http://git-wip-us.apache.org/repos/asf/hbase-site/blob/fd35fe02/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.html b/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.html index 8e48140..2676386 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.html @@ -33,855 +33,860 @@ <span class="sourceLineNo">025</span>import java.util.Iterator;<a name="line.25"></a> <span class="sourceLineNo">026</span>import java.util.List;<a name="line.26"></a> <span class="sourceLineNo">027</span>import java.util.Map;<a name="line.27"></a> -<span class="sourceLineNo">028</span>import java.util.Set;<a name="line.28"></a> -<span class="sourceLineNo">029</span>import java.util.SortedSet;<a name="line.29"></a> -<span class="sourceLineNo">030</span>import java.util.TreeSet;<a name="line.30"></a> -<span class="sourceLineNo">031</span>import java.util.UUID;<a name="line.31"></a> -<span class="sourceLineNo">032</span>import java.util.concurrent.ConcurrentHashMap;<a name="line.32"></a> -<span class="sourceLineNo">033</span>import java.util.concurrent.ConcurrentMap;<a name="line.33"></a> -<span class="sourceLineNo">034</span>import java.util.concurrent.Future;<a name="line.34"></a> -<span class="sourceLineNo">035</span>import java.util.concurrent.LinkedBlockingQueue;<a name="line.35"></a> -<span class="sourceLineNo">036</span>import java.util.concurrent.RejectedExecutionException;<a name="line.36"></a> -<span class="sourceLineNo">037</span>import java.util.concurrent.ThreadLocalRandom;<a name="line.37"></a> -<span class="sourceLineNo">038</span>import java.util.concurrent.ThreadPoolExecutor;<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.AtomicLong;<a name="line.40"></a> -<span class="sourceLineNo">041</span>import java.util.stream.Collectors;<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.FileSystem;<a name="line.43"></a> -<span class="sourceLineNo">044</span>import org.apache.hadoop.fs.Path;<a name="line.44"></a> -<span class="sourceLineNo">045</span>import org.apache.hadoop.hbase.HConstants;<a name="line.45"></a> -<span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.Server;<a name="line.46"></a> -<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.ServerName;<a name="line.47"></a> -<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.TableName;<a name="line.48"></a> -<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.replication.ReplicationException;<a name="line.49"></a> -<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.replication.ReplicationListener;<a name="line.50"></a> -<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.replication.ReplicationPeer;<a name="line.51"></a> -<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.replication.ReplicationPeer.PeerState;<a name="line.52"></a> -<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.replication.ReplicationPeers;<a name="line.53"></a> -<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.replication.ReplicationQueueInfo;<a name="line.54"></a> -<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.replication.ReplicationQueueStorage;<a name="line.55"></a> -<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.replication.ReplicationTracker;<a name="line.56"></a> -<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.util.Pair;<a name="line.57"></a> -<span class="sourceLineNo">058</span>import org.apache.hadoop.hbase.wal.AbstractFSWALProvider;<a name="line.58"></a> -<span class="sourceLineNo">059</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.59"></a> -<span class="sourceLineNo">060</span>import org.slf4j.Logger;<a name="line.60"></a> -<span class="sourceLineNo">061</span>import org.slf4j.LoggerFactory;<a name="line.61"></a> -<span class="sourceLineNo">062</span><a name="line.62"></a> -<span class="sourceLineNo">063</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.63"></a> -<span class="sourceLineNo">064</span>import org.apache.hbase.thirdparty.com.google.common.util.concurrent.ThreadFactoryBuilder;<a name="line.64"></a> -<span class="sourceLineNo">065</span><a name="line.65"></a> -<span class="sourceLineNo">066</span>/**<a name="line.66"></a> -<span class="sourceLineNo">067</span> * This class is responsible to manage all the replication sources. There are two classes of<a name="line.67"></a> -<span class="sourceLineNo">068</span> * sources:<a name="line.68"></a> -<span class="sourceLineNo">069</span> * <ul><a name="line.69"></a> -<span class="sourceLineNo">070</span> * <li>Normal sources are persistent and one per peer cluster</li><a name="line.70"></a> -<span class="sourceLineNo">071</span> * <li>Old sources are recovered from a failed region server and our only goal is to finish<a name="line.71"></a> -<span class="sourceLineNo">072</span> * replicating the WAL queue it had</li><a name="line.72"></a> -<span class="sourceLineNo">073</span> * </ul><a name="line.73"></a> -<span class="sourceLineNo">074</span> * <p><a name="line.74"></a> -<span class="sourceLineNo">075</span> * When a region server dies, this class uses a watcher to get notified and it tries to grab a lock<a name="line.75"></a> -<span class="sourceLineNo">076</span> * in order to transfer all the queues in a local old source.<a name="line.76"></a> -<span class="sourceLineNo">077</span> * <p><a name="line.77"></a> -<span class="sourceLineNo">078</span> * Synchronization specification:<a name="line.78"></a> -<span class="sourceLineNo">079</span> * <ul><a name="line.79"></a> -<span class="sourceLineNo">080</span> * <li>No need synchronized on {@link #sources}. {@link #sources} is a ConcurrentHashMap and there<a name="line.80"></a> -<span class="sourceLineNo">081</span> * is a Lock for peer id in {@link PeerProcedureHandlerImpl}. So there is no race for peer<a name="line.81"></a> -<span class="sourceLineNo">082</span> * operations.</li><a name="line.82"></a> -<span class="sourceLineNo">083</span> * <li>Need synchronized on {@link #walsById}. There are four methods which modify it,<a name="line.83"></a> -<span class="sourceLineNo">084</span> * {@link #addPeer(String)}, {@link #removePeer(String)},<a name="line.84"></a> -<span class="sourceLineNo">085</span> * {@link #cleanOldLogs(SortedSet, String, String)} and {@link #preLogRoll(Path)}.<a name="line.85"></a> -<span class="sourceLineNo">086</span> * {@link #walsById} is a ConcurrentHashMap and there is a Lock for peer id in<a name="line.86"></a> -<span class="sourceLineNo">087</span> * {@link PeerProcedureHandlerImpl}. So there is no race between {@link #addPeer(String)} and<a name="line.87"></a> -<span class="sourceLineNo">088</span> * {@link #removePeer(String)}. {@link #cleanOldLogs(SortedSet, String, String)} is called by<a name="line.88"></a> -<span class="sourceLineNo">089</span> * {@link ReplicationSourceInterface}. So no race with {@link #addPeer(String)}.<a name="line.89"></a> -<span class="sourceLineNo">090</span> * {@link #removePeer(String)} will terminate the {@link ReplicationSourceInterface} firstly, then<a name="line.90"></a> -<span class="sourceLineNo">091</span> * remove the wals from {@link #walsById}. So no race with {@link #removePeer(String)}. The only<a name="line.91"></a> -<span class="sourceLineNo">092</span> * case need synchronized is {@link #cleanOldLogs(SortedSet, String, String)} and<a name="line.92"></a> -<span class="sourceLineNo">093</span> * {@link #preLogRoll(Path)}.</li><a name="line.93"></a> -<span class="sourceLineNo">094</span> * <li>No need synchronized on {@link #walsByIdRecoveredQueues}. There are three methods which<a name="line.94"></a> -<span class="sourceLineNo">095</span> * modify it, {@link #removePeer(String)} , {@link #cleanOldLogs(SortedSet, String, String)} and<a name="line.95"></a> -<span class="sourceLineNo">096</span> * {@link ReplicationSourceManager.NodeFailoverWorker#run()}.<a name="line.96"></a> -<span class="sourceLineNo">097</span> * {@link #cleanOldLogs(SortedSet, String, String)} is called by {@link ReplicationSourceInterface}.<a name="line.97"></a> -<span class="sourceLineNo">098</span> * {@link #removePeer(String)} will terminate the {@link ReplicationSourceInterface} firstly, then<a name="line.98"></a> -<span class="sourceLineNo">099</span> * remove the wals from {@link #walsByIdRecoveredQueues}. And<a name="line.99"></a> -<span class="sourceLineNo">100</span> * {@link ReplicationSourceManager.NodeFailoverWorker#run()} will add the wals to<a name="line.100"></a> -<span class="sourceLineNo">101</span> * {@link #walsByIdRecoveredQueues} firstly, then start up a {@link ReplicationSourceInterface}. So<a name="line.101"></a> -<span class="sourceLineNo">102</span> * there is no race here. For {@link ReplicationSourceManager.NodeFailoverWorker#run()} and<a name="line.102"></a> -<span class="sourceLineNo">103</span> * {@link #removePeer(String)}, there is already synchronized on {@link #oldsources}. So no need<a name="line.103"></a> -<span class="sourceLineNo">104</span> * synchronized on {@link #walsByIdRecoveredQueues}.</li><a name="line.104"></a> -<span class="sourceLineNo">105</span> * <li>Need synchronized on {@link #latestPaths} to avoid the new open source miss new log.</li><a name="line.105"></a> -<span class="sourceLineNo">106</span> * <li>Need synchronized on {@link #oldsources} to avoid adding recovered source for the<a name="line.106"></a> -<span class="sourceLineNo">107</span> * to-be-removed peer.</li><a name="line.107"></a> -<span class="sourceLineNo">108</span> * </ul><a name="line.108"></a> -<span class="sourceLineNo">109</span> */<a name="line.109"></a> -<span class="sourceLineNo">110</span>@InterfaceAudience.Private<a name="line.110"></a> -<span class="sourceLineNo">111</span>public class ReplicationSourceManager implements ReplicationListener {<a name="line.111"></a> -<span class="sourceLineNo">112</span> private static final Logger LOG = LoggerFactory.getLogger(ReplicationSourceManager.class);<a name="line.112"></a> -<span class="sourceLineNo">113</span> // all the sources that read this RS's logs and every peer only has one replication source<a name="line.113"></a> -<span class="sourceLineNo">114</span> private final ConcurrentMap<String, ReplicationSourceInterface> sources;<a name="line.114"></a> -<span class="sourceLineNo">115</span> // List of all the sources we got from died RSs<a name="line.115"></a> -<span class="sourceLineNo">116</span> private final List<ReplicationSourceInterface> oldsources;<a name="line.116"></a> -<span class="sourceLineNo">117</span> private final ReplicationQueueStorage queueStorage;<a name="line.117"></a> -<span class="sourceLineNo">118</span> private final ReplicationTracker replicationTracker;<a name="line.118"></a> -<span class="sourceLineNo">119</span> private final ReplicationPeers replicationPeers;<a name="line.119"></a> -<span class="sourceLineNo">120</span> // UUID for this cluster<a name="line.120"></a> -<span class="sourceLineNo">121</span> private final UUID clusterId;<a name="line.121"></a> -<span class="sourceLineNo">122</span> // All about stopping<a name="line.122"></a> -<span class="sourceLineNo">123</span> private final Server server;<a name="line.123"></a> -<span class="sourceLineNo">124</span><a name="line.124"></a> -<span class="sourceLineNo">125</span> // All logs we are currently tracking<a name="line.125"></a> -<span class="sourceLineNo">126</span> // Index structure of the map is: queue_id->logPrefix/logGroup->logs<a name="line.126"></a> -<span class="sourceLineNo">127</span> // For normal replication source, the peer id is same with the queue id<a name="line.127"></a> -<span class="sourceLineNo">128</span> private final ConcurrentMap<String, Map<String, SortedSet<String>>> walsById;<a name="line.128"></a> -<span class="sourceLineNo">129</span> // Logs for recovered sources we are currently tracking<a name="line.129"></a> -<span class="sourceLineNo">130</span> // the map is: queue_id->logPrefix/logGroup->logs<a name="line.130"></a> -<span class="sourceLineNo">131</span> // For recovered source, the queue id's format is peer_id-servername-*<a name="line.131"></a> -<span class="sourceLineNo">132</span> private final ConcurrentMap<String, Map<String, SortedSet<String>>> walsByIdRecoveredQueues;<a name="line.132"></a> -<span class="sourceLineNo">133</span><a name="line.133"></a> -<span class="sourceLineNo">134</span> private final Configuration conf;<a name="line.134"></a> -<span class="sourceLineNo">135</span> private final FileSystem fs;<a name="line.135"></a> -<span class="sourceLineNo">136</span> // The paths to the latest log of each wal group, for new coming peers<a name="line.136"></a> -<span class="sourceLineNo">137</span> private final Set<Path> latestPaths;<a name="line.137"></a> -<span class="sourceLineNo">138</span> // Path to the wals directories<a name="line.138"></a> -<span class="sourceLineNo">139</span> private final Path logDir;<a name="line.139"></a> -<span class="sourceLineNo">140</span> // Path to the wal archive<a name="line.140"></a> -<span class="sourceLineNo">141</span> private final Path oldLogDir;<a name="line.141"></a> -<span class="sourceLineNo">142</span> private final WALFileLengthProvider walFileLengthProvider;<a name="line.142"></a> -<span class="sourceLineNo">143</span> // The number of ms that we wait before moving znodes, HBASE-3596<a name="line.143"></a> -<span class="sourceLineNo">144</span> private final long sleepBeforeFailover;<a name="line.144"></a> -<span class="sourceLineNo">145</span> // Homemade executer service for replication<a name="line.145"></a> -<span class="sourceLineNo">146</span> private final ThreadPoolExecutor executor;<a name="line.146"></a> -<span class="sourceLineNo">147</span><a name="line.147"></a> -<span class="sourceLineNo">148</span> private final boolean replicationForBulkLoadDataEnabled;<a name="line.148"></a> -<span class="sourceLineNo">149</span><a name="line.149"></a> +<span class="sourceLineNo">028</span>import java.util.NavigableSet;<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.SortedSet;<a name="line.30"></a> +<span class="sourceLineNo">031</span>import java.util.TreeSet;<a name="line.31"></a> +<span class="sourceLineNo">032</span>import java.util.UUID;<a name="line.32"></a> +<span class="sourceLineNo">033</span>import java.util.concurrent.ConcurrentHashMap;<a name="line.33"></a> +<span class="sourceLineNo">034</span>import java.util.concurrent.ConcurrentMap;<a name="line.34"></a> +<span class="sourceLineNo">035</span>import java.util.concurrent.Future;<a name="line.35"></a> +<span class="sourceLineNo">036</span>import java.util.concurrent.LinkedBlockingQueue;<a name="line.36"></a> +<span class="sourceLineNo">037</span>import java.util.concurrent.RejectedExecutionException;<a name="line.37"></a> +<span class="sourceLineNo">038</span>import java.util.concurrent.ThreadLocalRandom;<a name="line.38"></a> +<span class="sourceLineNo">039</span>import java.util.concurrent.ThreadPoolExecutor;<a name="line.39"></a> +<span class="sourceLineNo">040</span>import java.util.concurrent.TimeUnit;<a name="line.40"></a> +<span class="sourceLineNo">041</span>import java.util.concurrent.atomic.AtomicLong;<a name="line.41"></a> +<span class="sourceLineNo">042</span>import java.util.stream.Collectors;<a name="line.42"></a> +<span class="sourceLineNo">043</span>import org.apache.hadoop.conf.Configuration;<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.HConstants;<a name="line.46"></a> +<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.Server;<a name="line.47"></a> +<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.ServerName;<a name="line.48"></a> +<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.TableName;<a name="line.49"></a> +<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.replication.ReplicationException;<a name="line.50"></a> +<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.replication.ReplicationListener;<a name="line.51"></a> +<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.replication.ReplicationPeer;<a name="line.52"></a> +<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.replication.ReplicationPeer.PeerState;<a name="line.53"></a> +<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.replication.ReplicationPeers;<a name="line.54"></a> +<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.replication.ReplicationQueueInfo;<a name="line.55"></a> +<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.replication.ReplicationQueueStorage;<a name="line.56"></a> +<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.replication.ReplicationTracker;<a name="line.57"></a> +<span class="sourceLineNo">058</span>import org.apache.hadoop.hbase.util.Pair;<a name="line.58"></a> +<span class="sourceLineNo">059</span>import org.apache.hadoop.hbase.wal.AbstractFSWALProvider;<a name="line.59"></a> +<span class="sourceLineNo">060</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.60"></a> +<span class="sourceLineNo">061</span>import org.slf4j.Logger;<a name="line.61"></a> +<span class="sourceLineNo">062</span>import org.slf4j.LoggerFactory;<a name="line.62"></a> +<span class="sourceLineNo">063</span><a name="line.63"></a> +<span class="sourceLineNo">064</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.64"></a> +<span class="sourceLineNo">065</span>import org.apache.hbase.thirdparty.com.google.common.util.concurrent.ThreadFactoryBuilder;<a name="line.65"></a> +<span class="sourceLineNo">066</span><a name="line.66"></a> +<span class="sourceLineNo">067</span>/**<a name="line.67"></a> +<span class="sourceLineNo">068</span> * This class is responsible to manage all the replication sources. There are two classes of<a name="line.68"></a> +<span class="sourceLineNo">069</span> * sources:<a name="line.69"></a> +<span class="sourceLineNo">070</span> * <ul><a name="line.70"></a> +<span class="sourceLineNo">071</span> * <li>Normal sources are persistent and one per peer cluster</li><a name="line.71"></a> +<span class="sourceLineNo">072</span> * <li>Old sources are recovered from a failed region server and our only goal is to finish<a name="line.72"></a> +<span class="sourceLineNo">073</span> * replicating the WAL queue it had</li><a name="line.73"></a> +<span class="sourceLineNo">074</span> * </ul><a name="line.74"></a> +<span class="sourceLineNo">075</span> * <p><a name="line.75"></a> +<span class="sourceLineNo">076</span> * When a region server dies, this class uses a watcher to get notified and it tries to grab a lock<a name="line.76"></a> +<span class="sourceLineNo">077</span> * in order to transfer all the queues in a local old source.<a name="line.77"></a> +<span class="sourceLineNo">078</span> * <p><a name="line.78"></a> +<span class="sourceLineNo">079</span> * Synchronization specification:<a name="line.79"></a> +<span class="sourceLineNo">080</span> * <ul><a name="line.80"></a> +<span class="sourceLineNo">081</span> * <li>No need synchronized on {@link #sources}. {@link #sources} is a ConcurrentHashMap and there<a name="line.81"></a> +<span class="sourceLineNo">082</span> * is a Lock for peer id in {@link PeerProcedureHandlerImpl}. So there is no race for peer<a name="line.82"></a> +<span class="sourceLineNo">083</span> * operations.</li><a name="line.83"></a> +<span class="sourceLineNo">084</span> * <li>Need synchronized on {@link #walsById}. There are four methods which modify it,<a name="line.84"></a> +<span class="sourceLineNo">085</span> * {@link #addPeer(String)}, {@link #removePeer(String)},<a name="line.85"></a> +<span class="sourceLineNo">086</span> * {@link #cleanOldLogs(NavigableSet, String, boolean, String)} and {@link #preLogRoll(Path)}.<a name="line.86"></a> +<span class="sourceLineNo">087</span> * {@link #walsById} is a ConcurrentHashMap and there is a Lock for peer id in<a name="line.87"></a> +<span class="sourceLineNo">088</span> * {@link PeerProcedureHandlerImpl}. So there is no race between {@link #addPeer(String)} and<a name="line.88"></a> +<span class="sourceLineNo">089</span> * {@link #removePeer(String)}. {@link #cleanOldLogs(NavigableSet, String, boolean, String)} is<a name="line.89"></a> +<span class="sourceLineNo">090</span> * called by {@link ReplicationSourceInterface}. So no race with {@link #addPeer(String)}.<a name="line.90"></a> +<span class="sourceLineNo">091</span> * {@link #removePeer(String)} will terminate the {@link ReplicationSourceInterface} firstly, then<a name="line.91"></a> +<span class="sourceLineNo">092</span> * remove the wals from {@link #walsById}. So no race with {@link #removePeer(String)}. The only<a name="line.92"></a> +<span class="sourceLineNo">093</span> * case need synchronized is {@link #cleanOldLogs(NavigableSet, String, boolean, String)} and<a name="line.93"></a> +<span class="sourceLineNo">094</span> * {@link #preLogRoll(Path)}.</li><a name="line.94"></a> +<span class="sourceLineNo">095</span> * <li>No need synchronized on {@link #walsByIdRecoveredQueues}. There are three methods which<a name="line.95"></a> +<span class="sourceLineNo">096</span> * modify it, {@link #removePeer(String)} ,<a name="line.96"></a> +<span class="sourceLineNo">097</span> * {@link #cleanOldLogs(NavigableSet, String, boolean, String)} and<a name="line.97"></a> +<span class="sourceLineNo">098</span> * {@link ReplicationSourceManager.NodeFailoverWorker#run()}.<a name="line.98"></a> +<span class="sourceLineNo">099</span> * {@link #cleanOldLogs(NavigableSet, String, boolean, String)} is called by<a name="line.99"></a> +<span class="sourceLineNo">100</span> * {@link ReplicationSourceInterface}. {@link #removePeer(String)} will terminate the<a name="line.100"></a> +<span class="sourceLineNo">101</span> * {@link ReplicationSourceInterface} firstly, then remove the wals from<a name="line.101"></a> +<span class="sourceLineNo">102</span> * {@link #walsByIdRecoveredQueues}. And {@link ReplicationSourceManager.NodeFailoverWorker#run()}<a name="line.102"></a> +<span class="sourceLineNo">103</span> * will add the wals to {@link #walsByIdRecoveredQueues} firstly, then start up a<a name="line.103"></a> +<span class="sourceLineNo">104</span> * {@link ReplicationSourceInterface}. So there is no race here. For<a name="line.104"></a> +<span class="sourceLineNo">105</span> * {@link ReplicationSourceManager.NodeFailoverWorker#run()} and {@link #removePeer(String)}, there<a name="line.105"></a> +<span class="sourceLineNo">106</span> * is already synchronized on {@link #oldsources}. So no need synchronized on<a name="line.106"></a> +<span class="sourceLineNo">107</span> * {@link #walsByIdRecoveredQueues}.</li><a name="line.107"></a> +<span class="sourceLineNo">108</span> * <li>Need synchronized on {@link #latestPaths} to avoid the new open source miss new log.</li><a name="line.108"></a> +<span class="sourceLineNo">109</span> * <li>Need synchronized on {@link #oldsources} to avoid adding recovered source for the<a name="line.109"></a> +<span class="sourceLineNo">110</span> * to-be-removed peer.</li><a name="line.110"></a> +<span class="sourceLineNo">111</span> * </ul><a name="line.111"></a> +<span class="sourceLineNo">112</span> */<a name="line.112"></a> +<span class="sourceLineNo">113</span>@InterfaceAudience.Private<a name="line.113"></a> +<span class="sourceLineNo">114</span>public class ReplicationSourceManager implements ReplicationListener {<a name="line.114"></a> +<span class="sourceLineNo">115</span> private static final Logger LOG = LoggerFactory.getLogger(ReplicationSourceManager.class);<a name="line.115"></a> +<span class="sourceLineNo">116</span> // all the sources that read this RS's logs and every peer only has one replication source<a name="line.116"></a> +<span class="sourceLineNo">117</span> private final ConcurrentMap<String, ReplicationSourceInterface> sources;<a name="line.117"></a> +<span class="sourceLineNo">118</span> // List of all the sources we got from died RSs<a name="line.118"></a> +<span class="sourceLineNo">119</span> private final List<ReplicationSourceInterface> oldsources;<a name="line.119"></a> +<span class="sourceLineNo">120</span> private final ReplicationQueueStorage queueStorage;<a name="line.120"></a> +<span class="sourceLineNo">121</span> private final ReplicationTracker replicationTracker;<a name="line.121"></a> +<span class="sourceLineNo">122</span> private final ReplicationPeers replicationPeers;<a name="line.122"></a> +<span class="sourceLineNo">123</span> // UUID for this cluster<a name="line.123"></a> +<span class="sourceLineNo">124</span> private final UUID clusterId;<a name="line.124"></a> +<span class="sourceLineNo">125</span> // All about stopping<a name="line.125"></a> +<span class="sourceLineNo">126</span> private final Server server;<a name="line.126"></a> +<span class="sourceLineNo">127</span><a name="line.127"></a> +<span class="sourceLineNo">128</span> // All logs we are currently tracking<a name="line.128"></a> +<span class="sourceLineNo">129</span> // Index structure of the map is: queue_id->logPrefix/logGroup->logs<a name="line.129"></a> +<span class="sourceLineNo">130</span> // For normal replication source, the peer id is same with the queue id<a name="line.130"></a> +<span class="sourceLineNo">131</span> private final ConcurrentMap<String, Map<String, NavigableSet<String>>> walsById;<a name="line.131"></a> +<span class="sourceLineNo">132</span> // Logs for recovered sources we are currently tracking<a name="line.132"></a> +<span class="sourceLineNo">133</span> // the map is: queue_id->logPrefix/logGroup->logs<a name="line.133"></a> +<span class="sourceLineNo">134</span> // For recovered source, the queue id's format is peer_id-servername-*<a name="line.134"></a> +<span class="sourceLineNo">135</span> private final ConcurrentMap<String, Map<String, NavigableSet<String>>> walsByIdRecoveredQueues;<a name="line.135"></a> +<span class="sourceLineNo">136</span><a name="line.136"></a> +<span class="sourceLineNo">137</span> private final Configuration conf;<a name="line.137"></a> +<span class="sourceLineNo">138</span> private final FileSystem fs;<a name="line.138"></a> +<span class="sourceLineNo">139</span> // The paths to the latest log of each wal group, for new coming peers<a name="line.139"></a> +<span class="sourceLineNo">140</span> private final Set<Path> latestPaths;<a name="line.140"></a> +<span class="sourceLineNo">141</span> // Path to the wals directories<a name="line.141"></a> +<span class="sourceLineNo">142</span> private final Path logDir;<a name="line.142"></a> +<span class="sourceLineNo">143</span> // Path to the wal archive<a name="line.143"></a> +<span class="sourceLineNo">144</span> private final Path oldLogDir;<a name="line.144"></a> +<span class="sourceLineNo">145</span> private final WALFileLengthProvider walFileLengthProvider;<a name="line.145"></a> +<span class="sourceLineNo">146</span> // The number of ms that we wait before moving znodes, HBASE-3596<a name="line.146"></a> +<span class="sourceLineNo">147</span> private final long sleepBeforeFailover;<a name="line.147"></a> +<span class="sourceLineNo">148</span> // Homemade executer service for replication<a name="line.148"></a> +<span class="sourceLineNo">149</span> private final ThreadPoolExecutor executor;<a name="line.149"></a> <span class="sourceLineNo">150</span><a name="line.150"></a> -<span class="sourceLineNo">151</span> private AtomicLong totalBufferUsed = new AtomicLong();<a name="line.151"></a> +<span class="sourceLineNo">151</span> private final boolean replicationForBulkLoadDataEnabled;<a name="line.151"></a> <span class="sourceLineNo">152</span><a name="line.152"></a> -<span class="sourceLineNo">153</span> /**<a name="line.153"></a> -<span class="sourceLineNo">154</span> * Creates a replication manager and sets the watch on all the other registered region servers<a name="line.154"></a> -<span class="sourceLineNo">155</span> * @param queueStorage the interface for manipulating replication queues<a name="line.155"></a> -<span class="sourceLineNo">156</span> * @param replicationPeers<a name="line.156"></a> -<span class="sourceLineNo">157</span> * @param replicationTracker<a name="line.157"></a> -<span class="sourceLineNo">158</span> * @param conf the configuration to use<a name="line.158"></a> -<span class="sourceLineNo">159</span> * @param server the server for this region server<a name="line.159"></a> -<span class="sourceLineNo">160</span> * @param fs the file system to use<a name="line.160"></a> -<span class="sourceLineNo">161</span> * @param logDir the directory that contains all wal directories of live RSs<a name="line.161"></a> -<span class="sourceLineNo">162</span> * @param oldLogDir the directory where old logs are archived<a name="line.162"></a> -<span class="sourceLineNo">163</span> * @param clusterId<a name="line.163"></a> -<span class="sourceLineNo">164</span> */<a name="line.164"></a> -<span class="sourceLineNo">165</span> public ReplicationSourceManager(ReplicationQueueStorage queueStorage,<a name="line.165"></a> -<span class="sourceLineNo">166</span> ReplicationPeers replicationPeers, ReplicationTracker replicationTracker, Configuration conf,<a name="line.166"></a> -<span class="sourceLineNo">167</span> Server server, FileSystem fs, Path logDir, Path oldLogDir, UUID clusterId,<a name="line.167"></a> -<span class="sourceLineNo">168</span> WALFileLengthProvider walFileLengthProvider) throws IOException {<a name="line.168"></a> -<span class="sourceLineNo">169</span> // CopyOnWriteArrayList is thread-safe.<a name="line.169"></a> -<span class="sourceLineNo">170</span> // Generally, reading is more than modifying.<a name="line.170"></a> -<span class="sourceLineNo">171</span> this.sources = new ConcurrentHashMap<>();<a name="line.171"></a> -<span class="sourceLineNo">172</span> this.queueStorage = queueStorage;<a name="line.172"></a> -<span class="sourceLineNo">173</span> this.replicationPeers = replicationPeers;<a name="line.173"></a> -<span class="sourceLineNo">174</span> this.replicationTracker = replicationTracker;<a name="line.174"></a> -<span class="sourceLineNo">175</span> this.server = server;<a name="line.175"></a> -<span class="sourceLineNo">176</span> this.walsById = new ConcurrentHashMap<>();<a name="line.176"></a> -<span class="sourceLineNo">177</span> this.walsByIdRecoveredQueues = new ConcurrentHashMap<>();<a name="line.177"></a> -<span class="sourceLineNo">178</span> this.oldsources = new ArrayList<>();<a name="line.178"></a> -<span class="sourceLineNo">179</span> this.conf = conf;<a name="line.179"></a> -<span class="sourceLineNo">180</span> this.fs = fs;<a name="line.180"></a> -<span class="sourceLineNo">181</span> this.logDir = logDir;<a name="line.181"></a> -<span class="sourceLineNo">182</span> this.oldLogDir = oldLogDir;<a name="line.182"></a> -<span class="sourceLineNo">183</span> this.sleepBeforeFailover = conf.getLong("replication.sleep.before.failover", 30000); // 30<a name="line.183"></a> -<span class="sourceLineNo">184</span> // seconds<a name="line.184"></a> -<span class="sourceLineNo">185</span> this.clusterId = clusterId;<a name="line.185"></a> -<span class="sourceLineNo">186</span> this.walFileLengthProvider = walFileLengthProvider;<a name="line.186"></a> -<span class="sourceLineNo">187</span> this.replicationTracker.registerListener(this);<a name="line.187"></a> -<span class="sourceLineNo">188</span> // It's preferable to failover 1 RS at a time, but with good zk servers<a name="line.188"></a> -<span class="sourceLineNo">189</span> // more could be processed at the same time.<a name="line.189"></a> -<span class="sourceLineNo">190</span> int nbWorkers = conf.getInt("replication.executor.workers", 1);<a name="line.190"></a> -<span class="sourceLineNo">191</span> // use a short 100ms sleep since this could be done inline with a RS startup<a name="line.191"></a> -<span class="sourceLineNo">192</span> // even if we fail, other region servers can take care of it<a name="line.192"></a> -<span class="sourceLineNo">193</span> this.executor = new ThreadPoolExecutor(nbWorkers, nbWorkers, 100, TimeUnit.MILLISECONDS,<a name="line.193"></a> -<span class="sourceLineNo">194</span> new LinkedBlockingQueue<>());<a name="line.194"></a> -<span class="sourceLineNo">195</span> ThreadFactoryBuilder tfb = new ThreadFactoryBuilder();<a name="line.195"></a> -<span class="sourceLineNo">196</span> tfb.setNameFormat("ReplicationExecutor-%d");<a name="line.196"></a> -<span class="sourceLineNo">197</span> tfb.setDaemon(true);<a name="line.197"></a> -<span class="sourceLineNo">198</span> this.executor.setThreadFactory(tfb.build());<a name="line.198"></a> -<span class="sourceLineNo">199</span> this.latestPaths = new HashSet<Path>();<a name="line.199"></a> -<span class="sourceLineNo">200</span> replicationForBulkLoadDataEnabled = conf.getBoolean(HConstants.REPLICATION_BULKLOAD_ENABLE_KEY,<a name="line.200"></a> -<span class="sourceLineNo">201</span> HConstants.REPLICATION_BULKLOAD_ENABLE_DEFAULT);<a name="line.201"></a> -<span class="sourceLineNo">202</span> }<a name="line.202"></a> -<span class="sourceLineNo">203</span><a name="line.203"></a> -<span class="sourceLineNo">204</span> /**<a name="line.204"></a> -<span class="sourceLineNo">205</span> * Adds a normal source per registered peer cluster and tries to process all old region server wal<a name="line.205"></a> -<span class="sourceLineNo">206</span> * queues<a name="line.206"></a> -<span class="sourceLineNo">207</span> * <p><a name="line.207"></a> -<span class="sourceLineNo">208</span> * The returned future is for adoptAbandonedQueues task.<a name="line.208"></a> -<span class="sourceLineNo">209</span> */<a name="line.209"></a> -<span class="sourceLineNo">210</span> Future<?> init() throws IOException {<a name="line.210"></a> -<span class="sourceLineNo">211</span> for (String id : this.replicationPeers.getAllPeerIds()) {<a name="line.211"></a> -<span class="sourceLineNo">212</span> addSource(id);<a name="line.212"></a> -<span class="sourceLineNo">213</span> if (replicationForBulkLoadDataEnabled) {<a name="line.213"></a> -<span class="sourceLineNo">214</span> // Check if peer exists in hfile-refs queue, if not add it. This can happen in the case<a name="line.214"></a> -<span class="sourceLineNo">215</span> // when a peer was added before replication for bulk loaded data was enabled.<a name="line.215"></a> -<span class="sourceLineNo">216</span> throwIOExceptionWhenFail(() -> this.queueStorage.addPeerToHFileRefs(id));<a name="line.216"></a> -<span class="sourceLineNo">217</span> }<a name="line.217"></a> -<span class="sourceLineNo">218</span> }<a name="line.218"></a> -<span class="sourceLineNo">219</span> return this.executor.submit(this::adoptAbandonedQueues);<a name="line.219"></a> -<span class="sourceLineNo">220</span> }<a name="line.220"></a> -<span class="sourceLineNo">221</span><a name="line.221"></a> -<span class="sourceLineNo">222</span> private void adoptAbandonedQueues() {<a name="line.222"></a> -<span class="sourceLineNo">223</span> List<ServerName> currentReplicators = null;<a name="line.223"></a> -<span class="sourceLineNo">224</span> try {<a name="line.224"></a> -<span class="sourceLineNo">225</span> currentReplicators = queueStorage.getListOfReplicators();<a name="line.225"></a> -<span class="sourceLineNo">226</span> } catch (ReplicationException e) {<a name="line.226"></a> -<span class="sourceLineNo">227</span> server.abort("Failed to get all replicators", e);<a name="line.227"></a> -<span class="sourceLineNo">228</span> return;<a name="line.228"></a> -<span class="sourceLineNo">229</span> }<a name="line.229"></a> -<span class="sourceLineNo">230</span> if (currentReplicators == null || currentReplicators.isEmpty()) {<a name="line.230"></a> +<span class="sourceLineNo">153</span><a name="line.153"></a> +<span class="sourceLineNo">154</span> private AtomicLong totalBufferUsed = new AtomicLong();<a name="line.154"></a> +<span class="sourceLineNo">155</span><a name="line.155"></a> +<span class="sourceLineNo">156</span> /**<a name="line.156"></a> +<span class="sourceLineNo">157</span> * Creates a replication manager and sets the watch on all the other registered region servers<a name="line.157"></a> +<span class="sourceLineNo">158</span> * @param queueStorage the interface for manipulating replication queues<a name="line.158"></a> +<span class="sourceLineNo">159</span> * @param replicationPeers<a name="line.159"></a> +<span class="sourceLineNo">160</span> * @param replicationTracker<a name="line.160"></a> +<span class="sourceLineNo">161</span> * @param conf the configuration to use<a name="line.161"></a> +<span class="sourceLineNo">162</span> * @param server the server for this region server<a name="line.162"></a> +<span class="sourceLineNo">163</span> * @param fs the file system to use<a name="line.163"></a> +<span class="sourceLineNo">164</span> * @param logDir the directory that contains all wal directories of live RSs<a name="line.164"></a> +<span class="sourceLineNo">165</span> * @param oldLogDir the directory where old logs are archived<a name="line.165"></a> +<span class="sourceLineNo">166</span> * @param clusterId<a name="line.166"></a> +<span class="sourceLineNo">167</span> */<a name="line.167"></a> +<span class="sourceLineNo">168</span> public ReplicationSourceManager(ReplicationQueueStorage queueStorage,<a name="line.168"></a> +<span class="sourceLineNo">169</span> ReplicationPeers replicationPeers, ReplicationTracker replicationTracker, Configuration conf,<a name="line.169"></a> +<span class="sourceLineNo">170</span> Server server, FileSystem fs, Path logDir, Path oldLogDir, UUID clusterId,<a name="line.170"></a> +<span class="sourceLineNo">171</span> WALFileLengthProvider walFileLengthProvider) throws IOException {<a name="line.171"></a> +<span class="sourceLineNo">172</span> // CopyOnWriteArrayList is thread-safe.<a name="line.172"></a> +<span class="sourceLineNo">173</span> // Generally, reading is more than modifying.<a name="line.173"></a> +<span class="sourceLineNo">174</span> this.sources = new ConcurrentHashMap<>();<a name="line.174"></a> +<span class="sourceLineNo">175</span> this.queueStorage = queueStorage;<a name="line.175"></a> +<span class="sourceLineNo">176</span> this.replicationPeers = replicationPeers;<a name="line.176"></a> +<span class="sourceLineNo">177</span> this.replicationTracker = replicationTracker;<a name="line.177"></a> +<span class="sourceLineNo">178</span> this.server = server;<a name="line.178"></a> +<span class="sourceLineNo">179</span> this.walsById = new ConcurrentHashMap<>();<a name="line.179"></a> +<span class="sourceLineNo">180</span> this.walsByIdRecoveredQueues = new ConcurrentHashMap<>();<a name="line.180"></a> +<span class="sourceLineNo">181</span> this.oldsources = new ArrayList<>();<a name="line.181"></a> +<span class="sourceLineNo">182</span> this.conf = conf;<a name="line.182"></a> +<span class="sourceLineNo">183</span> this.fs = fs;<a name="line.183"></a> +<span class="sourceLineNo">184</span> this.logDir = logDir;<a name="line.184"></a> +<span class="sourceLineNo">185</span> this.oldLogDir = oldLogDir;<a name="line.185"></a> +<span class="sourceLineNo">186</span> this.sleepBeforeFailover = conf.getLong("replication.sleep.before.failover", 30000); // 30<a name="line.186"></a> +<span class="sourceLineNo">187</span> // seconds<a name="line.187"></a> +<span class="sourceLineNo">188</span> this.clusterId = clusterId;<a name="line.188"></a> +<span class="sourceLineNo">189</span> this.walFileLengthProvider = walFileLengthProvider;<a name="line.189"></a> +<span class="sourceLineNo">190</span> this.replicationTracker.registerListener(this);<a name="line.190"></a> +<span class="sourceLineNo">191</span> // It's preferable to failover 1 RS at a time, but with good zk servers<a name="line.191"></a> +<span class="sourceLineNo">192</span> // more could be processed at the same time.<a name="line.192"></a> +<span class="sourceLineNo">193</span> int nbWorkers = conf.getInt("replication.executor.workers", 1);<a name="line.193"></a> +<span class="sourceLineNo">194</span> // use a short 100ms sleep since this could be done inline with a RS startup<a name="line.194"></a> +<span class="sourceLineNo">195</span> // even if we fail, other region servers can take care of it<a name="line.195"></a> +<span class="sourceLineNo">196</span> this.executor = new ThreadPoolExecutor(nbWorkers, nbWorkers, 100, TimeUnit.MILLISECONDS,<a name="line.196"></a> +<span class="sourceLineNo">197</span> new LinkedBlockingQueue<>());<a name="line.197"></a> +<span class="sourceLineNo">198</span> ThreadFactoryBuilder tfb = new ThreadFactoryBuilder();<a name="line.198"></a> +<span class="sourceLineNo">199</span> tfb.setNameFormat("ReplicationExecutor-%d");<a name="line.199"></a> +<span class="sourceLineNo">200</span> tfb.setDaemon(true);<a name="line.200"></a> +<span class="sourceLineNo">201</span> this.executor.setThreadFactory(tfb.build());<a name="line.201"></a> +<span class="sourceLineNo">202</span> this.latestPaths = new HashSet<Path>();<a name="line.202"></a> +<span class="sourceLineNo">203</span> replicationForBulkLoadDataEnabled = conf.getBoolean(HConstants.REPLICATION_BULKLOAD_ENABLE_KEY,<a name="line.203"></a> +<span class="sourceLineNo">204</span> HConstants.REPLICATION_BULKLOAD_ENABLE_DEFAULT);<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> /**<a name="line.207"></a> +<span class="sourceLineNo">208</span> * Adds a normal source per registered peer cluster and tries to process all old region server wal<a name="line.208"></a> +<span class="sourceLineNo">209</span> * queues<a name="line.209"></a> +<span class="sourceLineNo">210</span> * <p><a name="line.210"></a> +<span class="sourceLineNo">211</span> * The returned future is for adoptAbandonedQueues task.<a name="line.211"></a> +<span class="sourceLineNo">212</span> */<a name="line.212"></a> +<span class="sourceLineNo">213</span> Future<?> init() throws IOException {<a name="line.213"></a> +<span class="sourceLineNo">214</span> for (String id : this.replicationPeers.getAllPeerIds()) {<a name="line.214"></a> +<span class="sourceLineNo">215</span> addSource(id);<a name="line.215"></a> +<span class="sourceLineNo">216</span> if (replicationForBulkLoadDataEnabled) {<a name="line.216"></a> +<span class="sourceLineNo">217</span> // Check if peer exists in hfile-refs queue, if not add it. This can happen in the case<a name="line.217"></a> +<span class="sourceLineNo">218</span> // when a peer was added before replication for bulk loaded data was enabled.<a name="line.218"></a> +<span class="sourceLineNo">219</span> throwIOExceptionWhenFail(() -> this.queueStorage.addPeerToHFileRefs(id));<a name="line.219"></a> +<span class="sourceLineNo">220</span> }<a name="line.220"></a> +<span class="sourceLineNo">221</span> }<a name="line.221"></a> +<span class="sourceLineNo">222</span> return this.executor.submit(this::adoptAbandonedQueues);<a name="line.222"></a> +<span class="sourceLineNo">223</span> }<a name="line.223"></a> +<span class="sourceLineNo">224</span><a name="line.224"></a> +<span class="sourceLineNo">225</span> private void adoptAbandonedQueues() {<a name="line.225"></a> +<span class="sourceLineNo">226</span> List<ServerName> currentReplicators = null;<a name="line.226"></a> +<span class="sourceLineNo">227</span> try {<a name="line.227"></a> +<span class="sourceLineNo">228</span> currentReplicators = queueStorage.getListOfReplicators();<a name="line.228"></a> +<span class="sourceLineNo">229</span> } catch (ReplicationException e) {<a name="line.229"></a> +<span class="sourceLineNo">230</span> server.abort("Failed to get all replicators", e);<a name="line.230"></a> <span class="sourceLineNo">231</span> return;<a name="line.231"></a> <span class="sourceLineNo">232</span> }<a name="line.232"></a> -<span class="sourceLineNo">233</span> List<ServerName> otherRegionServers = replicationTracker.getListOfRegionServers().stream()<a name="line.233"></a> -<span class="sourceLineNo">234</span> .map(ServerName::valueOf).collect(Collectors.toList());<a name="line.234"></a> -<span class="sourceLineNo">235</span> LOG.info(<a name="line.235"></a> -<span class="sourceLineNo">236</span> "Current list of replicators: " + currentReplicators + " other RSs: " + otherRegionServers);<a name="line.236"></a> -<span class="sourceLineNo">237</span><a name="line.237"></a> -<span class="sourceLineNo">238</span> // Look if there's anything to process after a restart<a name="line.238"></a> -<span class="sourceLineNo">239</span> for (ServerName rs : currentReplicators) {<a name="line.239"></a> -<span class="sourceLineNo">240</span> if (!otherRegionServers.contains(rs)) {<a name="line.240"></a> -<span class="sourceLineNo">241</span> transferQueues(rs);<a name="line.241"></a> -<span class="sourceLineNo">242</span> }<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><a name="line.245"></a> -<span class="sourceLineNo">246</span> /**<a name="line.246"></a> -<span class="sourceLineNo">247</span> * 1. Add peer to replicationPeers 2. Add the normal source and related replication queue 3. Add<a name="line.247"></a> -<span class="sourceLineNo">248</span> * HFile Refs<a name="line.248"></a> -<span class="sourceLineNo">249</span> * @param peerId the id of replication peer<a name="line.249"></a> -<span class="sourceLineNo">250</span> */<a name="line.250"></a> -<span class="sourceLineNo">251</span> public void addPeer(String peerId) throws IOException {<a name="line.251"></a> -<span class="sourceLineNo">252</span> boolean added = false;<a name="line.252"></a> -<span class="sourceLineNo">253</span> try {<a name="line.253"></a> -<span class="sourceLineNo">254</span> added = this.replicationPeers.addPeer(peerId);<a name="line.254"></a> -<span class="sourceLineNo">255</span> } catch (ReplicationException e) {<a name="line.255"></a> -<span class="sourceLineNo">256</span> throw new IOException(e);<a name="line.256"></a> -<span class="sourceLineNo">257</span> }<a name="line.257"></a> -<span class="sourceLineNo">258</span> if (added) {<a name="line.258"></a> -<span class="sourceLineNo">259</span> addSource(peerId);<a name="line.259"></a> -<span class="sourceLineNo">260</span> if (replicationForBulkLoadDataEnabled) {<a name="line.260"></a> -<span class="sourceLineNo">261</span> throwIOExceptionWhenFail(() -> this.queueStorage.addPeerToHFileRefs(peerId));<a name="line.261"></a> -<span class="sourceLineNo">262</span> }<a name="line.262"></a> -<span class="sourceLineNo">263</span> }<a name="line.263"></a> -<span class="sourceLineNo">264</span> }<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> * 1. Remove peer for replicationPeers 2. Remove all the recovered sources for the specified id<a name="line.267"></a> -<span class="sourceLineNo">268</span> * and related replication queues 3. Remove the normal source and related replication queue 4.<a name="line.268"></a> -<span class="sourceLineNo">269</span> * Remove HFile Refs<a name="line.269"></a> -<span class="sourceLineNo">270</span> * @param peerId the id of the replication peer<a name="line.270"></a> -<span class="sourceLineNo">271</span> */<a name="line.271"></a> -<span class="sourceLineNo">272</span> public void removePeer(String peerId) {<a name="line.272"></a> -<span class="sourceLineNo">273</span> replicationPeers.removePeer(peerId);<a name="line.273"></a> -<span class="sourceLineNo">274</span> String terminateMessage = "Replication stream was removed by a user";<a name="line.274"></a> -<span class="sourceLineNo">275</span> List<ReplicationSourceInterface> oldSourcesToDelete = new ArrayList<>();<a name="line.275"></a> -<span class="sourceLineNo">276</span> // synchronized on oldsources to avoid adding recovered source for the to-be-removed peer<a name="line.276"></a> -<span class="sourceLineNo">277</span> // see NodeFailoverWorker.run<a name="line.277"></a> -<span class="sourceLineNo">278</span> synchronized (this.oldsources) {<a name="line.278"></a> -<span class="sourceLineNo">279</span> // First close all the recovered sources for this peer<a name="line.279"></a> -<span class="sourceLineNo">280</span> for (ReplicationSourceInterface src : oldsources) {<a name="line.280"></a> -<span class="sourceLineNo">281</span> if (peerId.equals(src.getPeerId())) {<a name="line.281"></a> -<span class="sourceLineNo">282</span> oldSourcesToDelete.add(src);<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> for (ReplicationSourceInterface src : oldSourcesToDelete) {<a name="line.285"></a> -<span class="sourceLineNo">286</span> src.terminate(terminateMessage);<a name="line.286"></a> -<span class="sourceLineNo">287</span> removeRecoveredSource(src);<a name="line.287"></a> -<span class="sourceLineNo">288</span> }<a name="line.288"></a> -<span class="sourceLineNo">289</span> }<a name="line.289"></a> -<span class="sourceLineNo">290</span> LOG.info(<a name="line.290"></a> -<span class="sourceLineNo">291</span> "Number of deleted recovered sources for " + peerId + ": " + oldSourcesToDelete.size());<a name="line.291"></a> -<span class="sourceLineNo">292</span> // Now close the normal source for this peer<a name="line.292"></a> -<span class="sourceLineNo">293</span> ReplicationSourceInterface srcToRemove = this.sources.get(peerId);<a name="line.293"></a> -<span class="sourceLineNo">294</span> if (srcToRemove != null) {<a name="line.294"></a> -<span class="sourceLineNo">295</span> srcToRemove.terminate(terminateMessage);<a name="line.295"></a> -<span class="sourceLineNo">296</span> removeSource(srcToRemove);<a name="line.296"></a> -<span class="sourceLineNo">297</span> } else {<a name="line.297"></a> -<span class="sourceLineNo">298</span> // This only happened in unit test TestReplicationSourceManager#testPeerRemovalCleanup<a name="line.298"></a> -<span class="sourceLineNo">299</span> // Delete queue from storage and memory and queue id is same with peer id for normal<a name="line.299"></a> -<span class="sourceLineNo">300</span> // source<a name="line.300"></a> -<span class="sourceLineNo">301</span> deleteQueue(peerId);<a name="line.301"></a> -<span class="sourceLineNo">302</span> this.walsById.remove(peerId);<a name="line.302"></a> -<span class="sourceLineNo">303</span> }<a name="line.303"></a> -<span class="sourceLineNo">304</span><a name="line.304"></a> -<span class="sourceLineNo">305</span> // Remove HFile Refs<a name="line.305"></a> -<span class="sourceLineNo">306</span> abortWhenFail(() -> this.queueStorage.removePeerFromHFileRefs(peerId));<a name="line.306"></a> -<span class="sourceLineNo">307</span> }<a name="line.307"></a> -<span class="sourceLineNo">308</span><a name="line.308"></a> -<span class="sourceLineNo">309</span> /**<a name="line.309"></a> -<span class="sourceLineNo">310</span> * Factory method to create a replication source<a name="line.310"></a> -<span class="sourceLineNo">311</span> * @param queueId the id of the replication queue<a name="line.311"></a> -<span class="sourceLineNo">312</span> * @return the created source<a name="line.312"></a> -<span class="sourceLineNo">313</span> */<a name="line.313"></a> -<span class="sourceLineNo">314</span> private ReplicationSourceInterface createSource(String queueId, ReplicationPeer replicationPeer)<a name="line.314"></a> -<span class="sourceLineNo">315</span> throws IOException {<a name="line.315"></a> -<span class="sourceLineNo">316</span> ReplicationSourceInterface src = ReplicationSourceFactory.create(conf, queueId);<a name="line.316"></a> -<span class="sourceLineNo">317</span><a name="line.317"></a> -<span class="sourceLineNo">318</span> MetricsSource metrics = new MetricsSource(queueId);<a name="line.318"></a> -<span class="sourceLineNo">319</span> // init replication source<a name="line.319"></a> -<span class="sourceLineNo">320</span> src.init(conf, fs, this, queueStorage, replicationPeer, server, queueId, clusterId,<a name="line.320"></a> -<span class="sourceLineNo">321</span> walFileLengthProvider, metrics);<a name="line.321"></a> -<span class="sourceLineNo">322</span> return src;<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> * Add a normal source for the given peer on this region server. Meanwhile, add new replication<a name="line.326"></a> -<span class="sourceLineNo">327</span> * queue to storage. For the newly added peer, we only need to enqueue the latest log of each wal<a name="line.327"></a> -<span class="sourceLineNo">328</span> * group and do replication<a name="line.328"></a> -<span class="sourceLineNo">329</span> * @param peerId the id of the replication peer<a name="line.329"></a> -<span class="sourceLineNo">330</span> * @return the source that was created<a name="line.330"></a> -<span class="sourceLineNo">331</span> */<a name="line.331"></a> -<span class="sourceLineNo">332</span> @VisibleForTesting<a name="line.332"></a> -<span class="sourceLineNo">333</span> ReplicationSourceInterface addSource(String peerId) throws IOException {<a name="line.333"></a> -<span class="sourceLineNo">334</span> ReplicationPeer peer = replicationPeers.getPeer(peerId);<a name="line.334"></a> -<span class="sourceLineNo">335</span> ReplicationSourceInterface src = createSource(peerId, peer);<a name="line.335"></a> -<span class="sourceLineNo">336</span> // synchronized on latestPaths to avoid missing the new log<a name="line.336"></a> -<span class="sourceLineNo">337</span> synchronized (this.latestPaths) {<a name="line.337"></a> -<span class="sourceLineNo">338</span> this.sources.put(peerId, src);<a name="line.338"></a> -<span class="sourceLineNo">339</span> Map<String, SortedSet<String>> walsByGroup = new HashMap<>();<a name="line.339"></a> -<span class="sourceLineNo">340</span> this.walsById.put(peerId, walsByGroup);<a name="line.340"></a> -<span class="sourceLineNo">341</span> // Add the latest wal to that source's queue<a name="line.341"></a> -<span class="sourceLineNo">342</span> if (this.latestPaths.size() > 0) {<a name="line.342"></a> -<span class="sourceLineNo">343</span> for (Path logPath : latestPaths) {<a name="line.343"></a> -<span class="sourceLineNo">344</span> String name = logPath.getName();<a name="line.344"></a> -<span class="sourceLineNo">345</span> String walPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(name);<a name="line.345"></a> -<span class="sourceLineNo">346</span> SortedSet<String> logs = new TreeSet<>();<a name="line.346"></a> -<span class="sourceLineNo">347</span> logs.add(name);<a name="line.347"></a> -<span class="sourceLineNo">348</span> walsByGroup.put(walPrefix, logs);<a name="line.348"></a> -<span class="sourceLineNo">349</span> // Abort RS and throw exception to make add peer failed<a name="line.349"></a> -<span class="sourceLineNo">350</span> abortAndThrowIOExceptionWhenFail(<a name="line.350"></a> -<span class="sourceLineNo">351</span> () -> this.queueStorage.addWAL(server.getServerName(), peerId, name));<a name="line.351"></a> -<span class="sourceLineNo">352</span> src.enqueueLog(logPath);<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> }<a name="line.355"></a> -<span class="sourceLineNo">356</span> src.startup();<a name="line.356"></a> -<span class="sourceLineNo">357</span> return src;<a name="line.357"></a> -<span class="sourceLineNo">358</span> }<a name="line.358"></a> -<span class="sourceLineNo">359</span><a name="line.359"></a> -<span class="sourceLineNo">360</span> /**<a name="line.360"></a> -<span class="sourceLineNo">361</span> * Close the previous replication sources of this peer id and open new sources to trigger the new<a name="line.361"></a> -<span class="sourceLineNo">362</span> * replication state changes or new replication config changes. Here we don't need to change<a name="line.362"></a> -<span class="sourceLineNo">363</span> * replication queue storage and only to enqueue all logs to the new replication source<a name="line.363"></a> -<span class="sourceLineNo">364</span> * @param peerId the id of the replication peer<a name="line.364"></a> -<span class="sourceLineNo">365</span> * @throws IOException<a name="line.365"></a> -<span class="sourceLineNo">366</span> */<a name="line.366"></a> -<span class="sourceLineNo">367</span> public void refreshSources(String peerId) throws IOException {<a name="line.367"></a> -<span class="sourceLineNo">368</span> String terminateMessage = "Peer " + peerId +<a name="line.368"></a> -<span class="sourceLineNo">369</span> " state or config changed. Will close the previous replication source and open a new one";<a name="line.369"></a> -<span class="sourceLineNo">370</span> ReplicationPeer peer = replicationPeers.getPeer(peerId);<a name="line.370"></a> -<span class="sourceLineNo">371</span> ReplicationSourceInterface src = createSource(peerId, peer);<a name="line.371"></a> -<span class="sourceLineNo">372</span> // synchronized on latestPaths to avoid missing the new log<a name="line.372"></a> -<span class="sourceLineNo">373</span> synchronized (this.latestPaths) {<a name="line.373"></a> -<span class="sourceLineNo">374</span> ReplicationSourceInterface toRemove = this.sources.put(peerId, src);<a name="line.374"></a> -<span class="sourceLineNo">375</span> if (toRemove != null) {<a name="line.375"></a> -<span class="sourceLineNo">376</span> LOG.info("Terminate replication source for " + toRemove.getPeerId());<a name="line.376"></a> -<span class="sourceLineNo">377</span> toRemove.terminate(terminateMessage);<a name="line.377"></a> -<span class="sourceLineNo">378</span> }<a name="line.378"></a> -<span class="sourceLineNo">379</span> for (SortedSet<String> walsByGroup : walsById.get(peerId).values()) {<a name="line.379"></a> -<span class="sourceLineNo">380</span> walsByGroup.forEach(wal -> src.enqueueLog(new Path(this.logDir, wal)));<a name="line.380"></a> +<span class="sourceLineNo">233</span> if (currentReplicators == null || currentReplicators.isEmpty()) {<a name="line.233"></a> +<span class="sourceLineNo">234</span> return;<a name="line.234"></a> +<span class="sourceLineNo">235</span> }<a name="line.235"></a> +<span class="sourceLineNo">236</span> List<ServerName> otherRegionServers = replicationTracker.getListOfRegionServers().stream()<a name="line.236"></a> +<span class="sourceLineNo">237</span> .map(ServerName::valueOf).collect(Collectors.toList());<a name="line.237"></a> +<span class="sourceLineNo">238</span> LOG.info(<a name="line.238"></a> +<span class="sourceLineNo">239</span> "Current list of replicators: " + currentReplicators + " other RSs: " + otherRegionServers);<a name="line.239"></a> +<span class="sourceLineNo">240</span><a name="line.240"></a> +<span class="sourceLineNo">241</span> // Look if there's anything to process after a restart<a name="line.241"></a> +<span class="sourceLineNo">242</span> for (ServerName rs : currentReplicators) {<a name="line.242"></a> +<span class="sourceLineNo">243</span> if (!otherRegionServers.contains(rs)) {<a name="line.243"></a> +<span class="sourceLineNo">244</span> transferQueues(rs);<a name="line.244"></a> +<span class="sourceLineNo">245</span> }<a name="line.245"></a> +<span class="sourceLineNo">246</span> }<a name="line.246"></a> +<span class="sourceLineNo">247</span> }<a name="line.247"></a> +<span class="sourceLineNo">248</span><a name="line.248"></a> +<span class="sourceLineNo">249</span> /**<a name="line.249"></a> +<span class="sourceLineNo">250</span> * 1. Add peer to replicationPeers 2. Add the normal source and related replication queue 3. Add<a name="line.250"></a> +<span class="sourceLineNo">251</span> * HFile Refs<a name="line.251"></a> +<span class="sourceLineNo">252</span> * @param peerId the id of replication peer<a name="line.252"></a> +<span class="sourceLineNo">253</span> */<a name="line.253"></a> +<span class="sourceLineNo">254</span> public void addPeer(String peerId) throws IOException {<a name="line.254"></a> +<span class="sourceLineNo">255</span> boolean added = false;<a name="line.255"></a> +<span class="sourceLineNo">256</span> try {<a name="line.256"></a> +<span class="sourceLineNo">257</span> added = this.replicationPeers.addPeer(peerId);<a name="line.257"></a> +<span class="sourceLineNo">258</span> } catch (ReplicationException e) {<a name="line.258"></a> +<span class="sourceLineNo">259</span> throw new IOException(e);<a name="line.259"></a> +<span class="sourceLineNo">260</span> }<a name="line.260"></a> +<span class="sourceLineNo">261</span> if (added) {<a name="line.261"></a> +<span class="sourceLineNo">262</span> addSource(peerId);<a name="line.262"></a> +<span class="sourceLineNo">263</span> if (replicationForBulkLoadDataEnabled) {<a name="line.263"></a> +<span class="sourceLineNo">264</span> throwIOExceptionWhenFail(() -> this.queueStorage.addPeerToHFileRefs(peerId));<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><a name="line.268"></a> +<span class="sourceLineNo">269</span> /**<a name="line.269"></a> +<span class="sourceLineNo">270</span> * 1. Remove peer for replicationPeers 2. Remove all the recovered sources for the specified id<a name="line.270"></a> +<span class="sourceLineNo">271</span> * and related replication queues 3. Remove the normal source and related replication queue 4.<a name="line.271"></a> +<span class="sourceLineNo">272</span> * Remove HFile Refs<a name="line.272"></a> +<span class="sourceLineNo">273</span> * @param peerId the id of the replication peer<a name="line.273"></a> +<span class="sourceLineNo">274</span> */<a name="line.274"></a> +<span class="sourceLineNo">275</span> public void removePeer(String peerId) {<a name="line.275"></a> +<span class="sourceLineNo">276</span> replicationPeers.removePeer(peerId);<a name="line.276"></a> +<span class="sourceLineNo">277</span> String terminateMessage = "Replication stream was removed by a user";<a name="line.277"></a> +<span class="sourceLineNo">278</span> List<ReplicationSourceInterface> oldSourcesToDelete = new ArrayList<>();<a name="line.278"></a> +<span class="sourceLineNo">279</span> // synchronized on oldsources to avoid adding recovered source for the to-be-removed peer<a name="line.279"></a> +<span class="sourceLineNo">280</span> // see NodeFailoverWorker.run<a name="line.280"></a> +<span class="sourceLineNo">281</span> synchronized (this.oldsources) {<a name="line.281"></a> +<span class="sourceLineNo">282</span> // First close all the recovered sources for this peer<a name="line.282"></a> +<span class="sourceLineNo">283</span> for (ReplicationSourceInterface src : oldsources) {<a name="line.283"></a> +<span class="sourceLineNo">284</span> if (peerId.equals(src.getPeerId())) {<a name="line.284"></a> +<span class="sourceLineNo">285</span> oldSourcesToDelete.add(src);<a name="line.285"></a> +<span class="sourceLineNo">286</span> }<a name="line.286"></a> +<span class="sourceLineNo">287</span> }<a name="line.287"></a> +<span class="sourceLineNo">288</span> for (ReplicationSourceInterface src : oldSourcesToDelete) {<a name="line.288"></a> +<span class="sourceLineNo">289</span> src.terminate(terminateMessage);<a name="line.289"></a> +<span class="sourceLineNo">290</span> removeRecoveredSource(src);<a name="line.290"></a> +<span class="sourceLineNo">291</span> }<a name="line.291"></a> +<span class="sourceLineNo">292</span> }<a name="line.292"></a> +<span class="sourceLineNo">293</span> LOG.info(<a name="line.293"></a> +<span class="sourceLineNo">294</span> "Number of deleted recovered sources for " + peerId + ": " + oldSourcesToDelete.size());<a name="line.294"></a> +<span class="sourceLineNo">295</span> // Now close the normal source for this peer<a name="line.295"></a> +<span class="sourceLineNo">296</span> ReplicationSourceInterface srcToRemove = this.sources.get(peerId);<a name="line.296"></a> +<span class="sourceLineNo">297</span> if (srcToRemove != null) {<a name="line.297"></a> +<span class="sourceLineNo">298</span> srcToRemove.terminate(terminateMessage);<a name="line.298"></a> +<span class="sourceLineNo">299</span> removeSource(srcToRemove);<a name="line.299"></a> +<span class="sourceLineNo">300</span> } else {<a name="line.300"></a> +<span class="sourceLineNo">301</span> // This only happened in unit test TestReplicationSourceManager#testPeerRemovalCleanup<a name="line.301"></a> +<span class="sourceLineNo">302</span> // Delete queue from storage and memory and queue id is same with peer id for normal<a name="line.302"></a> +<span class="sourceLineNo">303</span> // source<a name="line.303"></a> +<span class="sourceLineNo">304</span> deleteQueue(peerId);<a name="line.304"></a> +<span class="sourceLineNo">305</span> this.walsById.remove(peerId);<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> // Remove HFile Refs<a name="line.308"></a> +<span class="sourceLineNo">309</span> abortWhenFail(() -> this.queueStorage.removePeerFromHFileRefs(peerId));<a name="line.309"></a> +<span class="sourceLineNo">310</span> }<a name="line.310"></a> +<span class="sourceLineNo">311</span><a name="line.311"></a> +<span class="sourceLineNo">312</span> /**<a name="line.312"></a> +<span class="sourceLineNo">313</span> * Factory method to create a replication source<a name="line.313"></a> +<span class="sourceLineNo">314</span> * @param queueId the id of the replication queue<a name="line.314"></a> +<span class="sourceLineNo">315</span> * @return the created source<a name="line.315"></a> +<span class="sourceLineNo">316</span> */<a name="line.316"></a> +<span class="sourceLineNo">317</span> private ReplicationSourceInterface createSource(String queueId, ReplicationPeer replicationPeer)<a name="line.317"></a> +<span class="sourceLineNo">318</span> throws IOException {<a name="line.318"></a> +<span class="sourceLineNo">319</span> ReplicationSourceInterface src = ReplicationSourceFactory.create(conf, queueId);<a name="line.319"></a> +<span class="sourceLineNo">320</span><a name="line.320"></a> +<span class="sourceLineNo">321</span> MetricsSource metrics = new MetricsSource(queueId);<a name="line.321"></a> +<span class="sourceLineNo">322</span> // init replication source<a name="line.322"></a> +<span class="sourceLineNo">323</span> src.init(conf, fs, this, queueStorage, replicationPeer, server, queueId, clusterId,<a name="line.323"></a> +<span class="sourceLineNo">324</span> walFileLengthProvider, metrics);<a name="line.324"></a> +<span class="sourceLineNo">325</span> return src;<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> * Add a normal source for the given peer on this region server. Meanwhile, add new replication<a name="line.329"></a> +<span class="sourceLineNo">330</span> * queue to storage. For the newly added peer, we only need to enqueue the latest log of each wal<a name="line.330"></a> +<span class="sourceLineNo">331</span> * group and do replication<a name="line.331"></a> +<span class="sourceLineNo">332</span> * @param peerId the id of the replication peer<a name="line.332"></a> +<span class="sourceLineNo">333</span> * @return the source that was created<a name="line.333"></a> +<span class="sourceLineNo">334</span> */<a name="line.334"></a> +<span class="sourceLineNo">335</span> @VisibleForTesting<a name="line.335"></a> +<span class="sourceLineNo">336</span> ReplicationSourceInterface addSource(String peerId) throws IOException {<a name="line.336"></a> +<span class="sourceLineNo">337</span> ReplicationPeer peer = replicationPeers.getPeer(peerId);<a name="line.337"></a> +<span class="sourceLineNo">338</span> ReplicationSourceInterface src = createSource(peerId, peer);<a name="line.338"></a> +<span class="sourceLineNo">339</span> // synchronized on latestPaths to avoid missing the new log<a name="line.339"></a> +<span class="sourceLineNo">340</span> synchronized (this.latestPaths) {<a name="line.340"></a> +<span class="sourceLineNo">341</span> this.sources.put(peerId, src);<a name="line.341"></a> +<span class="sourceLineNo">342</span> Map<String, NavigableSet<String>> walsByGroup = new HashMap<>();<a name="line.342"></a> +<span class="sourceLineNo">343</span> this.walsById.put(peerId, walsByGroup);<a name="line.343"></a> +<span class="sourceLineNo">344</span> // Add the latest wal to that source's queue<a name="line.344"></a> +<span class="sourceLineNo">345</span> if (this.latestPaths.size() > 0) {<a name="line.345"></a> +<span class="sourceLineNo">346</span> for (Path logPath : latestPaths) {<a name="line.346"></a> +<span class="sourceLineNo">347</span> String name = logPath.getName();<a name="line.347"></a> +<span class="sourceLineNo">348</span> String walPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(name);<a name="line.348"></a> +<span class="sourceLineNo">349</span> NavigableSet<String> logs = new TreeSet<>();<a name="line.349"></a> +<span class="sourceLineNo">350</span> logs.add(name);<a name="line.350"></a> +<span class="sourceLineNo">351</span> walsByGroup.put(walPrefix, logs);<a name="line.351"></a> +<span class="sourceLineNo">352</span> // Abort RS and throw exception to make add peer failed<a name="line.352"></a> +<span class="sourceLineNo">353</span> abortAndThrowIOExceptionWhenFail(<a name="line.353"></a> +<span class="sourceLineNo">354</span> () -> this.queueStorage.addWAL(server.getServerName(), peerId, name));<a name="line.354"></a> +<span class="sourceLineNo">355</span> src.enqueueLog(logPath);<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> }<a name="line.358"></a> +<span class="sourceLineNo">359</span> src.startup();<a name="line.359"></a> +<span class="sourceLineNo">360</span> return src;<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> /**<a name="line.363"></a> +<span class="sourceLineNo">364</span> * Close the previous replication sources of this peer id and open new sources to trigger the new<a name="line.364"></a> +<span class="sourceLineNo">365</span> * replication state changes or new replication config changes. Here we don't need to change<a name="line.365"></a> +<span class="sourceLineNo">366</span> * replication queue storage and only to enqueue all logs to the new replication source<a name="line.366"></a> +<span class="sourceLineNo">367</span> * @param peerId the id of the replication peer<a name="line.367"></a> +<span class="sourceLineNo">368</span> * @throws IOException<a name="line.368"></a> +<span class="sourceLineNo">369</span> */<a name="line.369"></a> +<span class="sourceLineNo">370</span> public void refreshSources(String peerId) throws IOException {<a name="line.370"></a> +<span class="sourceLineNo">371</span> String terminateMessage = "Peer " + peerId +<a name="line.371"></a> +<span class="sourceLineNo">372</span> " state or config changed. Will close the previous replication source and open a new one";<a name="line.372"></a> +<span class="sourceLineNo">373</span> ReplicationPeer peer = replicationPeers.getPeer(peerId);<a name="line.373"></a> +<span class="sourceLineNo">374</span> ReplicationSourceInterface src = createSource(peerId, peer);<a name="line.374"></a> +<span class="sourceLineNo">375</span> // synchronized on latestPaths to avoid missing the new log<a name="line.375"></a> +<span class="sourceLineNo">376</span> synchronized (this.latestPaths) {<a name="line.376"></a> +<span class="sourceLineNo">377</span> ReplicationSourceInterface toRemove = this.sources.put(peerId, src);<a name="line.377"></a> +<span class="sourceLineNo">378</span> if (toRemove != null) {<a name="line.378"></a> +<span class="sourceLineNo">379</span> LOG.info("Terminate replication source for " + toRemove.getPeerId());<a name="line.379"></a> +<span class="sourceLineNo">380</span> toRemove.terminate(terminateMessage);<a name="line.380"></a> <span class="sourceLineNo">381</span> }<a name="line.381"></a> -<span class="sourceLineNo">382</span> }<a name="line.382"></a> -<span class="sourceLineNo">383</span> LOG.info("Startup replication source for " + src.getPeerId());<a name="line.383"></a> -<span class="sourceLineNo">384</span> src.startup();<a name="line.384"></a> -<span class="sourceLineNo">385</span><a name="line.385"></a> -<span class="sourceLineNo">386</span> List<ReplicationSourceInterface> toStartup = new ArrayList<>();<a name="line.386"></a> -<span class="sourceLineNo">387</span> // synchronized on oldsources to avoid race with NodeFailoverWorker<a name="line.387"></a> -<span class="sourceLineNo">388</span> synchronized (this.oldsources) {<a name="line.388"></a> -<span class="sourceLineNo">389</span> List<String> previousQueueIds = new ArrayList<>();<a name="line.389"></a> -<span class="sourceLineNo">390</span> for (ReplicationSourceInterface oldSource : this.oldsources) {<a name="line.390"></a> -<span class="sourceLineNo">391</span> if (oldSource.getPeerId().equals(peerId)) {<a name="line.391"></a> -<span class="sourceLineNo">392</span> previousQueueIds.add(oldSource.getQueueId());<a name="line.392"></a> -<span class="sourceLineNo">393</span> oldSource.terminate(terminateMessage);<a name="line.393"></a> -<span class="sourceLineNo">394</span> this.oldsources.remove(oldSource);<a name="line.394"></a> -<span class="sourceLineNo">395</span> }<a name="line.395"></a> -<span class="sourceLineNo">396</span> }<a name="line.396"></a> -<span class="sourceLineNo">397</span> for (String queueId : previousQueueIds) {<a name="line.397"></a> -<span class="sourceLineNo">398</span> ReplicationSourceInterface replicationSource = createSource(queueId, peer);<a name="line.398"></a> -<span class="sourceLineNo">399</span> this.oldsources.add(replicationSource);<a name="line.399"></a> -<span class="sourceLineNo">400</span> for (SortedSet<String> walsByGroup : walsByIdRecoveredQueues.get(queueId).values()) {<a name="line.400"></a> -<span class="sourceLineNo">401</span> walsByGroup.forEach(wal -> src.enqueueLog(new Path(wal)));<a name="line.401"></a> -<span class="sourceLineNo">402</span> }<a name="line.402"></a> -<span class="sourceLineNo">403</span> toStartup.add(replicationSource);<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> for (ReplicationSourceInterface replicationSource : toStartup) {<a name="line.406"></a> -<span class="sourceLineNo">407</span> replicationSource.startup();<a name="line.407"></a> +<span class="sourceLineNo">382</span> for (SortedSet<String> walsByGroup : walsById.get(peerId).values()) {<a name="line.382"></a> +<span class="sourceLineNo">383</span> walsByGroup.forEach(wal -> src.enqueueLog(new Path(this.logDir, wal)));<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> LOG.info("Startup replication source for " + src.getPeerId());<a name="line.386"></a> +<span class="sourceLineNo">387</span> src.startup();<a name="line.387"></a> +<span class="sourceLineNo">388</span><a name="line.388"></a> +<span class="sourceLineNo">389</span> List<ReplicationSourceInterface> toStartup = new ArrayList<>();<a name="line.389"></a> +<span class="sourceLineNo">390</span> // synchronized on oldsources to avoid race with NodeFailoverWorker<a name="line.390"></a> +<span class="sourceLineNo">391</span> synchronized (this.oldsources) {<a name="line.391"></a> +<span class="sourceLineNo">392</span> List<String> previousQueueIds = new ArrayList<>();<a name="line.392"></a> +<span class="sourceLineNo">393</span> for (ReplicationSourceInterface oldSource : this.oldsources) {<a name="line.393"></a> +<span class="sourceLineNo">394</span> if (oldSource.getPeerId().equals(peerId)) {<a name="line.394"></a> +<span class="sourceLineNo">395</span> previousQueueIds.add(oldSource.getQueueId());<a name="line.395"></a> +<span class="sourceLineNo">396</span> oldSource.terminate(terminateMessage);<a name="line.396"></a> +<span class="sourceLineNo">397</span> this.oldsources.remove(oldSource);<a name="line.397"></a> +<span class="sourceLineNo">398</span> }<a name="line.398"></a> +<span class="sourceLineNo">399</span> }<a name="line.399"></a> +<span class="sourceLineNo">400</span> for (String queueId : previousQueueIds) {<a name="line.400"></a> +<span class="sourceLineNo">401</span> ReplicationSourceInterface replicationSource = createSource(queueId, peer);<a name="line.401"></a> +<span class="sourceLineNo">402</span> this.oldsources.add(replicationSource);<a name="line.402"></a> +<span class="sourceLineNo">403</span> for (SortedSet<String> walsByGroup : walsByIdRecoveredQueues.get(queueId).values()) {<a name="line.403"></a> +<span class="sourceLineNo">404</span> walsByGroup.forEach(wal -> src.enqueueLog(new Path(wal)));<a name="line.404"></a> +<span class="sourceLineNo">405</span> }<a name="line.405"></a> +<span class="sourceLineNo">406</span> toStartup.add(replicationSource);<a name="line.406"></a> +<span class="sourceLineNo">407</span> }<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> * Clear the metrics and related replication queue of the specified old source<a name="line.412"></a> -<span class="sourceLineNo">413</span> * @param src source to clear<a name="line.413"></a> -<span class="sourceLineNo">414</span> */<a name="line.414"></a> -<span class="sourceLineNo">415</span> void removeRecoveredSource(ReplicationSourceInterface src) {<a name="line.415"></a> -<span class="sourceLineNo">416</span> LOG.info("Done with the recovered queue " + src.getQueueId());<a name="line.416"></a> -<span class="sourceLineNo">417</span> src.getSourceMetrics().clear();<a name="line.417"></a> -<span class="sourceLineNo">418</span> this.oldsources.remove(src);<a name="line.418"></a> -<span class="sourceLineNo">419</span> // Delete queue from storage and memory<a name="line.419"></a> -<span class="sourceLineNo">420</span> deleteQueue(src.getQueueId());<a name="line.420"></a> -<span class="sourceLineNo">421</span> this.walsByIdRecoveredQueues.remove(src.getQueueId());<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> * Clear the metrics and related replication queue of the specified old source<a name="line.425"></a> -<span class="sourceLineNo">426</span> * @param src source to clear<a name="line.426"></a> -<span class="sourceLineNo">427</span> */<a name="line.427"></a> -<span class="sourceLineNo">428</span> void removeSource(ReplicationSourceInterface src) {<a name="line.428"></a> -<span class="sourceLineNo">429</span> LOG.info("Done with the queue " + src.getQueueId());<a name="line.429"></a> -<span class="sourceLineNo">430</span> src.getSourceMetrics().clear();<a name="line.430"></a> -<span class="sourceLineNo">431</span> this.sources.remove(src.getPeerId());<a name="line.431"></a> -<span class="sourceLineNo">432</span> // Delete queue from storage and memory<a name="line.432"></a> -<span class="sourceLineNo">433</span> deleteQueue(src.getQueueId());<a name="line.433"></a> -<span class="sourceLineNo">434</span> this.walsById.remove(src.getQueueId());<a name="line.434"></a> -<span class="sourceLineNo">435</span> }<a name="line.435"></a> -<span class="sourceLineNo">436</span><a name="line.436"></a> -<span class="sourceLineNo">437</span> /**<a name="line.437"></a> -<span class="sourceLineNo">438</span> * Delete a complete queue of wals associated with a replication source<a name="line.438"></a> -<span class="sourceLineNo">439</span> * @param queueId the id of replication queue to delete<a name="line.439"></a> -<span class="sourceLineNo">440</span> */<a name="line.440"></a> -<span class="sourceLineNo">441</span> private void deleteQueue(String queueId) {<a name="line.441"></a> -<span class="sourceLineNo">442</span> abortWhenFail(() -> this.queueStorage.removeQueue(server.getServerName(), queueId));<a name="line.442"></a> -<span class="sourceLineNo">443</span> }<a name="line.443"></a> -<span class="sourceLineNo">444</span><a name="line.444"></a> -<span class="sourceLineNo">445</span> @FunctionalInterface<a name="line.445"></a> -<span class="sourceLineNo">446</span> private interface ReplicationQueueOperation {<a name="line.446"></a> -<span class="sourceLineNo">447</span> void exec() throws ReplicationException;<a name="line.447"></a> -<span class="sourceLineNo">448</span> }<a name="line.448"></a> -<span class="sourceLineNo">449</span><a name="line.449"></a> -<span class="sourceLineNo">450</span> private void abortWhenFail(ReplicationQueueOperation op) {<a name="line.450"></a> -<span class="sourceLineNo">451</span> try {<a name="line.451"></a> -<span class="sourceLineNo">452</span> op.exec();<a name="line.452"></a> -<span class="sourceLineNo">453</span> } catch (ReplicationException e) {<a name="line.453"></a> -<span class="sourceLineNo">454</span> server.abort("Failed to operate on replication queue", e);<a name="line.454"></a> -<span class="sourceLineNo">455</span> }<a name="line.455"></a> -<span class="sourceLineNo">456</span> }<a name="line.456"></a> -<span class="sourceLineNo">457</span><a name="line.457"></a> -<span class="sourceLineNo">458</span> private void throwIOExceptionWhenFail(ReplicationQueueOperation op) throws IOException {<a name="line.458"></a> -<span class="sourceLineNo">459</span> try {<a name="line.459"></a> -<span class="sourceLineNo">460</span> op.exec();<a name="line.460"></a> -<span class="sourceLineNo">461</span> } catch (ReplicationException e) {<a name="line.461"></a> -<span class="sourceLineNo">462</span> throw new IOException(e);<a name="line.462"></a> -<span class="sourceLineNo">463</span> }<a name="line.463"></a> -<span class="sourceLineNo">464</span> }<a name="line.464"></a> -<span class="sourceLineNo">465</span><a name="line.465"></a> -<span class="sourceLineNo">466</span> private void abortAndThrowIOExceptionWhenFail(ReplicationQueueOperation op) throws IOException {<a name="line.466"></a> -<span class="sourceLineNo">467</span> try {<a name="line.467"></a> -<span class="sourceLineNo">468</span> op.exec();<a name="line.468"></a> -<span class="sourceLineNo">469</span> } catch (ReplicationException e) {<a name="line.469"></a> -<span class="sourceLineNo">470</span> server.abort("Failed to operate on replication queue", e);<a name="line.470"></a> -<span class="sourceLineNo">471</span> throw new IOException(e);<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> /**<a name="line.475"></a> -<span class="sourceLineNo">476</span> * This method will log the current position to storage. And also clean old logs from the<a name="line.476"></a> -<span class="sourceLineNo">477</span> * replication queue.<a name="line.477"></a> -<span class="sourceLineNo">478</span> * @param log Path to the log currently being replicated<a name="line.478"></a> -<span class="sourceLineNo">479</span> * @param queueId id of the replication queue<a name="line.479"></a> -<span class="sourceLineNo">480</span> * @param position current location in the log<a name="line.480"></a> -<span class="sourceLineNo">481</span> * @param queueRecovered indicates if this queue comes from another region server<a name="line.481"></a> -<span class="sourceLineNo">482</span> */<a name="line.482"></a> -<span class="sourceLineNo">483</span> public void logPositionAndClean
<TRUNCATED>
