198</span>  @edu.umd.cs.findbugs.annotations.SuppressWarnings(value="MS_SHOULD_BE_FINAL")<a name="line.198"></a>
+<span class="sourceLineNo">199</span>  public static boolean TEST_SKIP_SPLIT_HANDLING = false;<a name="line.199"></a>
+<span class="sourceLineNo">200</span><a name="line.200"></a>
+<span class="sourceLineNo">201</span>  /** Listeners that are called on assignment events. */<a name="line.201"></a>
+<span class="sourceLineNo">202</span>  private List&lt;AssignmentListener&gt; listeners = new CopyOnWriteArrayList&lt;AssignmentListener&gt;();<a name="line.202"></a>
+<span class="sourceLineNo">203</span><a name="line.203"></a>
+<span class="sourceLineNo">204</span>  private RegionStateListener regionStateListener;<a name="line.204"></a>
+<span class="sourceLineNo">205</span><a name="line.205"></a>
+<span class="sourceLineNo">206</span>  /**<a name="line.206"></a>
+<span class="sourceLineNo">207</span>   * Constructs a new assignment manager.<a name="line.207"></a>
+<span class="sourceLineNo">208</span>   *<a name="line.208"></a>
+<span class="sourceLineNo">209</span>   * @param server instance of HMaster this AM running inside<a name="line.209"></a>
+<span class="sourceLineNo">210</span>   * @param serverManager serverManager for associated HMaster<a name="line.210"></a>
+<span class="sourceLineNo">211</span>   * @param balancer implementation of {@link LoadBalancer}<a name="line.211"></a>
+<span class="sourceLineNo">212</span>   * @param service Executor service<a name="line.212"></a>
+<span class="sourceLineNo">213</span>   * @param metricsMaster metrics manager<a name="line.213"></a>
+<span class="sourceLineNo">214</span>   * @param tableLockManager TableLock manager<a name="line.214"></a>
+<span class="sourceLineNo">215</span>   * @throws IOException<a name="line.215"></a>
+<span class="sourceLineNo">216</span>   */<a name="line.216"></a>
+<span class="sourceLineNo">217</span>  public AssignmentManager(MasterServices server, ServerManager serverManager,<a name="line.217"></a>
+<span class="sourceLineNo">218</span>      final LoadBalancer balancer,<a name="line.218"></a>
+<span class="sourceLineNo">219</span>      final ExecutorService service, MetricsMaster metricsMaster,<a name="line.219"></a>
+<span class="sourceLineNo">220</span>      final TableLockManager tableLockManager,<a name="line.220"></a>
+<span class="sourceLineNo">221</span>      final TableStateManager tableStateManager)<a name="line.221"></a>
+<span class="sourceLineNo">222</span>          throws IOException {<a name="line.222"></a>
+<span class="sourceLineNo">223</span>    this.server = server;<a name="line.223"></a>
+<span class="sourceLineNo">224</span>    this.serverManager = serverManager;<a name="line.224"></a>
+<span class="sourceLineNo">225</span>    this.executorService = service;<a name="line.225"></a>
+<span class="sourceLineNo">226</span>    this.regionStateStore = new RegionStateStore(server);<a name="line.226"></a>
+<span class="sourceLineNo">227</span>    this.regionsToReopen = Collections.synchronizedMap<a name="line.227"></a>
+<span class="sourceLineNo">228</span>                           (new HashMap&lt;String, HRegionInfo&gt; ());<a name="line.228"></a>
+<span class="sourceLineNo">229</span>    Configuration conf = server.getConfiguration();<a name="line.229"></a>
+<span class="sourceLineNo">230</span>    // Only read favored nodes if using the favored nodes load balancer.<a name="line.230"></a>
+<span class="sourceLineNo">231</span>    this.shouldAssignRegionsWithFavoredNodes = conf.getClass(<a name="line.231"></a>
+<span class="sourceLineNo">232</span>           HConstants.HBASE_MASTER_LOADBALANCER_CLASS, Object.class).equals(<a name="line.232"></a>
+<span class="sourceLineNo">233</span>           FavoredNodeLoadBalancer.class);<a name="line.233"></a>
+<span class="sourceLineNo">234</span><a name="line.234"></a>
+<span class="sourceLineNo">235</span>    this.tableStateManager = tableStateManager;<a name="line.235"></a>
+<span class="sourceLineNo">236</span><a name="line.236"></a>
+<span class="sourceLineNo">237</span>    // This is the max attempts, not retries, so it should be at least 1.<a name="line.237"></a>
+<span class="sourceLineNo">238</span>    this.maximumAttempts = Math.max(1,<a name="line.238"></a>
+<span class="sourceLineNo">239</span>      this.server.getConfiguration().getInt("hbase.assignment.maximum.attempts", 10));<a name="line.239"></a>
+<span class="sourceLineNo">240</span>    this.sleepTimeBeforeRetryingMetaAssignment = this.server.getConfiguration().getLong(<a name="line.240"></a>
+<span class="sourceLineNo">241</span>        "hbase.meta.assignment.retry.sleeptime", 1000l);<a name="line.241"></a>
+<span class="sourceLineNo">242</span>    this.balancer = balancer;<a name="line.242"></a>
+<span class="sourceLineNo">243</span>    int maxThreads = conf.getInt("hbase.assignment.threads.max", 30);<a name="line.243"></a>
+<span class="sourceLineNo">244</span>    this.threadPoolExecutorService = Threads.getBoundedCachedThreadPool(<a name="line.244"></a>
+<span class="sourceLineNo">245</span>      maxThreads, 60L, TimeUnit.SECONDS, Threads.newDaemonThreadFactory("AM."));<a name="line.245"></a>
+<span class="sourceLineNo">246</span>    this.regionStates = new RegionStates(<a name="line.246"></a>
+<span class="sourceLineNo">247</span>      server, tableStateManager, serverManager, regionStateStore);<a name="line.247"></a>
+<span class="sourceLineNo">248</span><a name="line.248"></a>
+<span class="sourceLineNo">249</span>    this.bulkAssignWaitTillAllAssigned =<a name="line.249"></a>
+<span class="sourceLineNo">250</span>      conf.getBoolean("hbase.bulk.assignment.waittillallassigned", false);<a name="line.250"></a>
+<span class="sourceLineNo">251</span>    this.bulkAssignThresholdRegions = conf.getInt("hbase.bulk.assignment.threshold.regions", 7);<a name="line.251"></a>
+<span class="sourceLineNo">252</span>    this.bulkAssignThresholdServers = conf.getInt("hbase.bulk.assignment.threshold.servers", 3);<a name="line.252"></a>
+<span class="sourceLineNo">253</span>    this.bulkPerRegionOpenTimeGuesstimate =<a name="line.253"></a>
+<span class="sourceLineNo">254</span>      conf.getInt("hbase.bulk.assignment.perregion.open.time", 10000);<a name="line.254"></a>
+<span class="sourceLineNo">255</span><a name="line.255"></a>
+<span class="sourceLineNo">256</span>    this.metricsAssignmentManager = new MetricsAssignmentManager();<a name="line.256"></a>
+<span class="sourceLineNo">257</span>    this.tableLockManager = tableLockManager;<a name="line.257"></a>
+<span class="sourceLineNo">258</span>  }<a name="line.258"></a>
+<span class="sourceLineNo">259</span><a name="line.259"></a>
+<span class="sourceLineNo">260</span>  /**<a name="line.260"></a>
+<span class="sourceLineNo">261</span>   * Add the listener to the notification list.<a name="line.261"></a>
+<span class="sourceLineNo">262</span>   * @param listener The AssignmentListener to register<a name="line.262"></a>
+<span class="sourceLineNo">263</span>   */<a name="line.263"></a>
+<span class="sourceLineNo">264</span>  public void registerListener(final AssignmentListener listener) {<a name="line.264"></a>
+<span class="sourceLineNo">265</span>    this.listeners.add(listener);<a name="line.265"></a>
+<span class="sourceLineNo">266</span>  }<a name="line.266"></a>
+<span class="sourceLineNo">267</span><a name="line.267"></a>
+<span class="sourceLineNo">268</span>  /**<a name="line.268"></a>
+<span class="sourceLineNo">269</span>   * Remove the listener from the notification list.<a name="line.269"></a>
+<span class="sourceLineNo">270</span>   * @param listener The AssignmentListener to unregister<a name="line.270"></a>
+<span class="sourceLineNo">271</span>   */<a name="line.271"></a>
+<span class="sourceLineNo">272</span>  public boolean unregisterListener(final AssignmentListener listener) {<a name="line.272"></a>
+<span class="sourceLineNo">273</span>    return this.listeners.remove(listener);<a name="line.273"></a>
+<span class="sourceLineNo">274</span>  }<a name="line.274"></a>
+<span class="sourceLineNo">275</span><a name="line.275"></a>
+<span class="sourceLineNo">276</span>  /**<a name="line.276"></a>
+<span class="sourceLineNo">277</span>   * @return Instance of ZKTableStateManager.<a name="line.277"></a>
+<span class="sourceLineNo">278</span>   */<a name="line.278"></a>
+<span class="sourceLineNo">279</span>  public TableStateManager getTableStateManager() {<a name="line.279"></a>
+<span class="sourceLineNo">280</span>    // These are 'expensive' to make involving trip to zk ensemble so allow<a name="line.280"></a>
+<span class="sourceLineNo">281</span>    // sharing.<a name="line.281"></a>
+<span class="sourceLineNo">282</span>    return this.tableStateManager;<a name="line.282"></a>
+<span class="sourceLineNo">283</span>  }<a name="line.283"></a>
+<span class="sourceLineNo">284</span><a name="line.284"></a>
+<span class="sourceLineNo">285</span>  /**<a name="line.285"></a>
+<span class="sourceLineNo">286</span>   * This SHOULD not be public. It is public now<a name="line.286"></a>
+<span class="sourceLineNo">287</span>   * because of some unit tests.<a name="line.287"></a>
+<span class="sourceLineNo">288</span>   *<a name="line.288"></a>
+<span class="sourceLineNo">289</span>   * TODO: make it package private and keep RegionStates in the master package<a name="line.289"></a>
+<span class="sourceLineNo">290</span>   */<a name="line.290"></a>
+<span class="sourceLineNo">291</span>  public RegionStates getRegionStates() {<a name="line.291"></a>
+<span class="sourceLineNo">292</span>    return regionStates;<a name="line.292"></a>
+<span class="sourceLineNo">293</span>  }<a name="line.293"></a>
+<span class="sourceLineNo">294</span><a name="line.294"></a>
+<span class="sourceLineNo">295</span>  /**<a name="line.295"></a>
+<span class="sourceLineNo">296</span>   * Used in some tests to mock up region state in meta<a name="line.296"></a>
+<span class="sourceLineNo">297</span>   */<a name="line.297"></a>
+<span class="sourceLineNo">298</span>  @VisibleForTesting<a name="line.298"></a>
+<span class="sourceLineNo">299</span>  RegionStateStore getRegionStateStore() {<a name="line.299"></a>
+<span class="sourceLineNo">300</span>    return regionStateStore;<a name="line.300"></a>
+<span class="sourceLineNo">301</span>  }<a name="line.301"></a>
+<span class="sourceLineNo">302</span><a name="line.302"></a>
+<span class="sourceLineNo">303</span>  public RegionPlan getRegionReopenPlan(HRegionInfo hri) {<a name="line.303"></a>
+<span class="sourceLineNo">304</span>    return new RegionPlan(hri, null, regionStates.getRegionServerOfRegion(hri));<a name="line.304"></a>
+<span class="sourceLineNo">305</span>  }<a name="line.305"></a>
+<span class="sourceLineNo">306</span><a name="line.306"></a>
+<span class="sourceLineNo">307</span>  /**<a name="line.307"></a>
+<span class="sourceLineNo">308</span>   * Add a regionPlan for the specified region.<a name="line.308"></a>
+<span class="sourceLineNo">309</span>   * @param encodedName<a name="line.309"></a>
+<span class="sourceLineNo">310</span>   * @param plan<a name="line.310"></a>
+<span class="sourceLineNo">311</span>   */<a name="line.311"></a>
+<span class="sourceLineNo">312</span>  public void addPlan(String encodedName, RegionPlan plan) {<a name="line.312"></a>
+<span class="sourceLineNo">313</span>    synchronized (regionPlans) {<a name="line.313"></a>
+<span class="sourceLineNo">314</span>      regionPlans.put(encodedName, plan);<a name="line.314"></a>
+<span class="sourceLineNo">315</span>    }<a name="line.315"></a>
+<span class="sourceLineNo">316</span>  }<a name="line.316"></a>
+<span class="sourceLineNo">317</span><a name="line.317"></a>
+<span class="sourceLineNo">318</span>  /**<a name="line.318"></a>
+<span class="sourceLineNo">319</span>   * Add a map of region plans.<a name="line.319"></a>
+<span class="sourceLineNo">320</span>   */<a name="line.320"></a>
+<span class="sourceLineNo">321</span>  public void addPlans(Map&lt;String, RegionPlan&gt; plans) {<a name="line.321"></a>
+<span class="sourceLineNo">322</span>    synchronized (regionPlans) {<a name="line.322"></a>
+<span class="sourceLineNo">323</span>      regionPlans.putAll(plans);<a name="line.323"></a>
+<span class="sourceLineNo">324</span>    }<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>   * Set the list of regions that will be reopened<a name="line.328"></a>
+<span class="sourceLineNo">329</span>   * because of an update in table schema<a name="line.329"></a>
+<span class="sourceLineNo">330</span>   *<a name="line.330"></a>
+<span class="sourceLineNo">331</span>   * @param regions<a name="line.331"></a>
+<span class="sourceLineNo">332</span>   *          list of regions that should be tracked for reopen<a name="line.332"></a>
+<span class="sourceLineNo">333</span>   */<a name="line.333"></a>
+<span class="sourceLineNo">334</span>  public void setRegionsToReopen(List &lt;HRegionInfo&gt; regions) {<a name="line.334"></a>
+<span class="sourceLineNo">335</span>    for(HRegionInfo hri : regions) {<a name="line.335"></a>
+<span class="sourceLineNo">336</span>      regionsToReopen.put(hri.getEncodedName(), hri);<a name="line.336"></a>
+<span class="sourceLineNo">337</span>    }<a name="line.337"></a>
+<span class="sourceLineNo">338</span>  }<a name="line.338"></a>
+<span class="sourceLineNo">339</span><a name="line.339"></a>
+<span class="sourceLineNo">340</span>  /**<a name="line.340"></a>
+<span class="sourceLineNo">341</span>   * Used by the client to identify if all regions have the schema updates<a name="line.341"></a>
+<span class="sourceLineNo">342</span>   *<a name="line.342"></a>
+<span class="sourceLineNo">343</span>   * @param tableName<a name="line.343"></a>
+<span class="sourceLineNo">344</span>   * @return Pair indicating the status of the alter command<a name="line.344"></a>
+<span class="sourceLineNo">345</span>   * @throws IOException<a name="line.345"></a>
+<span class="sourceLineNo">346</span>   */<a name="line.346"></a>
+<span class="sourceLineNo">347</span>  public Pair&lt;Integer, Integer&gt; getReopenStatus(TableName tableName)<a name="line.347"></a>
+<span class="sourceLineNo">348</span>      throws IOException {<a name="line.348"></a>
+<span class="sourceLineNo">349</span>    List&lt;HRegionInfo&gt; hris;<a name="line.349"></a>
+<span class="sourceLineNo">350</span>    if (TableName.META_TABLE_NAME.equals(tableName)) {<a name="line.350"></a>
+<span class="sourceLineNo">351</span>      hris = new MetaTableLocator().getMetaRegions(server.getZooKeeper());<a name="line.351"></a>
+<span class="sourceLineNo">352</span>    } else {<a name="line.352"></a>
+<span class="sourceLineNo">353</span>      hris = MetaTableAccessor.getTableRegions(server.getConnection(), tableName, true);<a name="line.353"></a>
+<span class="sourceLineNo">354</span>    }<a name="line.354"></a>
+<span class="sourceLineNo">355</span><a name="line.355"></a>
+<span class="sourceLineNo">356</span>    Integer pending = 0;<a name="line.356"></a>
+<span class="sourceLineNo">357</span>    for (HRegionInfo hri : hris) {<a name="line.357"></a>
+<span class="sourceLineNo">358</span>      String name = hri.getEncodedName();<a name="line.358"></a>
+<span class="sourceLineNo">359</span>      // no lock concurrent access ok: sequential consistency respected.<a name="line.359"></a>
+<span class="sourceLineNo">360</span>      if (regionsToReopen.containsKey(name)<a name="line.360"></a>
+<span class="sourceLineNo">361</span>          || regionStates.isRegionInTransition(name)) {<a name="line.361"></a>
+<span class="sourceLineNo">362</span>        pending++;<a name="line.362"></a>
+<span class="sourceLineNo">363</span>      }<a name="line.363"></a>
+<span class="sourceLineNo">364</span>    }<a name="line.364"></a>
+<span class="sourceLineNo">365</span>    return new Pair&lt;Integer, Integer&gt;(pending, hris.size());<a name="line.365"></a>
+<span class="sourceLineNo">366</span>  }<a name="line.366"></a>
+<span class="sourceLineNo">367</span><a name="line.367"></a>
+<span class="sourceLineNo">368</span>  /**<a name="line.368"></a>
+<span class="sourceLineNo">369</span>   * Used by ServerShutdownHandler to make sure AssignmentManager has completed<a name="line.369"></a>
+<span class="sourceLineNo">370</span>   * the failover cleanup before re-assigning regions of dead servers. So that<a name="line.370"></a>
+<span class="sourceLineNo">371</span>   * when re-assignment happens, AssignmentManager has proper region states.<a name="line.371"></a>
+<span class="sourceLineNo">372</span>   */<a name="line.372"></a>
+<span class="sourceLineNo">373</span>  public boolean isFailoverCleanupDone() {<a name="line.373"></a>
+<span class="sourceLineNo">374</span>    return failoverCleanupDone.get();<a name="line.374"></a>
+<span class="sourceLineNo">375</span>  }<a name="line.375"></a>
+<span class="sourceLineNo">376</span><a name="line.376"></a>
+<span class="sourceLineNo">377</span>  /**<a name="line.377"></a>
+<span class="sourceLineNo">378</span>   * To avoid racing with AM, external entities may need to lock a region,<a name="line.378"></a>
+<span class="sourceLineNo">379</span>   * for example, when SSH checks what regions to skip re-assigning.<a name="line.379"></a>
+<span class="sourceLineNo">380</span>   */<a name="line.380"></a>
+<span class="sourceLineNo">381</span>  public Lock acquireRegionLock(final String encodedName) {<a name="line.381"></a>
+<span class="sourceLineNo">382</span>    return locker.acquireLock(encodedName);<a name="line.382"></a>
+<span class="sourceLineNo">383</span>  }<a name="line.383"></a>
+<span class="sourceLineNo">384</span><a name="line.384"></a>
+<span class="sourceLineNo">385</span>  /**<a name="line.385"></a>
+<span class="sourceLineNo">386</span>   * Now, failover cleanup is completed. Notify server manager to<a name="line.386"></a>
+<span class="sourceLineNo">387</span>   * process queued up dead servers processing, if any.<a name="line.387"></a>
+<span class="sourceLineNo">388</span>   */<a name="line.388"></a>
+<span class="sourceLineNo">389</span>  void failoverCleanupDone() {<a name="line.389"></a>
+<span class="sourceLineNo">390</span>    failoverCleanupDone.set(true);<a name="line.390"></a>
+<span class="sourceLineNo">391</span>    serverManager.processQueuedDeadServers();<a name="line.391"></a>
+<span class="sourceLineNo">392</span>  }<a name="line.392"></a>
+<span class="sourceLineNo">393</span><a name="line.393"></a>
+<span class="sourceLineNo">394</span>  /**<a name="line.394"></a>
+<span class="sourceLineNo">395</span>   * Called on startup.<a name="line.395"></a>
+<span class="sourceLineNo">396</span>   * Figures whether a fresh cluster start of we are joining extant running cluster.<a name="line.396"></a>
+<span class="sourceLineNo">397</span>   * @throws IOException<a name="line.397"></a>
+<span class="sourceLineNo">398</span>   * @throws KeeperException<a name="line.398"></a>
+<span class="sourceLineNo">399</span>   * @throws InterruptedException<a name="line.399"></a>
+<span class="sourceLineNo">400</span>   * @throws CoordinatedStateException<a name="line.400"></a>
+<span class="sourceLineNo">401</span>   */<a name="line.401"></a>
+<span class="sourceLineNo">402</span>  void joinCluster()<a name="line.402"></a>
+<span class="sourceLineNo">403</span>  throws IOException, KeeperException, InterruptedException, CoordinatedStateException {<a name="line.403"></a>
+<span class="sourceLineNo">404</span>    long startTime = System.currentTimeMillis();<a name="line.404"></a>
+<span class="sourceLineNo">405</span>    // Concurrency note: In the below the accesses on regionsInTransition are<a name="line.405"></a>
+<span class="sourceLineNo">406</span>    // outside of a synchronization block where usually all acces

<TRUNCATED>

Mime
View raw message