http://git-wip-us.apache.org/repos/asf/hbase-site/blob/65565d77/devapidocs/src-html/org/apache/hadoop/hbase/master/MasterWalManager.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/master/MasterWalManager.html b/devapidocs/src-html/org/apache/hadoop/hbase/master/MasterWalManager.html index dd2def7..498695e 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/master/MasterWalManager.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/master/MasterWalManager.html @@ -34,282 +34,308 @@ <span class="sourceLineNo">026</span>import java.util.Set;<a name="line.26"></a> <span class="sourceLineNo">027</span>import java.util.concurrent.locks.Lock;<a name="line.27"></a> <span class="sourceLineNo">028</span>import java.util.concurrent.locks.ReentrantLock;<a name="line.28"></a> -<span class="sourceLineNo">029</span><a name="line.29"></a> -<span class="sourceLineNo">030</span>import org.apache.hadoop.conf.Configuration;<a name="line.30"></a> -<span class="sourceLineNo">031</span>import org.apache.hadoop.fs.FileStatus;<a name="line.31"></a> -<span class="sourceLineNo">032</span>import org.apache.hadoop.fs.FileSystem;<a name="line.32"></a> -<span class="sourceLineNo">033</span>import org.apache.hadoop.fs.Path;<a name="line.33"></a> -<span class="sourceLineNo">034</span>import org.apache.hadoop.fs.PathFilter;<a name="line.34"></a> -<span class="sourceLineNo">035</span>import org.apache.hadoop.hbase.HConstants;<a name="line.35"></a> -<span class="sourceLineNo">036</span>import org.apache.hadoop.hbase.ServerName;<a name="line.36"></a> -<span class="sourceLineNo">037</span>import org.apache.hadoop.hbase.util.EnvironmentEdgeManager;<a name="line.37"></a> -<span class="sourceLineNo">038</span>import org.apache.hadoop.hbase.util.FSUtils;<a name="line.38"></a> -<span class="sourceLineNo">039</span>import org.apache.hadoop.hbase.wal.AbstractFSWALProvider;<a name="line.39"></a> -<span class="sourceLineNo">040</span>import org.apache.hadoop.hbase.wal.WALSplitter;<a name="line.40"></a> -<span class="sourceLineNo">041</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.41"></a> -<span class="sourceLineNo">042</span>import org.slf4j.Logger;<a name="line.42"></a> -<span class="sourceLineNo">043</span>import org.slf4j.LoggerFactory;<a name="line.43"></a> -<span class="sourceLineNo">044</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.44"></a> -<span class="sourceLineNo">045</span><a name="line.45"></a> -<span class="sourceLineNo">046</span>/**<a name="line.46"></a> -<span class="sourceLineNo">047</span> * This class abstracts a bunch of operations the HMaster needs<a name="line.47"></a> -<span class="sourceLineNo">048</span> * when splitting log files e.g. finding log files, dirs etc.<a name="line.48"></a> -<span class="sourceLineNo">049</span> */<a name="line.49"></a> -<span class="sourceLineNo">050</span>@InterfaceAudience.Private<a name="line.50"></a> -<span class="sourceLineNo">051</span>public class MasterWalManager {<a name="line.51"></a> -<span class="sourceLineNo">052</span> private static final Logger LOG = LoggerFactory.getLogger(MasterWalManager.class);<a name="line.52"></a> -<span class="sourceLineNo">053</span><a name="line.53"></a> -<span class="sourceLineNo">054</span> final static PathFilter META_FILTER = new PathFilter() {<a name="line.54"></a> -<span class="sourceLineNo">055</span> @Override<a name="line.55"></a> -<span class="sourceLineNo">056</span> public boolean accept(Path p) {<a name="line.56"></a> -<span class="sourceLineNo">057</span> return AbstractFSWALProvider.isMetaFile(p);<a name="line.57"></a> -<span class="sourceLineNo">058</span> }<a name="line.58"></a> -<span class="sourceLineNo">059</span> };<a name="line.59"></a> -<span class="sourceLineNo">060</span><a name="line.60"></a> -<span class="sourceLineNo">061</span> final static PathFilter NON_META_FILTER = new PathFilter() {<a name="line.61"></a> -<span class="sourceLineNo">062</span> @Override<a name="line.62"></a> -<span class="sourceLineNo">063</span> public boolean accept(Path p) {<a name="line.63"></a> -<span class="sourceLineNo">064</span> return !AbstractFSWALProvider.isMetaFile(p);<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><a name="line.67"></a> -<span class="sourceLineNo">068</span> // metrics for master<a name="line.68"></a> -<span class="sourceLineNo">069</span> // TODO: Rename it, since those metrics are split-manager related<a name="line.69"></a> -<span class="sourceLineNo">070</span> private final MetricsMasterFileSystem metricsMasterFilesystem = new MetricsMasterFileSystem();<a name="line.70"></a> -<span class="sourceLineNo">071</span><a name="line.71"></a> -<span class="sourceLineNo">072</span> // Keep around for convenience.<a name="line.72"></a> -<span class="sourceLineNo">073</span> private final MasterServices services;<a name="line.73"></a> -<span class="sourceLineNo">074</span> private final Configuration conf;<a name="line.74"></a> -<span class="sourceLineNo">075</span> private final FileSystem fs;<a name="line.75"></a> -<span class="sourceLineNo">076</span><a name="line.76"></a> -<span class="sourceLineNo">077</span> // The Path to the old logs dir<a name="line.77"></a> -<span class="sourceLineNo">078</span> private final Path oldLogDir;<a name="line.78"></a> -<span class="sourceLineNo">079</span> private final Path rootDir;<a name="line.79"></a> -<span class="sourceLineNo">080</span><a name="line.80"></a> -<span class="sourceLineNo">081</span> // create the split log lock<a name="line.81"></a> -<span class="sourceLineNo">082</span> private final Lock splitLogLock = new ReentrantLock();<a name="line.82"></a> -<span class="sourceLineNo">083</span> private final SplitLogManager splitLogManager;<a name="line.83"></a> -<span class="sourceLineNo">084</span><a name="line.84"></a> -<span class="sourceLineNo">085</span> // Is the fileystem ok?<a name="line.85"></a> -<span class="sourceLineNo">086</span> private volatile boolean fsOk = true;<a name="line.86"></a> -<span class="sourceLineNo">087</span><a name="line.87"></a> -<span class="sourceLineNo">088</span> public MasterWalManager(MasterServices services) throws IOException {<a name="line.88"></a> -<span class="sourceLineNo">089</span> this(services.getConfiguration(), services.getMasterFileSystem().getWALFileSystem(),<a name="line.89"></a> -<span class="sourceLineNo">090</span> services.getMasterFileSystem().getWALRootDir(), services);<a name="line.90"></a> -<span class="sourceLineNo">091</span> }<a name="line.91"></a> -<span class="sourceLineNo">092</span><a name="line.92"></a> -<span class="sourceLineNo">093</span> public MasterWalManager(Configuration conf, FileSystem fs, Path rootDir, MasterServices services)<a name="line.93"></a> -<span class="sourceLineNo">094</span> throws IOException {<a name="line.94"></a> -<span class="sourceLineNo">095</span> this.fs = fs;<a name="line.95"></a> -<span class="sourceLineNo">096</span> this.conf = conf;<a name="line.96"></a> -<span class="sourceLineNo">097</span> this.rootDir = rootDir;<a name="line.97"></a> -<span class="sourceLineNo">098</span> this.services = services;<a name="line.98"></a> -<span class="sourceLineNo">099</span> this.splitLogManager = new SplitLogManager(services, conf);<a name="line.99"></a> -<span class="sourceLineNo">100</span><a name="line.100"></a> -<span class="sourceLineNo">101</span> this.oldLogDir = new Path(rootDir, HConstants.HREGION_OLDLOGDIR_NAME);<a name="line.101"></a> -<span class="sourceLineNo">102</span> }<a name="line.102"></a> -<span class="sourceLineNo">103</span><a name="line.103"></a> -<span class="sourceLineNo">104</span> public void stop() {<a name="line.104"></a> -<span class="sourceLineNo">105</span> if (splitLogManager != null) {<a name="line.105"></a> -<span class="sourceLineNo">106</span> splitLogManager.stop();<a name="line.106"></a> -<span class="sourceLineNo">107</span> }<a name="line.107"></a> -<span class="sourceLineNo">108</span> }<a name="line.108"></a> -<span class="sourceLineNo">109</span><a name="line.109"></a> -<span class="sourceLineNo">110</span> @VisibleForTesting<a name="line.110"></a> -<span class="sourceLineNo">111</span> SplitLogManager getSplitLogManager() {<a name="line.111"></a> -<span class="sourceLineNo">112</span> return this.splitLogManager;<a name="line.112"></a> -<span class="sourceLineNo">113</span> }<a name="line.113"></a> -<span class="sourceLineNo">114</span><a name="line.114"></a> -<span class="sourceLineNo">115</span> /**<a name="line.115"></a> -<span class="sourceLineNo">116</span> * Get the directory where old logs go<a name="line.116"></a> -<span class="sourceLineNo">117</span> * @return the dir<a name="line.117"></a> -<span class="sourceLineNo">118</span> */<a name="line.118"></a> -<span class="sourceLineNo">119</span> Path getOldLogDir() {<a name="line.119"></a> -<span class="sourceLineNo">120</span> return this.oldLogDir;<a name="line.120"></a> -<span class="sourceLineNo">121</span> }<a name="line.121"></a> -<span class="sourceLineNo">122</span><a name="line.122"></a> -<span class="sourceLineNo">123</span> public FileSystem getFileSystem() {<a name="line.123"></a> -<span class="sourceLineNo">124</span> return this.fs;<a name="line.124"></a> -<span class="sourceLineNo">125</span> }<a name="line.125"></a> -<span class="sourceLineNo">126</span><a name="line.126"></a> -<span class="sourceLineNo">127</span> /**<a name="line.127"></a> -<span class="sourceLineNo">128</span> * Checks to see if the file system is still accessible.<a name="line.128"></a> -<span class="sourceLineNo">129</span> * If not, sets closed<a name="line.129"></a> -<span class="sourceLineNo">130</span> * @return false if file system is not available<a name="line.130"></a> -<span class="sourceLineNo">131</span> */<a name="line.131"></a> -<span class="sourceLineNo">132</span> private boolean checkFileSystem() {<a name="line.132"></a> -<span class="sourceLineNo">133</span> if (this.fsOk) {<a name="line.133"></a> -<span class="sourceLineNo">134</span> try {<a name="line.134"></a> -<span class="sourceLineNo">135</span> FSUtils.checkFileSystemAvailable(this.fs);<a name="line.135"></a> -<span class="sourceLineNo">136</span> FSUtils.checkDfsSafeMode(this.conf);<a name="line.136"></a> -<span class="sourceLineNo">137</span> } catch (IOException e) {<a name="line.137"></a> -<span class="sourceLineNo">138</span> services.abort("Shutting down HBase cluster: file system not available", e);<a name="line.138"></a> -<span class="sourceLineNo">139</span> this.fsOk = false;<a name="line.139"></a> -<span class="sourceLineNo">140</span> }<a name="line.140"></a> -<span class="sourceLineNo">141</span> }<a name="line.141"></a> -<span class="sourceLineNo">142</span> return this.fsOk;<a name="line.142"></a> -<span class="sourceLineNo">143</span> }<a name="line.143"></a> -<span class="sourceLineNo">144</span><a name="line.144"></a> -<span class="sourceLineNo">145</span> /**<a name="line.145"></a> -<span class="sourceLineNo">146</span> * Inspect the log directory to find dead servers which need recovery work<a name="line.146"></a> -<span class="sourceLineNo">147</span> * @return A set of ServerNames which aren't running but still have WAL files left in file system<a name="line.147"></a> -<span class="sourceLineNo">148</span> */<a name="line.148"></a> -<span class="sourceLineNo">149</span> public Set<ServerName> getFailedServersFromLogFolders() {<a name="line.149"></a> -<span class="sourceLineNo">150</span> boolean retrySplitting = !conf.getBoolean("hbase.hlog.split.skip.errors",<a name="line.150"></a> -<span class="sourceLineNo">151</span> WALSplitter.SPLIT_SKIP_ERRORS_DEFAULT);<a name="line.151"></a> -<span class="sourceLineNo">152</span><a name="line.152"></a> -<span class="sourceLineNo">153</span> Set<ServerName> serverNames = new HashSet<>();<a name="line.153"></a> -<span class="sourceLineNo">154</span> Path logsDirPath = new Path(this.rootDir, HConstants.HREGION_LOGDIR_NAME);<a name="line.154"></a> -<span class="sourceLineNo">155</span><a name="line.155"></a> -<span class="sourceLineNo">156</span> do {<a name="line.156"></a> -<span class="sourceLineNo">157</span> if (services.isStopped()) {<a name="line.157"></a> -<span class="sourceLineNo">158</span> LOG.warn("Master stopped while trying to get failed servers.");<a name="line.158"></a> -<span class="sourceLineNo">159</span> break;<a name="line.159"></a> +<span class="sourceLineNo">029</span>import java.util.stream.Collectors;<a name="line.29"></a> +<span class="sourceLineNo">030</span>import java.util.stream.Stream;<a name="line.30"></a> +<span class="sourceLineNo">031</span>import org.apache.hadoop.conf.Configuration;<a name="line.31"></a> +<span class="sourceLineNo">032</span>import org.apache.hadoop.fs.FileStatus;<a name="line.32"></a> +<span class="sourceLineNo">033</span>import org.apache.hadoop.fs.FileSystem;<a name="line.33"></a> +<span class="sourceLineNo">034</span>import org.apache.hadoop.fs.Path;<a name="line.34"></a> +<span class="sourceLineNo">035</span>import org.apache.hadoop.fs.PathFilter;<a name="line.35"></a> +<span class="sourceLineNo">036</span>import org.apache.hadoop.hbase.HConstants;<a name="line.36"></a> +<span class="sourceLineNo">037</span>import org.apache.hadoop.hbase.ServerName;<a name="line.37"></a> +<span class="sourceLineNo">038</span>import org.apache.hadoop.hbase.util.EnvironmentEdgeManager;<a name="line.38"></a> +<span class="sourceLineNo">039</span>import org.apache.hadoop.hbase.util.FSUtils;<a name="line.39"></a> +<span class="sourceLineNo">040</span>import org.apache.hadoop.hbase.wal.AbstractFSWALProvider;<a name="line.40"></a> +<span class="sourceLineNo">041</span>import org.apache.hadoop.hbase.wal.WALSplitter;<a name="line.41"></a> +<span class="sourceLineNo">042</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.42"></a> +<span class="sourceLineNo">043</span>import org.slf4j.Logger;<a name="line.43"></a> +<span class="sourceLineNo">044</span>import org.slf4j.LoggerFactory;<a name="line.44"></a> +<span class="sourceLineNo">045</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.45"></a> +<span class="sourceLineNo">046</span><a name="line.46"></a> +<span class="sourceLineNo">047</span>/**<a name="line.47"></a> +<span class="sourceLineNo">048</span> * This class abstracts a bunch of operations the HMaster needs<a name="line.48"></a> +<span class="sourceLineNo">049</span> * when splitting log files e.g. finding log files, dirs etc.<a name="line.49"></a> +<span class="sourceLineNo">050</span> */<a name="line.50"></a> +<span class="sourceLineNo">051</span>@InterfaceAudience.Private<a name="line.51"></a> +<span class="sourceLineNo">052</span>public class MasterWalManager {<a name="line.52"></a> +<span class="sourceLineNo">053</span> private static final Logger LOG = LoggerFactory.getLogger(MasterWalManager.class);<a name="line.53"></a> +<span class="sourceLineNo">054</span><a name="line.54"></a> +<span class="sourceLineNo">055</span> final static PathFilter META_FILTER = new PathFilter() {<a name="line.55"></a> +<span class="sourceLineNo">056</span> @Override<a name="line.56"></a> +<span class="sourceLineNo">057</span> public boolean accept(Path p) {<a name="line.57"></a> +<span class="sourceLineNo">058</span> return AbstractFSWALProvider.isMetaFile(p);<a name="line.58"></a> +<span class="sourceLineNo">059</span> }<a name="line.59"></a> +<span class="sourceLineNo">060</span> };<a name="line.60"></a> +<span class="sourceLineNo">061</span><a name="line.61"></a> +<span class="sourceLineNo">062</span> final static PathFilter NON_META_FILTER = new PathFilter() {<a name="line.62"></a> +<span class="sourceLineNo">063</span> @Override<a name="line.63"></a> +<span class="sourceLineNo">064</span> public boolean accept(Path p) {<a name="line.64"></a> +<span class="sourceLineNo">065</span> return !AbstractFSWALProvider.isMetaFile(p);<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 name="line.68"></a> +<span class="sourceLineNo">069</span> // metrics for master<a name="line.69"></a> +<span class="sourceLineNo">070</span> // TODO: Rename it, since those metrics are split-manager related<a name="line.70"></a> +<span class="sourceLineNo">071</span> private final MetricsMasterFileSystem metricsMasterFilesystem = new MetricsMasterFileSystem();<a name="line.71"></a> +<span class="sourceLineNo">072</span><a name="line.72"></a> +<span class="sourceLineNo">073</span> // Keep around for convenience.<a name="line.73"></a> +<span class="sourceLineNo">074</span> private final MasterServices services;<a name="line.74"></a> +<span class="sourceLineNo">075</span> private final Configuration conf;<a name="line.75"></a> +<span class="sourceLineNo">076</span> private final FileSystem fs;<a name="line.76"></a> +<span class="sourceLineNo">077</span><a name="line.77"></a> +<span class="sourceLineNo">078</span> // The Path to the old logs dir<a name="line.78"></a> +<span class="sourceLineNo">079</span> private final Path oldLogDir;<a name="line.79"></a> +<span class="sourceLineNo">080</span> private final Path rootDir;<a name="line.80"></a> +<span class="sourceLineNo">081</span><a name="line.81"></a> +<span class="sourceLineNo">082</span> // create the split log lock<a name="line.82"></a> +<span class="sourceLineNo">083</span> private final Lock splitLogLock = new ReentrantLock();<a name="line.83"></a> +<span class="sourceLineNo">084</span> private final SplitLogManager splitLogManager;<a name="line.84"></a> +<span class="sourceLineNo">085</span><a name="line.85"></a> +<span class="sourceLineNo">086</span> // Is the fileystem ok?<a name="line.86"></a> +<span class="sourceLineNo">087</span> private volatile boolean fsOk = true;<a name="line.87"></a> +<span class="sourceLineNo">088</span><a name="line.88"></a> +<span class="sourceLineNo">089</span> public MasterWalManager(MasterServices services) throws IOException {<a name="line.89"></a> +<span class="sourceLineNo">090</span> this(services.getConfiguration(), services.getMasterFileSystem().getWALFileSystem(),<a name="line.90"></a> +<span class="sourceLineNo">091</span> services.getMasterFileSystem().getWALRootDir(), services);<a name="line.91"></a> +<span class="sourceLineNo">092</span> }<a name="line.92"></a> +<span class="sourceLineNo">093</span><a name="line.93"></a> +<span class="sourceLineNo">094</span> public MasterWalManager(Configuration conf, FileSystem fs, Path rootDir, MasterServices services)<a name="line.94"></a> +<span class="sourceLineNo">095</span> throws IOException {<a name="line.95"></a> +<span class="sourceLineNo">096</span> this.fs = fs;<a name="line.96"></a> +<span class="sourceLineNo">097</span> this.conf = conf;<a name="line.97"></a> +<span class="sourceLineNo">098</span> this.rootDir = rootDir;<a name="line.98"></a> +<span class="sourceLineNo">099</span> this.services = services;<a name="line.99"></a> +<span class="sourceLineNo">100</span> this.splitLogManager = new SplitLogManager(services, conf);<a name="line.100"></a> +<span class="sourceLineNo">101</span><a name="line.101"></a> +<span class="sourceLineNo">102</span> this.oldLogDir = new Path(rootDir, HConstants.HREGION_OLDLOGDIR_NAME);<a name="line.102"></a> +<span class="sourceLineNo">103</span> }<a name="line.103"></a> +<span class="sourceLineNo">104</span><a name="line.104"></a> +<span class="sourceLineNo">105</span> public void stop() {<a name="line.105"></a> +<span class="sourceLineNo">106</span> if (splitLogManager != null) {<a name="line.106"></a> +<span class="sourceLineNo">107</span> splitLogManager.stop();<a name="line.107"></a> +<span class="sourceLineNo">108</span> }<a name="line.108"></a> +<span class="sourceLineNo">109</span> }<a name="line.109"></a> +<span class="sourceLineNo">110</span><a name="line.110"></a> +<span class="sourceLineNo">111</span> @VisibleForTesting<a name="line.111"></a> +<span class="sourceLineNo">112</span> SplitLogManager getSplitLogManager() {<a name="line.112"></a> +<span class="sourceLineNo">113</span> return this.splitLogManager;<a name="line.113"></a> +<span class="sourceLineNo">114</span> }<a name="line.114"></a> +<span class="sourceLineNo">115</span><a name="line.115"></a> +<span class="sourceLineNo">116</span> /**<a name="line.116"></a> +<span class="sourceLineNo">117</span> * Get the directory where old logs go<a name="line.117"></a> +<span class="sourceLineNo">118</span> * @return the dir<a name="line.118"></a> +<span class="sourceLineNo">119</span> */<a name="line.119"></a> +<span class="sourceLineNo">120</span> Path getOldLogDir() {<a name="line.120"></a> +<span class="sourceLineNo">121</span> return this.oldLogDir;<a name="line.121"></a> +<span class="sourceLineNo">122</span> }<a name="line.122"></a> +<span class="sourceLineNo">123</span><a name="line.123"></a> +<span class="sourceLineNo">124</span> public FileSystem getFileSystem() {<a name="line.124"></a> +<span class="sourceLineNo">125</span> return this.fs;<a name="line.125"></a> +<span class="sourceLineNo">126</span> }<a name="line.126"></a> +<span class="sourceLineNo">127</span><a name="line.127"></a> +<span class="sourceLineNo">128</span> /**<a name="line.128"></a> +<span class="sourceLineNo">129</span> * Checks to see if the file system is still accessible.<a name="line.129"></a> +<span class="sourceLineNo">130</span> * If not, sets closed<a name="line.130"></a> +<span class="sourceLineNo">131</span> * @return false if file system is not available<a name="line.131"></a> +<span class="sourceLineNo">132</span> */<a name="line.132"></a> +<span class="sourceLineNo">133</span> private boolean checkFileSystem() {<a name="line.133"></a> +<span class="sourceLineNo">134</span> if (this.fsOk) {<a name="line.134"></a> +<span class="sourceLineNo">135</span> try {<a name="line.135"></a> +<span class="sourceLineNo">136</span> FSUtils.checkFileSystemAvailable(this.fs);<a name="line.136"></a> +<span class="sourceLineNo">137</span> FSUtils.checkDfsSafeMode(this.conf);<a name="line.137"></a> +<span class="sourceLineNo">138</span> } catch (IOException e) {<a name="line.138"></a> +<span class="sourceLineNo">139</span> services.abort("Shutting down HBase cluster: file system not available", e);<a name="line.139"></a> +<span class="sourceLineNo">140</span> this.fsOk = false;<a name="line.140"></a> +<span class="sourceLineNo">141</span> }<a name="line.141"></a> +<span class="sourceLineNo">142</span> }<a name="line.142"></a> +<span class="sourceLineNo">143</span> return this.fsOk;<a name="line.143"></a> +<span class="sourceLineNo">144</span> }<a name="line.144"></a> +<span class="sourceLineNo">145</span><a name="line.145"></a> +<span class="sourceLineNo">146</span> public Set<ServerName> getLiveServersFromWALDir() throws IOException {<a name="line.146"></a> +<span class="sourceLineNo">147</span> Path walDirPath = new Path(rootDir, HConstants.HREGION_LOGDIR_NAME);<a name="line.147"></a> +<span class="sourceLineNo">148</span> FileStatus[] walDirForLiveServers = FSUtils.listStatus(fs, walDirPath,<a name="line.148"></a> +<span class="sourceLineNo">149</span> p -> !p.getName().endsWith(AbstractFSWALProvider.SPLITTING_EXT));<a name="line.149"></a> +<span class="sourceLineNo">150</span> if (walDirForLiveServers == null) {<a name="line.150"></a> +<span class="sourceLineNo">151</span> return Collections.emptySet();<a name="line.151"></a> +<span class="sourceLineNo">152</span> }<a name="line.152"></a> +<span class="sourceLineNo">153</span> return Stream.of(walDirForLiveServers).map(s -> {<a name="line.153"></a> +<span class="sourceLineNo">154</span> ServerName serverName = AbstractFSWALProvider.getServerNameFromWALDirectoryName(s.getPath());<a name="line.154"></a> +<span class="sourceLineNo">155</span> if (serverName == null) {<a name="line.155"></a> +<span class="sourceLineNo">156</span> LOG.warn("Log folder {} doesn't look like its name includes a " +<a name="line.156"></a> +<span class="sourceLineNo">157</span> "region server name; leaving in place. If you see later errors about missing " +<a name="line.157"></a> +<span class="sourceLineNo">158</span> "write ahead logs they may be saved in this location.", s.getPath());<a name="line.158"></a> +<span class="sourceLineNo">159</span> return null;<a name="line.159"></a> <span class="sourceLineNo">160</span> }<a name="line.160"></a> -<span class="sourceLineNo">161</span> try {<a name="line.161"></a> -<span class="sourceLineNo">162</span> if (!this.fs.exists(logsDirPath)) return serverNames;<a name="line.162"></a> -<span class="sourceLineNo">163</span> FileStatus[] logFolders = FSUtils.listStatus(this.fs, logsDirPath, null);<a name="line.163"></a> -<span class="sourceLineNo">164</span> // Get online servers after getting log folders to avoid log folder deletion of newly<a name="line.164"></a> -<span class="sourceLineNo">165</span> // checked in region servers . see HBASE-5916<a name="line.165"></a> -<span class="sourceLineNo">166</span> Set<ServerName> onlineServers = services.getServerManager().getOnlineServers().keySet();<a name="line.166"></a> -<span class="sourceLineNo">167</span><a name="line.167"></a> -<span class="sourceLineNo">168</span> if (logFolders == null || logFolders.length == 0) {<a name="line.168"></a> -<span class="sourceLineNo">169</span> LOG.debug("No log files to split, proceeding...");<a name="line.169"></a> -<span class="sourceLineNo">170</span> return serverNames;<a name="line.170"></a> -<span class="sourceLineNo">171</span> }<a name="line.171"></a> -<span class="sourceLineNo">172</span> for (FileStatus status : logFolders) {<a name="line.172"></a> -<span class="sourceLineNo">173</span> FileStatus[] curLogFiles = FSUtils.listStatus(this.fs, status.getPath(), null);<a name="line.173"></a> -<span class="sourceLineNo">174</span> if (curLogFiles == null || curLogFiles.length == 0) {<a name="line.174"></a> -<span class="sourceLineNo">175</span> // Empty log folder. No recovery needed<a name="line.175"></a> -<span class="sourceLineNo">176</span> continue;<a name="line.176"></a> -<span class="sourceLineNo">177</span> }<a name="line.177"></a> -<span class="sourceLineNo">178</span> final ServerName serverName = AbstractFSWALProvider.getServerNameFromWALDirectoryName(<a name="line.178"></a> -<span class="sourceLineNo">179</span> status.getPath());<a name="line.179"></a> -<span class="sourceLineNo">180</span> if (null == serverName) {<a name="line.180"></a> -<span class="sourceLineNo">181</span> LOG.warn("Log folder " + status.getPath() + " doesn't look like its name includes a " +<a name="line.181"></a> -<span class="sourceLineNo">182</span> "region server name; leaving in place. If you see later errors about missing " +<a name="line.182"></a> -<span class="sourceLineNo">183</span> "write ahead logs they may be saved in this location.");<a name="line.183"></a> -<span class="sourceLineNo">184</span> } else if (!onlineServers.contains(serverName)) {<a name="line.184"></a> -<span class="sourceLineNo">185</span> LOG.info("Log folder " + status.getPath() + " doesn't belong "<a name="line.185"></a> -<span class="sourceLineNo">186</span> + "to a known region server, splitting");<a name="line.186"></a> -<span class="sourceLineNo">187</span> serverNames.add(serverName);<a name="line.187"></a> -<span class="sourceLineNo">188</span> } else {<a name="line.188"></a> -<span class="sourceLineNo">189</span> LOG.info("Log folder " + status.getPath() + " belongs to an existing region server");<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> retrySplitting = false;<a name="line.192"></a> -<span class="sourceLineNo">193</span> } catch (IOException ioe) {<a name="line.193"></a> -<span class="sourceLineNo">194</span> LOG.warn("Failed getting failed servers to be recovered.", ioe);<a name="line.194"></a> -<span class="sourceLineNo">195</span> if (!checkFileSystem()) {<a name="line.195"></a> -<span class="sourceLineNo">196</span> LOG.warn("Bad Filesystem, exiting");<a name="line.196"></a> -<span class="sourceLineNo">197</span> Runtime.getRuntime().halt(1);<a name="line.197"></a> -<span class="sourceLineNo">198</span> }<a name="line.198"></a> -<span class="sourceLineNo">199</span> try {<a name="line.199"></a> -<span class="sourceLineNo">200</span> if (retrySplitting) {<a name="line.200"></a> -<span class="sourceLineNo">201</span> Thread.sleep(conf.getInt("hbase.hlog.split.failure.retry.interval", 30 * 1000));<a name="line.201"></a> +<span class="sourceLineNo">161</span> return serverName;<a name="line.161"></a> +<span class="sourceLineNo">162</span> }).filter(s -> s != null).collect(Collectors.toSet());<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> /**<a name="line.165"></a> +<span class="sourceLineNo">166</span> * Inspect the log directory to find dead servers which need recovery work<a name="line.166"></a> +<span class="sourceLineNo">167</span> * @return A set of ServerNames which aren't running but still have WAL files left in file system<a name="line.167"></a> +<span class="sourceLineNo">168</span> * @deprecated With proc-v2, we can record the crash server with procedure store, so do not need<a name="line.168"></a> +<span class="sourceLineNo">169</span> * to scan the wal directory to find out the splitting wal directory any more. Leave<a name="line.169"></a> +<span class="sourceLineNo">170</span> * it here only because {@code RecoverMetaProcedure}(which is also deprecated) uses<a name="line.170"></a> +<span class="sourceLineNo">171</span> * it.<a name="line.171"></a> +<span class="sourceLineNo">172</span> */<a name="line.172"></a> +<span class="sourceLineNo">173</span> @Deprecated<a name="line.173"></a> +<span class="sourceLineNo">174</span> public Set<ServerName> getFailedServersFromLogFolders() {<a name="line.174"></a> +<span class="sourceLineNo">175</span> boolean retrySplitting = !conf.getBoolean("hbase.hlog.split.skip.errors",<a name="line.175"></a> +<span class="sourceLineNo">176</span> WALSplitter.SPLIT_SKIP_ERRORS_DEFAULT);<a name="line.176"></a> +<span class="sourceLineNo">177</span><a name="line.177"></a> +<span class="sourceLineNo">178</span> Set<ServerName> serverNames = new HashSet<>();<a name="line.178"></a> +<span class="sourceLineNo">179</span> Path logsDirPath = new Path(this.rootDir, HConstants.HREGION_LOGDIR_NAME);<a name="line.179"></a> +<span class="sourceLineNo">180</span><a name="line.180"></a> +<span class="sourceLineNo">181</span> do {<a name="line.181"></a> +<span class="sourceLineNo">182</span> if (services.isStopped()) {<a name="line.182"></a> +<span class="sourceLineNo">183</span> LOG.warn("Master stopped while trying to get failed servers.");<a name="line.183"></a> +<span class="sourceLineNo">184</span> break;<a name="line.184"></a> +<span class="sourceLineNo">185</span> }<a name="line.185"></a> +<span class="sourceLineNo">186</span> try {<a name="line.186"></a> +<span class="sourceLineNo">187</span> if (!this.fs.exists(logsDirPath)) return serverNames;<a name="line.187"></a> +<span class="sourceLineNo">188</span> FileStatus[] logFolders = FSUtils.listStatus(this.fs, logsDirPath, null);<a name="line.188"></a> +<span class="sourceLineNo">189</span> // Get online servers after getting log folders to avoid log folder deletion of newly<a name="line.189"></a> +<span class="sourceLineNo">190</span> // checked in region servers . see HBASE-5916<a name="line.190"></a> +<span class="sourceLineNo">191</span> Set<ServerName> onlineServers = services.getServerManager().getOnlineServers().keySet();<a name="line.191"></a> +<span class="sourceLineNo">192</span><a name="line.192"></a> +<span class="sourceLineNo">193</span> if (logFolders == null || logFolders.length == 0) {<a name="line.193"></a> +<span class="sourceLineNo">194</span> LOG.debug("No log files to split, proceeding...");<a name="line.194"></a> +<span class="sourceLineNo">195</span> return serverNames;<a name="line.195"></a> +<span class="sourceLineNo">196</span> }<a name="line.196"></a> +<span class="sourceLineNo">197</span> for (FileStatus status : logFolders) {<a name="line.197"></a> +<span class="sourceLineNo">198</span> FileStatus[] curLogFiles = FSUtils.listStatus(this.fs, status.getPath(), null);<a name="line.198"></a> +<span class="sourceLineNo">199</span> if (curLogFiles == null || curLogFiles.length == 0) {<a name="line.199"></a> +<span class="sourceLineNo">200</span> // Empty log folder. No recovery needed<a name="line.200"></a> +<span class="sourceLineNo">201</span> continue;<a name="line.201"></a> <span class="sourceLineNo">202</span> }<a name="line.202"></a> -<span class="sourceLineNo">203</span> } catch (InterruptedException e) {<a name="line.203"></a> -<span class="sourceLineNo">204</span> LOG.warn("Interrupted, aborting since cannot return w/o splitting");<a name="line.204"></a> -<span class="sourceLineNo">205</span> Thread.currentThread().interrupt();<a name="line.205"></a> -<span class="sourceLineNo">206</span> retrySplitting = false;<a name="line.206"></a> -<span class="sourceLineNo">207</span> Runtime.getRuntime().halt(1);<a name="line.207"></a> -<span class="sourceLineNo">208</span> }<a name="line.208"></a> -<span class="sourceLineNo">209</span> }<a name="line.209"></a> -<span class="sourceLineNo">210</span> } while (retrySplitting);<a name="line.210"></a> -<span class="sourceLineNo">211</span><a name="line.211"></a> -<span class="sourceLineNo">212</span> return serverNames;<a name="line.212"></a> -<span class="sourceLineNo">213</span> }<a name="line.213"></a> -<span class="sourceLineNo">214</span><a name="line.214"></a> -<span class="sourceLineNo">215</span> public void splitLog(final ServerName serverName) throws IOException {<a name="line.215"></a> -<span class="sourceLineNo">216</span> splitLog(Collections.<ServerName>singleton(serverName));<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> /**<a name="line.219"></a> -<span class="sourceLineNo">220</span> * Specialized method to handle the splitting for meta WAL<a name="line.220"></a> -<span class="sourceLineNo">221</span> * @param serverName logs belonging to this server will be split<a name="line.221"></a> -<span class="sourceLineNo">222</span> */<a name="line.222"></a> -<span class="sourceLineNo">223</span> public void splitMetaLog(final ServerName serverName) throws IOException {<a name="line.223"></a> -<span class="sourceLineNo">224</span> splitMetaLog(Collections.<ServerName>singleton(serverName));<a name="line.224"></a> -<span class="sourceLineNo">225</span> }<a name="line.225"></a> -<span class="sourceLineNo">226</span><a name="line.226"></a> -<span class="sourceLineNo">227</span> /**<a name="line.227"></a> -<span class="sourceLineNo">228</span> * Specialized method to handle the splitting for meta WAL<a name="line.228"></a> -<span class="sourceLineNo">229</span> * @param serverNames logs belonging to these servers will be split<a name="line.229"></a> -<span class="sourceLineNo">230</span> */<a name="line.230"></a> -<span class="sourceLineNo">231</span> public void splitMetaLog(final Set<ServerName> serverNames) throws IOException {<a name="line.231"></a> -<span class="sourceLineNo">232</span> splitLog(serverNames, META_FILTER);<a name="line.232"></a> -<span class="sourceLineNo">233</span> }<a name="line.233"></a> -<span class="sourceLineNo">234</span><a name="line.234"></a> -<span class="sourceLineNo">235</span> @edu.umd.cs.findbugs.annotations.SuppressWarnings(value="UL_UNRELEASED_LOCK", justification=<a name="line.235"></a> -<span class="sourceLineNo">236</span> "We only release this lock when we set it. Updates to code that uses it should verify use " +<a name="line.236"></a> -<span class="sourceLineNo">237</span> "of the guard boolean.")<a name="line.237"></a> -<span class="sourceLineNo">238</span> private List<Path> getLogDirs(final Set<ServerName> serverNames) throws IOException {<a name="line.238"></a> -<span class="sourceLineNo">239</span> List<Path> logDirs = new ArrayList<>();<a name="line.239"></a> -<span class="sourceLineNo">240</span> boolean needReleaseLock = false;<a name="line.240"></a> -<span class="sourceLineNo">241</span> if (!this.services.isInitialized()) {<a name="line.241"></a> -<span class="sourceLineNo">242</span> // during master initialization, we could have multiple places splitting a same wal<a name="line.242"></a> -<span class="sourceLineNo">243</span> this.splitLogLock.lock();<a name="line.243"></a> -<span class="sourceLineNo">244</span> needReleaseLock = true;<a name="line.244"></a> -<span class="sourceLineNo">245</span> }<a name="line.245"></a> -<span class="sourceLineNo">246</span> try {<a name="line.246"></a> -<span class="sourceLineNo">247</span> for (ServerName serverName : serverNames) {<a name="line.247"></a> -<span class="sourceLineNo">248</span> Path logDir = new Path(this.rootDir,<a name="line.248"></a> -<span class="sourceLineNo">249</span> AbstractFSWALProvider.getWALDirectoryName(serverName.toString()));<a name="line.249"></a> -<span class="sourceLineNo">250</span> Path splitDir = logDir.suffix(AbstractFSWALProvider.SPLITTING_EXT);<a name="line.250"></a> -<span class="sourceLineNo">251</span> // Rename the directory so a rogue RS doesn't create more WALs<a name="line.251"></a> -<span class="sourceLineNo">252</span> if (fs.exists(logDir)) {<a name="line.252"></a> -<span class="sourceLineNo">253</span> if (!this.fs.rename(logDir, splitDir)) {<a name="line.253"></a> -<span class="sourceLineNo">254</span> throw new IOException("Failed fs.rename for log split: " + logDir);<a name="line.254"></a> -<span class="sourceLineNo">255</span> }<a name="line.255"></a> -<span class="sourceLineNo">256</span> logDir = splitDir;<a name="line.256"></a> -<span class="sourceLineNo">257</span> LOG.debug("Renamed region directory: " + splitDir);<a name="line.257"></a> -<span class="sourceLineNo">258</span> } else if (!fs.exists(splitDir)) {<a name="line.258"></a> -<span class="sourceLineNo">259</span> LOG.info("Log dir for server " + serverName + " does not exist");<a name="line.259"></a> -<span class="sourceLineNo">260</span> continue;<a name="line.260"></a> -<span class="sourceLineNo">261</span> }<a name="line.261"></a> -<span class="sourceLineNo">262</span> logDirs.add(splitDir);<a name="line.262"></a> -<span class="sourceLineNo">263</span> }<a name="line.263"></a> -<span class="sourceLineNo">264</span> } catch (IOException ioe) {<a name="line.264"></a> -<span class="sourceLineNo">265</span> if (!checkFileSystem()) {<a name="line.265"></a> -<span class="sourceLineNo">266</span> this.services.abort("Aborting due to filesystem unavailable", ioe);<a name="line.266"></a> -<span class="sourceLineNo">267</span> throw ioe;<a name="line.267"></a> -<span class="sourceLineNo">268</span> }<a name="line.268"></a> -<span class="sourceLineNo">269</span> } finally {<a name="line.269"></a> -<span class="sourceLineNo">270</span> if (needReleaseLock) {<a name="line.270"></a> -<span class="sourceLineNo">271</span> this.splitLogLock.unlock();<a name="line.271"></a> -<span class="sourceLineNo">272</span> }<a name="line.272"></a> -<span class="sourceLineNo">273</span> }<a name="line.273"></a> -<span class="sourceLineNo">274</span> return logDirs;<a name="line.274"></a> -<span class="sourceLineNo">275</span> }<a name="line.275"></a> -<span class="sourceLineNo">276</span><a name="line.276"></a> -<span class="sourceLineNo">277</span> public void splitLog(final Set<ServerName> serverNames) throws IOException {<a name="line.277"></a> -<span class="sourceLineNo">278</span> splitLog(serverNames, NON_META_FILTER);<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> /**<a name="line.281"></a> -<span class="sourceLineNo">282</span> * This method is the base split method that splits WAL files matching a filter. Callers should<a name="line.282"></a> -<span class="sourceLineNo">283</span> * pass the appropriate filter for meta and non-meta WALs.<a name="line.283"></a> -<span class="sourceLineNo">284</span> * @param serverNames logs belonging to these servers will be split; this will rename the log<a name="line.284"></a> -<span class="sourceLineNo">285</span> * directory out from under a soft-failed server<a name="line.285"></a> -<span class="sourceLineNo">286</span> */<a name="line.286"></a> -<span class="sourceLineNo">287</span> public void splitLog(final Set<ServerName> serverNames, PathFilter filter) throws IOException {<a name="line.287"></a> -<span class="sourceLineNo">288</span> long splitTime = 0, splitLogSize = 0;<a name="line.288"></a> -<span class="sourceLineNo">289</span> List<Path> logDirs = getLogDirs(serverNames);<a name="line.289"></a> -<span class="sourceLineNo">290</span><a name="line.290"></a> -<span class="sourceLineNo">291</span> splitLogManager.handleDeadWorkers(serverNames);<a name="line.291"></a> -<span class="sourceLineNo">292</span> splitTime = EnvironmentEdgeManager.currentTime();<a name="line.292"></a> -<span class="sourceLineNo">293</span> splitLogSize = splitLogManager.splitLogDistributed(serverNames, logDirs, filter);<a name="line.293"></a> -<span class="sourceLineNo">294</span> splitTime = EnvironmentEdgeManager.currentTime() - splitTime;<a name="line.294"></a> -<span class="sourceLineNo">295</span><a name="line.295"></a> -<span class="sourceLineNo">296</span> if (this.metricsMasterFilesystem != null) {<a name="line.296"></a> -<span class="sourceLineNo">297</span> if (filter == META_FILTER) {<a name="line.297"></a> -<span class="sourceLineNo">298</span> this.metricsMasterFilesystem.addMetaWALSplit(splitTime, splitLogSize);<a name="line.298"></a> -<span class="sourceLineNo">299</span> } else {<a name="line.299"></a> -<span class="sourceLineNo">300</span> this.metricsMasterFilesystem.addSplit(splitTime, splitLogSize);<a name="line.300"></a> -<span class="sourceLineNo">301</span> }<a name="line.301"></a> -<span class="sourceLineNo">302</span> }<a name="line.302"></a> -<span class="sourceLineNo">303</span> }<a name="line.303"></a> -<span class="sourceLineNo">304</span>}<a name="line.304"></a> +<span class="sourceLineNo">203</span> final ServerName serverName = AbstractFSWALProvider.getServerNameFromWALDirectoryName(<a name="line.203"></a> +<span class="sourceLineNo">204</span> status.getPath());<a name="line.204"></a> +<span class="sourceLineNo">205</span> if (null == serverName) {<a name="line.205"></a> +<span class="sourceLineNo">206</span> LOG.warn("Log folder " + status.getPath() + " doesn't look like its name includes a " +<a name="line.206"></a> +<span class="sourceLineNo">207</span> "region server name; leaving in place. If you see later errors about missing " +<a name="line.207"></a> +<span class="sourceLineNo">208</span> "write ahead logs they may be saved in this location.");<a name="line.208"></a> +<span class="sourceLineNo">209</span> } else if (!onlineServers.contains(serverName)) {<a name="line.209"></a> +<span class="sourceLineNo">210</span> LOG.info("Log folder " + status.getPath() + " doesn't belong "<a name="line.210"></a> +<span class="sourceLineNo">211</span> + "to a known region server, splitting");<a name="line.211"></a> +<span class="sourceLineNo">212</span> serverNames.add(serverName);<a name="line.212"></a> +<span class="sourceLineNo">213</span> } else {<a name="line.213"></a> +<span class="sourceLineNo">214</span> LOG.info("Log folder " + status.getPath() + " belongs to an existing region server");<a name="line.214"></a> +<span class="sourceLineNo">215</span> }<a name="line.215"></a> +<span class="sourceLineNo">216</span> }<a name="line.216"></a> +<span class="sourceLineNo">217</span> retrySplitting = false;<a name="line.217"></a> +<span class="sourceLineNo">218</span> } catch (IOException ioe) {<a name="line.218"></a> +<span class="sourceLineNo">219</span> LOG.warn("Failed getting failed servers to be recovered.", ioe);<a name="line.219"></a> +<span class="sourceLineNo">220</span> if (!checkFileSystem()) {<a name="line.220"></a> +<span class="sourceLineNo">221</span> LOG.warn("Bad Filesystem, exiting");<a name="line.221"></a> +<span class="sourceLineNo">222</span> Runtime.getRuntime().halt(1);<a name="line.222"></a> +<span class="sourceLineNo">223</span> }<a name="line.223"></a> +<span class="sourceLineNo">224</span> try {<a name="line.224"></a> +<span class="sourceLineNo">225</span> if (retrySplitting) {<a name="line.225"></a> +<span class="sourceLineNo">226</span> Thread.sleep(conf.getInt("hbase.hlog.split.failure.retry.interval", 30 * 1000));<a name="line.226"></a> +<span class="sourceLineNo">227</span> }<a name="line.227"></a> +<span class="sourceLineNo">228</span> } catch (InterruptedException e) {<a name="line.228"></a> +<span class="sourceLineNo">229</span> LOG.warn("Interrupted, aborting since cannot return w/o splitting");<a name="line.229"></a> +<span class="sourceLineNo">230</span> Thread.currentThread().interrupt();<a name="line.230"></a> +<span class="sourceLineNo">231</span> retrySplitting = false;<a name="line.231"></a> +<span class="sourceLineNo">232</span> Runtime.getRuntime().halt(1);<a name="line.232"></a> +<span class="sourceLineNo">233</span> }<a name="line.233"></a> +<span class="sourceLineNo">234</span> }<a name="line.234"></a> +<span class="sourceLineNo">235</span> } while (retrySplitting);<a name="line.235"></a> +<span class="sourceLineNo">236</span><a name="line.236"></a> +<span class="sourceLineNo">237</span> return serverNames;<a name="line.237"></a> +<span class="sourceLineNo">238</span> }<a name="line.238"></a> +<span class="sourceLineNo">239</span><a name="line.239"></a> +<span class="sourceLineNo">240</span> public void splitLog(final ServerName serverName) throws IOException {<a name="line.240"></a> +<span class="sourceLineNo">241</span> splitLog(Collections.<ServerName>singleton(serverName));<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> * Specialized method to handle the splitting for meta WAL<a name="line.245"></a> +<span class="sourceLineNo">246</span> * @param serverName logs belonging to this server will be split<a name="line.246"></a> +<span class="sourceLineNo">247</span> */<a name="line.247"></a> +<span class="sourceLineNo">248</span> public void splitMetaLog(final ServerName serverName) throws IOException {<a name="line.248"></a> +<span class="sourceLineNo">249</span> splitMetaLog(Collections.<ServerName>singleton(serverName));<a name="line.249"></a> +<span class="sourceLineNo">250</span> }<a name="line.250"></a> +<span class="sourceLineNo">251</span><a name="line.251"></a> +<span class="sourceLineNo">252</span> /**<a name="line.252"></a> +<span class="sourceLineNo">253</span> * Specialized method to handle the splitting for meta WAL<a name="line.253"></a> +<span class="sourceLineNo">254</span> * @param serverNames logs belonging to these servers will be split<a name="line.254"></a> +<span class="sourceLineNo">255</span> */<a name="line.255"></a> +<span class="sourceLineNo">256</span> public void splitMetaLog(final Set<ServerName> serverNames) throws IOException {<a name="line.256"></a> +<span class="sourceLineNo">257</span> splitLog(serverNames, META_FILTER);<a name="line.257"></a> +<span class="sourceLineNo">258</span> }<a name="line.258"></a> +<span class="sourceLineNo">259</span><a name="line.259"></a> +<span class="sourceLineNo">260</span> @edu.umd.cs.findbugs.annotations.SuppressWarnings(value="UL_UNRELEASED_LOCK", justification=<a name="line.260"></a> +<span class="sourceLineNo">261</span> "We only release this lock when we set it. Updates to code that uses it should verify use " +<a name="line.261"></a> +<span class="sourceLineNo">262</span> "of the guard boolean.")<a name="line.262"></a> +<span class="sourceLineNo">263</span> private List<Path> getLogDirs(final Set<ServerName> serverNames) throws IOException {<a name="line.263"></a> +<span class="sourceLineNo">264</span> List<Path> logDirs = new ArrayList<>();<a name="line.264"></a> +<span class="sourceLineNo">265</span> boolean needReleaseLock = false;<a name="line.265"></a> +<span class="sourceLineNo">266</span> if (!this.services.isInitialized()) {<a name="line.266"></a> +<span class="sourceLineNo">267</span> // during master initialization, we could have multiple places splitting a same wal<a name="line.267"></a> +<span class="sourceLineNo">268</span> // XXX: Does this still exist after we move to proc-v2?<a name="line.268"></a> +<span class="sourceLineNo">269</span> this.splitLogLock.lock();<a name="line.269"></a> +<span class="sourceLineNo">270</span> needReleaseLock = true;<a name="line.270"></a> +<span class="sourceLineNo">271</span> }<a name="line.271"></a> +<span class="sourceLineNo">272</span> try {<a name="line.272"></a> +<span class="sourceLineNo">273</span> for (ServerName serverName : serverNames) {<a name="line.273"></a> +<span class="sourceLineNo">274</span> Path logDir = new Path(this.rootDir,<a name="line.274"></a> +<span class="sourceLineNo">275</span> AbstractFSWALProvider.getWALDirectoryName(serverName.toString()));<a name="line.275"></a> +<span class="sourceLineNo">276</span> Path splitDir = logDir.suffix(AbstractFSWALProvider.SPLITTING_EXT);<a name="line.276"></a> +<span class="sourceLineNo">277</span> // Rename the directory so a rogue RS doesn't create more WALs<a name="line.277"></a> +<span class="sourceLineNo">278</span> if (fs.exists(logDir)) {<a name="line.278"></a> +<span class="sourceLineNo">279</span> if (!this.fs.rename(logDir, splitDir)) {<a name="line.279"></a> +<span class="sourceLineNo">280</span> throw new IOException("Failed fs.rename for log split: " + logDir);<a name="line.280"></a> +<span class="sourceLineNo">281</span> }<a name="line.281"></a> +<span class="sourceLineNo">282</span> logDir = splitDir;<a name="line.282"></a> +<span class="sourceLineNo">283</span> LOG.debug("Renamed region directory: " + splitDir);<a name="line.283"></a> +<span class="sourceLineNo">284</span> } else if (!fs.exists(splitDir)) {<a name="line.284"></a> +<span class="sourceLineNo">285</span> LOG.info("Log dir for server " + serverName + " does not exist");<a name="line.285"></a> +<span class="sourceLineNo">286</span> continue;<a name="line.286"></a> +<span class="sourceLineNo">287</span> }<a name="line.287"></a> +<span class="sourceLineNo">288</span> logDirs.add(splitDir);<a name="line.288"></a> +<span class="sourceLineNo">289</span> }<a name="line.289"></a> +<span class="sourceLineNo">290</span> } catch (IOException ioe) {<a name="line.290"></a> +<span class="sourceLineNo">291</span> if (!checkFileSystem()) {<a name="line.291"></a> +<span class="sourceLineNo">292</span> this.services.abort("Aborting due to filesystem unavailable", ioe);<a name="line.292"></a> +<span class="sourceLineNo">293</span> throw ioe;<a name="line.293"></a> +<span class="sourceLineNo">294</span> }<a name="line.294"></a> +<span class="sourceLineNo">295</span> } finally {<a name="line.295"></a> +<span class="sourceLineNo">296</span> if (needReleaseLock) {<a name="line.296"></a> +<span class="sourceLineNo">297</span> this.splitLogLock.unlock();<a name="line.297"></a> +<span class="sourceLineNo">298</span> }<a name="line.298"></a> +<span class="sourceLineNo">299</span> }<a name="line.299"></a> +<span class="sourceLineNo">300</span> return logDirs;<a name="line.300"></a> +<span class="sourceLineNo">301</span> }<a name="line.301"></a> +<span class="sourceLineNo">302</span><a name="line.302"></a> +<span class="sourceLineNo">303</span> public void splitLog(final Set<ServerName> serverNames) throws IOException {<a name="line.303"></a> +<span class="sourceLineNo">304</span> splitLog(serverNames, NON_META_FILTER);<a name="line.304"></a> +<span class="sourceLineNo">305</span> }<a name="line.305"></a> +<span class="sourceLineNo">306</span><a name="line.306"></a> +<span class="sourceLineNo">307</span> /**<a name="line.307"></a> +<span class="sourceLineNo">308</span> * This method is the base split method that splits WAL files matching a filter. Callers should<a name="line.308"></a> +<span class="sourceLineNo">309</span> * pass the appropriate filter for meta and non-meta WALs.<a name="line.309"></a> +<span class="sourceLineNo">310</span> * @param serverNames logs belonging to these servers will be split; this will rename the log<a name="line.310"></a> +<span class="sourceLineNo">311</span> * directory out from under a soft-failed server<a name="line.311"></a> +<span class="sourceLineNo">312</span> */<a name="line.312"></a> +<span class="sourceLineNo">313</span> public void splitLog(final Set<ServerName> serverNames, PathFilter filter) throws IOException {<a name="line.313"></a> +<span class="sourceLineNo">314</span> long splitTime = 0, splitLogSize = 0;<a name="line.314"></a> +<span class="sourceLineNo">315</span> List<Path> logDirs = getLogDirs(serverNames);<a name="line.315"></a> +<span class="sourceLineNo">316</span><a name="line.316"></a> +<span class="sourceLineNo">317</span> splitLogManager.handleDeadWorkers(serverNames);<a name="line.317"></a> +<span class="sourceLineNo">318</span> splitTime = EnvironmentEdgeManager.currentTime();<a name="line.318"></a> +<span class="sourceLineNo">319</span> splitLogSize = splitLogManager.splitLogDistributed(serverNames, logDirs, filter);<a name="line.319"></a> +<span class="sourceLineNo">320</span> splitTime = EnvironmentEdgeManager.currentTime() - splitTime;<a name="line.320"></a> +<span class="sourceLineNo">321</span><a name="line.321"></a> +<span class="sourceLineNo">322</span> if (this.metricsMasterFilesystem != null) {<a name="line.322"></a> +<span class="sourceLineNo">323</span> if (filter == META_FILTER) {<a name="line.323"></a> +<span class="sourceLineNo">324</span> this.metricsMasterFilesystem.addMetaWALSplit(splitTime, splitLogSize);<a name="line.324"></a> +<span class="sourceLineNo">325</span> } else {<a name="line.325"></a> +<span class="sourceLineNo">326</span> this.metricsMasterFilesystem.addSplit(splitTime, splitLogSize);<a name="line.326"></a> +<span class="sourceLineNo">327</span> }<a name="line.327"></a> +<span class="sourceLineNo">328</span> }<a name="line.328"></a> +<span class="sourceLineNo">329</span> }<a name="line.329"></a> +<span class="sourceLineNo">330</span>}<a name="line.330"></a>
http://git-wip-us.apache.org/repos/asf/hbase-site/blob/65565d77/devapidocs/src-html/org/apache/hadoop/hbase/master/RegionServerTracker.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/master/RegionServerTracker.html b/devapidocs/src-html/org/apache/hadoop/hbase/master/RegionServerTracker.html index cc0dba8..06b90f3 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/master/RegionServerTracker.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/master/RegionServerTracker.html @@ -54,143 +54,145 @@ <span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.HBaseProtos.RegionServerInfo;<a name="line.46"></a> <span class="sourceLineNo">047</span><a name="line.47"></a> <span class="sourceLineNo">048</span>/**<a name="line.48"></a> -<span class="sourceLineNo">049</span> * <p><a name="line.49"></a> -<span class="sourceLineNo">050</span> * Tracks the online region servers via ZK.<a name="line.50"></a> -<span class="sourceLineNo">051</span> * </p><a name="line.51"></a> -<span class="sourceLineNo">052</span> * <p><a name="line.52"></a> -<span class="sourceLineNo">053</span> * Handling of new RSs checking in is done via RPC. This class is only responsible for watching for<a name="line.53"></a> -<span class="sourceLineNo">054</span> * expired nodes. It handles listening for changes in the RS node list. The only exception is when<a name="line.54"></a> -<span class="sourceLineNo">055</span> * master restart, we will use the list fetched from zk to construct the initial set of live region<a name="line.55"></a> -<span class="sourceLineNo">056</span> * servers.<a name="line.56"></a> -<span class="sourceLineNo">057</span> * </p><a name="line.57"></a> -<span class="sourceLineNo">058</span> * <p><a name="line.58"></a> -<span class="sourceLineNo">059</span> * If an RS node gets deleted, this automatically handles calling of<a name="line.59"></a> -<span class="sourceLineNo">060</span> * {@link ServerManager#expireServer(ServerName)}<a name="line.60"></a> -<span class="sourceLineNo">061</span> * </p><a name="line.61"></a> -<span class="sourceLineNo">062</span> */<a name="line.62"></a> -<span class="sourceLineNo">063</span>@InterfaceAudience.Private<a name="line.63"></a> -<span class="sourceLineNo">064</span>public class RegionServerTracker extends ZKListener {<a name="line.64"></a> -<span class="sourceLineNo">065</span> private static final Logger LOG = LoggerFactory.getLogger(RegionServerTracker.class);<a name="line.65"></a> -<span class="sourceLineNo">066</span> private final Set<ServerName> regionServers = new HashSet<>();<a name="line.66"></a> -<span class="sourceLineNo">067</span> private final ServerManager serverManager;<a name="line.67"></a> -<span class="sourceLineNo">068</span> private final MasterServices server;<a name="line.68"></a> -<span class="sourceLineNo">069</span> // As we need to send request to zk when processing the nodeChildrenChanged event, we'd better<a name="line.69"></a> -<span class="sourceLineNo">070</span> // move the operation to a single threaded thread pool in order to not block the zk event<a name="line.70"></a> -<span class="sourceLineNo">071</span> // processing since all the zk listener across HMaster will be called in one thread sequentially.<a name="line.71"></a> -<span class="sourceLineNo">072</span> private final ExecutorService executor;<a name="line.72"></a> -<span class="sourceLineNo">073</span><a name="line.73"></a> -<span class="sourceLineNo">074</span> public RegionServerTracker(ZKWatcher watcher, MasterServices server,<a name="line.74"></a> -<span class="sourceLineNo">075</span> ServerManager serverManager) {<a name="line.75"></a> -<span class="sourceLineNo">076</span> super(watcher);<a name="line.76"></a> -<span class="sourceLineNo">077</span> this.server = server;<a name="line.77"></a> -<span class="sourceLineNo">078</span> this.serverManager = serverManager;<a name="line.78"></a> -<span class="sourceLineNo">079</span> executor = Executors.newSingleThreadExecutor(<a name="line.79"></a> -<span class="sourceLineNo">080</span> new ThreadFactoryBuilder().setDaemon(true).setNameFormat("RegionServerTracker-%d").build());<a name="line.80"></a> -<span class="sourceLineNo">081</span> }<a name="line.81"></a> -<span class="sourceLineNo">082</span><a name="line.82"></a> -<span class="sourceLineNo">083</span> private Pair<ServerName, RegionServerInfo> getServerInfo(String name)<a name="line.83"></a> -<span class="sourceLineNo">084</span> throws KeeperException, IOException {<a name="line.84"></a> -<span class="sourceLineNo">085</span> ServerName serverName = ServerName.parseServerName(name);<a name="line.85"></a> -<span class="sourceLineNo">086</span> String nodePath = ZNodePaths.joinZNode(watcher.getZNodePaths().rsZNode, name);<a name="line.86"></a> -<span class="sourceLineNo">087</span> byte[] data;<a name="line.87"></a> -<span class="sourceLineNo">088</span> try {<a name="line.88"></a> -<span class="sourceLineNo">089</span> data = ZKUtil.getData(watcher, nodePath);<a name="line.89"></a> -<span class="sourceLineNo">090</span> } catch (InterruptedException e) {<a name="line.90"></a> -<span class="sourceLineNo">091</span> throw (InterruptedIOException) new InterruptedIOException().initCause(e);<a name="line.91"></a> -<span class="sourceLineNo">092</span> }<a name="line.92"></a> -<span class="sourceLineNo">093</span> if (data == null) {<a name="line.93"></a> -<span class="sourceLineNo">094</span> // we should receive a children changed event later and then we will expire it, so we still<a name="line.94"></a> -<span class="sourceLineNo">095</span> // need to add it to the region server set.<a name="line.95"></a> -<span class="sourceLineNo">096</span> LOG.warn("Server node {} does not exist, already dead?", name);<a name="line.96"></a> -<span class="sourceLineNo">097</span> return Pair.newPair(serverName, null);<a name="line.97"></a> -<span class="sourceLineNo">098</span> }<a name="line.98"></a> -<span class="sourceLineNo">099</span> if (data.length == 0 || !ProtobufUtil.isPBMagicPrefix(data)) {<a name="line.99"></a> -<span class="sourceLineNo">100</span> // this should not happen actually, unless we have bugs or someone has messed zk up.<a name="line.100"></a> -<span class="sourceLineNo">101</span> LOG.warn("Invalid data for region server node {} on zookeeper, data length = {}", name,<a name="line.101"></a> -<span class="sourceLineNo">102</span> data.length);<a name="line.102"></a> -<span class="sourceLineNo">103</span> return Pair.newPair(serverName, null);<a name="line.103"></a> -<span class="sourceLineNo">104</span> }<a name="line.104"></a> -<span class="sourceLineNo">105</span> RegionServerInfo.Builder builder = RegionServerInfo.newBuilder();<a name="line.105"></a> -<span class="sourceLineNo">106</span> int magicLen = ProtobufUtil.lengthOfPBMagic();<a name="line.106"></a> -<span class="sourceLineNo">107</span> ProtobufUtil.mergeFrom(builder, data, magicLen, data.length - magicLen);<a name="line.107"></a> -<span class="sourceLineNo">108</span> return Pair.newPair(serverName, builder.build());<a name="line.108"></a> -<span class="sourceLineNo">109</span> }<a name="line.109"></a> -<span class="sourceLineNo">110</span><a name="line.110"></a> -<span class="sourceLineNo">111</span> /**<a name="line.111"></a> -<span class="sourceLineNo">112</span> * <p><a name="line.112"></a> -<span class="sourceLineNo">113</span> * Starts the tracking of online RegionServers.<a name="line.113"></a> -<span class="sourceLineNo">114</span> * </p><a name="line.114"></a> -<span class="sourceLineNo">115</span> * <p><a name="line.115"></a> -<span class="sourceLineNo">116</span> * All RSs will be tracked after this method is called.<a name="line.116"></a> -<span class="sourceLineNo">117</span> * </p><a name="line.117"></a> +<span class="sourceLineNo">049</span> * Tracks the online region servers via ZK.<a name="line.49"></a> +<span class="sourceLineNo">050</span> * <p/><a name="line.50"></a> +<span class="sourceLineNo">051</span> * Handling of new RSs checking in is done via RPC. This class is only responsible for watching for<a name="line.51"></a> +<span class="sourceLineNo">052</span> * expired nodes. It handles listening for changes in the RS node list. The only exception is when<a name="line.52"></a> +<span class="sourceLineNo">053</span> * master restart, we will use the list fetched from zk to construct the initial set of live region<a name="line.53"></a> +<span class="sourceLineNo">054</span> * servers.<a name="line.54"></a> +<span class="sourceLineNo">055</span> * <p/><a name="line.55"></a> +<span class="sourceLineNo">056</span> * If an RS node gets deleted, this automatically handles calling of<a name="line.56"></a> +<span class="sourceLineNo">057</span> * {@link ServerManager#expireServer(ServerName)}<a name="line.57"></a> +<span class="sourceLineNo">058</span> */<a name="line.58"></a> +<span class="sourceLineNo">059</span>@InterfaceAudience.Private<a name="line.59"></a> +<span class="sourceLineNo">060</span>public class RegionServerTracker extends ZKListener {<a name="line.60"></a> +<span class="sourceLineNo">061</span> private static final Logger LOG = LoggerFactory.getLogger(RegionServerTracker.class);<a name="line.61"></a> +<span class="sourceLineNo">062</span> private final Set<ServerName> regionServers = new HashSet<>();<a name="line.62"></a> +<span class="sourceLineNo">063</span> private final ServerManager serverManager;<a name="line.63"></a> +<span class="sourceLineNo">064</span> private final MasterServices server;<a name="line.64"></a> +<span class="sourceLineNo">065</span> // As we need to send request to zk when processing the nodeChildrenChanged event, we'd better<a name="line.65"></a> +<span class="sourceLineNo">066</span> // move the operation to a single threaded thread pool in order to not block the zk event<a name="line.66"></a> +<span class="sourceLineNo">067</span> // processing since all the zk listener across HMaster will be called in one thread sequentially.<a name="line.67"></a> +<span class="sourceLineNo">068</span> private final ExecutorService executor;<a name="line.68"></a> +<span class="sourceLineNo">069</span><a name="line.69"></a> +<span class="sourceLineNo">070</span> public RegionServerTracker(ZKWatcher watcher, MasterServices server,<a name="line.70"></a> +<span class="sourceLineNo">071</span> ServerManager serverManager) {<a name="line.71"></a> +<span class="sourceLineNo">072</span> super(watcher);<a name="line.72"></a> +<span class="sourceLineNo">073</span> this.server = server;<a name="line.73"></a> +<span class="sourceLineNo">074</span> this.serverManager = serverManager;<a name="line.74"></a> +<span class="sourceLineNo">075</span> this.executor = Executors.newSingleThreadExecutor(<a name="line.75"></a> +<span class="sourceLineNo">076</span> new ThreadFactoryBuilder().setDaemon(true).setNameFormat("RegionServerTracker-%d").build());<a name="line.76"></a> +<span class="sourceLineNo">077</span> }<a name="line.77"></a> +<span class="sourceLineNo">078</span><a name="line.78"></a> +<span class="sourceLineNo">079</span> private Pair<ServerName, RegionServerInfo> getServerInfo(String name)<a name="line.79"></a> +<span class="sourceLineNo">080</span> throws KeeperException, IOException {<a name="line.80"></a> +<span class="sourceLineNo">081</span> ServerName serverName = ServerName.parseServerName(name);<a name="line.81"></a> +<span class="sourceLineNo">082</span> String nodePath = ZNodePaths.joinZNode(watcher.getZNodePaths().rsZNode, name);<a name="line.82"></a> +<span class="sourceLineNo">083</span> byte[] data;<a name="line.83"></a> +<span class="sourceLineNo">084</span> try {<a name="line.84"></a> +<span class="sourceLineNo">085</span> data = ZKUtil.getData(watcher, nodePath);<a name="line.85"></a> +<span class="sourceLineNo">086</span> } catch (InterruptedException e) {<a name="line.86"></a> +<span class="sourceLineNo">087</span> throw (InterruptedIOException) new InterruptedIOException().initCause(e);<a name="line.87"></a> +<span class="sourceLineNo">088</span> }<a name="line.88"></a> +<span class="sourceLineNo">089</span> if (data == null) {<a name="line.89"></a> +<span class="sourceLineNo">090</span> // we should receive a children changed event later and then we will expire it, so we still<a name="line.90"></a> +<span class="sourceLineNo">091</span> // need to add it to the region server set.<a name="line.91"></a> +<span class="sourceLineNo">092</span> LOG.warn("Server node {} does not exist, already dead?", name);<a name="line.92"></a> +<span class="sourceLineNo">093</span> return Pair.newPair(serverName, null);<a name="line.93"></a> +<span class="sourceLineNo">094</span> }<a name="line.94"></a> +<span class="sourceLineNo">095</span> if (data.length == 0 || !ProtobufUtil.isPBMagicPrefix(data)) {<a name="line.95"></a> +<span class="sourceLineNo">096</span> // this should not happen actually, unless we have bugs or someone has messed zk up.<a name="line.96"></a> +<span class="sourceLineNo">097</span> LOG.warn("Invalid data for region server node {} on zookeeper, data length = {}", name,<a name="line.97"></a> +<span class="sourceLineNo">098</span> data.length);<a name="line.98"></a> +<span class="sourceLineNo">099</span> return Pair.newPair(serverName, null);<a name="line.99"></a> +<span class="sourceLineNo">100</span> }<a name="line.100"></a> +<span class="sourceLineNo">101</span> RegionServerInfo.Builder builder = RegionServerInfo.newBuilder();<a name="line.101"></a> +<span class="sourceLineNo">102</span> int magicLen = ProtobufUtil.lengthOfPBMagic();<a name="line.102"></a> +<span class="sourceLineNo">103</span> ProtobufUtil.mergeFrom(builder, data, magicLen, data.length - magicLen);<a name="line.103"></a> +<span class="sourceLineNo">104</span> return Pair.newPair(serverName, builder.build());<a name="line.104"></a> +<span class="sourceLineNo">105</span> }<a name="line.105"></a> +<span class="sourceLineNo">106</span><a name="line.106"></a> +<span class="sourceLineNo">107</span> /**<a name="line.107"></a> +<span class="sourceLineNo">108</span> * Starts the tracking of online RegionServers. All RSes will be tracked after this method is<a name="line.108"></a> +<span class="sourceLineNo">109</span> * called.<a name="line.109"></a> +<span class="sourceLineNo">110</span> * <p/><a name="line.110"></a> +<span class="sourceLineNo">111</span> * In this method, we will also construct the region server sets in {@link ServerManager}. If a<a name="line.111"></a> +<span class="sourceLineNo">112</span> * region server is dead between the crash of the previous master instance and the start of the<a name="line.112"></a> +<span class="sourceLineNo">113</span> * current master instance, we will schedule a SCP for it. This is done in<a name="line.113"></a> +<span class="sourceLineNo">114</span> * {@link ServerManager#findOutDeadServersAndProcess(Set, Set)}, we call it here under the lock<a name="line.114"></a> +<span class="sourceLineNo">115</span> * protection to prevent concurrency issues with server expiration operation.<a name="line.115"></a> +<span class="sourceLineNo">116</span> * @param deadServersFromPE the region servers which already have SCP associated.<a name="line.116"></a> +<span class="sourceLineNo">117</span> * @param liveServersFromWALDir the live region servers from wal directory.<a name="line.117"></a> <span class="sourceLineNo">118</span> */<a name="line.118"></a> -<span class="sourceLineNo">119</span> public void start() throws KeeperException, IOException {<a name="line.119"></a> -<span class="sourceLineNo">120</span> watcher.registerListener(this);<a name="line.120"></a> -<span class="sourceLineNo">121</span> synchronized (this) {<a name="line.121"></a> -<span class="sourceLineNo">122</span> List<String> servers =<a name="line.122"></a> -<span class="sourceLineNo">123</span> ZKUtil.listChildrenAndWatchForNewChildren(watcher, watcher.getZNodePaths().rsZNode);<a name="line.123"></a> -<span class="sourceLineNo">124</span> for (String n : servers) {<a name="line.124"></a> -<span class="sourceLineNo">125</span> Pair<ServerName, RegionServerInfo> pair = getServerInfo(n);<a name="line.125"></a> -<span class="sourceLineNo">126</span> ServerName serverName = pair.getFirst();<a name="line.126"></a> -<span class="sourceLineNo">127</span> RegionServerInfo info = pair.getSecond();<a name="line.127"></a> -<span class="sourceLineNo">128</span> regionServers.add(serverName);<a name="line.128"></a> -<span class="sourceLineNo">129</span> ServerMetrics serverMetrics = info != null<a name="line.129"></a> -<span class="sourceLineNo">130</span> ? ServerMetricsBuilder.of(serverName,<a name="line.130"></a> -<span class="sourceLineNo">131</span> VersionInfoUtil.getVersionNumber(info.getVersionInfo()))<a name="line.131"></a> -<span class="sourceLineNo">132</span> : ServerMetricsBuilder.of(serverName);<a name="line.132"></a> -<span class="sourceLineNo">133</span> serverManager.checkAndRecordNewServer(serverName, serverMetrics);<a name="line.133"></a> -<span class="sourceLineNo">134</span> }<a name="line.134"></a> -<span class="sourceLineNo">135</span> }<a name="line.135"></a> -<span class="sourceLineNo">136</span> }<a name="line.136"></a> -<span class="sourceLineNo">137</span><a name="line.137"></a> -<span class="sourceLineNo">138</span> public void stop() {<a name="line.138"></a> -<span class="sourceLineNo">139</span> executor.shutdownNow();<a name="line.139"></a> -<span class="sourceLineNo">140</span> }<a name="line.140"></a> -<span class="sourceLineNo">141</span><a name="line.141"></a> -<span class="sourceLineNo">142</span> private synchronized void refresh() {<a name="line.142"></a> -<span class="sourceLineNo">143</span> List<String> names;<a name="line.143"></a> -<span class="sourceLineNo">144</span> try {<a name="line.144"></a> -<span class="sourceLineNo">145</span> names = ZKUtil.listChildrenAndWatchForNewChildren(watcher, watcher.getZNodePaths().rsZNode);<a name="line.145"></a> -<span class="sourceLineNo">146</span> } catch (KeeperException e) {<a name="line.146"></a> -<span class="sourceLineNo">147</span> // here we need to abort as we failed to set watcher on the rs node which means that we can<a name="line.147"></a> -<span class="sourceLineNo">148</span> // not track the node deleted evetnt any more.<a name="line.148"></a> -<span class="sourceLineNo">149</span> server.abort("Unexpected zk exception getting RS nodes", e);<a name="line.149"></a> -<span class="sourceLineNo">150</span> return;<a name="line.150"></a> -<span class="sourceLineNo">151</span> }<a name="line.151"></a> -<span class="sourceLineNo">152</span> Set<ServerName> servers =<a name="line.152"></a> -<span class="sourceLineNo">153</span> names.stream().map(ServerName::parseServerName).collect(Collectors.toSet());<a name="line.153"></a> -<span class="sourceLineNo">154</span> for (Iterator<ServerName> iter = regionServers.iterator(); iter.hasNext();) {<a name="line.154"></a> -<span class="sourceLineNo">155</span> ServerName sn = iter.next();<a name="line.155"></a> -<span class="sourceLineNo">156</span> if (!servers.contains(sn)) {<a name="line.156"></a> -<span class="sourceLineNo">157</span> LOG.info("RegionServer ephemeral node deleted, processing expiration [{}]", sn);<a name="line.157"></a> -<span class="sourceLineNo">158</span> serverManager.expireServer(sn);<a name="line.158"></a> -<span class="sourceLineNo">159</span> iter.remove();<a name="line.159"></a> -<span class="sourceLineNo">160</span> }<a name="line.160"></a> -<span class="sourceLineNo">161</span> }<a name="line.161"></a> -<span class="sourceLineNo">162</span> // here we do not need to parse the region server info as it is useless now, we only need the<a name="line.162"></a> -<span class="sourceLineNo">163</span> // server name.<a name="line.163"></a> -<span class="sourceLineNo">164</span> boolean newServerAdded = false;<a name="line.164"></a> -<span class="sourceLineNo">165</span> for (ServerName sn : servers) {<a name="line.165"></a> -<span class="sourceLineNo">166</span> if (regionServers.add(sn)) {<a name="line.166"></a> -<span class="sourceLineNo">167</span> newServerAdded = true;<a name="line.167"></a> -<span class="sourceLineNo">168</span> LOG.info("RegionServer ephemeral node created, adding [" + sn + "]");<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> if (newServerAdded && server.isInitialized()) {<a name="line.171"></a> -<span class="sourceLineNo">172</span> // Only call the check to move servers if a RegionServer was added to the cluster; in this<a name="line.172"></a> -<span class="sourceLineNo">173</span> // case it could be a server with a new version so it makes sense to run the check.<a name="line.173"></a> -<span class="sourceLineNo">174</span> server.checkIfShouldMoveSystemRegionAsync();<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><a name="line.177"></a> -<span class="sourceLineNo">178</span> @Override<a name="line.178"></a> -<span class="sourceLineNo">179</span> public void nodeChildrenChanged(String path) {<a name="line.179"></a> -<span class="sourceLineNo">180</span> if (path.equals(watcher.getZNodePaths().rsZNode) && !server.isAborted() &&<a name="line.180"></a> -<span class="sourceLineNo">181</span> !server.isStopped()) {<a name="line.181"></a> -<span class="sourceLineNo">182</span> executor.execute(this::refresh);<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">119</span> public void start(Set<ServerName> deadServersFromPE, Set<ServerName> liveServersFromWALDir)<a name="line.119"></a> +<span class="sourceLineNo">120</span> throws KeeperException, IOException {<a name="line.120"></a> +<span class="sourceLineNo">121</span> watcher.registerListener(this);<a name="line.121"></a> +<span class="sourceLineNo">122</span> synchronized (this) {<a name="line.122"></a> +<span class="sourceLineNo">123</span> List<String> servers =<a name="line.123"></a> +<span class="sourceLineNo">124</span> ZKUtil.listChildrenAndWatchForNewChildren(watcher, watcher.getZNodePaths().rsZNode);<a name="line.124"></a> +<span class="sourceLineNo">125</span> for (String n : servers) {<a name="line.125"></a> +<span class="sourceLineNo">126</span> Pair<ServerName, RegionServerInfo> pair = getServerInfo(n);<a name="line.126"></a> +<span class="sourceLineNo">127</span> ServerName serverName = pair.getFirst();<a name="line.127"></a> +<span class="sourceLineNo">128</span> RegionServerInfo info = pair.getSecond();<a name="line.128"></a> +<span class="sourceLineNo">129</span> regionServers.add(serverName);<a name="line.129"></a> +<span class="sourceLineNo">130</span> ServerMetrics serverMetrics = info != null<a name="line.130"></a> +<span class="sourceLineNo">131</span> ? ServerMetricsBuilder.of(serverName,<a name="line.131"></a> +<span class="sourceLineNo">132</span> VersionInfoUtil.getVersionNumber(info.getVersionInfo()))<a name="line.132"></a> +<span class="sourceLineNo">133</span> : ServerMetricsBuilder.of(serverName);<a name="line.133"></a> +<span class="sourceLineNo">134</span> serverManager.checkAndRecordNewServer(serverName, serverMetrics);<a name="line.134"></a> +<span class="sourceLineNo">135</span> }<a name="line.135"></a> +<span class="sourceLineNo">136</span> serverManager.findOutDeadServersAndProcess(deadServersFromPE, liveServersFromWALDir);<a name="line.136"></a> +<span class="sourceLineNo">137</span> }<a name="line.137"></a> +<span class="sourceLineNo">138</span> }<a name="line.138"></a> +<span class="sourceLineNo">139</span><a name="line.139"></a> +<span class="sourceLineNo">140</span> public void stop() {<a name="line.140"></a> +<span class="sourceLineNo">141</span> executor.shutdownNow();<a name="line.141"></a> +<span class="sourceLineNo">142</span> }<a name="line.142"></a> +<span class="sourceLineNo">143</span><a name="line.143"></a> +<span class="sourceLineNo">144</span> private synchronized void refresh() {<a name="line.144"></a> +<span class="sourceLineNo">145</span> List<String> names;<a name="line.145"></a> +<span class="sourceLineNo">146</span> try {<a name="line.146"></a> +<span class="sourceLineNo">147</span> names = ZKUtil.listChildrenAndWatchForNewChildren(watcher, watcher.getZNodePaths().rsZNode);<a name="line.147"></a> +<span class="sourceLineNo">148</span> } catch (KeeperException e) {<a name="line.148"></a> +<span class="sourceLineNo">149</span> // here we need to abort as we failed to set watcher on the rs node which means that we can<a name="line.149"></a> +<span class="sourceLineNo">150</span> // not track the node deleted evetnt any more.<a name="line.150"></a> +<span class="sourceLineNo">151</span> server.abort("Unexpected zk exception getting RS nodes", e);<a name="line.151"></a> +<span class="sourceLineNo">152</span> return;<a name="line.152"></a> +<span class="sourceLineNo">153</span> }<a name="line.153"></a> +<span class="sourceLineNo">154</span> Set<ServerName> servers =<a name="line.154"></a> +<span class="sourceLineNo">155</span> names.stream().map(ServerName::parseServerName).collect(Collectors.toSet());<a name="line.155"></a> +<span class="sourceLineNo">156</span> for (Iterator<ServerName> iter = regionServers.iterator(); iter.hasNext();) {<a name="line.156"></a> +<span class="sourceLineNo">157</span> ServerName sn = iter.next();<a name="line.157"></a> +<span class="sourceLineNo">158</span> if (!servers.contains(sn)) {<a name="line.158"></a> +<span class="sourceLineNo">159</span> LOG.info("RegionServer ephemeral node deleted, processing expiration [{}]", sn);<a name="line.159"></a> +<span class="sourceLineNo">160</span> serverManager.expireServer(sn);<a name="line.160"></a> +<span class="sourceLineNo">161</span> iter.remove();<a name="line.161"></a> +<span class="sourceLineNo">162</span> }<a name="line.162"></a> +<span class="sourceLineNo">163</span> }<a name="line.163"></a> +<span class="sourceLineNo">164</span> // here we do not need to parse the region server info as it is useless now, we only need the<a name="line.164"></a> +<span class="sourceLineNo">165</span> // server name.<a name="line.165"></a> +<span class="sourceLineNo">166</span> boolean newServerAdded = false;<a name="line.166"></a> +<span class="sourceLineNo">167</span> for (ServerName sn : servers) {<a name="line.167"></a> +<span class="sourceLineNo">168</span> if (regionServers.add(sn)) {<a name="line.168"></a> +<span class="sourceLineNo">169</span> newServerAdded = true;<a name="line.169"></a> +<span class="sourceLineNo">170</span> LOG.info("RegionServer ephemeral node created, adding [" + sn + "]");<a name="line.170"></a> +<span class="sourceLineNo">171</span> }<a name="line.171"></a> +<span class="sourceLineNo">172</span> }<a name="line.172"></a> +<span class="sourceLineNo">173</span> if (newServerAdded && server.isInitialized()) {<a name="line.173"></a> +<span class="sourceLineNo">174</span> // Only call the check to move servers if a RegionServer was added to the cluster; in this<a name="line.174"></a> +<span class="sourceLineNo">175</span> // case it could be a server with a new version so it makes sense to run the check.<a name="line.175"></a> +<span class="sourceLineNo">176</span> server.checkIfShouldMoveSystemRegionAsync();<a name="line.176"></a> +<span class="sourceLineNo">177</span> }<a name="line.177"></a> +<span class="sourceLineNo">178</span> }<a name="line.178"></a> +<span class="sourceLineNo">179</span><a name="line.179"></a> +<span class="sourceLineNo">180</span> @Override<a name="line.180"></a> +<span class="sourceLineNo">181</span> public void nodeChildrenChanged(String path) {<a name="line.181"></a> +<span class="sourceLineNo">182</span> if (path.equals(watcher.getZNodePaths().rsZNode) && !server.isAborted() &&<a name="line.182"></a> +<span class="sourceLineNo">183</span> !server.isStopped()) {<a name="line.183"></a> +<span class="sourceLineNo">184</span> executor.execute(this::refresh);<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>
