http://git-wip-us.apache.org/repos/asf/hbase-site/blob/6674e3ab/devapidocs/src-html/org/apache/hadoop/hbase/master/SplitLogManager.TimeoutMonitor.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/master/SplitLogManager.TimeoutMonitor.html b/devapidocs/src-html/org/apache/hadoop/hbase/master/SplitLogManager.TimeoutMonitor.html index 2939a56..681e263 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/master/SplitLogManager.TimeoutMonitor.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/master/SplitLogManager.TimeoutMonitor.html @@ -61,602 +61,608 @@ <span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.monitoring.TaskMonitor;<a name="line.53"></a> <span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.util.EnvironmentEdgeManager;<a name="line.54"></a> <span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.util.FSUtils;<a name="line.55"></a> -<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.wal.AbstractFSWALProvider;<a name="line.56"></a> -<span class="sourceLineNo">057</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.57"></a> -<span class="sourceLineNo">058</span>import org.slf4j.Logger;<a name="line.58"></a> -<span class="sourceLineNo">059</span>import org.slf4j.LoggerFactory;<a name="line.59"></a> -<span class="sourceLineNo">060</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.60"></a> -<span class="sourceLineNo">061</span><a name="line.61"></a> -<span class="sourceLineNo">062</span>/**<a name="line.62"></a> -<span class="sourceLineNo">063</span> * Distributes the task of log splitting to the available region servers.<a name="line.63"></a> -<span class="sourceLineNo">064</span> * Coordination happens via coordination engine. For every log file that has to be split a<a name="line.64"></a> -<span class="sourceLineNo">065</span> * task is created. SplitLogWorkers race to grab a task.<a name="line.65"></a> -<span class="sourceLineNo">066</span> *<a name="line.66"></a> -<span class="sourceLineNo">067</span> * <p>SplitLogManager monitors the tasks that it creates using the<a name="line.67"></a> -<span class="sourceLineNo">068</span> * timeoutMonitor thread. If a task's progress is slow then<a name="line.68"></a> -<span class="sourceLineNo">069</span> * {@link SplitLogManagerCoordination#checkTasks} will take away the<a name="line.69"></a> -<span class="sourceLineNo">070</span> * task from the owner {@link org.apache.hadoop.hbase.regionserver.SplitLogWorker}<a name="line.70"></a> -<span class="sourceLineNo">071</span> * and the task will be up for grabs again. When the task is done then it is<a name="line.71"></a> -<span class="sourceLineNo">072</span> * deleted by SplitLogManager.<a name="line.72"></a> -<span class="sourceLineNo">073</span> *<a name="line.73"></a> -<span class="sourceLineNo">074</span> * <p>Clients call {@link #splitLogDistributed(Path)} to split a region server's<a name="line.74"></a> -<span class="sourceLineNo">075</span> * log files. The caller thread waits in this method until all the log files<a name="line.75"></a> -<span class="sourceLineNo">076</span> * have been split.<a name="line.76"></a> -<span class="sourceLineNo">077</span> *<a name="line.77"></a> -<span class="sourceLineNo">078</span> * <p>All the coordination calls made by this class are asynchronous. This is mainly<a name="line.78"></a> -<span class="sourceLineNo">079</span> * to help reduce response time seen by the callers.<a name="line.79"></a> -<span class="sourceLineNo">080</span> *<a name="line.80"></a> -<span class="sourceLineNo">081</span> * <p>There is race in this design between the SplitLogManager and the<a name="line.81"></a> -<span class="sourceLineNo">082</span> * SplitLogWorker. SplitLogManager might re-queue a task that has in reality<a name="line.82"></a> -<span class="sourceLineNo">083</span> * already been completed by a SplitLogWorker. We rely on the idempotency of<a name="line.83"></a> -<span class="sourceLineNo">084</span> * the log splitting task for correctness.<a name="line.84"></a> -<span class="sourceLineNo">085</span> *<a name="line.85"></a> -<span class="sourceLineNo">086</span> * <p>It is also assumed that every log splitting task is unique and once<a name="line.86"></a> -<span class="sourceLineNo">087</span> * completed (either with success or with error) it will be not be submitted<a name="line.87"></a> -<span class="sourceLineNo">088</span> * again. If a task is resubmitted then there is a risk that old "delete task"<a name="line.88"></a> -<span class="sourceLineNo">089</span> * can delete the re-submission.<a name="line.89"></a> -<span class="sourceLineNo">090</span> */<a name="line.90"></a> -<span class="sourceLineNo">091</span>@InterfaceAudience.Private<a name="line.91"></a> -<span class="sourceLineNo">092</span>public class SplitLogManager {<a name="line.92"></a> -<span class="sourceLineNo">093</span> private static final Logger LOG = LoggerFactory.getLogger(SplitLogManager.class);<a name="line.93"></a> -<span class="sourceLineNo">094</span><a name="line.94"></a> -<span class="sourceLineNo">095</span> private final MasterServices server;<a name="line.95"></a> -<span class="sourceLineNo">096</span><a name="line.96"></a> -<span class="sourceLineNo">097</span> private final Configuration conf;<a name="line.97"></a> -<span class="sourceLineNo">098</span> private final ChoreService choreService;<a name="line.98"></a> -<span class="sourceLineNo">099</span><a name="line.99"></a> -<span class="sourceLineNo">100</span> public static final int DEFAULT_UNASSIGNED_TIMEOUT = (3 * 60 * 1000); // 3 min<a name="line.100"></a> -<span class="sourceLineNo">101</span><a name="line.101"></a> -<span class="sourceLineNo">102</span> private long unassignedTimeout;<a name="line.102"></a> -<span class="sourceLineNo">103</span> private long lastTaskCreateTime = Long.MAX_VALUE;<a name="line.103"></a> -<span class="sourceLineNo">104</span><a name="line.104"></a> -<span class="sourceLineNo">105</span> @VisibleForTesting<a name="line.105"></a> -<span class="sourceLineNo">106</span> final ConcurrentMap<String, Task> tasks = new ConcurrentHashMap<>();<a name="line.106"></a> -<span class="sourceLineNo">107</span> private TimeoutMonitor timeoutMonitor;<a name="line.107"></a> -<span class="sourceLineNo">108</span><a name="line.108"></a> -<span class="sourceLineNo">109</span> private volatile Set<ServerName> deadWorkers = null;<a name="line.109"></a> -<span class="sourceLineNo">110</span> private final Object deadWorkersLock = new Object();<a name="line.110"></a> -<span class="sourceLineNo">111</span><a name="line.111"></a> -<span class="sourceLineNo">112</span> /**<a name="line.112"></a> -<span class="sourceLineNo">113</span> * Its OK to construct this object even when region-servers are not online. It does lookup the<a name="line.113"></a> -<span class="sourceLineNo">114</span> * orphan tasks in coordination engine but it doesn't block waiting for them to be done.<a name="line.114"></a> -<span class="sourceLineNo">115</span> * @param master the master services<a name="line.115"></a> -<span class="sourceLineNo">116</span> * @param conf the HBase configuration<a name="line.116"></a> -<span class="sourceLineNo">117</span> * @throws IOException<a name="line.117"></a> -<span class="sourceLineNo">118</span> */<a name="line.118"></a> -<span class="sourceLineNo">119</span> public SplitLogManager(MasterServices master, Configuration conf)<a name="line.119"></a> -<span class="sourceLineNo">120</span> throws IOException {<a name="line.120"></a> -<span class="sourceLineNo">121</span> this.server = master;<a name="line.121"></a> -<span class="sourceLineNo">122</span> this.conf = conf;<a name="line.122"></a> -<span class="sourceLineNo">123</span> this.choreService = new ChoreService(master.getServerName() + "_splitLogManager_");<a name="line.123"></a> -<span class="sourceLineNo">124</span> if (server.getCoordinatedStateManager() != null) {<a name="line.124"></a> -<span class="sourceLineNo">125</span> SplitLogManagerCoordination coordination = getSplitLogManagerCoordination();<a name="line.125"></a> -<span class="sourceLineNo">126</span> Set<String> failedDeletions = Collections.synchronizedSet(new HashSet<String>());<a name="line.126"></a> -<span class="sourceLineNo">127</span> SplitLogManagerDetails details = new SplitLogManagerDetails(tasks, master, failedDeletions);<a name="line.127"></a> -<span class="sourceLineNo">128</span> coordination.setDetails(details);<a name="line.128"></a> -<span class="sourceLineNo">129</span> coordination.init();<a name="line.129"></a> -<span class="sourceLineNo">130</span> }<a name="line.130"></a> -<span class="sourceLineNo">131</span> this.unassignedTimeout =<a name="line.131"></a> -<span class="sourceLineNo">132</span> conf.getInt("hbase.splitlog.manager.unassigned.timeout", DEFAULT_UNASSIGNED_TIMEOUT);<a name="line.132"></a> -<span class="sourceLineNo">133</span> this.timeoutMonitor =<a name="line.133"></a> -<span class="sourceLineNo">134</span> new TimeoutMonitor(conf.getInt("hbase.splitlog.manager.timeoutmonitor.period", 1000),<a name="line.134"></a> -<span class="sourceLineNo">135</span> master);<a name="line.135"></a> -<span class="sourceLineNo">136</span> choreService.scheduleChore(timeoutMonitor);<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> private SplitLogManagerCoordination getSplitLogManagerCoordination() {<a name="line.139"></a> -<span class="sourceLineNo">140</span> return server.getCoordinatedStateManager().getSplitLogManagerCoordination();<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> private FileStatus[] getFileList(List<Path> logDirs, PathFilter filter) throws IOException {<a name="line.143"></a> -<span class="sourceLineNo">144</span> return getFileList(conf, logDirs, filter);<a name="line.144"></a> -<span class="sourceLineNo">145</span> }<a name="line.145"></a> -<span class="sourceLineNo">146</span><a name="line.146"></a> -<span class="sourceLineNo">147</span> /**<a name="line.147"></a> -<span class="sourceLineNo">148</span> * Get a list of paths that need to be split given a set of server-specific directories and<a name="line.148"></a> -<span class="sourceLineNo">149</span> * optionally a filter.<a name="line.149"></a> -<span class="sourceLineNo">150</span> *<a name="line.150"></a> -<span class="sourceLineNo">151</span> * See {@link AbstractFSWALProvider#getServerNameFromWALDirectoryName} for more info on directory<a name="line.151"></a> -<span class="sourceLineNo">152</span> * layout.<a name="line.152"></a> -<span class="sourceLineNo">153</span> *<a name="line.153"></a> -<span class="sourceLineNo">154</span> * Should be package-private, but is needed by<a name="line.154"></a> -<span class="sourceLineNo">155</span> * {@link org.apache.hadoop.hbase.wal.WALSplitter#split(Path, Path, Path, FileSystem,<a name="line.155"></a> -<span class="sourceLineNo">156</span> * Configuration, org.apache.hadoop.hbase.wal.WALFactory)} for tests.<a name="line.156"></a> -<span class="sourceLineNo">157</span> */<a name="line.157"></a> -<span class="sourceLineNo">158</span> @VisibleForTesting<a name="line.158"></a> -<span class="sourceLineNo">159</span> public static FileStatus[] getFileList(final Configuration conf, final List<Path> logDirs,<a name="line.159"></a> -<span class="sourceLineNo">160</span> final PathFilter filter)<a name="line.160"></a> -<span class="sourceLineNo">161</span> throws IOException {<a name="line.161"></a> -<span class="sourceLineNo">162</span> List<FileStatus> fileStatus = new ArrayList<>();<a name="line.162"></a> -<span class="sourceLineNo">163</span> for (Path logDir : logDirs) {<a name="line.163"></a> -<span class="sourceLineNo">164</span> final FileSystem fs = logDir.getFileSystem(conf);<a name="line.164"></a> -<span class="sourceLineNo">165</span> if (!fs.exists(logDir)) {<a name="line.165"></a> -<span class="sourceLineNo">166</span> LOG.warn(logDir + " doesn't exist. Nothing to do!");<a name="line.166"></a> -<span class="sourceLineNo">167</span> continue;<a name="line.167"></a> -<span class="sourceLineNo">168</span> }<a name="line.168"></a> -<span class="sourceLineNo">169</span> FileStatus[] logfiles = FSUtils.listStatus(fs, logDir, filter);<a name="line.169"></a> -<span class="sourceLineNo">170</span> if (logfiles == null || logfiles.length == 0) {<a name="line.170"></a> -<span class="sourceLineNo">171</span> LOG.info(logDir + " is empty dir, no logs to split");<a name="line.171"></a> -<span class="sourceLineNo">172</span> } else {<a name="line.172"></a> -<span class="sourceLineNo">173</span> Collections.addAll(fileStatus, logfiles);<a name="line.173"></a> +<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.util.HasThread;<a name="line.56"></a> +<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.wal.AbstractFSWALProvider;<a name="line.57"></a> +<span class="sourceLineNo">058</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.58"></a> +<span class="sourceLineNo">059</span>import org.slf4j.Logger;<a name="line.59"></a> +<span class="sourceLineNo">060</span>import org.slf4j.LoggerFactory;<a name="line.60"></a> +<span class="sourceLineNo">061</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.61"></a> +<span class="sourceLineNo">062</span><a name="line.62"></a> +<span class="sourceLineNo">063</span>/**<a name="line.63"></a> +<span class="sourceLineNo">064</span> * Distributes the task of log splitting to the available region servers.<a name="line.64"></a> +<span class="sourceLineNo">065</span> * Coordination happens via coordination engine. For every log file that has to be split a<a name="line.65"></a> +<span class="sourceLineNo">066</span> * task is created. SplitLogWorkers race to grab a task.<a name="line.66"></a> +<span class="sourceLineNo">067</span> *<a name="line.67"></a> +<span class="sourceLineNo">068</span> * <p>SplitLogManager monitors the tasks that it creates using the<a name="line.68"></a> +<span class="sourceLineNo">069</span> * timeoutMonitor thread. If a task's progress is slow then<a name="line.69"></a> +<span class="sourceLineNo">070</span> * {@link SplitLogManagerCoordination#checkTasks} will take away the<a name="line.70"></a> +<span class="sourceLineNo">071</span> * task from the owner {@link org.apache.hadoop.hbase.regionserver.SplitLogWorker}<a name="line.71"></a> +<span class="sourceLineNo">072</span> * and the task will be up for grabs again. When the task is done then it is<a name="line.72"></a> +<span class="sourceLineNo">073</span> * deleted by SplitLogManager.<a name="line.73"></a> +<span class="sourceLineNo">074</span> *<a name="line.74"></a> +<span class="sourceLineNo">075</span> * <p>Clients call {@link #splitLogDistributed(Path)} to split a region server's<a name="line.75"></a> +<span class="sourceLineNo">076</span> * log files. The caller thread waits in this method until all the log files<a name="line.76"></a> +<span class="sourceLineNo">077</span> * have been split.<a name="line.77"></a> +<span class="sourceLineNo">078</span> *<a name="line.78"></a> +<span class="sourceLineNo">079</span> * <p>All the coordination calls made by this class are asynchronous. This is mainly<a name="line.79"></a> +<span class="sourceLineNo">080</span> * to help reduce response time seen by the callers.<a name="line.80"></a> +<span class="sourceLineNo">081</span> *<a name="line.81"></a> +<span class="sourceLineNo">082</span> * <p>There is race in this design between the SplitLogManager and the<a name="line.82"></a> +<span class="sourceLineNo">083</span> * SplitLogWorker. SplitLogManager might re-queue a task that has in reality<a name="line.83"></a> +<span class="sourceLineNo">084</span> * already been completed by a SplitLogWorker. We rely on the idempotency of<a name="line.84"></a> +<span class="sourceLineNo">085</span> * the log splitting task for correctness.<a name="line.85"></a> +<span class="sourceLineNo">086</span> *<a name="line.86"></a> +<span class="sourceLineNo">087</span> * <p>It is also assumed that every log splitting task is unique and once<a name="line.87"></a> +<span class="sourceLineNo">088</span> * completed (either with success or with error) it will be not be submitted<a name="line.88"></a> +<span class="sourceLineNo">089</span> * again. If a task is resubmitted then there is a risk that old "delete task"<a name="line.89"></a> +<span class="sourceLineNo">090</span> * can delete the re-submission.<a name="line.90"></a> +<span class="sourceLineNo">091</span> */<a name="line.91"></a> +<span class="sourceLineNo">092</span>@InterfaceAudience.Private<a name="line.92"></a> +<span class="sourceLineNo">093</span>public class SplitLogManager {<a name="line.93"></a> +<span class="sourceLineNo">094</span> private static final Logger LOG = LoggerFactory.getLogger(SplitLogManager.class);<a name="line.94"></a> +<span class="sourceLineNo">095</span><a name="line.95"></a> +<span class="sourceLineNo">096</span> private final MasterServices server;<a name="line.96"></a> +<span class="sourceLineNo">097</span><a name="line.97"></a> +<span class="sourceLineNo">098</span> private final Configuration conf;<a name="line.98"></a> +<span class="sourceLineNo">099</span> private final ChoreService choreService;<a name="line.99"></a> +<span class="sourceLineNo">100</span><a name="line.100"></a> +<span class="sourceLineNo">101</span> public static final int DEFAULT_UNASSIGNED_TIMEOUT = (3 * 60 * 1000); // 3 min<a name="line.101"></a> +<span class="sourceLineNo">102</span><a name="line.102"></a> +<span class="sourceLineNo">103</span> private long unassignedTimeout;<a name="line.103"></a> +<span class="sourceLineNo">104</span> private long lastTaskCreateTime = Long.MAX_VALUE;<a name="line.104"></a> +<span class="sourceLineNo">105</span><a name="line.105"></a> +<span class="sourceLineNo">106</span> @VisibleForTesting<a name="line.106"></a> +<span class="sourceLineNo">107</span> final ConcurrentMap<String, Task> tasks = new ConcurrentHashMap<>();<a name="line.107"></a> +<span class="sourceLineNo">108</span> private TimeoutMonitor timeoutMonitor;<a name="line.108"></a> +<span class="sourceLineNo">109</span><a name="line.109"></a> +<span class="sourceLineNo">110</span> private volatile Set<ServerName> deadWorkers = null;<a name="line.110"></a> +<span class="sourceLineNo">111</span> private final Object deadWorkersLock = new Object();<a name="line.111"></a> +<span class="sourceLineNo">112</span><a name="line.112"></a> +<span class="sourceLineNo">113</span> /**<a name="line.113"></a> +<span class="sourceLineNo">114</span> * Its OK to construct this object even when region-servers are not online. It does lookup the<a name="line.114"></a> +<span class="sourceLineNo">115</span> * orphan tasks in coordination engine but it doesn't block waiting for them to be done.<a name="line.115"></a> +<span class="sourceLineNo">116</span> * @param master the master services<a name="line.116"></a> +<span class="sourceLineNo">117</span> * @param conf the HBase configuration<a name="line.117"></a> +<span class="sourceLineNo">118</span> * @throws IOException<a name="line.118"></a> +<span class="sourceLineNo">119</span> */<a name="line.119"></a> +<span class="sourceLineNo">120</span> public SplitLogManager(MasterServices master, Configuration conf)<a name="line.120"></a> +<span class="sourceLineNo">121</span> throws IOException {<a name="line.121"></a> +<span class="sourceLineNo">122</span> this.server = master;<a name="line.122"></a> +<span class="sourceLineNo">123</span> this.conf = conf;<a name="line.123"></a> +<span class="sourceLineNo">124</span> // Get Server Thread name. Sometimes the Server is mocked so may not implement HasThread.<a name="line.124"></a> +<span class="sourceLineNo">125</span> // For example, in tests.<a name="line.125"></a> +<span class="sourceLineNo">126</span> String name = master instanceof HasThread? ((HasThread)master).getName():<a name="line.126"></a> +<span class="sourceLineNo">127</span> master.getServerName().toShortString();<a name="line.127"></a> +<span class="sourceLineNo">128</span> this.choreService =<a name="line.128"></a> +<span class="sourceLineNo">129</span> new ChoreService(name + ".splitLogManager.");<a name="line.129"></a> +<span class="sourceLineNo">130</span> if (server.getCoordinatedStateManager() != null) {<a name="line.130"></a> +<span class="sourceLineNo">131</span> SplitLogManagerCoordination coordination = getSplitLogManagerCoordination();<a name="line.131"></a> +<span class="sourceLineNo">132</span> Set<String> failedDeletions = Collections.synchronizedSet(new HashSet<String>());<a name="line.132"></a> +<span class="sourceLineNo">133</span> SplitLogManagerDetails details = new SplitLogManagerDetails(tasks, master, failedDeletions);<a name="line.133"></a> +<span class="sourceLineNo">134</span> coordination.setDetails(details);<a name="line.134"></a> +<span class="sourceLineNo">135</span> coordination.init();<a name="line.135"></a> +<span class="sourceLineNo">136</span> }<a name="line.136"></a> +<span class="sourceLineNo">137</span> this.unassignedTimeout =<a name="line.137"></a> +<span class="sourceLineNo">138</span> conf.getInt("hbase.splitlog.manager.unassigned.timeout", DEFAULT_UNASSIGNED_TIMEOUT);<a name="line.138"></a> +<span class="sourceLineNo">139</span> this.timeoutMonitor =<a name="line.139"></a> +<span class="sourceLineNo">140</span> new TimeoutMonitor(conf.getInt("hbase.splitlog.manager.timeoutmonitor.period", 1000),<a name="line.140"></a> +<span class="sourceLineNo">141</span> master);<a name="line.141"></a> +<span class="sourceLineNo">142</span> choreService.scheduleChore(timeoutMonitor);<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> private SplitLogManagerCoordination getSplitLogManagerCoordination() {<a name="line.145"></a> +<span class="sourceLineNo">146</span> return server.getCoordinatedStateManager().getSplitLogManagerCoordination();<a name="line.146"></a> +<span class="sourceLineNo">147</span> }<a name="line.147"></a> +<span class="sourceLineNo">148</span><a name="line.148"></a> +<span class="sourceLineNo">149</span> private FileStatus[] getFileList(List<Path> logDirs, PathFilter filter) throws IOException {<a name="line.149"></a> +<span class="sourceLineNo">150</span> return getFileList(conf, logDirs, filter);<a name="line.150"></a> +<span class="sourceLineNo">151</span> }<a name="line.151"></a> +<span class="sourceLineNo">152</span><a name="line.152"></a> +<span class="sourceLineNo">153</span> /**<a name="line.153"></a> +<span class="sourceLineNo">154</span> * Get a list of paths that need to be split given a set of server-specific directories and<a name="line.154"></a> +<span class="sourceLineNo">155</span> * optionally a filter.<a name="line.155"></a> +<span class="sourceLineNo">156</span> *<a name="line.156"></a> +<span class="sourceLineNo">157</span> * See {@link AbstractFSWALProvider#getServerNameFromWALDirectoryName} for more info on directory<a name="line.157"></a> +<span class="sourceLineNo">158</span> * layout.<a name="line.158"></a> +<span class="sourceLineNo">159</span> *<a name="line.159"></a> +<span class="sourceLineNo">160</span> * Should be package-private, but is needed by<a name="line.160"></a> +<span class="sourceLineNo">161</span> * {@link org.apache.hadoop.hbase.wal.WALSplitter#split(Path, Path, Path, FileSystem,<a name="line.161"></a> +<span class="sourceLineNo">162</span> * Configuration, org.apache.hadoop.hbase.wal.WALFactory)} for tests.<a name="line.162"></a> +<span class="sourceLineNo">163</span> */<a name="line.163"></a> +<span class="sourceLineNo">164</span> @VisibleForTesting<a name="line.164"></a> +<span class="sourceLineNo">165</span> public static FileStatus[] getFileList(final Configuration conf, final List<Path> logDirs,<a name="line.165"></a> +<span class="sourceLineNo">166</span> final PathFilter filter)<a name="line.166"></a> +<span class="sourceLineNo">167</span> throws IOException {<a name="line.167"></a> +<span class="sourceLineNo">168</span> List<FileStatus> fileStatus = new ArrayList<>();<a name="line.168"></a> +<span class="sourceLineNo">169</span> for (Path logDir : logDirs) {<a name="line.169"></a> +<span class="sourceLineNo">170</span> final FileSystem fs = logDir.getFileSystem(conf);<a name="line.170"></a> +<span class="sourceLineNo">171</span> if (!fs.exists(logDir)) {<a name="line.171"></a> +<span class="sourceLineNo">172</span> LOG.warn(logDir + " doesn't exist. Nothing to do!");<a name="line.172"></a> +<span class="sourceLineNo">173</span> continue;<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> FileStatus[] a = new FileStatus[fileStatus.size()];<a name="line.176"></a> -<span class="sourceLineNo">177</span> return fileStatus.toArray(a);<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> /**<a name="line.180"></a> -<span class="sourceLineNo">181</span> * @param logDir one region sever wal dir path in .logs<a name="line.181"></a> -<span class="sourceLineNo">182</span> * @throws IOException if there was an error while splitting any log file<a name="line.182"></a> -<span class="sourceLineNo">183</span> * @return cumulative size of the logfiles split<a name="line.183"></a> -<span class="sourceLineNo">184</span> * @throws IOException<a name="line.184"></a> -<span class="sourceLineNo">185</span> */<a name="line.185"></a> -<span class="sourceLineNo">186</span> public long splitLogDistributed(final Path logDir) throws IOException {<a name="line.186"></a> -<span class="sourceLineNo">187</span> List<Path> logDirs = new ArrayList<>();<a name="line.187"></a> -<span class="sourceLineNo">188</span> logDirs.add(logDir);<a name="line.188"></a> -<span class="sourceLineNo">189</span> return splitLogDistributed(logDirs);<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> * The caller will block until all the log files of the given region server have been processed -<a name="line.193"></a> -<span class="sourceLineNo">194</span> * successfully split or an error is encountered - by an available worker region server. This<a name="line.194"></a> -<span class="sourceLineNo">195</span> * method must only be called after the region servers have been brought online.<a name="line.195"></a> -<span class="sourceLineNo">196</span> * @param logDirs List of log dirs to split<a name="line.196"></a> -<span class="sourceLineNo">197</span> * @throws IOException If there was an error while splitting any log file<a name="line.197"></a> -<span class="sourceLineNo">198</span> * @return cumulative size of the logfiles split<a name="line.198"></a> -<span class="sourceLineNo">199</span> */<a name="line.199"></a> -<span class="sourceLineNo">200</span> public long splitLogDistributed(final List<Path> logDirs) throws IOException {<a name="line.200"></a> -<span class="sourceLineNo">201</span> if (logDirs.isEmpty()) {<a name="line.201"></a> -<span class="sourceLineNo">202</span> return 0;<a name="line.202"></a> -<span class="sourceLineNo">203</span> }<a name="line.203"></a> -<span class="sourceLineNo">204</span> Set<ServerName> serverNames = new HashSet<>();<a name="line.204"></a> -<span class="sourceLineNo">205</span> for (Path logDir : logDirs) {<a name="line.205"></a> -<span class="sourceLineNo">206</span> try {<a name="line.206"></a> -<span class="sourceLineNo">207</span> ServerName serverName = AbstractFSWALProvider.getServerNameFromWALDirectoryName(logDir);<a name="line.207"></a> -<span class="sourceLineNo">208</span> if (serverName != null) {<a name="line.208"></a> -<span class="sourceLineNo">209</span> serverNames.add(serverName);<a name="line.209"></a> -<span class="sourceLineNo">210</span> }<a name="line.210"></a> -<span class="sourceLineNo">211</span> } catch (IllegalArgumentException e) {<a name="line.211"></a> -<span class="sourceLineNo">212</span> // ignore invalid format error.<a name="line.212"></a> -<span class="sourceLineNo">213</span> LOG.warn("Cannot parse server name from " + logDir);<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> return splitLogDistributed(serverNames, logDirs, null);<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> * The caller will block until all the hbase:meta log files of the given region server have been<a name="line.220"></a> -<span class="sourceLineNo">221</span> * processed - successfully split or an error is encountered - by an available worker region<a name="line.221"></a> -<span class="sourceLineNo">222</span> * server. This method must only be called after the region servers have been brought online.<a name="line.222"></a> -<span class="sourceLineNo">223</span> * @param logDirs List of log dirs to split<a name="line.223"></a> -<span class="sourceLineNo">224</span> * @param filter the Path filter to select specific files for considering<a name="line.224"></a> -<span class="sourceLineNo">225</span> * @throws IOException If there was an error while splitting any log file<a name="line.225"></a> -<span class="sourceLineNo">226</span> * @return cumulative size of the logfiles split<a name="line.226"></a> -<span class="sourceLineNo">227</span> */<a name="line.227"></a> -<span class="sourceLineNo">228</span> public long splitLogDistributed(final Set<ServerName> serverNames, final List<Path> logDirs,<a name="line.228"></a> -<span class="sourceLineNo">229</span> PathFilter filter) throws IOException {<a name="line.229"></a> -<span class="sourceLineNo">230</span> MonitoredTask status = TaskMonitor.get().createStatus("Doing distributed log split in " +<a name="line.230"></a> -<span class="sourceLineNo">231</span> logDirs + " for serverName=" + serverNames);<a name="line.231"></a> -<span class="sourceLineNo">232</span> FileStatus[] logfiles = getFileList(logDirs, filter);<a name="line.232"></a> -<span class="sourceLineNo">233</span> status.setStatus("Checking directory contents...");<a name="line.233"></a> -<span class="sourceLineNo">234</span> SplitLogCounters.tot_mgr_log_split_batch_start.increment();<a name="line.234"></a> -<span class="sourceLineNo">235</span> LOG.info("Started splitting " + logfiles.length + " logs in " + logDirs +<a name="line.235"></a> -<span class="sourceLineNo">236</span> " for " + serverNames);<a name="line.236"></a> -<span class="sourceLineNo">237</span> long t = EnvironmentEdgeManager.currentTime();<a name="line.237"></a> -<span class="sourceLineNo">238</span> long totalSize = 0;<a name="line.238"></a> -<span class="sourceLineNo">239</span> TaskBatch batch = new TaskBatch();<a name="line.239"></a> -<span class="sourceLineNo">240</span> for (FileStatus lf : logfiles) {<a name="line.240"></a> -<span class="sourceLineNo">241</span> // TODO If the log file is still being written to - which is most likely<a name="line.241"></a> -<span class="sourceLineNo">242</span> // the case for the last log file - then its length will show up here<a name="line.242"></a> -<span class="sourceLineNo">243</span> // as zero. The size of such a file can only be retrieved after<a name="line.243"></a> -<span class="sourceLineNo">244</span> // recover-lease is done. totalSize will be under in most cases and the<a name="line.244"></a> -<span class="sourceLineNo">245</span> // metrics that it drives will also be under-reported.<a name="line.245"></a> -<span class="sourceLineNo">246</span> totalSize += lf.getLen();<a name="line.246"></a> -<span class="sourceLineNo">247</span> String pathToLog = FSUtils.removeWALRootPath(lf.getPath(), conf);<a name="line.247"></a> -<span class="sourceLineNo">248</span> if (!enqueueSplitTask(pathToLog, batch)) {<a name="line.248"></a> -<span class="sourceLineNo">249</span> throw new IOException("duplicate log split scheduled for " + lf.getPath());<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> waitForSplittingCompletion(batch, status);<a name="line.252"></a> -<span class="sourceLineNo">253</span><a name="line.253"></a> -<span class="sourceLineNo">254</span> if (batch.done != batch.installed) {<a name="line.254"></a> -<span class="sourceLineNo">255</span> batch.isDead = true;<a name="line.255"></a> -<span class="sourceLineNo">256</span> SplitLogCounters.tot_mgr_log_split_batch_err.increment();<a name="line.256"></a> -<span class="sourceLineNo">257</span> LOG.warn("error while splitting logs in " + logDirs + " installed = " + batch.installed<a name="line.257"></a> -<span class="sourceLineNo">258</span> + " but only " + batch.done + " done");<a name="line.258"></a> -<span class="sourceLineNo">259</span> String msg = "error or interrupted while splitting logs in " + logDirs + " Task = " + batch;<a name="line.259"></a> -<span class="sourceLineNo">260</span> status.abort(msg);<a name="line.260"></a> -<span class="sourceLineNo">261</span> throw new IOException(msg);<a name="line.261"></a> -<span class="sourceLineNo">262</span> }<a name="line.262"></a> -<span class="sourceLineNo">263</span> for (Path logDir : logDirs) {<a name="line.263"></a> -<span class="sourceLineNo">264</span> status.setStatus("Cleaning up log directory...");<a name="line.264"></a> -<span class="sourceLineNo">265</span> final FileSystem fs = logDir.getFileSystem(conf);<a name="line.265"></a> -<span class="sourceLineNo">266</span> try {<a name="line.266"></a> -<span class="sourceLineNo">267</span> if (fs.exists(logDir) && !fs.delete(logDir, false)) {<a name="line.267"></a> -<span class="sourceLineNo">268</span> LOG.warn("Unable to delete log src dir. Ignoring. " + logDir);<a name="line.268"></a> -<span class="sourceLineNo">269</span> }<a name="line.269"></a> -<span class="sourceLineNo">270</span> } catch (IOException ioe) {<a name="line.270"></a> -<span class="sourceLineNo">271</span> FileStatus[] files = fs.listStatus(logDir);<a name="line.271"></a> -<span class="sourceLineNo">272</span> if (files != null && files.length > 0) {<a name="line.272"></a> -<span class="sourceLineNo">273</span> LOG.warn("Returning success without actually splitting and "<a name="line.273"></a> -<span class="sourceLineNo">274</span> + "deleting all the log files in path " + logDir + ": "<a name="line.274"></a> -<span class="sourceLineNo">275</span> + Arrays.toString(files), ioe);<a name="line.275"></a> -<span class="sourceLineNo">276</span> } else {<a name="line.276"></a> -<span class="sourceLineNo">277</span> LOG.warn("Unable to delete log src dir. Ignoring. " + logDir, ioe);<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> SplitLogCounters.tot_mgr_log_split_batch_success.increment();<a name="line.280"></a> -<span class="sourceLineNo">281</span> }<a name="line.281"></a> -<span class="sourceLineNo">282</span> String msg =<a name="line.282"></a> -<span class="sourceLineNo">283</span> "finished splitting (more than or equal to) " + totalSize + " bytes in " + batch.installed<a name="line.283"></a> -<span class="sourceLineNo">284</span> + " log files in " + logDirs + " in "<a name="line.284"></a> -<span class="sourceLineNo">285</span> + (EnvironmentEdgeManager.currentTime() - t) + "ms";<a name="line.285"></a> -<span class="sourceLineNo">286</span> status.markComplete(msg);<a name="line.286"></a> -<span class="sourceLineNo">287</span> LOG.info(msg);<a name="line.287"></a> -<span class="sourceLineNo">288</span> return totalSize;<a name="line.288"></a> -<span class="sourceLineNo">289</span> }<a name="line.289"></a> -<span class="sourceLineNo">290</span><a name="line.290"></a> -<span class="sourceLineNo">291</span> /**<a name="line.291"></a> -<span class="sourceLineNo">292</span> * Add a task entry to coordination if it is not already there.<a name="line.292"></a> -<span class="sourceLineNo">293</span> * @param taskname the path of the log to be split<a name="line.293"></a> -<span class="sourceLineNo">294</span> * @param batch the batch this task belongs to<a name="line.294"></a> -<span class="sourceLineNo">295</span> * @return true if a new entry is created, false if it is already there.<a name="line.295"></a> -<span class="sourceLineNo">296</span> */<a name="line.296"></a> -<span class="sourceLineNo">297</span> boolean enqueueSplitTask(String taskname, TaskBatch batch) {<a name="line.297"></a> -<span class="sourceLineNo">298</span> lastTaskCreateTime = EnvironmentEdgeManager.currentTime();<a name="line.298"></a> -<span class="sourceLineNo">299</span> String task = getSplitLogManagerCoordination().prepareTask(taskname);<a name="line.299"></a> -<span class="sourceLineNo">300</span> Task oldtask = createTaskIfAbsent(task, batch);<a name="line.300"></a> -<span class="sourceLineNo">301</span> if (oldtask == null) {<a name="line.301"></a> -<span class="sourceLineNo">302</span> // publish the task in the coordination engine<a name="line.302"></a> -<span class="sourceLineNo">303</span> getSplitLogManagerCoordination().submitTask(task);<a name="line.303"></a> -<span class="sourceLineNo">304</span> return true;<a name="line.304"></a> -<span class="sourceLineNo">305</span> }<a name="line.305"></a> -<span class="sourceLineNo">306</span> return false;<a name="line.306"></a> -<span class="sourceLineNo">307</span> }<a name="line.307"></a> -<span class="sourceLineNo">308</span><a name="line.308"></a> -<span class="sourceLineNo">309</span> private void waitForSplittingCompletion(TaskBatch batch, MonitoredTask status) {<a name="line.309"></a> -<span class="sourceLineNo">310</span> synchronized (batch) {<a name="line.310"></a> -<span class="sourceLineNo">311</span> while ((batch.done + batch.error) != batch.installed) {<a name="line.311"></a> -<span class="sourceLineNo">312</span> try {<a name="line.312"></a> -<span class="sourceLineNo">313</span> status.setStatus("Waiting for distributed tasks to finish. " + " scheduled="<a name="line.313"></a> -<span class="sourceLineNo">314</span> + batch.installed + " done=" + batch.done + " error=" + batch.error);<a name="line.314"></a> -<span class="sourceLineNo">315</span> int remaining = batch.installed - (batch.done + batch.error);<a name="line.315"></a> -<span class="sourceLineNo">316</span> int actual = activeTasks(batch);<a name="line.316"></a> -<span class="sourceLineNo">317</span> if (remaining != actual) {<a name="line.317"></a> -<span class="sourceLineNo">318</span> LOG.warn("Expected " + remaining + " active tasks, but actually there are " + actual);<a name="line.318"></a> -<span class="sourceLineNo">319</span> }<a name="line.319"></a> -<span class="sourceLineNo">320</span> int remainingTasks = getSplitLogManagerCoordination().remainingTasksInCoordination();<a name="line.320"></a> -<span class="sourceLineNo">321</span> if (remainingTasks >= 0 && actual > remainingTasks) {<a name="line.321"></a> -<span class="sourceLineNo">322</span> LOG.warn("Expected at least" + actual + " tasks remaining, but actually there are "<a name="line.322"></a> -<span class="sourceLineNo">323</span> + remainingTasks);<a name="line.323"></a> -<span class="sourceLineNo">324</span> }<a name="line.324"></a> -<span class="sourceLineNo">325</span> if (remainingTasks == 0 || actual == 0) {<a name="line.325"></a> -<span class="sourceLineNo">326</span> LOG.warn("No more task remaining, splitting "<a name="line.326"></a> -<span class="sourceLineNo">327</span> + "should have completed. Remaining tasks is " + remainingTasks<a name="line.327"></a> -<span class="sourceLineNo">328</span> + ", active tasks in map " + actual);<a name="line.328"></a> -<span class="sourceLineNo">329</span> if (remainingTasks == 0 && actual == 0) {<a name="line.329"></a> -<span class="sourceLineNo">330</span> return;<a name="line.330"></a> -<span class="sourceLineNo">331</span> }<a name="line.331"></a> -<span class="sourceLineNo">332</span> }<a name="line.332"></a> -<span class="sourceLineNo">333</span> batch.wait(100);<a name="line.333"></a> -<span class="sourceLineNo">334</span> if (server.isStopped()) {<a name="line.334"></a> -<span class="sourceLineNo">335</span> LOG.warn("Stopped while waiting for log splits to be completed");<a name="line.335"></a> -<span class="sourceLineNo">336</span> return;<a name="line.336"></a> -<span class="sourceLineNo">337</span> }<a name="line.337"></a> -<span class="sourceLineNo">338</span> } catch (InterruptedException e) {<a name="line.338"></a> -<span class="sourceLineNo">339</span> LOG.warn("Interrupted while waiting for log splits to be completed");<a name="line.339"></a> -<span class="sourceLineNo">340</span> Thread.currentThread().interrupt();<a name="line.340"></a> -<span class="sourceLineNo">341</span> return;<a name="line.341"></a> -<span class="sourceLineNo">342</span> }<a name="line.342"></a> -<span class="sourceLineNo">343</span> }<a name="line.343"></a> -<span class="sourceLineNo">344</span> }<a name="line.344"></a> -<span class="sourceLineNo">345</span> }<a name="line.345"></a> -<span class="sourceLineNo">346</span><a name="line.346"></a> -<span class="sourceLineNo">347</span> @VisibleForTesting<a name="line.347"></a> -<span class="sourceLineNo">348</span> ConcurrentMap<String, Task> getTasks() {<a name="line.348"></a> -<span class="sourceLineNo">349</span> return tasks;<a name="line.349"></a> -<span class="sourceLineNo">350</span> }<a name="line.350"></a> -<span class="sourceLineNo">351</span><a name="line.351"></a> -<span class="sourceLineNo">352</span> private int activeTasks(final TaskBatch batch) {<a name="line.352"></a> -<span class="sourceLineNo">353</span> int count = 0;<a name="line.353"></a> -<span class="sourceLineNo">354</span> for (Task t : tasks.values()) {<a name="line.354"></a> -<span class="sourceLineNo">355</span> if (t.batch == batch && t.status == TerminationStatus.IN_PROGRESS) {<a name="line.355"></a> -<span class="sourceLineNo">356</span> count++;<a name="line.356"></a> -<span class="sourceLineNo">357</span> }<a name="line.357"></a> -<span class="sourceLineNo">358</span> }<a name="line.358"></a> -<span class="sourceLineNo">359</span> return count;<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><a name="line.362"></a> -<span class="sourceLineNo">363</span> /**<a name="line.363"></a> -<span class="sourceLineNo">364</span> * @param path<a name="line.364"></a> -<span class="sourceLineNo">365</span> * @param batch<a name="line.365"></a> -<span class="sourceLineNo">366</span> * @return null on success, existing task on error<a name="line.366"></a> -<span class="sourceLineNo">367</span> */<a name="line.367"></a> -<span class="sourceLineNo">368</span> private Task createTaskIfAbsent(String path, TaskBatch batch) {<a name="line.368"></a> -<span class="sourceLineNo">369</span> Task oldtask;<a name="line.369"></a> -<span class="sourceLineNo">370</span> // batch.installed is only changed via this function and<a name="line.370"></a> -<span class="sourceLineNo">371</span> // a single thread touches batch.installed.<a name="line.371"></a> -<span class="sourceLineNo">372</span> Task newtask = new Task();<a name="line.372"></a> -<span class="sourceLineNo">373</span> newtask.batch = batch;<a name="line.373"></a> -<span class="sourceLineNo">374</span> oldtask = tasks.putIfAbsent(path, newtask);<a name="line.374"></a> -<span class="sourceLineNo">375</span> if (oldtask == null) {<a name="line.375"></a> -<span class="sourceLineNo">376</span> batch.installed++;<a name="line.376"></a> -<span class="sourceLineNo">377</span> return null;<a name="line.377"></a> -<span class="sourceLineNo">378</span> }<a name="line.378"></a> -<span class="sourceLineNo">379</span> // new task was not used.<a name="line.379"></a> -<span class="sourceLineNo">380</span> synchronized (oldtask) {<a name="line.380"></a> -<span class="sourceLineNo">381</span> if (oldtask.isOrphan()) {<a name="line.381"></a> -<span class="sourceLineNo">382</span> if (oldtask.status == SUCCESS) {<a name="line.382"></a> -<span class="sourceLineNo">383</span> // The task is already done. Do not install the batch for this<a name="line.383"></a> -<span class="sourceLineNo">384</span> // task because it might be too late for setDone() to update<a name="line.384"></a> -<span class="sourceLineNo">385</span> // batch.done. There is no need for the batch creator to wait for<a name="line.385"></a> -<span class="sourceLineNo">386</span> // this task to complete.<a name="line.386"></a> -<span class="sourceLineNo">387</span> return (null);<a name="line.387"></a> -<span class="sourceLineNo">388</span> }<a name="line.388"></a> -<span class="sourceLineNo">389</span> if (oldtask.status == IN_PROGRESS) {<a name="line.389"></a> -<span class="sourceLineNo">390</span> oldtask.batch = batch;<a name="line.390"></a> -<span class="sourceLineNo">391</span> batch.installed++;<a name="line.391"></a> -<span class="sourceLineNo">392</span> LOG.debug("Previously orphan task " + path + " is now being waited upon");<a name="line.392"></a> -<span class="sourceLineNo">393</span> return null;<a name="line.393"></a> +<span class="sourceLineNo">175</span> FileStatus[] logfiles = FSUtils.listStatus(fs, logDir, filter);<a name="line.175"></a> +<span class="sourceLineNo">176</span> if (logfiles == null || logfiles.length == 0) {<a name="line.176"></a> +<span class="sourceLineNo">177</span> LOG.info(logDir + " is empty dir, no logs to split");<a name="line.177"></a> +<span class="sourceLineNo">178</span> } else {<a name="line.178"></a> +<span class="sourceLineNo">179</span> Collections.addAll(fileStatus, logfiles);<a name="line.179"></a> +<span class="sourceLineNo">180</span> }<a name="line.180"></a> +<span class="sourceLineNo">181</span> }<a name="line.181"></a> +<span class="sourceLineNo">182</span> FileStatus[] a = new FileStatus[fileStatus.size()];<a name="line.182"></a> +<span class="sourceLineNo">183</span> return fileStatus.toArray(a);<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> * @param logDir one region sever wal dir path in .logs<a name="line.187"></a> +<span class="sourceLineNo">188</span> * @throws IOException if there was an error while splitting any log file<a name="line.188"></a> +<span class="sourceLineNo">189</span> * @return cumulative size of the logfiles split<a name="line.189"></a> +<span class="sourceLineNo">190</span> * @throws IOException<a name="line.190"></a> +<span class="sourceLineNo">191</span> */<a name="line.191"></a> +<span class="sourceLineNo">192</span> public long splitLogDistributed(final Path logDir) throws IOException {<a name="line.192"></a> +<span class="sourceLineNo">193</span> List<Path> logDirs = new ArrayList<>();<a name="line.193"></a> +<span class="sourceLineNo">194</span> logDirs.add(logDir);<a name="line.194"></a> +<span class="sourceLineNo">195</span> return splitLogDistributed(logDirs);<a name="line.195"></a> +<span class="sourceLineNo">196</span> }<a name="line.196"></a> +<span class="sourceLineNo">197</span><a name="line.197"></a> +<span class="sourceLineNo">198</span> /**<a name="line.198"></a> +<span class="sourceLineNo">199</span> * The caller will block until all the log files of the given region server have been processed -<a name="line.199"></a> +<span class="sourceLineNo">200</span> * successfully split or an error is encountered - by an available worker region server. This<a name="line.200"></a> +<span class="sourceLineNo">201</span> * method must only be called after the region servers have been brought online.<a name="line.201"></a> +<span class="sourceLineNo">202</span> * @param logDirs List of log dirs to split<a name="line.202"></a> +<span class="sourceLineNo">203</span> * @throws IOException If there was an error while splitting any log file<a name="line.203"></a> +<span class="sourceLineNo">204</span> * @return cumulative size of the logfiles split<a name="line.204"></a> +<span class="sourceLineNo">205</span> */<a name="line.205"></a> +<span class="sourceLineNo">206</span> public long splitLogDistributed(final List<Path> logDirs) throws IOException {<a name="line.206"></a> +<span class="sourceLineNo">207</span> if (logDirs.isEmpty()) {<a name="line.207"></a> +<span class="sourceLineNo">208</span> return 0;<a name="line.208"></a> +<span class="sourceLineNo">209</span> }<a name="line.209"></a> +<span class="sourceLineNo">210</span> Set<ServerName> serverNames = new HashSet<>();<a name="line.210"></a> +<span class="sourceLineNo">211</span> for (Path logDir : logDirs) {<a name="line.211"></a> +<span class="sourceLineNo">212</span> try {<a name="line.212"></a> +<span class="sourceLineNo">213</span> ServerName serverName = AbstractFSWALProvider.getServerNameFromWALDirectoryName(logDir);<a name="line.213"></a> +<span class="sourceLineNo">214</span> if (serverName != null) {<a name="line.214"></a> +<span class="sourceLineNo">215</span> serverNames.add(serverName);<a name="line.215"></a> +<span class="sourceLineNo">216</span> }<a name="line.216"></a> +<span class="sourceLineNo">217</span> } catch (IllegalArgumentException e) {<a name="line.217"></a> +<span class="sourceLineNo">218</span> // ignore invalid format error.<a name="line.218"></a> +<span class="sourceLineNo">219</span> LOG.warn("Cannot parse server name from " + logDir);<a name="line.219"></a> +<span class="sourceLineNo">220</span> }<a name="line.220"></a> +<span class="sourceLineNo">221</span> }<a name="line.221"></a> +<span class="sourceLineNo">222</span> return splitLogDistributed(serverNames, logDirs, null);<a name="line.222"></a> +<span class="sourceLineNo">223</span> }<a name="line.223"></a> +<span class="sourceLineNo">224</span><a name="line.224"></a> +<span class="sourceLineNo">225</span> /**<a name="line.225"></a> +<span class="sourceLineNo">226</span> * The caller will block until all the hbase:meta log files of the given region server have been<a name="line.226"></a> +<span class="sourceLineNo">227</span> * processed - successfully split or an error is encountered - by an available worker region<a name="line.227"></a> +<span class="sourceLineNo">228</span> * server. This method must only be called after the region servers have been brought online.<a name="line.228"></a> +<span class="sourceLineNo">229</span> * @param logDirs List of log dirs to split<a name="line.229"></a> +<span class="sourceLineNo">230</span> * @param filter the Path filter to select specific files for considering<a name="line.230"></a> +<span class="sourceLineNo">231</span> * @throws IOException If there was an error while splitting any log file<a name="line.231"></a> +<span class="sourceLineNo">232</span> * @return cumulative size of the logfiles split<a name="line.232"></a> +<span class="sourceLineNo">233</span> */<a name="line.233"></a> +<span class="sourceLineNo">234</span> public long splitLogDistributed(final Set<ServerName> serverNames, final List<Path> logDirs,<a name="line.234"></a> +<span class="sourceLineNo">235</span> PathFilter filter) throws IOException {<a name="line.235"></a> +<span class="sourceLineNo">236</span> MonitoredTask status = TaskMonitor.get().createStatus("Doing distributed log split in " +<a name="line.236"></a> +<span class="sourceLineNo">237</span> logDirs + " for serverName=" + serverNames);<a name="line.237"></a> +<span class="sourceLineNo">238</span> FileStatus[] logfiles = getFileList(logDirs, filter);<a name="line.238"></a> +<span class="sourceLineNo">239</span> status.setStatus("Checking directory contents...");<a name="line.239"></a> +<span class="sourceLineNo">240</span> SplitLogCounters.tot_mgr_log_split_batch_start.increment();<a name="line.240"></a> +<span class="sourceLineNo">241</span> LOG.info("Started splitting " + logfiles.length + " logs in " + logDirs +<a name="line.241"></a> +<span class="sourceLineNo">242</span> " for " + serverNames);<a name="line.242"></a> +<span class="sourceLineNo">243</span> long t = EnvironmentEdgeManager.currentTime();<a name="line.243"></a> +<span class="sourceLineNo">244</span> long totalSize = 0;<a name="line.244"></a> +<span class="sourceLineNo">245</span> TaskBatch batch = new TaskBatch();<a name="line.245"></a> +<span class="sourceLineNo">246</span> for (FileStatus lf : logfiles) {<a name="line.246"></a> +<span class="sourceLineNo">247</span> // TODO If the log file is still being written to - which is most likely<a name="line.247"></a> +<span class="sourceLineNo">248</span> // the case for the last log file - then its length will show up here<a name="line.248"></a> +<span class="sourceLineNo">249</span> // as zero. The size of such a file can only be retrieved after<a name="line.249"></a> +<span class="sourceLineNo">250</span> // recover-lease is done. totalSize will be under in most cases and the<a name="line.250"></a> +<span class="sourceLineNo">251</span> // metrics that it drives will also be under-reported.<a name="line.251"></a> +<span class="sourceLineNo">252</span> totalSize += lf.getLen();<a name="line.252"></a> +<span class="sourceLineNo">253</span> String pathToLog = FSUtils.removeWALRootPath(lf.getPath(), conf);<a name="line.253"></a> +<span class="sourceLineNo">254</span> if (!enqueueSplitTask(pathToLog, batch)) {<a name="line.254"></a> +<span class="sourceLineNo">255</span> throw new IOException("duplicate log split scheduled for " + lf.getPath());<a name="line.255"></a> +<span class="sourceLineNo">256</span> }<a name="line.256"></a> +<span class="sourceLineNo">257</span> }<a name="line.257"></a> +<span class="sourceLineNo">258</span> waitForSplittingCompletion(batch, status);<a name="line.258"></a> +<span class="sourceLineNo">259</span><a name="line.259"></a> +<span class="sourceLineNo">260</span> if (batch.done != batch.installed) {<a name="line.260"></a> +<span class="sourceLineNo">261</span> batch.isDead = true;<a name="line.261"></a> +<span class="sourceLineNo">262</span> SplitLogCounters.tot_mgr_log_split_batch_err.increment();<a name="line.262"></a> +<span class="sourceLineNo">263</span> LOG.warn("error while splitting logs in " + logDirs + " installed = " + batch.installed<a name="line.263"></a> +<span class="sourceLineNo">264</span> + " but only " + batch.done + " done");<a name="line.264"></a> +<span class="sourceLineNo">265</span> String msg = "error or interrupted while splitting logs in " + logDirs + " Task = " + batch;<a name="line.265"></a> +<span class="sourceLineNo">266</span> status.abort(msg);<a name="line.266"></a> +<span class="sourceLineNo">267</span> throw new IOException(msg);<a name="line.267"></a> +<span class="sourceLineNo">268</span> }<a name="line.268"></a> +<span class="sourceLineNo">269</span> for (Path logDir : logDirs) {<a name="line.269"></a> +<span class="sourceLineNo">270</span> status.setStatus("Cleaning up log directory...");<a name="line.270"></a> +<span class="sourceLineNo">271</span> final FileSystem fs = logDir.getFileSystem(conf);<a name="line.271"></a> +<span class="sourceLineNo">272</span> try {<a name="line.272"></a> +<span class="sourceLineNo">273</span> if (fs.exists(logDir) && !fs.delete(logDir, false)) {<a name="line.273"></a> +<span class="sourceLineNo">274</span> LOG.warn("Unable to delete log src dir. Ignoring. " + logDir);<a name="line.274"></a> +<span class="sourceLineNo">275</span> }<a name="line.275"></a> +<span class="sourceLineNo">276</span> } catch (IOException ioe) {<a name="line.276"></a> +<span class="sourceLineNo">277</span> FileStatus[] files = fs.listStatus(logDir);<a name="line.277"></a> +<span class="sourceLineNo">278</span> if (files != null && files.length > 0) {<a name="line.278"></a> +<span class="sourceLineNo">279</span> LOG.warn("Returning success without actually splitting and "<a name="line.279"></a> +<span class="sourceLineNo">280</span> + "deleting all the log files in path " + logDir + ": "<a name="line.280"></a> +<span class="sourceLineNo">281</span> + Arrays.toString(files), ioe);<a name="line.281"></a> +<span class="sourceLineNo">282</span> } else {<a name="line.282"></a> +<span class="sourceLineNo">283</span> LOG.warn("Unable to delete log src dir. Ignoring. " + logDir, ioe);<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> SplitLogCounters.tot_mgr_log_split_batch_success.increment();<a name="line.286"></a> +<span class="sourceLineNo">287</span> }<a name="line.287"></a> +<span class="sourceLineNo">288</span> String msg =<a name="line.288"></a> +<span class="sourceLineNo">289</span> "finished splitting (more than or equal to) " + totalSize + " bytes in " + batch.installed<a name="line.289"></a> +<span class="sourceLineNo">290</span> + " log files in " + logDirs + " in "<a name="line.290"></a> +<span class="sourceLineNo">291</span> + (EnvironmentEdgeManager.currentTime() - t) + "ms";<a name="line.291"></a> +<span class="sourceLineNo">292</span> status.markComplete(msg);<a name="line.292"></a> +<span class="sourceLineNo">293</span> LOG.info(msg);<a name="line.293"></a> +<span class="sourceLineNo">294</span> return totalSize;<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> * Add a task entry to coordination if it is not already there.<a name="line.298"></a> +<span class="sourceLineNo">299</span> * @param taskname the path of the log to be split<a name="line.299"></a> +<span class="sourceLineNo">300</span> * @param batch the batch this task belongs to<a name="line.300"></a> +<span class="sourceLineNo">301</span> * @return true if a new entry is created, false if it is already there.<a name="line.301"></a> +<span class="sourceLineNo">302</span> */<a name="line.302"></a> +<span class="sourceLineNo">303</span> boolean enqueueSplitTask(String taskname, TaskBatch batch) {<a name="line.303"></a> +<span class="sourceLineNo">304</span> lastTaskCreateTime = EnvironmentEdgeManager.currentTime();<a name="line.304"></a> +<span class="sourceLineNo">305</span> String task = getSplitLogManagerCoordination().prepareTask(taskname);<a name="line.305"></a> +<span class="sourceLineNo">306</span> Task oldtask = createTaskIfAbsent(task, batch);<a name="line.306"></a> +<span class="sourceLineNo">307</span> if (oldtask == null) {<a name="line.307"></a> +<span class="sourceLineNo">308</span> // publish the task in the coordination engine<a name="line.308"></a> +<span class="sourceLineNo">309</span> getSplitLogManagerCoordination().submitTask(task);<a name="line.309"></a> +<span class="sourceLineNo">310</span> return true;<a name="line.310"></a> +<span class="sourceLineNo">311</span> }<a name="line.311"></a> +<span class="sourceLineNo">312</span> return false;<a name="line.312"></a> +<span class="sourceLineNo">313</span> }<a name="line.313"></a> +<span class="sourceLineNo">314</span><a name="line.314"></a> +<span class="sourceLineNo">315</span> private void waitForSplittingCompletion(TaskBatch batch, MonitoredTask status) {<a name="line.315"></a> +<span class="sourceLineNo">316</span> synchronized (batch) {<a name="line.316"></a> +<span class="sourceLineNo">317</span> while ((batch.done + batch.error) != batch.installed) {<a name="line.317"></a> +<span class="sourceLineNo">318</span> try {<a name="line.318"></a> +<span class="sourceLineNo">319</span> status.setStatus("Waiting for distributed tasks to finish. " + " scheduled="<a name="line.319"></a> +<span class="sourceLineNo">320</span> + batch.installed + " done=" + batch.done + " error=" + batch.error);<a name="line.320"></a> +<span class="sourceLineNo">321</span> int remaining = batch.installed - (batch.done + batch.error);<a name="line.321"></a> +<span class="sourceLineNo">322</span> int actual = activeTasks(batch);<a name="line.322"></a> +<span class="sourceLineNo">323</span> if (remaining != actual) {<a name="line.323"></a> +<span class="sourceLineNo">324</span> LOG.warn("Expected " + remaining + " active tasks, but actually there are " + actual);<a name="line.324"></a> +<span class="sourceLineNo">325</span> }<a name="line.325"></a> +<span class="sourceLineNo">326</span> int remainingTasks = getSplitLogManagerCoordination().remainingTasksInCoordination();<a name="line.326"></a> +<span class="sourceLineNo">327</span> if (remainingTasks >= 0 && actual > remainingTasks) {<a name="line.327"></a> +<span class="sourceLineNo">328</span> LOG.warn("Expected at least" + actual + " tasks remaining, but actually there are "<a name="line.328"></a> +<span class="sourceLineNo">329</span> + remainingTasks);<a name="line.329"></a> +<span class="sourceLineNo">330</span> }<a name="line.330"></a> +<span class="sourceLineNo">331</span> if (remainingTasks == 0 || actual == 0) {<a name="line.331"></a> +<span class="sourceLineNo">332</span> LOG.warn("No more task remaining, splitting "<a name="line.332"></a> +<span class="sourceLineNo">333</span> + "should have completed. Remaining tasks is " + remainingTasks<a name="line.333"></a> +<span class="sourceLineNo">334</span> + ", active tasks in map " + actual);<a name="line.334"></a> +<span class="sourceLineNo">335</span> if (remainingTasks == 0 && actual == 0) {<a name="line.335"></a> +<span class="sourceLineNo">336</span> return;<a name="line.336"></a> +<span class="sourceLineNo">337</span> }<a name="line.337"></a> +<span class="sourceLineNo">338</span> }<a name="line.338"></a> +<span class="sourceLineNo">339</span> batch.wait(100);<a name="line.339"></a> +<span class="sourceLineNo">340</span> if (server.isStopped()) {<a name="line.340"></a> +<span class="sourceLineNo">341</span> LOG.warn("Stopped while waiting for log splits to be completed");<a name="line.341"></a> +<span class="sourceLineNo">342</span> return;<a name="line.342"></a> +<span class="sourceLineNo">343</span> }<a name="line.343"></a> +<span class="sourceLineNo">344</span> } catch (InterruptedException e) {<a name="line.344"></a> +<span class="sourceLineNo">345</span> LOG.warn("Interrupted while waiting for log splits to be completed");<a name="line.345"></a> +<span class="sourceLineNo">346</span> Thread.currentThread().interrupt();<a name="line.346"></a> +<span class="sourceLineNo">347</span> return;<a name="line.347"></a> +<span class="sourceLineNo">348</span> }<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> }<a name="line.351"></a> +<span class="sourceLineNo">352</span><a name="line.352"></a> +<span class="sourceLineNo">353</span> @VisibleForTesting<a name="line.353"></a> +<span class="sourceLineNo">354</span> ConcurrentMap<String, Task> getTasks() {<a name="line.354"></a> +<span class="sourceLineNo">355</span> return tasks;<a name="line.355"></a> +<span class="sourceLineNo">356</span> }<a name="line.356"></a> +<span class="sourceLineNo">357</span><a name="line.357"></a> +<span class="sourceLineNo">358</span> private int activeTasks(final TaskBatch batch) {<a name="line.358"></a> +<span class="sourceLineNo">359</span> int count = 0;<a name="line.359"></a> +<span class="sourceLineNo">360</span> for (Task t : tasks.values()) {<a name="line.360"></a> +<span class="sourceLineNo">361</span> if (t.batch == batch && t.status == TerminationStatus.IN_PROGRESS) {<a name="line.361"></a> +<span class="sourceLineNo">362</span> count++;<a name="line.362"></a> +<span class="sourceLineNo">363</span> }<a name="line.363"></a> +<span class="sourceLineNo">364</span> }<a name="line.364"></a> +<span class="sourceLineNo">365</span> return count;<a name="line.365"></a> +<span class="sourceLineNo">366</span><a name="line.366"></a> +<span class="sourceLineNo">367</span> }<a name="line.367"></a> +<span class="sourceLineNo">368</span><a name="line.368"></a> +<span class="sourceLineNo">369</span> /**<a name="line.369"></a> +<span class="sourceLineNo">370</span> * @param path<a name="line.370"></a> +<span class="sourceLineNo">371</span> * @param batch<a name="line.371"></a> +<span class="sourceLineNo">372</span> * @return null on success, existing task on error<a name="line.372"></a> +<span class="sourceLineNo">373</span> */<a name="line.373"></a> +<span class="sourceLineNo">374</span> private Task createTaskIfAbsent(String path, TaskBatch batch) {<a name="line.374"></a> +<span class="sourceLineNo">375</span> Task oldtask;<a name="line.375"></a> +<span class="sourceLineNo">376</span> // batch.installed is only changed via this function and<a name="line.376"></a> +<span class="sourceLineNo">377</span> // a single thread touches batch.installed.<a name="line.377"></a> +<span class="sourceLineNo">378</span> Task newtask = new Task();<a name="line.378"></a> +<span class="sourceLineNo">379</span> newtask.batch = batch;<a name="line.379"></a> +<span class="sourceLineNo">380</span> oldtask = tasks.putIfAbsent(path, newtask);<a name="line.380"></a> +<span class="sourceLineNo">381</span> if (oldtask == null) {<a name="line.381"></a> +<span class="sourceLineNo">382</span> batch.installed++;<a name="line.382"></a> +<span class="sourceLineNo">383</span> return null;<a name="line.383"></a> +<span class="sourceLineNo">384</span> }<a name="line.384"></a> +<span class="sourceLineNo">385</span> // new task was not used.<a name="line.385"></a> +<span class="sourceLineNo">386</span> synchronized (oldtask) {<a name="line.386"></a> +<span class="sourceLineNo">387</span> if (oldtask.isOrphan()) {<a name="line.387"></a> +<span class="sourceLineNo">388</span> if (oldtask.status == SUCCESS) {<a name="line.388"></a> +<span class="sourceLineNo">389</span> // The task is already done. Do not install the batch for this<a name="line.389"></a> +<span class="sourceLineNo">390</span> // task because it might be too late for setDone() to update<a name="line.390"></a> +<span class="sourceLineNo">391</span> // batch.done. There is no need for the batch creator to wait for<a name="line.391"></a> +<span class="sourceLineNo">392</span> // this task to complete.<a name="line.392"></a> +<span class="sourceLineNo">393</span> return (null);<a name="line.393"></a> <span class="sourceLineNo">394</span> }<a name="line.394"></a> -<span class="sourceLineNo">395</span> while (oldtask.status == FAILURE) {<a name="line.395"></a> -<span class="sourceLineNo">396</span> LOG.debug("wait for status of task " + path + " to change to DELETED");<a name="line.396"></a> -<span class="sourceLineNo">397</span> SplitLogCounters.tot_mgr_wait_for_zk_delete.increment();<a name="line.397"></a> -<span class="sourceLineNo">398</span> try {<a name="line.398"></a> -<span class="sourceLineNo">399</span> oldtask.wait();<a name="line.399"></a> -<span class="sourceLineNo">400</span> } catch (InterruptedException e) {<a name="line.400"></a> -<span class="sourceLineNo">401</span> Thread.currentThread().interrupt();<a name="line.401"></a> -<span class="sourceLineNo">402</span> LOG.warn("Interrupted when waiting for znode delete callback");<a name="line.402"></a> -<span class="sourceLineNo">403</span> // fall through to return failure<a name="line.403"></a> -<span class="sourceLineNo">404</span> break;<a name="line.404"></a> -<span class="sourceLineNo">405</span> }<a name="line.405"></a> -<span class="sourceLineNo">406</span> }<a name="line.406"></a> -<span class="sourceLineNo">407</span> if (oldtask.status != DELETED) {<a name="line.407"></a> -<span class="sourceLineNo">408</span> LOG.warn("Failure because previously failed task"<a name="line.408"></a> -<span class="sourceLineNo">409</span> + " state still present. Waiting for znode delete callback" + " path=" + path);<a name="line.409"></a> -<span class="sourceLineNo">410</span> return oldtask;<a name="line.410"></a> -<span class="sourceLineNo">411</span> }<a name="line.411"></a> -<span class="sourceLineNo">412</span> // reinsert the newTask and it must succeed this time<a name="line.412"></a> -<span class="sourceLineNo">413</span> Task t = tasks.putIfAbsent(path, newtask);<a name="line.413"></a> -<span class="sourceLineNo">414</span> if (t == null) {<a name="line.414"></a> -<span class="sourceLineNo">415</span> batch.installed++;<a name="line.415"></a> -<span class="sourceLineNo">416</span> return null;<a name="line.416"></a> +<span class="sourceLineNo">395</span> if (oldtask.status == IN_PROGRESS) {<a name="line.395"></a> +<span class="sourceLineNo">396</span> oldtask.batch = batch;<a name="line.396"></a> +<span class="sourceLineNo">397</span> batch.installed++;<a name="line.397"></a> +<span class="sourceLineNo">398</span> LOG.debug("Previously orphan task " + path + " is now being waited upon");<a name="line.398"></a> +<span class="sourceLineNo">399</span> return null;<a name="line.399"></a> +<span class="sourceLineNo">400</span> }<a name="line.400"></a> +<span class="sourceLineNo">401</span> while (oldtask.status == FAILURE) {<a name="line.401"></a> +<span class="sourceLineNo">402</span> LOG.debug("wait for status of task " + path + " to change to DELETED");<a name="line.402"></a> +<span class="sourceLineNo">403</span> SplitLogCounters.tot_mgr_wait_for_zk_delete.increment();<a name="line.403"></a> +<span class="sourceLineNo">404</span> try {<a name="line.404"></a> +<span class="sourceLineNo">405</span> oldtask.wait();<a name="line.405"></a> +<span class="sourceLineNo">406</span> } catch (InterruptedException e) {<a name="line.406"></a> +<span class="sourceLineNo">407</span> Thread.currentThread().interrupt();<a name="line.407"></a> +<span class="sourceLineNo">408</span> LOG.warn("Interrupted when waiting for znode delete callback");<a name="line.408"></a> +<span class="sourceLineNo">409</span> // fall through to return failure<a name="line.409"></a> +<span class="sourceLineNo">410</span> break;<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> if (oldtask.status != DELETED) {<a name="line.413"></a> +<span class="sourceLineNo">414</span> LOG.warn("Failure because previously failed task"<a name="line.414"></a> +<span class="sourceLineNo">415</span> + " state still present. Waiting for znode delete callback" + " path=" + path);<a name="line.415"></a> +<span class="sourceLineNo">416</span> return oldtask;<a name="line.416"></a> <span class="sourceLineNo">417</span> }<a name="line.417"></a> -<span class="sourceLineNo">418</span> LOG.error(HBaseMarkers.FATAL, "Logic error. Deleted task still present in tasks map");<a name="line.418"></a> -<span class="sourceLineNo">419</span> assert false : "Deleted task still present in tasks map";<a name="line.419"></a> -<span class="sourceLineNo">420</span> return t;<a name="line.420"></a> -<span class="sourceLineNo">421</span> }<a name="line.421"></a> -<span class="sourceLineNo">422</span> LOG.warn("Failure because two threads can't wait for the same task; path=" + path);<a name="line.422"></a> -<span class="sourceLineNo">423</span> return oldtask;<a name="line.423"></a> -<span class="sourceLineNo">424</span> }<a name="line.424"></a> -<span class="sourceLineNo">425</span> }<a name="line.425"></a> -<span class="sourceLineNo">426</span><a name="line.426"></a> -<span class="sourceLineNo">427</span> public void stop() {<a name="line.427"></a> -<span class="sourceLineNo">428</span> if (choreService != null) {<a name="line.428"></a> -<span class="sourceLineNo">429</span> choreService.shutdown();<a name="line.429"></a> +<span class="sourceLineNo">418</span> // reinsert the newTask and it must succeed this time<a name="line.418"></a> +<span class="sourceLineNo">419</span> Task t = tasks.putIfAbsent(path, newtask);<a name="line.419"></a> +<span class="sourceLineNo">420</span> if (t == null) {<a name="line.420"></a> +<span class="sourceLineNo">421</span> batch.installed++;<a name="line.421"></a> +<span class="sourceLineNo">422</span> return null;<a name="line.422"></a> +<span class="sourceLineNo">423</span> }<a name="line.423"></a> +<span class="sourceLineNo">424</span> LOG.error(HBaseMarkers.FATAL, "Logic error. Deleted task still present in tasks map");<a name="line.424"></a> +<span class="sourceLineNo">425</span> assert false : "Deleted task still present in tasks map";<a name="line.425"></a> +<span class="sourceLineNo">426</span> return t;<a name="line.426"></a> +<span class="sourceLineNo">427</span> }<a name="line.427"></a> +<span class="sourceLineNo">428</span> LOG.warn("Failure because two threads can't wait for the same task; path=" + path);<a name="line.428"></a> +<span class="sourceLineNo">429</span> return oldtask;<a name="line.429"></a> <span class="sourceLineNo">430</span> }<a name="line.430"></a> -<span class="sourceLineNo">431</span> if (timeoutMonitor != null) {<a name="line.431"></a> -<span class="sourceLineNo">432</span> timeoutMonitor.cancel(true);<a name="line.432"></a> -<span class="sourceLineNo">433</span> }<a name="line.433"></a> -<span class="sourceLineNo">434</span> }<a name="line.434"></a> -<span class="sourceLineNo">435</span><a name="line.435"></a> -<span class="sourceLineNo">436</span> void handleDeadWorker(ServerName workerName) {<a name="line.436"></a> -<span class="sourceLineNo">437</span> // resubmit the tasks on the TimeoutMonitor thread. Makes it easier<a name="line.437"></a> -<span class="sourceLineNo">438</span> // to reason about concurrency. Makes it easier to retry.<a name="line.438"></a> -<span class="sourceLineNo">439</span> synchronized (deadWorkersLock) {<a name="line.439"></a> -<span class="sourceLineNo">440</span> if (deadWorkers == null) {<a name="line.440"></a> -<span class="sourceLineNo">441</span> deadWorkers = new HashSet<>(100);<a name="line.441"></a> -<span class="sourceLineNo">442</span> }<a name="line.442"></a> -<span class="sourceLineNo">443</span> deadWorkers.add(workerName);<a name="line.443"></a> -<span class="sourceLineNo">444</span> }<a name="line.444"></a> -<span class="sourceLineNo">445</span> LOG.info("dead splitlog worker " + workerName);<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> void handleDeadWorkers(Set<ServerName> serverNames) {<a name="line.448"></a> -<span class="sourceLineNo">449</span> synchronized (deadWorkersLock) {<a name="line.449"></a> -<span class="sourceLineNo">450</span> if (deadWorkers == null) {<a name="line.450"></a> -<span class="sourceLineNo">451</span> deadWorkers = new HashSet<>(100);<a name="line.451"></a> -<span class="sourceLineNo">452</span> }<a name="line.452"></a> -<span class="sourceLineNo">453</span> deadWorkers.addAll(serverNames);<a name="line.453"></a> -<span class="sourceLineNo">454</span> }<a name="line.454"></a> -<span class="sourceLineNo">455</span> LOG.info("dead splitlog workers " + serverNames);<a name="line.455"></a> -<span class="sourceLineNo">456</span> }<a name="line.456"></a> -<span class="sourceLineNo">457</span><a name="line.457"></a> -<span class="sourceLineNo">458</span> /**<a name="line.458"></a> -<span class="sourceLineNo">459</span> * Keeps track of the batch of tasks submitted together by a caller in splitLogDistributed().<a name="line.459"></a> -<span class="sourceLineNo">460</span> * Clients threads use this object to wait for all their tasks to be done.<a name="line.460"></a> -<span class="sourceLineNo">461</span> * <p><a name="line.461"></a> -<span class="sourceLineNo">462</span> * All access is synchronized.<a name="line.462"></a> -<span class="sourceLineNo">463</span> */<a name="line.463"></a> -<span class="sourceLineNo">464</span> @InterfaceAudience.Private<a name="line.464"></a> -<span class="sourceLineNo">465</span> public static class TaskBatch {<a name="line.465"></a> -<span class="sourceLineNo">466</span> public int installed = 0;<a name="line.466"></a> -<span class="sourceLineNo">467</span> public int done = 0;<a name="line.467"></a> -<span class="sourceLineNo">468</span> public int error = 0;<a name="line.468"></a> -<span class="sourceLineNo">469</span> public volatile boolean isDead = false;<a name="line.469"></a> -<span class="sourceLineNo">470</span><a name="line.470"></a> -<span class="sourceLineNo">471</span> @Override<a name="line.471"></a> -<span class="sourceLineNo">472</span> public String toString() {<a name="line.472"></a> -<span class="sourceLineNo">473</span> return ("installed = " + installed + " done = " + done + " error = " + error);<a name="line.473"></a> -<span class="sourceLineNo">474</span> }<a name="line.474"></a> -<span class="sourceLineNo">475</span> }<a name="line.475"></a> +<span class="sourceLineNo">431</span> }<a name="line.431"></a> +<span class="sourceLineNo">432</span><a name="line.432"></a> +<span class="sourceLineNo">433</span> public void stop() {<a name="line.433"></a> +<span class="sourceLineNo">434</span> if (choreService != null) {<a name="line.434"></a> +<span class="sourceLineNo">435</span> choreService.shutdown();<a name="line.435"></a> +<span class="sourceLineNo">436</span> }<a name="line.436"></a> +<span class="sourceLineNo">437</span> if (timeoutMonitor != null) {<a name="line.437"></a> +<span class="sourceLineNo">438</span> timeoutMonitor.cancel(true);<a name="line.438"></a> +<span class="sourceLineNo">439</span> }<a name="line.439"></a> +<span class="sourceLineNo">440</span> }<a name="line.440"></a> +<span class="sourceLineNo">441</span><a name="line.441"></a> +<span class="sourceLineNo">442</span> void handleDeadWorker(ServerName workerName) {<a name="line.442"></a> +<span class="sourceLineNo">443</span> // resubmit the tasks on the TimeoutMonitor thread. Makes it easier<a name="line.443"></a> +<span class="sourceLineNo">444</span> // to reason about concurrency. Makes it easier to retry.<a name="line.444"></a> +<span class="sourceLineNo">445</span> synchronized (deadWorkersLock) {<a name="line.445"></a> +<span class="sourceLineNo">446</span> if (deadWorkers == null) {<a name="line.446"></a> +<span class="sourceLineNo">447</span> deadWorkers = new HashSet<>(100);<a name="line.447"></a> +<span class="sourceLineNo">448</span> }<a name="line.448"></a> +<span class="sourceLineNo">449</span> deadWorkers.add(workerName);<a name="line.449"></a> +<span class="sourceLineNo">450</span> }<a name="line.450"></a> +<span class="sourceLineNo">451</span> LOG.info("dead splitlog worker " + workerName);<a name="line.451"></a> +<span class="sourceLineNo">452</span> }<a name="line.452"></a> +<span class="sourceLineNo">453</span><a name="line.453"></a> +<span class="sourceLineNo">454</span> void handleDeadWorkers(Set<ServerName> serverNames) {<a name="line.454"></a> +<span class="sourceLineNo">455</span> synchronized (deadWorkersLock) {<a name="line.455"></a> +<span class="sourceLineNo">456</span> if (deadWorkers == null) {<a name="line.456"></a> +<span class="sourceLineNo">457</span> deadWorkers = new HashSet<>(100);<a name="line.457"></a> +<span class="sourceLineNo">458</span> }<a name="line.458"></a> +<span class="sourceLineNo">459</span> deadWorkers.addAll(serverNames);<a name="line.459"></a> +<span class="sourceLineNo">460</span> }<a name="line.460"></a> +<span class="sourceLineNo">461</span> LOG.info("dead splitlog workers " + serverNames);<a name="line.461"></a> +<span class="sourceLineNo">462</span> }<a name="line.462"></a> +<span class="sourceLineNo">463</span><a name="line.463"></a> +<span class="sourceLineNo">464</span> /**<a name="line.464"></a> +<span class="sourceLineNo">465</span> * Keeps track of the batch of tasks submitted together by a caller in splitLogDistributed().<a name="line.465"></a> +<span class="sourceLineNo">466</span> * Clients threads use this object to wait for all their tasks to be done.<a name="line.466"></a> +<span class="sourceLineNo">467</span> * <p><a name="line.467"></a> +<span class="sourceLineNo">468</span> * All access is synchronized.<a name="line.468"></a> +<span class="sourceLineNo">469</span> */<a name="line.469"></a> +<span class="sourceLineNo">470</span> @InterfaceAudience.Private<a name="line.470"></a> +<span class="sourceLineNo">471</span> public static class TaskBatch {<a name="line.471"></a> +<span class="sourceLineNo">472</span> public int installed = 0;<a name="line.472"></a> +<span class="sourceLineNo">473</span> public int done = 0;<a name="line.473"></a> +<span class="sourceLineNo">474</span> public int error = 0;<a name="line.474"></a> +<span class="sourceLineNo">475</span> public volatile boolean isDead = false;<a name="line.475"></a> <span class="sourceLineNo">476</span><a name="line.476"></a> -<span class="sourceLineNo">477</span> /**<a name="line.477"></a> -<span class="sourceLineNo">478</span> * in memory state of an active task.<a name="line.478"></a> -<span class="sourceLineNo">479</span> */<a name="line.479"></a> -<span class="sourceLineNo">480</span> @InterfaceAudience.Private<a name="line.480"></a> -<span class="sourceLineNo">481</span> public static class Task {<a name="line.481"></a> -<span class="sourceLineNo">482</span> public volatile long last_update;<a name="line.482"></a> -<span class="sourceLineNo">483</span> public volatile int last_version;<a name="line.483"></a> -<span class="sourceLineNo">484</span> public volatile ServerName cur_worker_name;<a name="line.484"></a> -<span class="sourceLineNo">485</span> public volatile TaskBatch batch;<a name="line.485"></a> -<span class="sourceLineNo">486</span> public volatile TerminationStatus status;<a name="line.486"></a> -<span class="sourceLineNo">487</span> public volatile AtomicInteger incarnation = new AtomicInteger(0);<a name="line.487"></a> -<span class="sourceLineNo">488</span> public final AtomicInteger unforcedResubmits = new AtomicInteger();<a name="line.488"></a> -<span class="sourceLineNo">489</span> public volatile boolean resubmitThresholdReached;<a name="line.489"></a> -<span class="sourceLineNo">490</span><a name="line.490"></a> -<span class="sourceLineNo">491</span> @Override<a name="line.491"></a> -<span class="sourceLineNo">492</span> public String toString() {<a name="line.492"></a> -<span class="sourceLineNo">493</span> return ("last_update = " + last_update + " last_version = " + last_version<a name="line.493"></a> -<span class="sourceLineNo">494</span> + " cur_worker_name = " + cur_worker_name + " status = " + status + " incarnation = "<a name="line.494"></a> -<span class="sourceLineNo">495</span> + incarnation + " resubmits = " + unforcedResubmits.get() + " batch = " + batch);<a name="line.495"></a> -<span class="sourceLineNo">496</span> }<a name="line.496"></a> -<span class="sourceLineNo">497</span><a name="line.497"></a> -<span class="sourceLineNo">498</span> public Task() {<a name="line.498"></a> -<span class="sourceLineNo">499</span> last_version = -1;<a name="line.499"></a> -<span class="sourceLineNo">500</span> status = IN_PROGRESS;<a name="line.500"></a> -<span class="sourceLineNo">501</span> setUnassigned();<a name="line.501"></a> +<span class="sourceLineNo">477</span> @Override<a name="line.477"></a> +<span class="sourceLineNo">478</span> public String toString() {<a name="line.478"></a> +<span class="sourceLineNo">479</span> return ("installed = " + installed + " done = " + done + " error = " + error);<a name="line.479"></a> +<span class="sourceLineNo">480</span> }<a name="line.480"></a> +<span class="sourceLineNo">481</span> }<a name="line.481"></a> +<span class="sourceLineNo">482</span><a name="line.482"></a> +<span class="sourceLineNo">483</span> /**<a name="line.483"></a> +<span class="sourceLineNo">484</span> * in memory state of an active task.<a name="line.484"></a> +<span class="sourceLineNo">485</span> */<a name="line.485"></a> +<span class="sourceLineNo">486</span> @InterfaceAudience.Private<a name="line.486"></a> +<span class="sourceLineNo">487</span> public static class Task {<a name="line.487"></a> +<span class="sourceLineNo">488</span> public volatile long last_update;<a name="line.488"></a> +<span class="sourceLineNo">489</span> public volatile int last_version;<a name="line.489"></a> +<span class="sourceLineNo">490</span> public volatile ServerName cur_worker_name;
<TRUNCATED>