http://git-wip-us.apache.org/repos/asf/hbase-site/blob/de18d468/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/HBaseInterClusterReplicationEndpoint.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/HBaseInterClusterReplicationEndpoint.html b/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/HBaseInterClusterReplicationEndpoint.html index d5f3932..166de59 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/HBaseInterClusterReplicationEndpoint.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/HBaseInterClusterReplicationEndpoint.html @@ -32,9 +32,9 @@ <span class="sourceLineNo">024</span>import java.net.UnknownHostException;<a name="line.24"></a> <span class="sourceLineNo">025</span>import java.util.ArrayList;<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.List;<a name="line.28"></a> -<span class="sourceLineNo">029</span>import java.util.Map;<a name="line.29"></a> +<span class="sourceLineNo">027</span>import java.util.List;<a name="line.27"></a> +<span class="sourceLineNo">028</span>import java.util.Map;<a name="line.28"></a> +<span class="sourceLineNo">029</span>import java.util.TreeMap;<a name="line.29"></a> <span class="sourceLineNo">030</span>import java.util.concurrent.Callable;<a name="line.30"></a> <span class="sourceLineNo">031</span>import java.util.concurrent.CompletionService;<a name="line.31"></a> <span class="sourceLineNo">032</span>import java.util.concurrent.ExecutionException;<a name="line.32"></a> @@ -45,473 +45,494 @@ <span class="sourceLineNo">037</span>import java.util.concurrent.TimeUnit;<a name="line.37"></a> <span class="sourceLineNo">038</span>import java.util.regex.Matcher;<a name="line.38"></a> <span class="sourceLineNo">039</span>import java.util.regex.Pattern;<a name="line.39"></a> -<span class="sourceLineNo">040</span>import org.apache.commons.lang3.StringUtils;<a name="line.40"></a> -<span class="sourceLineNo">041</span>import org.apache.hadoop.conf.Configuration;<a name="line.41"></a> -<span class="sourceLineNo">042</span>import org.apache.hadoop.fs.Path;<a name="line.42"></a> -<span class="sourceLineNo">043</span>import org.apache.hadoop.hbase.Abortable;<a name="line.43"></a> -<span class="sourceLineNo">044</span>import org.apache.hadoop.hbase.HBaseConfiguration;<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.TableName;<a name="line.46"></a> -<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.TableNotFoundException;<a name="line.47"></a> -<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.client.ClusterConnection;<a name="line.48"></a> -<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.client.Connection;<a name="line.49"></a> -<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.client.ConnectionFactory;<a name="line.50"></a> -<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.ipc.RpcServer;<a name="line.51"></a> -<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.protobuf.ReplicationProtbufUtil;<a name="line.52"></a> -<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.replication.HBaseReplicationEndpoint;<a name="line.53"></a> -<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.replication.regionserver.ReplicationSinkManager.SinkPeer;<a name="line.54"></a> -<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.55"></a> -<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.util.FSUtils;<a name="line.56"></a> -<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.wal.WAL.Entry;<a name="line.57"></a> -<span class="sourceLineNo">058</span>import org.apache.hadoop.ipc.RemoteException;<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><a name="line.64"></a> -<span class="sourceLineNo">065</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.AdminProtos.AdminService.BlockingInterface;<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> * A {@link org.apache.hadoop.hbase.replication.ReplicationEndpoint}<a name="line.68"></a> -<span class="sourceLineNo">069</span> * implementation for replicating to another HBase cluster.<a name="line.69"></a> -<span class="sourceLineNo">070</span> * For the slave cluster it selects a random number of peers<a name="line.70"></a> -<span class="sourceLineNo">071</span> * using a replication ratio. For example, if replication ration = 0.1<a name="line.71"></a> -<span class="sourceLineNo">072</span> * and slave cluster has 100 region servers, 10 will be selected.<a name="line.72"></a> -<span class="sourceLineNo">073</span> * <p><a name="line.73"></a> -<span class="sourceLineNo">074</span> * A stream is considered down when we cannot contact a region server on the<a name="line.74"></a> -<span class="sourceLineNo">075</span> * peer cluster for more than 55 seconds by default.<a name="line.75"></a> -<span class="sourceLineNo">076</span> * </p><a name="line.76"></a> -<span class="sourceLineNo">077</span> */<a name="line.77"></a> -<span class="sourceLineNo">078</span>@InterfaceAudience.Private<a name="line.78"></a> -<span class="sourceLineNo">079</span>public class HBaseInterClusterReplicationEndpoint extends HBaseReplicationEndpoint {<a name="line.79"></a> -<span class="sourceLineNo">080</span> private static final Logger LOG =<a name="line.80"></a> -<span class="sourceLineNo">081</span> LoggerFactory.getLogger(HBaseInterClusterReplicationEndpoint.class);<a name="line.81"></a> -<span class="sourceLineNo">082</span><a name="line.82"></a> -<span class="sourceLineNo">083</span> private static final long DEFAULT_MAX_TERMINATION_WAIT_MULTIPLIER = 2;<a name="line.83"></a> -<span class="sourceLineNo">084</span><a name="line.84"></a> -<span class="sourceLineNo">085</span> private ClusterConnection conn;<a name="line.85"></a> -<span class="sourceLineNo">086</span> private Configuration localConf;<a name="line.86"></a> -<span class="sourceLineNo">087</span> private Configuration conf;<a name="line.87"></a> -<span class="sourceLineNo">088</span> // How long should we sleep for each retry<a name="line.88"></a> -<span class="sourceLineNo">089</span> private long sleepForRetries;<a name="line.89"></a> -<span class="sourceLineNo">090</span> // Maximum number of retries before taking bold actions<a name="line.90"></a> -<span class="sourceLineNo">091</span> private int maxRetriesMultiplier;<a name="line.91"></a> -<span class="sourceLineNo">092</span> // Socket timeouts require even bolder actions since we don't want to DDOS<a name="line.92"></a> -<span class="sourceLineNo">093</span> private int socketTimeoutMultiplier;<a name="line.93"></a> -<span class="sourceLineNo">094</span> // Amount of time for shutdown to wait for all tasks to complete<a name="line.94"></a> -<span class="sourceLineNo">095</span> private long maxTerminationWait;<a name="line.95"></a> -<span class="sourceLineNo">096</span> // Size limit for replication RPCs, in bytes<a name="line.96"></a> -<span class="sourceLineNo">097</span> private int replicationRpcLimit;<a name="line.97"></a> -<span class="sourceLineNo">098</span> //Metrics for this source<a name="line.98"></a> -<span class="sourceLineNo">099</span> private MetricsSource metrics;<a name="line.99"></a> -<span class="sourceLineNo">100</span> // Handles connecting to peer region servers<a name="line.100"></a> -<span class="sourceLineNo">101</span> private ReplicationSinkManager replicationSinkMgr;<a name="line.101"></a> -<span class="sourceLineNo">102</span> private boolean peersSelected = false;<a name="line.102"></a> -<span class="sourceLineNo">103</span> private String replicationClusterId = "";<a name="line.103"></a> -<span class="sourceLineNo">104</span> private ThreadPoolExecutor exec;<a name="line.104"></a> -<span class="sourceLineNo">105</span> private int maxThreads;<a name="line.105"></a> -<span class="sourceLineNo">106</span> private Path baseNamespaceDir;<a name="line.106"></a> -<span class="sourceLineNo">107</span> private Path hfileArchiveDir;<a name="line.107"></a> -<span class="sourceLineNo">108</span> private boolean replicationBulkLoadDataEnabled;<a name="line.108"></a> -<span class="sourceLineNo">109</span> private Abortable abortable;<a name="line.109"></a> -<span class="sourceLineNo">110</span> private boolean dropOnDeletedTables;<a name="line.110"></a> -<span class="sourceLineNo">111</span><a name="line.111"></a> -<span class="sourceLineNo">112</span> @Override<a name="line.112"></a> -<span class="sourceLineNo">113</span> public void init(Context context) throws IOException {<a name="line.113"></a> -<span class="sourceLineNo">114</span> super.init(context);<a name="line.114"></a> -<span class="sourceLineNo">115</span> this.conf = HBaseConfiguration.create(ctx.getConfiguration());<a name="line.115"></a> -<span class="sourceLineNo">116</span> this.localConf = HBaseConfiguration.create(ctx.getLocalConfiguration());<a name="line.116"></a> -<span class="sourceLineNo">117</span> decorateConf();<a name="line.117"></a> -<span class="sourceLineNo">118</span> this.maxRetriesMultiplier = this.conf.getInt("replication.source.maxretriesmultiplier", 300);<a name="line.118"></a> -<span class="sourceLineNo">119</span> this.socketTimeoutMultiplier = this.conf.getInt("replication.source.socketTimeoutMultiplier",<a name="line.119"></a> -<span class="sourceLineNo">120</span> maxRetriesMultiplier);<a name="line.120"></a> -<span class="sourceLineNo">121</span> // A Replicator job is bound by the RPC timeout. We will wait this long for all Replicator<a name="line.121"></a> -<span class="sourceLineNo">122</span> // tasks to terminate when doStop() is called.<a name="line.122"></a> -<span class="sourceLineNo">123</span> long maxTerminationWaitMultiplier = this.conf.getLong(<a name="line.123"></a> -<span class="sourceLineNo">124</span> "replication.source.maxterminationmultiplier",<a name="line.124"></a> -<span class="sourceLineNo">125</span> DEFAULT_MAX_TERMINATION_WAIT_MULTIPLIER);<a name="line.125"></a> -<span class="sourceLineNo">126</span> this.maxTerminationWait = maxTerminationWaitMultiplier *<a name="line.126"></a> -<span class="sourceLineNo">127</span> this.conf.getLong(HConstants.HBASE_RPC_TIMEOUT_KEY, HConstants.DEFAULT_HBASE_RPC_TIMEOUT);<a name="line.127"></a> -<span class="sourceLineNo">128</span> // TODO: This connection is replication specific or we should make it particular to<a name="line.128"></a> -<span class="sourceLineNo">129</span> // replication and make replication specific settings such as compression or codec to use<a name="line.129"></a> -<span class="sourceLineNo">130</span> // passing Cells.<a name="line.130"></a> -<span class="sourceLineNo">131</span> this.conn = (ClusterConnection) ConnectionFactory.createConnection(this.conf);<a name="line.131"></a> -<span class="sourceLineNo">132</span> this.sleepForRetries =<a name="line.132"></a> -<span class="sourceLineNo">133</span> this.conf.getLong("replication.source.sleepforretries", 1000);<a name="line.133"></a> -<span class="sourceLineNo">134</span> this.metrics = context.getMetrics();<a name="line.134"></a> -<span class="sourceLineNo">135</span> // ReplicationQueueInfo parses the peerId out of the znode for us<a name="line.135"></a> -<span class="sourceLineNo">136</span> this.replicationSinkMgr = new ReplicationSinkManager(conn, ctx.getPeerId(), this, this.conf);<a name="line.136"></a> -<span class="sourceLineNo">137</span> // per sink thread pool<a name="line.137"></a> -<span class="sourceLineNo">138</span> this.maxThreads = this.conf.getInt(HConstants.REPLICATION_SOURCE_MAXTHREADS_KEY,<a name="line.138"></a> -<span class="sourceLineNo">139</span> HConstants.REPLICATION_SOURCE_MAXTHREADS_DEFAULT);<a name="line.139"></a> -<span class="sourceLineNo">140</span> this.exec = new ThreadPoolExecutor(maxThreads, maxThreads, 60, TimeUnit.SECONDS,<a name="line.140"></a> -<span class="sourceLineNo">141</span> new LinkedBlockingQueue<>());<a name="line.141"></a> -<span class="sourceLineNo">142</span> this.exec.allowCoreThreadTimeOut(true);<a name="line.142"></a> -<span class="sourceLineNo">143</span> this.abortable = ctx.getAbortable();<a name="line.143"></a> -<span class="sourceLineNo">144</span> // Set the size limit for replication RPCs to 95% of the max request size.<a name="line.144"></a> -<span class="sourceLineNo">145</span> // We could do with less slop if we have an accurate estimate of encoded size. Being<a name="line.145"></a> -<span class="sourceLineNo">146</span> // conservative for now.<a name="line.146"></a> -<span class="sourceLineNo">147</span> this.replicationRpcLimit = (int)(0.95 * conf.getLong(RpcServer.MAX_REQUEST_SIZE,<a name="line.147"></a> -<span class="sourceLineNo">148</span> RpcServer.DEFAULT_MAX_REQUEST_SIZE));<a name="line.148"></a> -<span class="sourceLineNo">149</span> this.dropOnDeletedTables =<a name="line.149"></a> -<span class="sourceLineNo">150</span> this.conf.getBoolean(HConstants.REPLICATION_DROP_ON_DELETED_TABLE_KEY, false);<a name="line.150"></a> -<span class="sourceLineNo">151</span><a name="line.151"></a> -<span class="sourceLineNo">152</span> this.replicationBulkLoadDataEnabled =<a name="line.152"></a> -<span class="sourceLineNo">153</span> conf.getBoolean(HConstants.REPLICATION_BULKLOAD_ENABLE_KEY,<a name="line.153"></a> -<span class="sourceLineNo">154</span> HConstants.REPLICATION_BULKLOAD_ENABLE_DEFAULT);<a name="line.154"></a> -<span class="sourceLineNo">155</span> if (this.replicationBulkLoadDataEnabled) {<a name="line.155"></a> -<span class="sourceLineNo">156</span> replicationClusterId = this.conf.get(HConstants.REPLICATION_CLUSTER_ID);<a name="line.156"></a> -<span class="sourceLineNo">157</span> }<a name="line.157"></a> -<span class="sourceLineNo">158</span> // Construct base namespace directory and hfile archive directory path<a name="line.158"></a> -<span class="sourceLineNo">159</span> Path rootDir = FSUtils.getRootDir(conf);<a name="line.159"></a> -<span class="sourceLineNo">160</span> Path baseNSDir = new Path(HConstants.BASE_NAMESPACE_DIR);<a name="line.160"></a> -<span class="sourceLineNo">161</span> baseNamespaceDir = new Path(rootDir, baseNSDir);<a name="line.161"></a> -<span class="sourceLineNo">162</span> hfileArchiveDir = new Path(rootDir, new Path(HConstants.HFILE_ARCHIVE_DIRECTORY, baseNSDir));<a name="line.162"></a> -<span class="sourceLineNo">163</span> }<a name="line.163"></a> -<span class="sourceLineNo">164</span><a name="line.164"></a> -<span class="sourceLineNo">165</span> private void decorateConf() {<a name="line.165"></a> -<span class="sourceLineNo">166</span> String replicationCodec = this.conf.get(HConstants.REPLICATION_CODEC_CONF_KEY);<a name="line.166"></a> -<span class="sourceLineNo">167</span> if (StringUtils.isNotEmpty(replicationCodec)) {<a name="line.167"></a> -<span class="sourceLineNo">168</span> this.conf.set(HConstants.RPC_CODEC_CONF_KEY, replicationCodec);<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><a name="line.171"></a> -<span class="sourceLineNo">172</span> private void connectToPeers() {<a name="line.172"></a> -<span class="sourceLineNo">173</span> getRegionServers();<a name="line.173"></a> -<span class="sourceLineNo">174</span><a name="line.174"></a> -<span class="sourceLineNo">175</span> int sleepMultiplier = 1;<a name="line.175"></a> +<span class="sourceLineNo">040</span>import java.util.stream.Collectors;<a name="line.40"></a> +<span class="sourceLineNo">041</span>import java.util.stream.Stream;<a name="line.41"></a> +<span class="sourceLineNo">042</span><a name="line.42"></a> +<span class="sourceLineNo">043</span>import org.apache.commons.lang3.StringUtils;<a name="line.43"></a> +<span class="sourceLineNo">044</span>import org.apache.hadoop.conf.Configuration;<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.Abortable;<a name="line.46"></a> +<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.HBaseConfiguration;<a name="line.47"></a> +<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.HConstants;<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.TableNotFoundException;<a name="line.50"></a> +<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.client.ClusterConnection;<a name="line.51"></a> +<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.client.Connection;<a name="line.52"></a> +<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.client.ConnectionFactory;<a name="line.53"></a> +<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.ipc.RpcServer;<a name="line.54"></a> +<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.protobuf.ReplicationProtbufUtil;<a name="line.55"></a> +<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.replication.HBaseReplicationEndpoint;<a name="line.56"></a> +<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.replication.regionserver.ReplicationSinkManager.SinkPeer;<a name="line.57"></a> +<span class="sourceLineNo">058</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.58"></a> +<span class="sourceLineNo">059</span>import org.apache.hadoop.hbase.util.FSUtils;<a name="line.59"></a> +<span class="sourceLineNo">060</span>import org.apache.hadoop.hbase.wal.WAL.Entry;<a name="line.60"></a> +<span class="sourceLineNo">061</span>import org.apache.hadoop.ipc.RemoteException;<a name="line.61"></a> +<span class="sourceLineNo">062</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.62"></a> +<span class="sourceLineNo">063</span>import org.slf4j.Logger;<a name="line.63"></a> +<span class="sourceLineNo">064</span>import org.slf4j.LoggerFactory;<a name="line.64"></a> +<span class="sourceLineNo">065</span><a name="line.65"></a> +<span class="sourceLineNo">066</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.66"></a> +<span class="sourceLineNo">067</span><a name="line.67"></a> +<span class="sourceLineNo">068</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.AdminProtos.AdminService.BlockingInterface;<a name="line.68"></a> +<span class="sourceLineNo">069</span><a name="line.69"></a> +<span class="sourceLineNo">070</span>/**<a name="line.70"></a> +<span class="sourceLineNo">071</span> * A {@link org.apache.hadoop.hbase.replication.ReplicationEndpoint}<a name="line.71"></a> +<span class="sourceLineNo">072</span> * implementation for replicating to another HBase cluster.<a name="line.72"></a> +<span class="sourceLineNo">073</span> * For the slave cluster it selects a random number of peers<a name="line.73"></a> +<span class="sourceLineNo">074</span> * using a replication ratio. For example, if replication ration = 0.1<a name="line.74"></a> +<span class="sourceLineNo">075</span> * and slave cluster has 100 region servers, 10 will be selected.<a name="line.75"></a> +<span class="sourceLineNo">076</span> * <p><a name="line.76"></a> +<span class="sourceLineNo">077</span> * A stream is considered down when we cannot contact a region server on the<a name="line.77"></a> +<span class="sourceLineNo">078</span> * peer cluster for more than 55 seconds by default.<a name="line.78"></a> +<span class="sourceLineNo">079</span> * </p><a name="line.79"></a> +<span class="sourceLineNo">080</span> */<a name="line.80"></a> +<span class="sourceLineNo">081</span>@InterfaceAudience.Private<a name="line.81"></a> +<span class="sourceLineNo">082</span>public class HBaseInterClusterReplicationEndpoint extends HBaseReplicationEndpoint {<a name="line.82"></a> +<span class="sourceLineNo">083</span> private static final Logger LOG =<a name="line.83"></a> +<span class="sourceLineNo">084</span> LoggerFactory.getLogger(HBaseInterClusterReplicationEndpoint.class);<a name="line.84"></a> +<span class="sourceLineNo">085</span><a name="line.85"></a> +<span class="sourceLineNo">086</span> private static final long DEFAULT_MAX_TERMINATION_WAIT_MULTIPLIER = 2;<a name="line.86"></a> +<span class="sourceLineNo">087</span><a name="line.87"></a> +<span class="sourceLineNo">088</span> private ClusterConnection conn;<a name="line.88"></a> +<span class="sourceLineNo">089</span> private Configuration localConf;<a name="line.89"></a> +<span class="sourceLineNo">090</span> private Configuration conf;<a name="line.90"></a> +<span class="sourceLineNo">091</span> // How long should we sleep for each retry<a name="line.91"></a> +<span class="sourceLineNo">092</span> private long sleepForRetries;<a name="line.92"></a> +<span class="sourceLineNo">093</span> // Maximum number of retries before taking bold actions<a name="line.93"></a> +<span class="sourceLineNo">094</span> private int maxRetriesMultiplier;<a name="line.94"></a> +<span class="sourceLineNo">095</span> // Socket timeouts require even bolder actions since we don't want to DDOS<a name="line.95"></a> +<span class="sourceLineNo">096</span> private int socketTimeoutMultiplier;<a name="line.96"></a> +<span class="sourceLineNo">097</span> // Amount of time for shutdown to wait for all tasks to complete<a name="line.97"></a> +<span class="sourceLineNo">098</span> private long maxTerminationWait;<a name="line.98"></a> +<span class="sourceLineNo">099</span> // Size limit for replication RPCs, in bytes<a name="line.99"></a> +<span class="sourceLineNo">100</span> private int replicationRpcLimit;<a name="line.100"></a> +<span class="sourceLineNo">101</span> //Metrics for this source<a name="line.101"></a> +<span class="sourceLineNo">102</span> private MetricsSource metrics;<a name="line.102"></a> +<span class="sourceLineNo">103</span> // Handles connecting to peer region servers<a name="line.103"></a> +<span class="sourceLineNo">104</span> private ReplicationSinkManager replicationSinkMgr;<a name="line.104"></a> +<span class="sourceLineNo">105</span> private boolean peersSelected = false;<a name="line.105"></a> +<span class="sourceLineNo">106</span> private String replicationClusterId = "";<a name="line.106"></a> +<span class="sourceLineNo">107</span> private ThreadPoolExecutor exec;<a name="line.107"></a> +<span class="sourceLineNo">108</span> private int maxThreads;<a name="line.108"></a> +<span class="sourceLineNo">109</span> private Path baseNamespaceDir;<a name="line.109"></a> +<span class="sourceLineNo">110</span> private Path hfileArchiveDir;<a name="line.110"></a> +<span class="sourceLineNo">111</span> private boolean replicationBulkLoadDataEnabled;<a name="line.111"></a> +<span class="sourceLineNo">112</span> private Abortable abortable;<a name="line.112"></a> +<span class="sourceLineNo">113</span> private boolean dropOnDeletedTables;<a name="line.113"></a> +<span class="sourceLineNo">114</span> private boolean isSerial = false;<a name="line.114"></a> +<span class="sourceLineNo">115</span><a name="line.115"></a> +<span class="sourceLineNo">116</span> @Override<a name="line.116"></a> +<span class="sourceLineNo">117</span> public void init(Context context) throws IOException {<a name="line.117"></a> +<span class="sourceLineNo">118</span> super.init(context);<a name="line.118"></a> +<span class="sourceLineNo">119</span> this.conf = HBaseConfiguration.create(ctx.getConfiguration());<a name="line.119"></a> +<span class="sourceLineNo">120</span> this.localConf = HBaseConfiguration.create(ctx.getLocalConfiguration());<a name="line.120"></a> +<span class="sourceLineNo">121</span> decorateConf();<a name="line.121"></a> +<span class="sourceLineNo">122</span> this.maxRetriesMultiplier = this.conf.getInt("replication.source.maxretriesmultiplier", 300);<a name="line.122"></a> +<span class="sourceLineNo">123</span> this.socketTimeoutMultiplier = this.conf.getInt("replication.source.socketTimeoutMultiplier",<a name="line.123"></a> +<span class="sourceLineNo">124</span> maxRetriesMultiplier);<a name="line.124"></a> +<span class="sourceLineNo">125</span> // A Replicator job is bound by the RPC timeout. We will wait this long for all Replicator<a name="line.125"></a> +<span class="sourceLineNo">126</span> // tasks to terminate when doStop() is called.<a name="line.126"></a> +<span class="sourceLineNo">127</span> long maxTerminationWaitMultiplier = this.conf.getLong(<a name="line.127"></a> +<span class="sourceLineNo">128</span> "replication.source.maxterminationmultiplier",<a name="line.128"></a> +<span class="sourceLineNo">129</span> DEFAULT_MAX_TERMINATION_WAIT_MULTIPLIER);<a name="line.129"></a> +<span class="sourceLineNo">130</span> this.maxTerminationWait = maxTerminationWaitMultiplier *<a name="line.130"></a> +<span class="sourceLineNo">131</span> this.conf.getLong(HConstants.HBASE_RPC_TIMEOUT_KEY, HConstants.DEFAULT_HBASE_RPC_TIMEOUT);<a name="line.131"></a> +<span class="sourceLineNo">132</span> // TODO: This connection is replication specific or we should make it particular to<a name="line.132"></a> +<span class="sourceLineNo">133</span> // replication and make replication specific settings such as compression or codec to use<a name="line.133"></a> +<span class="sourceLineNo">134</span> // passing Cells.<a name="line.134"></a> +<span class="sourceLineNo">135</span> this.conn = (ClusterConnection) ConnectionFactory.createConnection(this.conf);<a name="line.135"></a> +<span class="sourceLineNo">136</span> this.sleepForRetries =<a name="line.136"></a> +<span class="sourceLineNo">137</span> this.conf.getLong("replication.source.sleepforretries", 1000);<a name="line.137"></a> +<span class="sourceLineNo">138</span> this.metrics = context.getMetrics();<a name="line.138"></a> +<span class="sourceLineNo">139</span> // ReplicationQueueInfo parses the peerId out of the znode for us<a name="line.139"></a> +<span class="sourceLineNo">140</span> this.replicationSinkMgr = new ReplicationSinkManager(conn, ctx.getPeerId(), this, this.conf);<a name="line.140"></a> +<span class="sourceLineNo">141</span> // per sink thread pool<a name="line.141"></a> +<span class="sourceLineNo">142</span> this.maxThreads = this.conf.getInt(HConstants.REPLICATION_SOURCE_MAXTHREADS_KEY,<a name="line.142"></a> +<span class="sourceLineNo">143</span> HConstants.REPLICATION_SOURCE_MAXTHREADS_DEFAULT);<a name="line.143"></a> +<span class="sourceLineNo">144</span> this.exec = new ThreadPoolExecutor(maxThreads, maxThreads, 60, TimeUnit.SECONDS,<a name="line.144"></a> +<span class="sourceLineNo">145</span> new LinkedBlockingQueue<>());<a name="line.145"></a> +<span class="sourceLineNo">146</span> this.exec.allowCoreThreadTimeOut(true);<a name="line.146"></a> +<span class="sourceLineNo">147</span> this.abortable = ctx.getAbortable();<a name="line.147"></a> +<span class="sourceLineNo">148</span> // Set the size limit for replication RPCs to 95% of the max request size.<a name="line.148"></a> +<span class="sourceLineNo">149</span> // We could do with less slop if we have an accurate estimate of encoded size. Being<a name="line.149"></a> +<span class="sourceLineNo">150</span> // conservative for now.<a name="line.150"></a> +<span class="sourceLineNo">151</span> this.replicationRpcLimit = (int)(0.95 * conf.getLong(RpcServer.MAX_REQUEST_SIZE,<a name="line.151"></a> +<span class="sourceLineNo">152</span> RpcServer.DEFAULT_MAX_REQUEST_SIZE));<a name="line.152"></a> +<span class="sourceLineNo">153</span> this.dropOnDeletedTables =<a name="line.153"></a> +<span class="sourceLineNo">154</span> this.conf.getBoolean(HConstants.REPLICATION_DROP_ON_DELETED_TABLE_KEY, false);<a name="line.154"></a> +<span class="sourceLineNo">155</span><a name="line.155"></a> +<span class="sourceLineNo">156</span> this.replicationBulkLoadDataEnabled =<a name="line.156"></a> +<span class="sourceLineNo">157</span> conf.getBoolean(HConstants.REPLICATION_BULKLOAD_ENABLE_KEY,<a name="line.157"></a> +<span class="sourceLineNo">158</span> HConstants.REPLICATION_BULKLOAD_ENABLE_DEFAULT);<a name="line.158"></a> +<span class="sourceLineNo">159</span> if (this.replicationBulkLoadDataEnabled) {<a name="line.159"></a> +<span class="sourceLineNo">160</span> replicationClusterId = this.conf.get(HConstants.REPLICATION_CLUSTER_ID);<a name="line.160"></a> +<span class="sourceLineNo">161</span> }<a name="line.161"></a> +<span class="sourceLineNo">162</span> // Construct base namespace directory and hfile archive directory path<a name="line.162"></a> +<span class="sourceLineNo">163</span> Path rootDir = FSUtils.getRootDir(conf);<a name="line.163"></a> +<span class="sourceLineNo">164</span> Path baseNSDir = new Path(HConstants.BASE_NAMESPACE_DIR);<a name="line.164"></a> +<span class="sourceLineNo">165</span> baseNamespaceDir = new Path(rootDir, baseNSDir);<a name="line.165"></a> +<span class="sourceLineNo">166</span> hfileArchiveDir = new Path(rootDir, new Path(HConstants.HFILE_ARCHIVE_DIRECTORY, baseNSDir));<a name="line.166"></a> +<span class="sourceLineNo">167</span> isSerial = context.getPeerConfig().isSerial();<a name="line.167"></a> +<span class="sourceLineNo">168</span> }<a name="line.168"></a> +<span class="sourceLineNo">169</span><a name="line.169"></a> +<span class="sourceLineNo">170</span> private void decorateConf() {<a name="line.170"></a> +<span class="sourceLineNo">171</span> String replicationCodec = this.conf.get(HConstants.REPLICATION_CODEC_CONF_KEY);<a name="line.171"></a> +<span class="sourceLineNo">172</span> if (StringUtils.isNotEmpty(replicationCodec)) {<a name="line.172"></a> +<span class="sourceLineNo">173</span> this.conf.set(HConstants.RPC_CODEC_CONF_KEY, replicationCodec);<a name="line.173"></a> +<span class="sourceLineNo">174</span> }<a name="line.174"></a> +<span class="sourceLineNo">175</span> }<a name="line.175"></a> <span class="sourceLineNo">176</span><a name="line.176"></a> -<span class="sourceLineNo">177</span> // Connect to peer cluster first, unless we have to stop<a name="line.177"></a> -<span class="sourceLineNo">178</span> while (this.isRunning() && replicationSinkMgr.getNumSinks() == 0) {<a name="line.178"></a> -<span class="sourceLineNo">179</span> replicationSinkMgr.chooseSinks();<a name="line.179"></a> -<span class="sourceLineNo">180</span> if (this.isRunning() && replicationSinkMgr.getNumSinks() == 0) {<a name="line.180"></a> -<span class="sourceLineNo">181</span> if (sleepForRetries("Waiting for peers", sleepMultiplier)) {<a name="line.181"></a> -<span class="sourceLineNo">182</span> sleepMultiplier++;<a name="line.182"></a> -<span class="sourceLineNo">183</span> }<a name="line.183"></a> -<span class="sourceLineNo">184</span> }<a name="line.184"></a> -<span class="sourceLineNo">185</span> }<a name="line.185"></a> -<span class="sourceLineNo">186</span> }<a name="line.186"></a> -<span class="sourceLineNo">187</span><a name="line.187"></a> -<span class="sourceLineNo">188</span> /**<a name="line.188"></a> -<span class="sourceLineNo">189</span> * Do the sleeping logic<a name="line.189"></a> -<span class="sourceLineNo">190</span> * @param msg Why we sleep<a name="line.190"></a> -<span class="sourceLineNo">191</span> * @param sleepMultiplier by how many times the default sleeping time is augmented<a name="line.191"></a> -<span class="sourceLineNo">192</span> * @return True if <code>sleepMultiplier</code> is &lt; <code>maxRetriesMultiplier</code><a name="line.192"></a> -<span class="sourceLineNo">193</span> */<a name="line.193"></a> -<span class="sourceLineNo">194</span> protected boolean sleepForRetries(String msg, int sleepMultiplier) {<a name="line.194"></a> -<span class="sourceLineNo">195</span> try {<a name="line.195"></a> -<span class="sourceLineNo">196</span> if (LOG.isTraceEnabled()) {<a name="line.196"></a> -<span class="sourceLineNo">197</span> LOG.trace(msg + ", sleeping " + sleepForRetries + " times " + sleepMultiplier);<a name="line.197"></a> -<span class="sourceLineNo">198</span> }<a name="line.198"></a> -<span class="sourceLineNo">199</span> Thread.sleep(this.sleepForRetries * sleepMultiplier);<a name="line.199"></a> -<span class="sourceLineNo">200</span> } catch (InterruptedException e) {<a name="line.200"></a> -<span class="sourceLineNo">201</span> LOG.debug("Interrupted while sleeping between retries");<a name="line.201"></a> -<span class="sourceLineNo">202</span> }<a name="line.202"></a> -<span class="sourceLineNo">203</span> return sleepMultiplier < maxRetriesMultiplier;<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> private List<List<Entry>> createBatches(final List<Entry> entries) {<a name="line.206"></a> -<span class="sourceLineNo">207</span> int numSinks = Math.max(replicationSinkMgr.getNumSinks(), 1);<a name="line.207"></a> -<span class="sourceLineNo">208</span> int n = Math.min(Math.min(this.maxThreads, entries.size()/100+1), numSinks);<a name="line.208"></a> -<span class="sourceLineNo">209</span> // Maintains the current batch for a given partition index<a name="line.209"></a> -<span class="sourceLineNo">210</span> Map<Integer, List<Entry>> entryMap = new HashMap<>(n);<a name="line.210"></a> -<span class="sourceLineNo">211</span> List<List<Entry>> entryLists = new ArrayList<>();<a name="line.211"></a> -<span class="sourceLineNo">212</span> int[] sizes = new int[n];<a name="line.212"></a> -<span class="sourceLineNo">213</span><a name="line.213"></a> -<span class="sourceLineNo">214</span> for (int i = 0; i < n; i++) {<a name="line.214"></a> -<span class="sourceLineNo">215</span> entryMap.put(i, new ArrayList<Entry>(entries.size()/n+1));<a name="line.215"></a> -<span class="sourceLineNo">216</span> }<a name="line.216"></a> -<span class="sourceLineNo">217</span><a name="line.217"></a> -<span class="sourceLineNo">218</span> for (Entry e: entries) {<a name="line.218"></a> -<span class="sourceLineNo">219</span> int index = Math.abs(Bytes.hashCode(e.getKey().getEncodedRegionName())%n);<a name="line.219"></a> -<span class="sourceLineNo">220</span> int entrySize = (int)e.getKey().estimatedSerializedSizeOf() +<a name="line.220"></a> -<span class="sourceLineNo">221</span> (int)e.getEdit().estimatedSerializedSizeOf();<a name="line.221"></a> -<span class="sourceLineNo">222</span> // If this batch is oversized, add it to final list and initialize a new empty batch<a name="line.222"></a> -<span class="sourceLineNo">223</span> if (sizes[index] > 0 /* must include at least one entry */ &&<a name="line.223"></a> -<span class="sourceLineNo">224</span> sizes[index] + entrySize > replicationRpcLimit) {<a name="line.224"></a> -<span class="sourceLineNo">225</span> entryLists.add(entryMap.get(index));<a name="line.225"></a> -<span class="sourceLineNo">226</span> entryMap.put(index, new ArrayList<Entry>());<a name="line.226"></a> -<span class="sourceLineNo">227</span> sizes[index] = 0;<a name="line.227"></a> -<span class="sourceLineNo">228</span> }<a name="line.228"></a> -<span class="sourceLineNo">229</span> entryMap.get(index).add(e);<a name="line.229"></a> -<span class="sourceLineNo">230</span> sizes[index] += entrySize;<a name="line.230"></a> -<span class="sourceLineNo">231</span> }<a name="line.231"></a> -<span class="sourceLineNo">232</span><a name="line.232"></a> -<span class="sourceLineNo">233</span> entryLists.addAll(entryMap.values());<a name="line.233"></a> -<span class="sourceLineNo">234</span> return entryLists;<a name="line.234"></a> -<span class="sourceLineNo">235</span> }<a name="line.235"></a> -<span class="sourceLineNo">236</span><a name="line.236"></a> -<span class="sourceLineNo">237</span> private TableName parseTable(String msg) {<a name="line.237"></a> -<span class="sourceLineNo">238</span> // ... TableNotFoundException: '<table>'/n...<a name="line.238"></a> -<span class="sourceLineNo">239</span> Pattern p = Pattern.compile("TableNotFoundException: \\'([\\S]*)\\'");<a name="line.239"></a> -<span class="sourceLineNo">240</span> Matcher m = p.matcher(msg);<a name="line.240"></a> -<span class="sourceLineNo">241</span> if (m.find()) {<a name="line.241"></a> -<span class="sourceLineNo">242</span> String table = m.group(1);<a name="line.242"></a> -<span class="sourceLineNo">243</span> try {<a name="line.243"></a> -<span class="sourceLineNo">244</span> // double check that table is a valid table name<a name="line.244"></a> -<span class="sourceLineNo">245</span> TableName.valueOf(TableName.isLegalFullyQualifiedTableName(Bytes.toBytes(table)));<a name="line.245"></a> -<span class="sourceLineNo">246</span> return TableName.valueOf(table);<a name="line.246"></a> -<span class="sourceLineNo">247</span> } catch (IllegalArgumentException ignore) {<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> return null;<a name="line.250"></a> -<span class="sourceLineNo">251</span> }<a name="line.251"></a> -<span class="sourceLineNo">252</span><a name="line.252"></a> -<span class="sourceLineNo">253</span> // Filter a set of batches by TableName<a name="line.253"></a> -<span class="sourceLineNo">254</span> private List<List<Entry>> filterBatches(final List<List<Entry>> oldEntryList, TableName table) {<a name="line.254"></a> -<span class="sourceLineNo">255</span> List<List<Entry>> entryLists = new ArrayList<>();<a name="line.255"></a> -<span class="sourceLineNo">256</span> for (List<Entry> entries : oldEntryList) {<a name="line.256"></a> -<span class="sourceLineNo">257</span> ArrayList<Entry> thisList = new ArrayList<Entry>(entries.size());<a name="line.257"></a> -<span class="sourceLineNo">258</span> entryLists.add(thisList);<a name="line.258"></a> -<span class="sourceLineNo">259</span> for (Entry e : entries) {<a name="line.259"></a> -<span class="sourceLineNo">260</span> if (!e.getKey().getTableName().equals(table)) {<a name="line.260"></a> -<span class="sourceLineNo">261</span> thisList.add(e);<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> return entryLists;<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> private void reconnectToPeerCluster() {<a name="line.268"></a> -<span class="sourceLineNo">269</span> ClusterConnection connection = null;<a name="line.269"></a> -<span class="sourceLineNo">270</span> try {<a name="line.270"></a> -<span class="sourceLineNo">271</span> connection = (ClusterConnection) ConnectionFactory.createConnection(this.conf);<a name="line.271"></a> -<span class="sourceLineNo">272</span> } catch (IOException ioe) {<a name="line.272"></a> -<span class="sourceLineNo">273</span> LOG.warn("Failed to create connection for peer cluster", ioe);<a name="line.273"></a> +<span class="sourceLineNo">177</span> private void connectToPeers() {<a name="line.177"></a> +<span class="sourceLineNo">178</span> getRegionServers();<a name="line.178"></a> +<span class="sourceLineNo">179</span><a name="line.179"></a> +<span class="sourceLineNo">180</span> int sleepMultiplier = 1;<a name="line.180"></a> +<span class="sourceLineNo">181</span><a name="line.181"></a> +<span class="sourceLineNo">182</span> // Connect to peer cluster first, unless we have to stop<a name="line.182"></a> +<span class="sourceLineNo">183</span> while (this.isRunning() && replicationSinkMgr.getNumSinks() == 0) {<a name="line.183"></a> +<span class="sourceLineNo">184</span> replicationSinkMgr.chooseSinks();<a name="line.184"></a> +<span class="sourceLineNo">185</span> if (this.isRunning() && replicationSinkMgr.getNumSinks() == 0) {<a name="line.185"></a> +<span class="sourceLineNo">186</span> if (sleepForRetries("Waiting for peers", sleepMultiplier)) {<a name="line.186"></a> +<span class="sourceLineNo">187</span> sleepMultiplier++;<a name="line.187"></a> +<span class="sourceLineNo">188</span> }<a name="line.188"></a> +<span class="sourceLineNo">189</span> }<a name="line.189"></a> +<span class="sourceLineNo">190</span> }<a name="line.190"></a> +<span class="sourceLineNo">191</span> }<a name="line.191"></a> +<span class="sourceLineNo">192</span><a name="line.192"></a> +<span class="sourceLineNo">193</span> /**<a name="line.193"></a> +<span class="sourceLineNo">194</span> * Do the sleeping logic<a name="line.194"></a> +<span class="sourceLineNo">195</span> * @param msg Why we sleep<a name="line.195"></a> +<span class="sourceLineNo">196</span> * @param sleepMultiplier by how many times the default sleeping time is augmented<a name="line.196"></a> +<span class="sourceLineNo">197</span> * @return True if <code>sleepMultiplier</code> is &lt; <code>maxRetriesMultiplier</code><a name="line.197"></a> +<span class="sourceLineNo">198</span> */<a name="line.198"></a> +<span class="sourceLineNo">199</span> protected boolean sleepForRetries(String msg, int sleepMultiplier) {<a name="line.199"></a> +<span class="sourceLineNo">200</span> try {<a name="line.200"></a> +<span class="sourceLineNo">201</span> if (LOG.isTraceEnabled()) {<a name="line.201"></a> +<span class="sourceLineNo">202</span> LOG.trace(msg + ", sleeping " + sleepForRetries + " times " + sleepMultiplier);<a name="line.202"></a> +<span class="sourceLineNo">203</span> }<a name="line.203"></a> +<span class="sourceLineNo">204</span> Thread.sleep(this.sleepForRetries * sleepMultiplier);<a name="line.204"></a> +<span class="sourceLineNo">205</span> } catch (InterruptedException e) {<a name="line.205"></a> +<span class="sourceLineNo">206</span> LOG.debug("Interrupted while sleeping between retries");<a name="line.206"></a> +<span class="sourceLineNo">207</span> }<a name="line.207"></a> +<span class="sourceLineNo">208</span> return sleepMultiplier < maxRetriesMultiplier;<a name="line.208"></a> +<span class="sourceLineNo">209</span> }<a name="line.209"></a> +<span class="sourceLineNo">210</span><a name="line.210"></a> +<span class="sourceLineNo">211</span> private int getEstimatedEntrySize(Entry e) {<a name="line.211"></a> +<span class="sourceLineNo">212</span> long size = e.getKey().estimatedSerializedSizeOf() + e.getEdit().estimatedSerializedSizeOf();<a name="line.212"></a> +<span class="sourceLineNo">213</span> return (int) size;<a name="line.213"></a> +<span class="sourceLineNo">214</span> }<a name="line.214"></a> +<span class="sourceLineNo">215</span><a name="line.215"></a> +<span class="sourceLineNo">216</span> private List<List<Entry>> createParallelBatches(final List<Entry> entries) {<a name="line.216"></a> +<span class="sourceLineNo">217</span> int numSinks = Math.max(replicationSinkMgr.getNumSinks(), 1);<a name="line.217"></a> +<span class="sourceLineNo">218</span> int n = Math.min(Math.min(this.maxThreads, entries.size() / 100 + 1), numSinks);<a name="line.218"></a> +<span class="sourceLineNo">219</span> List<List<Entry>> entryLists =<a name="line.219"></a> +<span class="sourceLineNo">220</span> Stream.generate(ArrayList<Entry>::new).limit(n).collect(Collectors.toList());<a name="line.220"></a> +<span class="sourceLineNo">221</span> int[] sizes = new int[n];<a name="line.221"></a> +<span class="sourceLineNo">222</span> for (Entry e : entries) {<a name="line.222"></a> +<span class="sourceLineNo">223</span> int index = Math.abs(Bytes.hashCode(e.getKey().getEncodedRegionName()) % n);<a name="line.223"></a> +<span class="sourceLineNo">224</span> int entrySize = getEstimatedEntrySize(e);<a name="line.224"></a> +<span class="sourceLineNo">225</span> // If this batch has at least one entry and is over sized, move it to the tail of list and<a name="line.225"></a> +<span class="sourceLineNo">226</span> // initialize the entryLists[index] to be a empty list.<a name="line.226"></a> +<span class="sourceLineNo">227</span> if (sizes[index] > 0 && sizes[index] + entrySize > replicationRpcLimit) {<a name="line.227"></a> +<span class="sourceLineNo">228</span> entryLists.add(entryLists.get(index));<a name="line.228"></a> +<span class="sourceLineNo">229</span> entryLists.set(index, new ArrayList<>());<a name="line.229"></a> +<span class="sourceLineNo">230</span> sizes[index] = 0;<a name="line.230"></a> +<span class="sourceLineNo">231</span> }<a name="line.231"></a> +<span class="sourceLineNo">232</span> entryLists.get(index).add(e);<a name="line.232"></a> +<span class="sourceLineNo">233</span> sizes[index] += entrySize;<a name="line.233"></a> +<span class="sourceLineNo">234</span> }<a name="line.234"></a> +<span class="sourceLineNo">235</span> return entryLists;<a name="line.235"></a> +<span class="sourceLineNo">236</span> }<a name="line.236"></a> +<span class="sourceLineNo">237</span><a name="line.237"></a> +<span class="sourceLineNo">238</span> private List<List<Entry>> createSerialBatches(final List<Entry> entries) {<a name="line.238"></a> +<span class="sourceLineNo">239</span> Map<byte[], List<Entry>> regionEntries = new TreeMap<>(Bytes.BYTES_COMPARATOR);<a name="line.239"></a> +<span class="sourceLineNo">240</span> for (Entry e : entries) {<a name="line.240"></a> +<span class="sourceLineNo">241</span> regionEntries.computeIfAbsent(e.getKey().getEncodedRegionName(), key -> new ArrayList<>())<a name="line.241"></a> +<span class="sourceLineNo">242</span> .add(e);<a name="line.242"></a> +<span class="sourceLineNo">243</span> }<a name="line.243"></a> +<span class="sourceLineNo">244</span> return new ArrayList<>(regionEntries.values());<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> * Divide the entries into multiple batches, so that we can replicate each batch in a thread pool<a name="line.248"></a> +<span class="sourceLineNo">249</span> * concurrently. Note that, for serial replication, we need to make sure that entries from the<a name="line.249"></a> +<span class="sourceLineNo">250</span> * same region to be replicated serially, so entries from the same region consist of a batch, and<a name="line.250"></a> +<span class="sourceLineNo">251</span> * we will divide a batch into several batches by replicationRpcLimit in method<a name="line.251"></a> +<span class="sourceLineNo">252</span> * serialReplicateRegionEntries()<a name="line.252"></a> +<span class="sourceLineNo">253</span> */<a name="line.253"></a> +<span class="sourceLineNo">254</span> private List<List<Entry>> createBatches(final List<Entry> entries) {<a name="line.254"></a> +<span class="sourceLineNo">255</span> if (isSerial) {<a name="line.255"></a> +<span class="sourceLineNo">256</span> return createSerialBatches(entries);<a name="line.256"></a> +<span class="sourceLineNo">257</span> } else {<a name="line.257"></a> +<span class="sourceLineNo">258</span> return createParallelBatches(entries);<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><a name="line.261"></a> +<span class="sourceLineNo">262</span> private TableName parseTable(String msg) {<a name="line.262"></a> +<span class="sourceLineNo">263</span> // ... TableNotFoundException: '<table>'/n...<a name="line.263"></a> +<span class="sourceLineNo">264</span> Pattern p = Pattern.compile("TableNotFoundException: '([\\S]*)'");<a name="line.264"></a> +<span class="sourceLineNo">265</span> Matcher m = p.matcher(msg);<a name="line.265"></a> +<span class="sourceLineNo">266</span> if (m.find()) {<a name="line.266"></a> +<span class="sourceLineNo">267</span> String table = m.group(1);<a name="line.267"></a> +<span class="sourceLineNo">268</span> try {<a name="line.268"></a> +<span class="sourceLineNo">269</span> // double check that table is a valid table name<a name="line.269"></a> +<span class="sourceLineNo">270</span> TableName.valueOf(TableName.isLegalFullyQualifiedTableName(Bytes.toBytes(table)));<a name="line.270"></a> +<span class="sourceLineNo">271</span> return TableName.valueOf(table);<a name="line.271"></a> +<span class="sourceLineNo">272</span> } catch (IllegalArgumentException ignore) {<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> if (connection != null) {<a name="line.275"></a> -<span class="sourceLineNo">276</span> this.conn = connection;<a name="line.276"></a> -<span class="sourceLineNo">277</span> }<a name="line.277"></a> -<span class="sourceLineNo">278</span> }<a name="line.278"></a> -<span class="sourceLineNo">279</span><a name="line.279"></a> -<span class="sourceLineNo">280</span> /**<a name="line.280"></a> -<span class="sourceLineNo">281</span> * Do the shipping logic<a name="line.281"></a> -<span class="sourceLineNo">282</span> */<a name="line.282"></a> -<span class="sourceLineNo">283</span> @Override<a name="line.283"></a> -<span class="sourceLineNo">284</span> public boolean replicate(ReplicateContext replicateContext) {<a name="line.284"></a> -<span class="sourceLineNo">285</span> CompletionService<Integer> pool = new ExecutorCompletionService<>(this.exec);<a name="line.285"></a> -<span class="sourceLineNo">286</span> List<List<Entry>> batches;<a name="line.286"></a> -<span class="sourceLineNo">287</span> String walGroupId = replicateContext.getWalGroupId();<a name="line.287"></a> -<span class="sourceLineNo">288</span> int sleepMultiplier = 1;<a name="line.288"></a> -<span class="sourceLineNo">289</span><a name="line.289"></a> -<span class="sourceLineNo">290</span> if (!peersSelected && this.isRunning()) {<a name="line.290"></a> -<span class="sourceLineNo">291</span> connectToPeers();<a name="line.291"></a> -<span class="sourceLineNo">292</span> peersSelected = true;<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> int numSinks = replicationSinkMgr.getNumSinks();<a name="line.295"></a> -<span class="sourceLineNo">296</span> if (numSinks == 0) {<a name="line.296"></a> -<span class="sourceLineNo">297</span> LOG.warn("No replication sinks found, returning without replicating. The source should retry"<a name="line.297"></a> -<span class="sourceLineNo">298</span> + " with the same set of edits.");<a name="line.298"></a> -<span class="sourceLineNo">299</span> return false;<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> batches = createBatches(replicateContext.getEntries());<a name="line.302"></a> -<span class="sourceLineNo">303</span><a name="line.303"></a> -<span class="sourceLineNo">304</span> while (this.isRunning() && !exec.isShutdown()) {<a name="line.304"></a> -<span class="sourceLineNo">305</span> if (!isPeerEnabled()) {<a name="line.305"></a> -<span class="sourceLineNo">306</span> if (sleepForRetries("Replication is disabled", sleepMultiplier)) {<a name="line.306"></a> -<span class="sourceLineNo">307</span> sleepMultiplier++;<a name="line.307"></a> -<span class="sourceLineNo">308</span> }<a name="line.308"></a> -<span class="sourceLineNo">309</span> continue;<a name="line.309"></a> -<span class="sourceLineNo">310</span> }<a name="line.310"></a> -<span class="sourceLineNo">311</span> if (this.conn == null || this.conn.isClosed()) {<a name="line.311"></a> -<span class="sourceLineNo">312</span> reconnectToPeerCluster();<a name="line.312"></a> -<span class="sourceLineNo">313</span> }<a name="line.313"></a> -<span class="sourceLineNo">314</span> try {<a name="line.314"></a> -<span class="sourceLineNo">315</span> int futures = 0;<a name="line.315"></a> -<span class="sourceLineNo">316</span> for (int i=0; i<batches.size(); i++) {<a name="line.316"></a> -<span class="sourceLineNo">317</span> List<Entry> entries = batches.get(i);<a name="line.317"></a> -<span class="sourceLineNo">318</span> if (!entries.isEmpty()) {<a name="line.318"></a> -<span class="sourceLineNo">319</span> if (LOG.isTraceEnabled()) {<a name="line.319"></a> -<span class="sourceLineNo">320</span> LOG.trace("Submitting " + entries.size() +<a name="line.320"></a> -<span class="sourceLineNo">321</span> " entries of total size " + replicateContext.getSize());<a name="line.321"></a> -<span class="sourceLineNo">322</span> }<a name="line.322"></a> -<span class="sourceLineNo">323</span> // RuntimeExceptions encountered here bubble up and are handled in ReplicationSource<a name="line.323"></a> -<span class="sourceLineNo">324</span> pool.submit(createReplicator(entries, i));<a name="line.324"></a> -<span class="sourceLineNo">325</span> futures++;<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> IOException iox = null;<a name="line.328"></a> -<span class="sourceLineNo">329</span><a name="line.329"></a> -<span class="sourceLineNo">330</span> long lastWriteTime = 0;<a name="line.330"></a> -<span class="sourceLineNo">331</span> for (int i=0; i<futures; i++) {<a name="line.331"></a> -<span class="sourceLineNo">332</span> try {<a name="line.332"></a> -<span class="sourceLineNo">333</span> // wait for all futures, remove successful parts<a name="line.333"></a> -<span class="sourceLineNo">334</span> // (only the remaining parts will be retried)<a name="line.334"></a> -<span class="sourceLineNo">335</span> Future<Integer> f = pool.take();<a name="line.335"></a> -<span class="sourceLineNo">336</span> int index = f.get().intValue();<a name="line.336"></a> -<span class="sourceLineNo">337</span> List<Entry> batch = batches.get(index);<a name="line.337"></a> -<span class="sourceLineNo">338</span> batches.set(index, Collections.<Entry>emptyList()); // remove successful batch<a name="line.338"></a> -<span class="sourceLineNo">339</span> // Find the most recent write time in the batch<a name="line.339"></a> -<span class="sourceLineNo">340</span> long writeTime = batch.get(batch.size() - 1).getKey().getWriteTime();<a name="line.340"></a> -<span class="sourceLineNo">341</span> if (writeTime > lastWriteTime) {<a name="line.341"></a> -<span class="sourceLineNo">342</span> lastWriteTime = writeTime;<a name="line.342"></a> -<span class="sourceLineNo">343</span> }<a name="line.343"></a> -<span class="sourceLineNo">344</span> } catch (InterruptedException ie) {<a name="line.344"></a> -<span class="sourceLineNo">345</span> iox = new IOException(ie);<a name="line.345"></a> -<span class="sourceLineNo">346</span> } catch (ExecutionException ee) {<a name="line.346"></a> -<span class="sourceLineNo">347</span> // cause must be an IOException<a name="line.347"></a> -<span class="sourceLineNo">348</span> iox = (IOException)ee.getCause();<a name="line.348"></a> -<span class="sourceLineNo">349</span> }<a name="line.349"></a> -<span class="sourceLineNo">350</span> }<a name="line.350"></a> -<span class="sourceLineNo">351</span> if (iox != null) {<a name="line.351"></a> -<span class="sourceLineNo">352</span> // if we had any exceptions, try again<a name="line.352"></a> -<span class="sourceLineNo">353</span> throw iox;<a name="line.353"></a> -<span class="sourceLineNo">354</span> }<a name="line.354"></a> -<span class="sourceLineNo">355</span> // update metrics<a name="line.355"></a> -<span class="sourceLineNo">356</span> if (lastWriteTime > 0) {<a name="line.356"></a> -<span class="sourceLineNo">357</span> this.metrics.setAgeOfLastShippedOp(lastWriteTime, walGroupId);<a name="line.357"></a> -<span class="sourceLineNo">358</span> }<a name="line.358"></a> -<span class="sourceLineNo">359</span> return true;<a name="line.359"></a> -<span class="sourceLineNo">360</span><a name="line.360"></a> -<span class="sourceLineNo">361</span> } catch (IOException ioe) {<a name="line.361"></a> -<span class="sourceLineNo">362</span> // Didn't ship anything, but must still age the last time we did<a name="line.362"></a> -<span class="sourceLineNo">363</span> this.metrics.refreshAgeOfLastShippedOp(walGroupId);<a name="line.363"></a> -<span class="sourceLineNo">364</span> if (ioe instanceof RemoteException) {<a name="line.364"></a> -<span class="sourceLineNo">365</span> ioe = ((RemoteException) ioe).unwrapRemoteException();<a name="line.365"></a> -<span class="sourceLineNo">366</span> LOG.warn("Can't replicate because of an error on the remote cluster: ", ioe);<a name="line.366"></a> -<span class="sourceLineNo">367</span> if (ioe instanceof TableNotFoundException) {<a name="line.367"></a> -<span class="sourceLineNo">368</span> if (dropOnDeletedTables) {<a name="line.368"></a> -<span class="sourceLineNo">369</span> // this is a bit fragile, but cannot change how TNFE is serialized<a name="line.369"></a> -<span class="sourceLineNo">370</span> // at least check whether the table name is legal<a name="line.370"></a> -<span class="sourceLineNo">371</span> TableName table = parseTable(ioe.getMessage());<a name="line.371"></a> -<span class="sourceLineNo">372</span> if (table != null) {<a name="line.372"></a> -<span class="sourceLineNo">373</span> try (Connection localConn =<a name="line.373"></a> -<span class="sourceLineNo">374</span> ConnectionFactory.createConnection(ctx.getLocalConfiguration())) {<a name="line.374"></a> -<span class="sourceLineNo">375</span> if (!localConn.getAdmin().tableExists(table)) {<a name="line.375"></a> -<span class="sourceLineNo">376</span> // Would potentially be better to retry in one of the outer loops<a name="line.376"></a> -<span class="sourceLineNo">377</span> // and add a table filter there; but that would break the encapsulation,<a name="line.377"></a> -<span class="sourceLineNo">378</span> // so we're doing the filtering here.<a name="line.378"></a> -<span class="sourceLineNo">379</span> LOG.info("Missing table detected at sink, local table also does not exist, filtering edits for '"+table+"'");<a name="line.379"></a> -<span class="sourceLineNo">380</span> batches = filterBatches(batches, table);<a name="line.380"></a> -<span class="sourceLineNo">381</span> continue;<a name="line.381"></a> -<span class="sourceLineNo">382</span> }<a name="line.382"></a> -<span class="sourceLineNo">383</span> } catch (IOException iox) {<a name="line.383"></a> -<span class="sourceLineNo">384</span> LOG.warn("Exception checking for local table: ", iox);<a name="line.384"></a> -<span class="sourceLineNo">385</span> }<a name="line.385"></a> -<span class="sourceLineNo">386</span> }<a name="line.386"></a> -<span class="sourceLineNo">387</span> }<a name="line.387"></a> -<span class="sourceLineNo">388</span> // fall through and sleep below<a name="line.388"></a> -<span class="sourceLineNo">389</span> } else {<a name="line.389"></a> -<span class="sourceLineNo">390</span> LOG.warn("Peer encountered RemoteException, rechecking all sinks: ", ioe);<a name="line.390"></a> -<span class="sourceLineNo">391</span> replicationSinkMgr.chooseSinks();<a name="line.391"></a> -<span class="sourceLineNo">392</span> }<a name="line.392"></a> -<span class="sourceLineNo">393</span> } else {<a name="line.393"></a> -<span class="sourceLineNo">394</span> if (ioe instanceof SocketTimeoutException) {<a name="line.394"></a> -<span class="sourceLineNo">395</span> // This exception means we waited for more than 60s and nothing<a name="line.395"></a> -<span class="sourceLineNo">396</span> // happened, the cluster is alive and calling it right away<a name="line.396"></a> -<span class="sourceLineNo">397</span> // even for a test just makes things worse.<a name="line.397"></a> -<span class="sourceLineNo">398</span> sleepForRetries("Encountered a SocketTimeoutException. Since the " +<a name="line.398"></a> -<span class="sourceLineNo">399</span> "call to the remote cluster timed out, which is usually " +<a name="line.399"></a> -<span class="sourceLineNo">400</span> "caused by a machine failure or a massive slowdown",<a name="line.400"></a> -<span class="sourceLineNo">401</span> this.socketTimeoutMultiplier);<a name="line.401"></a> -<span class="sourceLineNo">402</span> } else if (ioe instanceof ConnectException || ioe instanceof UnknownHostException) {<a name="line.402"></a> -<span class="sourceLineNo">403</span> LOG.warn("Peer is unavailable, rechecking all sinks: ", ioe);<a name="line.403"></a> -<span class="sourceLineNo">404</span> replicationSinkMgr.chooseSinks();<a name="line.404"></a> -<span class="sourceLineNo">405</span> } else {<a name="line.405"></a> -<span class="sourceLineNo">406</span> LOG.warn("Can't replicate because of a local or network error: ", ioe);<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> if (sleepForRetries("Since we are unable to replicate", sleepMultiplier)) {<a name="line.409"></a> -<span class="sourceLineNo">410</span> sleepMultiplier++;<a name="line.410"></a> -<span class="sourceLineNo">411</span> }<a name="line.411"></a> -<span class="sourceLineNo">412</span> }<a name="line.412"></a> -<span class="sourceLineNo">413</span> }<a name="line.413"></a> -<span class="sourceLineNo">414</span> return false; // in case we exited before replicating<a name="line.414"></a> -<span class="sourceLineNo">415</span> }<a name="line.415"></a> -<span class="sourceLineNo">416</span><a name="line.416"></a> -<span class="sourceLineNo">417</span> protected boolean isPeerEnabled() {<a name="line.417"></a> -<span class="sourceLineNo">418</span> return ctx.getReplicationPeer().isPeerEnabled();<a name="line.418"></a> -<span class="sourceLineNo">419</span> }<a name="line.419"></a> -<span class="sourceLineNo">420</span><a name="line.420"></a> -<span class="sourceLineNo">421</span> @Override<a name="line.421"></a> -<span class="sourceLineNo">422</span> protected void doStop() {<a name="line.422"></a> -<span class="sourceLineNo">423</span> disconnect(); //don't call super.doStop()<a name="line.423"></a> -<span class="sourceLineNo">424</span> if (this.conn != null) {<a name="line.424"></a> -<span class="sourceLineNo">425</span> try {<a name="line.425"></a> -<span class="sourceLineNo">426</span> this.conn.close();<a name="line.426"></a> -<span class="sourceLineNo">427</span> this.conn = null;<a name="line.427"></a> -<span class="sourceLineNo">428</span> } catch (IOException e) {<a name="line.428"></a> -<span class="sourceLineNo">429</span> LOG.warn("Failed to close the connection");<a name="line.429"></a> -<span class="sourceLineNo">430</span> }<a name="line.430"></a> -<span class="sourceLineNo">431</span> }<a name="line.431"></a> -<span class="sourceLineNo">432</span> // Allow currently running replication tasks to finish<a name="line.432"></a> -<span class="sourceLineNo">433</span> exec.shutdown();<a name="line.433"></a> -<span class="sourceLineNo">434</span> try {<a name="line.434"></a> -<span class="sourceLineNo">435</span> exec.awaitTermination(maxTerminationWait, TimeUnit.MILLISECONDS);<a name="line.435"></a> -<span class="sourceLineNo">436</span> } catch (InterruptedException e) {<a name="line.436"></a> +<span class="sourceLineNo">275</span> return null;<a name="line.275"></a> +<span class="sourceLineNo">276</span> }<a name="line.276"></a> +<span class="sourceLineNo">277</span><a name="line.277"></a> +<span class="sourceLineNo">278</span> // Filter a set of batches by TableName<a name="line.278"></a> +<span class="sourceLineNo">279</span> private List<List<Entry>> filterBatches(final List<List<Entry>> oldEntryList, TableName table) {<a name="line.279"></a> +<span class="sourceLineNo">280</span> return oldEntryList<a name="line.280"></a> +<span class="sourceLineNo">281</span> .stream().map(entries -> entries.stream()<a name="line.281"></a> +<span class="sourceLineNo">282</span> .filter(e -> !e.getKey().getTableName().equals(table)).collect(Collectors.toList()))<a name="line.282"></a> +<span class="sourceLineNo">283</span> .collect(Collectors.toList());<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> private void reconnectToPeerCluster() {<a name="line.286"></a> +<span class="sourceLineNo">287</span> ClusterConnection connection = null;<a name="line.287"></a> +<span class="sourceLineNo">288</span> try {<a name="line.288"></a> +<span class="sourceLineNo">289</span> connection = (ClusterConnection) ConnectionFactory.createConnection(this.conf);<a name="line.289"></a> +<span class="sourceLineNo">290</span> } catch (IOException ioe) {<a name="line.290"></a> +<span class="sourceLineNo">291</span> LOG.warn("Failed to create connection for peer cluster", ioe);<a name="line.291"></a> +<span class="sourceLineNo">292</span> }<a name="line.292"></a> +<span class="sourceLineNo">293</span> if (connection != null) {<a name="line.293"></a> +<span class="sourceLineNo">294</span> this.conn = connection;<a name="line.294"></a> +<span class="sourceLineNo">295</span> }<a name="line.295"></a> +<span class="sourceLineNo">296</span> }<a name="line.296"></a> +<span class="sourceLineNo">297</span><a name="line.297"></a> +<span class="sourceLineNo">298</span> private long parallelReplicate(CompletionService<Integer> pool, ReplicateContext replicateContext,<a name="line.298"></a> +<span class="sourceLineNo">299</span> List<List<Entry>> batches) throws IOException {<a name="line.299"></a> +<span class="sourceLineNo">300</span> int futures = 0;<a name="line.300"></a> +<span class="sourceLineNo">301</span> for (int i = 0; i < batches.size(); i++) {<a name="line.301"></a> +<span class="sourceLineNo">302</span> List<Entry> entries = batches.get(i);<a name="line.302"></a> +<span class="sourceLineNo">303</span> if (!entries.isEmpty()) {<a name="line.303"></a> +<span class="sourceLineNo">304</span> LOG.trace("Submitting {} entries of total size {}", entries.size(),<a name="line.304"></a> +<span class="sourceLineNo">305</span> replicateContext.getSize());<a name="line.305"></a> +<span class="sourceLineNo">306</span> // RuntimeExceptions encountered here bubble up and are handled in ReplicationSource<a name="line.306"></a> +<span class="sourceLineNo">307</span> pool.submit(createReplicator(entries, i));<a name="line.307"></a> +<span class="sourceLineNo">308</span> futures++;<a name="line.308"></a> +<span class="sourceLineNo">309</span> }<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> IOException iox = null;<a name="line.312"></a> +<span class="sourceLineNo">313</span> long lastWriteTime = 0;<a name="line.313"></a> +<span class="sourceLineNo">314</span> for (int i = 0; i < futures; i++) {<a name="line.314"></a> +<span class="sourceLineNo">315</span> try {<a name="line.315"></a> +<span class="sourceLineNo">316</span> // wait for all futures, remove successful parts<a name="line.316"></a> +<span class="sourceLineNo">317</span> // (only the remaining parts will be retried)<a name="line.317"></a> +<span class="sourceLineNo">318</span> Future<Integer> f = pool.take();<a name="line.318"></a> +<span class="sourceLineNo">319</span> int index = f.get();<a name="line.319"></a> +<span class="sourceLineNo">320</span> List<Entry> batch = batches.get(index);<a name="line.320"></a> +<span class="sourceLineNo">321</span> batches.set(index, Collections.emptyList()); // remove successful batch<a name="line.321"></a> +<span class="sourceLineNo">322</span> // Find the most recent write time in the batch<a name="line.322"></a> +<span class="sourceLineNo">323</span> long writeTime = batch.get(batch.size() - 1).getKey().getWriteTime();<a name="line.323"></a> +<span class="sourceLineNo">324</span> if (writeTime > lastWriteTime) {<a name="line.324"></a> +<span class="sourceLineNo">325</span> lastWriteTime = writeTime;<a name="line.325"></a> +<span class="sourceLineNo">326</span> }<a name="line.326"></a> +<span class="sourceLineNo">327</span> } catch (InterruptedException ie) {<a name="line.327"></a> +<span class="sourceLineNo">328</span> iox = new IOException(ie);<a name="line.328"></a> +<span class="sourceLineNo">329</span> } catch (ExecutionException ee) {<a name="line.329"></a> +<span class="sourceLineNo">330</span> // cause must be an IOException<a name="line.330"></a> +<span class="sourceLineNo">331</span> iox = (IOException) ee.getCause();<a name="line.331"></a> +<span class="sourceLineNo">332</span> }<a name="line.332"></a> +<span class="sourceLineNo">333</span> }<a name="line.333"></a> +<span class="sourceLineNo">334</span> if (iox != null) {<a name="line.334"></a> +<span class="sourceLineNo">335</span> // if we had any exceptions, try again<a name="line.335"></a> +<span class="sourceLineNo">336</span> throw iox;<a name="line.336"></a> +<span class="sourceLineNo">337</span> }<a name="line.337"></a> +<span class="sourceLineNo">338</span> return lastWriteTime;<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> /**<a name="line.341"></a> +<span class="sourceLineNo">342</span> * Do the shipping logic<a name="line.342"></a> +<span class="sourceLineNo">343</span> */<a name="line.343"></a> +<span class="sourceLineNo">344</span> @Override<a name="line.344"></a> +<span class="sourceLineNo">345</span> public boolean replicate(ReplicateContext replicateContext) {<a name="line.345"></a> +<span class="sourceLineNo">346</span> CompletionService<Integer> pool = new ExecutorCompletionService<>(this.exec);<a name="line.346"></a> +<span class="sourceLineNo">347</span> String walGroupId = replicateContext.getWalGroupId();<a name="line.347"></a> +<span class="sourceLineNo">348</span> int sleepMultiplier = 1;<a name="line.348"></a> +<span class="sourceLineNo">349</span><a name="line.349"></a> +<span class="sourceLineNo">350</span> if (!peersSelected && this.isRunning()) {<a name="line.350"></a> +<span class="sourceLineNo">351</span> connectToPeers();<a name="line.351"></a> +<span class="sourceLineNo">352</span> peersSelected = 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> int numSinks = replicationSinkMgr.getNumSinks();<a name="line.355"></a> +<span class="sourceLineNo">356</span> if (numSinks == 0) {<a name="line.356"></a> +<span class="sourceLineNo">357</span> LOG.warn("No replication sinks found, returning without replicating. The source should " +<a name="line.357"></a> +<span class="sourceLineNo">358</span> "retry with the same set of edits.");<a name="line.358"></a> +<span class="sourceLineNo">359</span> return false;<a name="line.359"></a> +<span class="sourceLineNo">360</span> }<a name="line.360"></a> +<span class="sourceLineNo">361</span><a name="line.361"></a> +<span class="sourceLineNo">362</span> List<List<Entry>> batches = createBatches(replicateContext.getEntries());<a name="line.362"></a> +<span class="sourceLineNo">363</span> while (this.isRunning() && !exec.isShutdown()) {<a name="line.363"></a> +<span class="sourceLineNo">364</span> if (!isPeerEnabled()) {<a name="line.364"></a> +<span class="sourceLineNo">365</span> if (sleepForRetries("Replication is disabled", sleepMultiplier)) {<a name="line.365"></a> +<span class="sourceLineNo">366</span> sleepMultiplier++;<a name="line.366"></a> +<span class="sourceLineNo">367</span> }<a name="line.367"></a> +<span class="sourceLineNo">368</span> continue;<a name="line.368"></a> +<span class="sourceLineNo">369</span> }<a name="line.369"></a> +<span class="sourceLineNo">370</span> if (this.conn == null || this.conn.isClosed()) {<a name="line.370"></a> +<span class="sourceLineNo">371</span> reconnectToPeerCluster();<a name="line.371"></a> +<span class="sourceLineNo">372</span> }<a name="line.372"></a> +<span class="sourceLineNo">373</span> try {<a name="line.373"></a> +<span class="sourceLineNo">374</span> long lastWriteTime;<a name="line.374"></a> +<span class="sourceLineNo">375</span><a name="line.375"></a> +<span class="sourceLineNo">376</span> // replicate the batches to sink side.<a name="line.376"></a> +<span class="sourceLineNo">377</span> lastWriteTime = parallelReplicate(pool, replicateContext, batches);<a name="line.377"></a> +<span class="sourceLineNo">378</span><a name="line.378"></a> +<span class="sourceLineNo">379</span> // update metrics<a name="line.379"></a> +<span class="sourceLineNo">380</span> if (lastWriteTime > 0) {<a name="line.380"></a> +<span class="sourceLineNo">381</span> this.metrics.setAgeOfLastShippedOp(lastWriteTime, walGroupId);<a name="line.381"></a> +<span class="sourceLineNo">382</span> }<a name="line.382"></a> +<span class="sourceLineNo">383</span> return true;<a name="line.383"></a> +<span class="sourceLineNo">384</span> } catch (IOException ioe) {<a name="line.384"></a> +<span class="sourceLineNo">385</span> // Didn't ship anything, but must still age the last time we did<a name="line.385"></a> +<span class="sourceLineNo">386</span> this.metrics.refreshAgeOfLastShippedOp(walGroupId);<a name="line.386"></a> +<span class="sourceLineNo">387</span> if (ioe instanceof RemoteException) {<a name="line.387"></a> +<span class="sourceLineNo">388</span> ioe = ((RemoteException) ioe).unwrapRemoteException();<a name="line.388"></a> +<span class="sourceLineNo">389</span> LOG.warn("Can't replicate because of an error on the remote cluster: ", ioe);<a name="line.389"></a> +<span class="sourceLineNo">390</span> if (ioe instanceof TableNotFoundException) {<a name="line.390"></a> +<span class="sourceLineNo">391</span> if (dropOnDeletedTables) {<a name="line.391"></a> +<span class="sourceLineNo">392</span> // this is a bit fragile, but cannot change how TNFE is serialized<a name="line.392"></a> +<span class="sourceLineNo">393</span> // at least check whether the table name is legal<a name="line.393"></a> +<span class="sourceLineNo">394</span> TableName table = parseTable(ioe.getMessage());<a name="line.394"></a> +<span class="sourceLineNo">395</span> if (table != null) {<a name="line.395"></a> +<span class="sourceLineNo">396</span> try (Connection localConn =<a name="line.396"></a> +<span class="sourceLineNo">397</span> ConnectionFactory.createConnection(ctx.getLocalConfiguration())) {<a name="line.397"></a> +<span class="sourceLineNo">398</span> if (!localConn.getAdmin().tableExists(table)) {<a name="line.398"></a> +<span class="sourceLineNo">399</span> // Would potentially be better to retry in one of the outer loops<a name="line.399"></a> +<span class="sourceLineNo">400</span> // and add a table filter there; but that would break the encapsulation,<a name="line.400"></a> +<span class="sourceLineNo">401</span> // so we're doing the filtering here.<a name="line.401"></a> +<span class="sourceLineNo">402</span> LOG.info("Missing table detected at sink, local table also does not exist, " +<a name="line.402"></a> +<span class="sourceLineNo">403</span> "filtering edits for '" + table + "'");<a name="line.403"></a> +<span class="sourceLineNo">404</span> batches = filterBatches(batches, table);<a name="line.404"></a> +<span class="sourceLineNo">405</span> continue;<a name="line.405"></a> +<span class="sourceLineNo">406</span> }<a name="line.406"></a> +<span class="sourceLineNo">407</span> } catch (IOException iox) {<a name="line.407"></a> +<span class="sourceLineNo">408</span> LOG.warn("Exception checking for local table: ", iox);<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> // fall through and sleep below<a name="line.412"></a> +<span class="sourceLineNo">413</span> } else {<a name="line.413"></a> +<span class="sourceLineNo">414</span> LOG.warn("Peer encountered RemoteException, rechecking all sinks: ", ioe);<a name="line.414"></a> +<span class="sourceLineNo">415</span> replicationSinkMgr.chooseSinks();<a name="line.415"></a> +<span class="sourceLineNo">416</span> }<a name="line.416"></a> +<span class="sourceLineNo">417</span> } else {<a name="line.417"></a> +<span class="sourceLineNo">418</span> if (ioe instanceof SocketTimeoutException) {<a name="line.418"></a> +<span class="sourceLineNo">419</span> // This exception means we waited for more than 60s and nothing<a name="line.419"></a> +<span class="sourceLineNo">420</span> // happened, the cluster is alive and calling it right away<a name="line.420"></a> +<span class="sourceLineNo">421</span> // even for a test just makes things worse.<a name="line.421"></a> +<span class="sourceLineNo">422</span> sleepForRetries("Encountered a SocketTimeoutException. Since the " +<a name="line.422"></a> +<span class="sourceLineNo">423</span> "call to the remote cluster timed out, which is usually " +<a name="line.423"></a> +<span class="sourceLineNo">424</span> "caused by a machine failure or a massive slowdown",<a name="line.424"></a> +<span class="sourceLineNo">425</span> this.socketTimeoutMultiplier);<a name="line.425"></a> +<span class="sourceLineNo">426</span> } else if (ioe instanceof ConnectException || ioe instanceof UnknownHostException) {<a name="line.426"></a> +<span class="sourceLineNo">427</span> LOG.warn("Peer is unavailable, rechecking all sinks: ", ioe);<a name="line.427"></a> +<span class="sourceLineNo">428</span> replicationSinkMgr.chooseSinks();<a name="line.428"></a> +<span class="sourceLineNo">429</span> } else {<a name="line.429"></a> +<span class="sourceLineNo">430</span> LOG.warn("Can't replicate because of a local or network error: ", ioe);<a name="line.430"></a> +<span class="sourceLineNo">431</span> }<a name="line.431"></a> +<span class="sourceLineNo">432</span> }<a name="line.432"></a> +<span class="sourceLineNo">433</span> if (sleepForRetries("Since we are unable to replicate", sleepMultiplier)) {<a name="line.433"></a> +<span class="sourceLineNo">434</span> sleepMultiplier++;<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> // Abort if the tasks did not terminate in time<a name="line.438"></a> -<span class="sourceLineNo">439</span> if (!exec.isTerminated()) {<a name="line.439"></a> -<span class="sourceLineNo">440</span> String errMsg = "HBaseInterClusterReplicationEndpoint termination failed. The " +<a name="line.440"></a> -<span class="sourceLineNo">441</span> "ThreadPoolExecutor failed to finish all tasks within " + maxTerminationWait + "ms. " +<a name="line.441"></a> -<span class="sourceLineNo">442</span> "Aborting to prevent Replication from deadlocking. See HBASE-16081.";<a name="line.442"></a> -<span class="sourceLineNo">443</span> abortable.abort(errMsg, new IOException(errMsg));<a name="line.443"></a> -<span class="sourceLineNo">444</span> }<a name="line.444"></a> -<span class="sourceLineNo">445</span> notifyStopped();<a name="line.445"></a> -<span class="sourceLineNo">446</span> }<a name="line.446"></a> -<span class="sourceLineNo">447</span><a name="line.447"></a> -<span class="sourceLineNo">448</span> @VisibleForTesting<a name="line.448"></a> -<span class="sourceLineNo">449</span> protected Replicator createReplicator(List<Entry> entries, int ordinal) {<a name="line.449"></a> -<span class="sourceLineNo">450</span> return new Replicator(entries, ordinal);<a name="line.450"></a> -<span class="sourceLineNo">451</span> }<a name="line.451"></a> -<span class="sourceLineNo">452</span><a name="line.452"></a> -<span class="sourceLineNo">453</span> @VisibleForTesting<a name="line.453"></a> -<span class="sourceLineNo">454</span> protected class Replicator implements Callable<Integer> {<a name="line.454"></a> -<span class="sourceLineNo">455</span> private List<Entry> entries;<a name="line.455"></a> -<span class="sourceLineNo">456</span> private int ordinal;<a name="line.456"></a> -<span class="sourceLineNo">457</span> public Replicator(List<Entry> entries, int ordinal) {<a name="line.457"></a> -<span class="sourceLineNo">458</span> this.entries = entries;<a name="line.458"></a> -<span class="sourceLineNo">459</span> this.ordinal = ordinal;<a name="line.459"></a> -<span class="sourceLineNo">460</span> }<a name="line.460"></a> -<span class="sourceLineNo">461</span><a name="line.461"></a> -<span class="sourceLineNo">462</span> protected void replicateEntries(BlockingInterface rrs, final List<Entry> batch,<a name="line.462"></a> -<span class="sourceLineNo">463</span> String replicationClusterId, Path baseNamespaceDir, Path hfileArchiveDir)<a name="line.463"></a> -<span class="sourceLineNo">464</span> throws IOException {<a name="line.464"></a> -<span class="sourceLineNo">465</span> if (LOG.isTraceEnabled()) {<a name="line.465"></a> -<span class="sourceLineNo">466</span> long size = 0;<a name="line.466"></a> -<span class="sourceLineNo">467</span> for (Entry e: entries) {<a name="line.467"></a> -<span class="sourceLineNo">468</span> size += e.getKey().estimatedSerializedSizeOf();<a name="line.468"></a> -<span class="sourceLineNo">469</span> size += e.getEdit().estimatedSerializedSizeOf();<a name="line.469"></a> -<span class="sourceLineNo">470</span> }<a name="line.470"></a> -<span class="sourceLineNo">471</span> LOG.trace("Replicating batch " + System.identityHashCode(entries) + " of " +<a name="line.471"></a> -<span class="sourceLineNo">472</span> entries.size() + " entries with total size " + size + " bytes to " +<a name="line.472"></a> -<span class="sourceLineNo">473</span> replicationClusterId);<a name="line.473"></a> -<span class="sourceLineNo">474</span> }<a name="line.474"></a> -<span class="sourceLineNo">475</span> try {<a name="line.475"></a> -<span class="sourceLineNo">476</span> ReplicationProtbufUtil.replicateWALEntry(rrs, batch.toArray(new Entry[batch.size()]),<a name="line.476"></a> -<span class="sourceLineNo">477</span> replicationClusterId, baseNamespaceDir, hfileArchiveDir);<a name="line.477"></a> -<span class="sourceLineNo">478</span> if (LOG.isTraceEnabled()) {<a name="line.478"></a> -<span class="sourceLineNo">479</span> LOG.trace("Completed replicating batch " + System.identityHashCode(entries));<a name="line.479"></a> -<span class="sourceLineNo">480</span> }<a name="line.480"></a> -<span class="sourceLineNo">481</span> } catch (IOException e) {<a name="line.481"></a> -<span class="sourceLineNo">482</span> if (LOG.isTraceEnabled()) {<a name="line.482"></a> -<span class="sourceLineNo">483</span> LOG.trace("Failed replicating batch " + System.identityHashCode(entries), e);<a name="line.483"></a> -<span class="sourceLineNo">484</span> }<a name="line.484"></a> -<span class="sourceLineNo">485</span> throw e;<a name="line.485"></a> -<span class="sourceLineNo">486</span> }<a name="line.486"></a> -<span class="sourceLineNo">487</span> }<a name="line.487"></a> -<span class="sourceLineNo">488</span><a name="line.488"></a> -<span class="sourceLineNo">489</span> @Override<a name="line.489"></a> -<span class="sourceLin
<TRUNCATED>
