http://git-wip-us.apache.org/repos/asf/hbase-site/blob/d1341859/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/store/wal/WALProcedureStore.PushType.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/store/wal/WALProcedureStore.PushType.html b/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/store/wal/WALProcedureStore.PushType.html index e1b183b..b456cd2 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/store/wal/WALProcedureStore.PushType.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/store/wal/WALProcedureStore.PushType.html @@ -53,1338 +53,1354 @@ <span class="sourceLineNo">045</span>import org.apache.hadoop.hbase.log.HBaseMarkers;<a name="line.45"></a> <span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.procedure2.Procedure;<a name="line.46"></a> <span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.procedure2.ProcedureExecutor;<a name="line.47"></a> -<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.procedure2.store.ProcedureStoreBase;<a name="line.48"></a> -<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.procedure2.store.ProcedureStoreTracker;<a name="line.49"></a> -<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.procedure2.util.ByteSlot;<a name="line.50"></a> -<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.procedure2.util.StringUtils;<a name="line.51"></a> -<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.util.CommonFSUtils;<a name="line.52"></a> -<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.util.Threads;<a name="line.53"></a> -<span class="sourceLineNo">054</span>import org.apache.hadoop.ipc.RemoteException;<a name="line.54"></a> -<span class="sourceLineNo">055</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.55"></a> -<span class="sourceLineNo">056</span>import org.slf4j.Logger;<a name="line.56"></a> -<span class="sourceLineNo">057</span>import org.slf4j.LoggerFactory;<a name="line.57"></a> -<span class="sourceLineNo">058</span><a name="line.58"></a> -<span class="sourceLineNo">059</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.59"></a> -<span class="sourceLineNo">060</span>import org.apache.hbase.thirdparty.org.apache.commons.collections4.queue.CircularFifoQueue;<a name="line.60"></a> -<span class="sourceLineNo">061</span><a name="line.61"></a> -<span class="sourceLineNo">062</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.ProcedureProtos.ProcedureWALHeader;<a name="line.62"></a> -<span class="sourceLineNo">063</span><a name="line.63"></a> -<span class="sourceLineNo">064</span>/**<a name="line.64"></a> -<span class="sourceLineNo">065</span> * WAL implementation of the ProcedureStore.<a name="line.65"></a> -<span class="sourceLineNo">066</span> * <p/><a name="line.66"></a> -<span class="sourceLineNo">067</span> * When starting, the upper layer will first call {@link #start(int)}, then {@link #recoverLease()},<a name="line.67"></a> -<span class="sourceLineNo">068</span> * then {@link #load(ProcedureLoader)}.<a name="line.68"></a> -<span class="sourceLineNo">069</span> * <p/><a name="line.69"></a> -<span class="sourceLineNo">070</span> * In {@link #recoverLease()}, we will get the lease by closing all the existing wal files(by<a name="line.70"></a> -<span class="sourceLineNo">071</span> * calling recoverFileLease), and creating a new wal writer. And we will also get the list of all<a name="line.71"></a> -<span class="sourceLineNo">072</span> * the old wal files.<a name="line.72"></a> -<span class="sourceLineNo">073</span> * <p/><a name="line.73"></a> -<span class="sourceLineNo">074</span> * FIXME: notice that the current recover lease implementation is problematic, it can not deal with<a name="line.74"></a> -<span class="sourceLineNo">075</span> * the races if there are two master both wants to acquire the lease...<a name="line.75"></a> -<span class="sourceLineNo">076</span> * <p/><a name="line.76"></a> -<span class="sourceLineNo">077</span> * In {@link #load(ProcedureLoader)} method, we will load all the active procedures. See the<a name="line.77"></a> -<span class="sourceLineNo">078</span> * comments of this method for more details.<a name="line.78"></a> -<span class="sourceLineNo">079</span> * <p/><a name="line.79"></a> -<span class="sourceLineNo">080</span> * The actual logging way is a bit like our FileSystem based WAL implementation as RS side. There is<a name="line.80"></a> -<span class="sourceLineNo">081</span> * a {@link #slots}, which is more like the ring buffer, and in the insert, update and delete<a name="line.81"></a> -<span class="sourceLineNo">082</span> * methods we will put thing into the {@link #slots} and wait. And there is a background sync<a name="line.82"></a> -<span class="sourceLineNo">083</span> * thread(see the {@link #syncLoop()} method) which get data from the {@link #slots} and write them<a name="line.83"></a> -<span class="sourceLineNo">084</span> * to the FileSystem, and notify the caller that we have finished.<a name="line.84"></a> -<span class="sourceLineNo">085</span> * <p/><a name="line.85"></a> -<span class="sourceLineNo">086</span> * TODO: try using disruptor to increase performance and simplify the logic?<a name="line.86"></a> -<span class="sourceLineNo">087</span> * <p/><a name="line.87"></a> -<span class="sourceLineNo">088</span> * The {@link #storeTracker} keeps track of the modified procedures in the newest wal file, which is<a name="line.88"></a> -<span class="sourceLineNo">089</span> * also the one being written currently. And the deleted bits in it are for all the procedures, not<a name="line.89"></a> -<span class="sourceLineNo">090</span> * only the ones in the newest wal file. And when rolling a log, we will first store it in the<a name="line.90"></a> -<span class="sourceLineNo">091</span> * trailer of the current wal file, and then reset its modified bits, so that it can start to track<a name="line.91"></a> -<span class="sourceLineNo">092</span> * the modified procedures for the new wal file.<a name="line.92"></a> -<span class="sourceLineNo">093</span> * <p/><a name="line.93"></a> -<span class="sourceLineNo">094</span> * The {@link #holdingCleanupTracker} is used to test whether we are safe to delete the oldest wal<a name="line.94"></a> -<span class="sourceLineNo">095</span> * file. When there are log rolling and there are more than 1 wal files, we will make use of it. It<a name="line.95"></a> -<span class="sourceLineNo">096</span> * will first be initialized to the oldest file's tracker(which is stored in the trailer), using the<a name="line.96"></a> -<span class="sourceLineNo">097</span> * method {@link ProcedureStoreTracker#resetTo(ProcedureStoreTracker, boolean)}, and then merge it<a name="line.97"></a> -<span class="sourceLineNo">098</span> * with the tracker of every newer wal files, using the<a name="line.98"></a> -<span class="sourceLineNo">099</span> * {@link ProcedureStoreTracker#setDeletedIfModifiedInBoth(ProcedureStoreTracker)}. If we find out<a name="line.99"></a> -<span class="sourceLineNo">100</span> * that all the modified procedures for the oldest wal file are modified or deleted in newer wal<a name="line.100"></a> -<span class="sourceLineNo">101</span> * files, then we can delete it.<a name="line.101"></a> -<span class="sourceLineNo">102</span> * @see ProcedureWALPrettyPrinter for printing content of a single WAL.<a name="line.102"></a> -<span class="sourceLineNo">103</span> * @see #main(String[]) to parse a directory of MasterWALProcs.<a name="line.103"></a> -<span class="sourceLineNo">104</span> */<a name="line.104"></a> -<span class="sourceLineNo">105</span>@InterfaceAudience.Private<a name="line.105"></a> -<span class="sourceLineNo">106</span>public class WALProcedureStore extends ProcedureStoreBase {<a name="line.106"></a> -<span class="sourceLineNo">107</span> private static final Logger LOG = LoggerFactory.getLogger(WALProcedureStore.class);<a name="line.107"></a> -<span class="sourceLineNo">108</span> public static final String LOG_PREFIX = "pv2-";<a name="line.108"></a> -<span class="sourceLineNo">109</span> /** Used to construct the name of the log directory for master procedures */<a name="line.109"></a> -<span class="sourceLineNo">110</span> public static final String MASTER_PROCEDURE_LOGDIR = "MasterProcWALs";<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> public interface LeaseRecovery {<a name="line.113"></a> -<span class="sourceLineNo">114</span> void recoverFileLease(FileSystem fs, Path path) throws IOException;<a name="line.114"></a> -<span class="sourceLineNo">115</span> }<a name="line.115"></a> +<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.procedure2.store.ProcedureStore;<a name="line.48"></a> +<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.procedure2.store.ProcedureStoreBase;<a name="line.49"></a> +<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.procedure2.store.ProcedureStoreTracker;<a name="line.50"></a> +<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.procedure2.util.ByteSlot;<a name="line.51"></a> +<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.procedure2.util.StringUtils;<a name="line.52"></a> +<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.util.CommonFSUtils;<a name="line.53"></a> +<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.util.Threads;<a name="line.54"></a> +<span class="sourceLineNo">055</span>import org.apache.hadoop.ipc.RemoteException;<a name="line.55"></a> +<span class="sourceLineNo">056</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.56"></a> +<span class="sourceLineNo">057</span>import org.slf4j.Logger;<a name="line.57"></a> +<span class="sourceLineNo">058</span>import org.slf4j.LoggerFactory;<a name="line.58"></a> +<span class="sourceLineNo">059</span><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>import org.apache.hbase.thirdparty.org.apache.commons.collections4.queue.CircularFifoQueue;<a name="line.61"></a> +<span class="sourceLineNo">062</span><a name="line.62"></a> +<span class="sourceLineNo">063</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.ProcedureProtos.ProcedureWALHeader;<a name="line.63"></a> +<span class="sourceLineNo">064</span><a name="line.64"></a> +<span class="sourceLineNo">065</span>/**<a name="line.65"></a> +<span class="sourceLineNo">066</span> * WAL implementation of the ProcedureStore.<a name="line.66"></a> +<span class="sourceLineNo">067</span> * <p/><a name="line.67"></a> +<span class="sourceLineNo">068</span> * When starting, the upper layer will first call {@link #start(int)}, then {@link #recoverLease()},<a name="line.68"></a> +<span class="sourceLineNo">069</span> * then {@link #load(ProcedureLoader)}.<a name="line.69"></a> +<span class="sourceLineNo">070</span> * <p/><a name="line.70"></a> +<span class="sourceLineNo">071</span> * In {@link #recoverLease()}, we will get the lease by closing all the existing wal files(by<a name="line.71"></a> +<span class="sourceLineNo">072</span> * calling recoverFileLease), and creating a new wal writer. And we will also get the list of all<a name="line.72"></a> +<span class="sourceLineNo">073</span> * the old wal files.<a name="line.73"></a> +<span class="sourceLineNo">074</span> * <p/><a name="line.74"></a> +<span class="sourceLineNo">075</span> * FIXME: notice that the current recover lease implementation is problematic, it can not deal with<a name="line.75"></a> +<span class="sourceLineNo">076</span> * the races if there are two master both wants to acquire the lease...<a name="line.76"></a> +<span class="sourceLineNo">077</span> * <p/><a name="line.77"></a> +<span class="sourceLineNo">078</span> * In {@link #load(ProcedureLoader)} method, we will load all the active procedures. See the<a name="line.78"></a> +<span class="sourceLineNo">079</span> * comments of this method for more details.<a name="line.79"></a> +<span class="sourceLineNo">080</span> * <p/><a name="line.80"></a> +<span class="sourceLineNo">081</span> * The actual logging way is a bit like our FileSystem based WAL implementation as RS side. There is<a name="line.81"></a> +<span class="sourceLineNo">082</span> * a {@link #slots}, which is more like the ring buffer, and in the insert, update and delete<a name="line.82"></a> +<span class="sourceLineNo">083</span> * methods we will put thing into the {@link #slots} and wait. And there is a background sync<a name="line.83"></a> +<span class="sourceLineNo">084</span> * thread(see the {@link #syncLoop()} method) which get data from the {@link #slots} and write them<a name="line.84"></a> +<span class="sourceLineNo">085</span> * to the FileSystem, and notify the caller that we have finished.<a name="line.85"></a> +<span class="sourceLineNo">086</span> * <p/><a name="line.86"></a> +<span class="sourceLineNo">087</span> * TODO: try using disruptor to increase performance and simplify the logic?<a name="line.87"></a> +<span class="sourceLineNo">088</span> * <p/><a name="line.88"></a> +<span class="sourceLineNo">089</span> * The {@link #storeTracker} keeps track of the modified procedures in the newest wal file, which is<a name="line.89"></a> +<span class="sourceLineNo">090</span> * also the one being written currently. And the deleted bits in it are for all the procedures, not<a name="line.90"></a> +<span class="sourceLineNo">091</span> * only the ones in the newest wal file. And when rolling a log, we will first store it in the<a name="line.91"></a> +<span class="sourceLineNo">092</span> * trailer of the current wal file, and then reset its modified bits, so that it can start to track<a name="line.92"></a> +<span class="sourceLineNo">093</span> * the modified procedures for the new wal file.<a name="line.93"></a> +<span class="sourceLineNo">094</span> * <p/><a name="line.94"></a> +<span class="sourceLineNo">095</span> * The {@link #holdingCleanupTracker} is used to test whether we are safe to delete the oldest wal<a name="line.95"></a> +<span class="sourceLineNo">096</span> * file. When there are log rolling and there are more than 1 wal files, we will make use of it. It<a name="line.96"></a> +<span class="sourceLineNo">097</span> * will first be initialized to the oldest file's tracker(which is stored in the trailer), using the<a name="line.97"></a> +<span class="sourceLineNo">098</span> * method {@link ProcedureStoreTracker#resetTo(ProcedureStoreTracker, boolean)}, and then merge it<a name="line.98"></a> +<span class="sourceLineNo">099</span> * with the tracker of every newer wal files, using the<a name="line.99"></a> +<span class="sourceLineNo">100</span> * {@link ProcedureStoreTracker#setDeletedIfModifiedInBoth(ProcedureStoreTracker)}. If we find out<a name="line.100"></a> +<span class="sourceLineNo">101</span> * that all the modified procedures for the oldest wal file are modified or deleted in newer wal<a name="line.101"></a> +<span class="sourceLineNo">102</span> * files, then we can delete it. This is because that, every time we call<a name="line.102"></a> +<span class="sourceLineNo">103</span> * {@link ProcedureStore#insert(Procedure[])} or {@link ProcedureStore#update(Procedure)}, we will<a name="line.103"></a> +<span class="sourceLineNo">104</span> * persist the full state of a Procedure, so the earlier wal records for this procedure can all be<a name="line.104"></a> +<span class="sourceLineNo">105</span> * deleted.<a name="line.105"></a> +<span class="sourceLineNo">106</span> * @see ProcedureWALPrettyPrinter for printing content of a single WAL.<a name="line.106"></a> +<span class="sourceLineNo">107</span> * @see #main(String[]) to parse a directory of MasterWALProcs.<a name="line.107"></a> +<span class="sourceLineNo">108</span> */<a name="line.108"></a> +<span class="sourceLineNo">109</span>@InterfaceAudience.Private<a name="line.109"></a> +<span class="sourceLineNo">110</span>public class WALProcedureStore extends ProcedureStoreBase {<a name="line.110"></a> +<span class="sourceLineNo">111</span> private static final Logger LOG = LoggerFactory.getLogger(WALProcedureStore.class);<a name="line.111"></a> +<span class="sourceLineNo">112</span> public static final String LOG_PREFIX = "pv2-";<a name="line.112"></a> +<span class="sourceLineNo">113</span> /** Used to construct the name of the log directory for master procedures */<a name="line.113"></a> +<span class="sourceLineNo">114</span> public static final String MASTER_PROCEDURE_LOGDIR = "MasterProcWALs";<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> public static final String WAL_COUNT_WARN_THRESHOLD_CONF_KEY =<a name="line.117"></a> -<span class="sourceLineNo">118</span> "hbase.procedure.store.wal.warn.threshold";<a name="line.118"></a> -<span class="sourceLineNo">119</span> private static final int DEFAULT_WAL_COUNT_WARN_THRESHOLD = 64;<a name="line.119"></a> +<span class="sourceLineNo">117</span> public interface LeaseRecovery {<a name="line.117"></a> +<span class="sourceLineNo">118</span> void recoverFileLease(FileSystem fs, Path path) throws IOException;<a name="line.118"></a> +<span class="sourceLineNo">119</span> }<a name="line.119"></a> <span class="sourceLineNo">120</span><a name="line.120"></a> -<span class="sourceLineNo">121</span> public static final String EXEC_WAL_CLEANUP_ON_LOAD_CONF_KEY =<a name="line.121"></a> -<span class="sourceLineNo">122</span> "hbase.procedure.store.wal.exec.cleanup.on.load";<a name="line.122"></a> -<span class="sourceLineNo">123</span> private static final boolean DEFAULT_EXEC_WAL_CLEANUP_ON_LOAD_CONF_KEY = true;<a name="line.123"></a> +<span class="sourceLineNo">121</span> public static final String WAL_COUNT_WARN_THRESHOLD_CONF_KEY =<a name="line.121"></a> +<span class="sourceLineNo">122</span> "hbase.procedure.store.wal.warn.threshold";<a name="line.122"></a> +<span class="sourceLineNo">123</span> private static final int DEFAULT_WAL_COUNT_WARN_THRESHOLD = 10;<a name="line.123"></a> <span class="sourceLineNo">124</span><a name="line.124"></a> -<span class="sourceLineNo">125</span> public static final String MAX_RETRIES_BEFORE_ROLL_CONF_KEY =<a name="line.125"></a> -<span class="sourceLineNo">126</span> "hbase.procedure.store.wal.max.retries.before.roll";<a name="line.126"></a> -<span class="sourceLineNo">127</span> private static final int DEFAULT_MAX_RETRIES_BEFORE_ROLL = 3;<a name="line.127"></a> +<span class="sourceLineNo">125</span> public static final String EXEC_WAL_CLEANUP_ON_LOAD_CONF_KEY =<a name="line.125"></a> +<span class="sourceLineNo">126</span> "hbase.procedure.store.wal.exec.cleanup.on.load";<a name="line.126"></a> +<span class="sourceLineNo">127</span> private static final boolean DEFAULT_EXEC_WAL_CLEANUP_ON_LOAD_CONF_KEY = true;<a name="line.127"></a> <span class="sourceLineNo">128</span><a name="line.128"></a> -<span class="sourceLineNo">129</span> public static final String WAIT_BEFORE_ROLL_CONF_KEY =<a name="line.129"></a> -<span class="sourceLineNo">130</span> "hbase.procedure.store.wal.wait.before.roll";<a name="line.130"></a> -<span class="sourceLineNo">131</span> private static final int DEFAULT_WAIT_BEFORE_ROLL = 500;<a name="line.131"></a> +<span class="sourceLineNo">129</span> public static final String MAX_RETRIES_BEFORE_ROLL_CONF_KEY =<a name="line.129"></a> +<span class="sourceLineNo">130</span> "hbase.procedure.store.wal.max.retries.before.roll";<a name="line.130"></a> +<span class="sourceLineNo">131</span> private static final int DEFAULT_MAX_RETRIES_BEFORE_ROLL = 3;<a name="line.131"></a> <span class="sourceLineNo">132</span><a name="line.132"></a> -<span class="sourceLineNo">133</span> public static final String ROLL_RETRIES_CONF_KEY =<a name="line.133"></a> -<span class="sourceLineNo">134</span> "hbase.procedure.store.wal.max.roll.retries";<a name="line.134"></a> -<span class="sourceLineNo">135</span> private static final int DEFAULT_ROLL_RETRIES = 3;<a name="line.135"></a> +<span class="sourceLineNo">133</span> public static final String WAIT_BEFORE_ROLL_CONF_KEY =<a name="line.133"></a> +<span class="sourceLineNo">134</span> "hbase.procedure.store.wal.wait.before.roll";<a name="line.134"></a> +<span class="sourceLineNo">135</span> private static final int DEFAULT_WAIT_BEFORE_ROLL = 500;<a name="line.135"></a> <span class="sourceLineNo">136</span><a name="line.136"></a> -<span class="sourceLineNo">137</span> public static final String MAX_SYNC_FAILURE_ROLL_CONF_KEY =<a name="line.137"></a> -<span class="sourceLineNo">138</span> "hbase.procedure.store.wal.sync.failure.roll.max";<a name="line.138"></a> -<span class="sourceLineNo">139</span> private static final int DEFAULT_MAX_SYNC_FAILURE_ROLL = 3;<a name="line.139"></a> +<span class="sourceLineNo">137</span> public static final String ROLL_RETRIES_CONF_KEY =<a name="line.137"></a> +<span class="sourceLineNo">138</span> "hbase.procedure.store.wal.max.roll.retries";<a name="line.138"></a> +<span class="sourceLineNo">139</span> private static final int DEFAULT_ROLL_RETRIES = 3;<a name="line.139"></a> <span class="sourceLineNo">140</span><a name="line.140"></a> -<span class="sourceLineNo">141</span> public static final String PERIODIC_ROLL_CONF_KEY =<a name="line.141"></a> -<span class="sourceLineNo">142</span> "hbase.procedure.store.wal.periodic.roll.msec";<a name="line.142"></a> -<span class="sourceLineNo">143</span> private static final int DEFAULT_PERIODIC_ROLL = 60 * 60 * 1000; // 1h<a name="line.143"></a> +<span class="sourceLineNo">141</span> public static final String MAX_SYNC_FAILURE_ROLL_CONF_KEY =<a name="line.141"></a> +<span class="sourceLineNo">142</span> "hbase.procedure.store.wal.sync.failure.roll.max";<a name="line.142"></a> +<span class="sourceLineNo">143</span> private static final int DEFAULT_MAX_SYNC_FAILURE_ROLL = 3;<a name="line.143"></a> <span class="sourceLineNo">144</span><a name="line.144"></a> -<span class="sourceLineNo">145</span> public static final String SYNC_WAIT_MSEC_CONF_KEY = "hbase.procedure.store.wal.sync.wait.msec";<a name="line.145"></a> -<span class="sourceLineNo">146</span> private static final int DEFAULT_SYNC_WAIT_MSEC = 100;<a name="line.146"></a> -<span class="sourceLineNo">147</span><a name="line.147"></a> -<span class="sourceLineNo">148</span> public static final String USE_HSYNC_CONF_KEY = "hbase.procedure.store.wal.use.hsync";<a name="line.148"></a> -<span class="sourceLineNo">149</span> private static final boolean DEFAULT_USE_HSYNC = true;<a name="line.149"></a> -<span class="sourceLineNo">150</span><a name="line.150"></a> -<span class="sourceLineNo">151</span> public static final String ROLL_THRESHOLD_CONF_KEY = "hbase.procedure.store.wal.roll.threshold";<a name="line.151"></a> -<span class="sourceLineNo">152</span> private static final long DEFAULT_ROLL_THRESHOLD = 32 * 1024 * 1024; // 32M<a name="line.152"></a> -<span class="sourceLineNo">153</span><a name="line.153"></a> -<span class="sourceLineNo">154</span> public static final String STORE_WAL_SYNC_STATS_COUNT =<a name="line.154"></a> -<span class="sourceLineNo">155</span> "hbase.procedure.store.wal.sync.stats.count";<a name="line.155"></a> -<span class="sourceLineNo">156</span> private static final int DEFAULT_SYNC_STATS_COUNT = 10;<a name="line.156"></a> +<span class="sourceLineNo">145</span> public static final String PERIODIC_ROLL_CONF_KEY =<a name="line.145"></a> +<span class="sourceLineNo">146</span> "hbase.procedure.store.wal.periodic.roll.msec";<a name="line.146"></a> +<span class="sourceLineNo">147</span> private static final int DEFAULT_PERIODIC_ROLL = 60 * 60 * 1000; // 1h<a name="line.147"></a> +<span class="sourceLineNo">148</span><a name="line.148"></a> +<span class="sourceLineNo">149</span> public static final String SYNC_WAIT_MSEC_CONF_KEY = "hbase.procedure.store.wal.sync.wait.msec";<a name="line.149"></a> +<span class="sourceLineNo">150</span> private static final int DEFAULT_SYNC_WAIT_MSEC = 100;<a name="line.150"></a> +<span class="sourceLineNo">151</span><a name="line.151"></a> +<span class="sourceLineNo">152</span> public static final String USE_HSYNC_CONF_KEY = "hbase.procedure.store.wal.use.hsync";<a name="line.152"></a> +<span class="sourceLineNo">153</span> private static final boolean DEFAULT_USE_HSYNC = true;<a name="line.153"></a> +<span class="sourceLineNo">154</span><a name="line.154"></a> +<span class="sourceLineNo">155</span> public static final String ROLL_THRESHOLD_CONF_KEY = "hbase.procedure.store.wal.roll.threshold";<a name="line.155"></a> +<span class="sourceLineNo">156</span> private static final long DEFAULT_ROLL_THRESHOLD = 32 * 1024 * 1024; // 32M<a name="line.156"></a> <span class="sourceLineNo">157</span><a name="line.157"></a> -<span class="sourceLineNo">158</span> private final LinkedList<ProcedureWALFile> logs = new LinkedList<>();<a name="line.158"></a> -<span class="sourceLineNo">159</span> private final ProcedureStoreTracker holdingCleanupTracker = new ProcedureStoreTracker();<a name="line.159"></a> -<span class="sourceLineNo">160</span> private final ProcedureStoreTracker storeTracker = new ProcedureStoreTracker();<a name="line.160"></a> -<span class="sourceLineNo">161</span> private final ReentrantLock lock = new ReentrantLock();<a name="line.161"></a> -<span class="sourceLineNo">162</span> private final Condition waitCond = lock.newCondition();<a name="line.162"></a> -<span class="sourceLineNo">163</span> private final Condition slotCond = lock.newCondition();<a name="line.163"></a> -<span class="sourceLineNo">164</span> private final Condition syncCond = lock.newCondition();<a name="line.164"></a> -<span class="sourceLineNo">165</span><a name="line.165"></a> -<span class="sourceLineNo">166</span> private final LeaseRecovery leaseRecovery;<a name="line.166"></a> -<span class="sourceLineNo">167</span> private final Configuration conf;<a name="line.167"></a> -<span class="sourceLineNo">168</span> private final FileSystem fs;<a name="line.168"></a> -<span class="sourceLineNo">169</span> private final Path walDir;<a name="line.169"></a> -<span class="sourceLineNo">170</span> private final Path walArchiveDir;<a name="line.170"></a> -<span class="sourceLineNo">171</span> private final boolean enforceStreamCapability;<a name="line.171"></a> -<span class="sourceLineNo">172</span><a name="line.172"></a> -<span class="sourceLineNo">173</span> private final AtomicReference<Throwable> syncException = new AtomicReference<>();<a name="line.173"></a> -<span class="sourceLineNo">174</span> private final AtomicBoolean loading = new AtomicBoolean(true);<a name="line.174"></a> -<span class="sourceLineNo">175</span> private final AtomicBoolean inSync = new AtomicBoolean(false);<a name="line.175"></a> -<span class="sourceLineNo">176</span> private final AtomicLong totalSynced = new AtomicLong(0);<a name="line.176"></a> -<span class="sourceLineNo">177</span> private final AtomicLong lastRollTs = new AtomicLong(0);<a name="line.177"></a> -<span class="sourceLineNo">178</span> private final AtomicLong syncId = new AtomicLong(0);<a name="line.178"></a> -<span class="sourceLineNo">179</span><a name="line.179"></a> -<span class="sourceLineNo">180</span> private LinkedTransferQueue<ByteSlot> slotsCache = null;<a name="line.180"></a> -<span class="sourceLineNo">181</span> private Set<ProcedureWALFile> corruptedLogs = null;<a name="line.181"></a> -<span class="sourceLineNo">182</span> private FSDataOutputStream stream = null;<a name="line.182"></a> -<span class="sourceLineNo">183</span> private int runningProcCount = 1;<a name="line.183"></a> -<span class="sourceLineNo">184</span> private long flushLogId = 0;<a name="line.184"></a> -<span class="sourceLineNo">185</span> private int syncMaxSlot = 1;<a name="line.185"></a> -<span class="sourceLineNo">186</span> private int slotIndex = 0;<a name="line.186"></a> -<span class="sourceLineNo">187</span> private Thread syncThread;<a name="line.187"></a> -<span class="sourceLineNo">188</span> private ByteSlot[] slots;<a name="line.188"></a> -<span class="sourceLineNo">189</span><a name="line.189"></a> -<span class="sourceLineNo">190</span> private int walCountWarnThreshold;<a name="line.190"></a> -<span class="sourceLineNo">191</span> private int maxRetriesBeforeRoll;<a name="line.191"></a> -<span class="sourceLineNo">192</span> private int maxSyncFailureRoll;<a name="line.192"></a> -<span class="sourceLineNo">193</span> private int waitBeforeRoll;<a name="line.193"></a> -<span class="sourceLineNo">194</span> private int rollRetries;<a name="line.194"></a> -<span class="sourceLineNo">195</span> private int periodicRollMsec;<a name="line.195"></a> -<span class="sourceLineNo">196</span> private long rollThreshold;<a name="line.196"></a> -<span class="sourceLineNo">197</span> private boolean useHsync;<a name="line.197"></a> -<span class="sourceLineNo">198</span> private int syncWaitMsec;<a name="line.198"></a> -<span class="sourceLineNo">199</span><a name="line.199"></a> -<span class="sourceLineNo">200</span> // Variables used for UI display<a name="line.200"></a> -<span class="sourceLineNo">201</span> private CircularFifoQueue<SyncMetrics> syncMetricsQueue;<a name="line.201"></a> -<span class="sourceLineNo">202</span><a name="line.202"></a> -<span class="sourceLineNo">203</span> public static class SyncMetrics {<a name="line.203"></a> -<span class="sourceLineNo">204</span> private long timestamp;<a name="line.204"></a> -<span class="sourceLineNo">205</span> private long syncWaitMs;<a name="line.205"></a> -<span class="sourceLineNo">206</span> private long totalSyncedBytes;<a name="line.206"></a> -<span class="sourceLineNo">207</span> private int syncedEntries;<a name="line.207"></a> -<span class="sourceLineNo">208</span> private float syncedPerSec;<a name="line.208"></a> -<span class="sourceLineNo">209</span><a name="line.209"></a> -<span class="sourceLineNo">210</span> public long getTimestamp() {<a name="line.210"></a> -<span class="sourceLineNo">211</span> return timestamp;<a name="line.211"></a> -<span class="sourceLineNo">212</span> }<a name="line.212"></a> +<span class="sourceLineNo">158</span> public static final String STORE_WAL_SYNC_STATS_COUNT =<a name="line.158"></a> +<span class="sourceLineNo">159</span> "hbase.procedure.store.wal.sync.stats.count";<a name="line.159"></a> +<span class="sourceLineNo">160</span> private static final int DEFAULT_SYNC_STATS_COUNT = 10;<a name="line.160"></a> +<span class="sourceLineNo">161</span><a name="line.161"></a> +<span class="sourceLineNo">162</span> private final LinkedList<ProcedureWALFile> logs = new LinkedList<>();<a name="line.162"></a> +<span class="sourceLineNo">163</span> private final ProcedureStoreTracker holdingCleanupTracker = new ProcedureStoreTracker();<a name="line.163"></a> +<span class="sourceLineNo">164</span> private final ProcedureStoreTracker storeTracker = new ProcedureStoreTracker();<a name="line.164"></a> +<span class="sourceLineNo">165</span> private final ReentrantLock lock = new ReentrantLock();<a name="line.165"></a> +<span class="sourceLineNo">166</span> private final Condition waitCond = lock.newCondition();<a name="line.166"></a> +<span class="sourceLineNo">167</span> private final Condition slotCond = lock.newCondition();<a name="line.167"></a> +<span class="sourceLineNo">168</span> private final Condition syncCond = lock.newCondition();<a name="line.168"></a> +<span class="sourceLineNo">169</span><a name="line.169"></a> +<span class="sourceLineNo">170</span> private final LeaseRecovery leaseRecovery;<a name="line.170"></a> +<span class="sourceLineNo">171</span> private final Configuration conf;<a name="line.171"></a> +<span class="sourceLineNo">172</span> private final FileSystem fs;<a name="line.172"></a> +<span class="sourceLineNo">173</span> private final Path walDir;<a name="line.173"></a> +<span class="sourceLineNo">174</span> private final Path walArchiveDir;<a name="line.174"></a> +<span class="sourceLineNo">175</span> private final boolean enforceStreamCapability;<a name="line.175"></a> +<span class="sourceLineNo">176</span><a name="line.176"></a> +<span class="sourceLineNo">177</span> private final AtomicReference<Throwable> syncException = new AtomicReference<>();<a name="line.177"></a> +<span class="sourceLineNo">178</span> private final AtomicBoolean loading = new AtomicBoolean(true);<a name="line.178"></a> +<span class="sourceLineNo">179</span> private final AtomicBoolean inSync = new AtomicBoolean(false);<a name="line.179"></a> +<span class="sourceLineNo">180</span> private final AtomicLong totalSynced = new AtomicLong(0);<a name="line.180"></a> +<span class="sourceLineNo">181</span> private final AtomicLong lastRollTs = new AtomicLong(0);<a name="line.181"></a> +<span class="sourceLineNo">182</span> private final AtomicLong syncId = new AtomicLong(0);<a name="line.182"></a> +<span class="sourceLineNo">183</span><a name="line.183"></a> +<span class="sourceLineNo">184</span> private LinkedTransferQueue<ByteSlot> slotsCache = null;<a name="line.184"></a> +<span class="sourceLineNo">185</span> private Set<ProcedureWALFile> corruptedLogs = null;<a name="line.185"></a> +<span class="sourceLineNo">186</span> private FSDataOutputStream stream = null;<a name="line.186"></a> +<span class="sourceLineNo">187</span> private int runningProcCount = 1;<a name="line.187"></a> +<span class="sourceLineNo">188</span> private long flushLogId = 0;<a name="line.188"></a> +<span class="sourceLineNo">189</span> private int syncMaxSlot = 1;<a name="line.189"></a> +<span class="sourceLineNo">190</span> private int slotIndex = 0;<a name="line.190"></a> +<span class="sourceLineNo">191</span> private Thread syncThread;<a name="line.191"></a> +<span class="sourceLineNo">192</span> private ByteSlot[] slots;<a name="line.192"></a> +<span class="sourceLineNo">193</span><a name="line.193"></a> +<span class="sourceLineNo">194</span> private int walCountWarnThreshold;<a name="line.194"></a> +<span class="sourceLineNo">195</span> private int maxRetriesBeforeRoll;<a name="line.195"></a> +<span class="sourceLineNo">196</span> private int maxSyncFailureRoll;<a name="line.196"></a> +<span class="sourceLineNo">197</span> private int waitBeforeRoll;<a name="line.197"></a> +<span class="sourceLineNo">198</span> private int rollRetries;<a name="line.198"></a> +<span class="sourceLineNo">199</span> private int periodicRollMsec;<a name="line.199"></a> +<span class="sourceLineNo">200</span> private long rollThreshold;<a name="line.200"></a> +<span class="sourceLineNo">201</span> private boolean useHsync;<a name="line.201"></a> +<span class="sourceLineNo">202</span> private int syncWaitMsec;<a name="line.202"></a> +<span class="sourceLineNo">203</span><a name="line.203"></a> +<span class="sourceLineNo">204</span> // Variables used for UI display<a name="line.204"></a> +<span class="sourceLineNo">205</span> private CircularFifoQueue<SyncMetrics> syncMetricsQueue;<a name="line.205"></a> +<span class="sourceLineNo">206</span><a name="line.206"></a> +<span class="sourceLineNo">207</span> public static class SyncMetrics {<a name="line.207"></a> +<span class="sourceLineNo">208</span> private long timestamp;<a name="line.208"></a> +<span class="sourceLineNo">209</span> private long syncWaitMs;<a name="line.209"></a> +<span class="sourceLineNo">210</span> private long totalSyncedBytes;<a name="line.210"></a> +<span class="sourceLineNo">211</span> private int syncedEntries;<a name="line.211"></a> +<span class="sourceLineNo">212</span> private float syncedPerSec;<a name="line.212"></a> <span class="sourceLineNo">213</span><a name="line.213"></a> -<span class="sourceLineNo">214</span> public long getSyncWaitMs() {<a name="line.214"></a> -<span class="sourceLineNo">215</span> return syncWaitMs;<a name="line.215"></a> +<span class="sourceLineNo">214</span> public long getTimestamp() {<a name="line.214"></a> +<span class="sourceLineNo">215</span> return timestamp;<a name="line.215"></a> <span class="sourceLineNo">216</span> }<a name="line.216"></a> <span class="sourceLineNo">217</span><a name="line.217"></a> -<span class="sourceLineNo">218</span> public long getTotalSyncedBytes() {<a name="line.218"></a> -<span class="sourceLineNo">219</span> return totalSyncedBytes;<a name="line.219"></a> +<span class="sourceLineNo">218</span> public long getSyncWaitMs() {<a name="line.218"></a> +<span class="sourceLineNo">219</span> return syncWaitMs;<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> public long getSyncedEntries() {<a name="line.222"></a> -<span class="sourceLineNo">223</span> return syncedEntries;<a name="line.223"></a> +<span class="sourceLineNo">222</span> public long getTotalSyncedBytes() {<a name="line.222"></a> +<span class="sourceLineNo">223</span> return totalSyncedBytes;<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> public float getSyncedPerSec() {<a name="line.226"></a> -<span class="sourceLineNo">227</span> return syncedPerSec;<a name="line.227"></a> +<span class="sourceLineNo">226</span> public long getSyncedEntries() {<a name="line.226"></a> +<span class="sourceLineNo">227</span> return syncedEntries;<a name="line.227"></a> <span class="sourceLineNo">228</span> }<a name="line.228"></a> -<span class="sourceLineNo">229</span> }<a name="line.229"></a> -<span class="sourceLineNo">230</span><a name="line.230"></a> -<span class="sourceLineNo">231</span> public WALProcedureStore(final Configuration conf, final LeaseRecovery leaseRecovery)<a name="line.231"></a> -<span class="sourceLineNo">232</span> throws IOException {<a name="line.232"></a> -<span class="sourceLineNo">233</span> this(conf,<a name="line.233"></a> -<span class="sourceLineNo">234</span> new Path(CommonFSUtils.getWALRootDir(conf), MASTER_PROCEDURE_LOGDIR),<a name="line.234"></a> -<span class="sourceLineNo">235</span> new Path(CommonFSUtils.getWALRootDir(conf), HConstants.HREGION_OLDLOGDIR_NAME),<a name="line.235"></a> -<span class="sourceLineNo">236</span> leaseRecovery);<a name="line.236"></a> -<span class="sourceLineNo">237</span> }<a name="line.237"></a> -<span class="sourceLineNo">238</span><a name="line.238"></a> -<span class="sourceLineNo">239</span> @VisibleForTesting<a name="line.239"></a> -<span class="sourceLineNo">240</span> public WALProcedureStore(final Configuration conf, final Path walDir, final Path walArchiveDir,<a name="line.240"></a> -<span class="sourceLineNo">241</span> final LeaseRecovery leaseRecovery) throws IOException {<a name="line.241"></a> -<span class="sourceLineNo">242</span> this.conf = conf;<a name="line.242"></a> -<span class="sourceLineNo">243</span> this.leaseRecovery = leaseRecovery;<a name="line.243"></a> -<span class="sourceLineNo">244</span> this.walDir = walDir;<a name="line.244"></a> -<span class="sourceLineNo">245</span> this.walArchiveDir = walArchiveDir;<a name="line.245"></a> -<span class="sourceLineNo">246</span> this.fs = walDir.getFileSystem(conf);<a name="line.246"></a> -<span class="sourceLineNo">247</span> this.enforceStreamCapability = conf.getBoolean(CommonFSUtils.UNSAFE_STREAM_CAPABILITY_ENFORCE, true);<a name="line.247"></a> -<span class="sourceLineNo">248</span><a name="line.248"></a> -<span class="sourceLineNo">249</span> // Create the log directory for the procedure store<a name="line.249"></a> -<span class="sourceLineNo">250</span> if (!fs.exists(walDir)) {<a name="line.250"></a> -<span class="sourceLineNo">251</span> if (!fs.mkdirs(walDir)) {<a name="line.251"></a> -<span class="sourceLineNo">252</span> throw new IOException("Unable to mkdir " + walDir);<a name="line.252"></a> -<span class="sourceLineNo">253</span> }<a name="line.253"></a> -<span class="sourceLineNo">254</span> }<a name="line.254"></a> -<span class="sourceLineNo">255</span> // Now that it exists, set the log policy<a name="line.255"></a> -<span class="sourceLineNo">256</span> String storagePolicy =<a name="line.256"></a> -<span class="sourceLineNo">257</span> conf.get(HConstants.WAL_STORAGE_POLICY, HConstants.DEFAULT_WAL_STORAGE_POLICY);<a name="line.257"></a> -<span class="sourceLineNo">258</span> CommonFSUtils.setStoragePolicy(fs, walDir, storagePolicy);<a name="line.258"></a> -<span class="sourceLineNo">259</span><a name="line.259"></a> -<span class="sourceLineNo">260</span> // Create archive dir up front. Rename won't work w/o it up on HDFS.<a name="line.260"></a> -<span class="sourceLineNo">261</span> if (this.walArchiveDir != null && !this.fs.exists(this.walArchiveDir)) {<a name="line.261"></a> -<span class="sourceLineNo">262</span> if (this.fs.mkdirs(this.walArchiveDir)) {<a name="line.262"></a> -<span class="sourceLineNo">263</span> LOG.debug("Created Procedure Store WAL archive dir {}", this.walArchiveDir);<a name="line.263"></a> -<span class="sourceLineNo">264</span> } else {<a name="line.264"></a> -<span class="sourceLineNo">265</span> LOG.warn("Failed create of {}", this.walArchiveDir);<a name="line.265"></a> -<span class="sourceLineNo">266</span> }<a name="line.266"></a> -<span class="sourceLineNo">267</span> }<a name="line.267"></a> -<span class="sourceLineNo">268</span> }<a name="line.268"></a> -<span class="sourceLineNo">269</span><a name="line.269"></a> -<span class="sourceLineNo">270</span> @Override<a name="line.270"></a> -<span class="sourceLineNo">271</span> public void start(int numSlots) throws IOException {<a name="line.271"></a> -<span class="sourceLineNo">272</span> if (!setRunning(true)) {<a name="line.272"></a> -<span class="sourceLineNo">273</span> return;<a name="line.273"></a> -<span class="sourceLineNo">274</span> }<a name="line.274"></a> -<span class="sourceLineNo">275</span><a name="line.275"></a> -<span class="sourceLineNo">276</span> // Init buffer slots<a name="line.276"></a> -<span class="sourceLineNo">277</span> loading.set(true);<a name="line.277"></a> -<span class="sourceLineNo">278</span> runningProcCount = numSlots;<a name="line.278"></a> -<span class="sourceLineNo">279</span> syncMaxSlot = numSlots;<a name="line.279"></a> -<span class="sourceLineNo">280</span> slots = new ByteSlot[numSlots];<a name="line.280"></a> -<span class="sourceLineNo">281</span> slotsCache = new LinkedTransferQueue<>();<a name="line.281"></a> -<span class="sourceLineNo">282</span> while (slotsCache.size() < numSlots) {<a name="line.282"></a> -<span class="sourceLineNo">283</span> slotsCache.offer(new ByteSlot());<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> // Tunings<a name="line.286"></a> -<span class="sourceLineNo">287</span> walCountWarnThreshold =<a name="line.287"></a> -<span class="sourceLineNo">288</span> conf.getInt(WAL_COUNT_WARN_THRESHOLD_CONF_KEY, DEFAULT_WAL_COUNT_WARN_THRESHOLD);<a name="line.288"></a> -<span class="sourceLineNo">289</span> maxRetriesBeforeRoll =<a name="line.289"></a> -<span class="sourceLineNo">290</span> conf.getInt(MAX_RETRIES_BEFORE_ROLL_CONF_KEY, DEFAULT_MAX_RETRIES_BEFORE_ROLL);<a name="line.290"></a> -<span class="sourceLineNo">291</span> maxSyncFailureRoll = conf.getInt(MAX_SYNC_FAILURE_ROLL_CONF_KEY, DEFAULT_MAX_SYNC_FAILURE_ROLL);<a name="line.291"></a> -<span class="sourceLineNo">292</span> waitBeforeRoll = conf.getInt(WAIT_BEFORE_ROLL_CONF_KEY, DEFAULT_WAIT_BEFORE_ROLL);<a name="line.292"></a> -<span class="sourceLineNo">293</span> rollRetries = conf.getInt(ROLL_RETRIES_CONF_KEY, DEFAULT_ROLL_RETRIES);<a name="line.293"></a> -<span class="sourceLineNo">294</span> rollThreshold = conf.getLong(ROLL_THRESHOLD_CONF_KEY, DEFAULT_ROLL_THRESHOLD);<a name="line.294"></a> -<span class="sourceLineNo">295</span> periodicRollMsec = conf.getInt(PERIODIC_ROLL_CONF_KEY, DEFAULT_PERIODIC_ROLL);<a name="line.295"></a> -<span class="sourceLineNo">296</span> syncWaitMsec = conf.getInt(SYNC_WAIT_MSEC_CONF_KEY, DEFAULT_SYNC_WAIT_MSEC);<a name="line.296"></a> -<span class="sourceLineNo">297</span> useHsync = conf.getBoolean(USE_HSYNC_CONF_KEY, DEFAULT_USE_HSYNC);<a name="line.297"></a> -<span class="sourceLineNo">298</span><a name="line.298"></a> -<span class="sourceLineNo">299</span> // WebUI<a name="line.299"></a> -<span class="sourceLineNo">300</span> syncMetricsQueue = new CircularFifoQueue<>(<a name="line.300"></a> -<span class="sourceLineNo">301</span> conf.getInt(STORE_WAL_SYNC_STATS_COUNT, DEFAULT_SYNC_STATS_COUNT));<a name="line.301"></a> +<span class="sourceLineNo">229</span><a name="line.229"></a> +<span class="sourceLineNo">230</span> public float getSyncedPerSec() {<a name="line.230"></a> +<span class="sourceLineNo">231</span> return syncedPerSec;<a name="line.231"></a> +<span class="sourceLineNo">232</span> }<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> public WALProcedureStore(final Configuration conf, final LeaseRecovery leaseRecovery)<a name="line.235"></a> +<span class="sourceLineNo">236</span> throws IOException {<a name="line.236"></a> +<span class="sourceLineNo">237</span> this(conf,<a name="line.237"></a> +<span class="sourceLineNo">238</span> new Path(CommonFSUtils.getWALRootDir(conf), MASTER_PROCEDURE_LOGDIR),<a name="line.238"></a> +<span class="sourceLineNo">239</span> new Path(CommonFSUtils.getWALRootDir(conf), HConstants.HREGION_OLDLOGDIR_NAME),<a name="line.239"></a> +<span class="sourceLineNo">240</span> leaseRecovery);<a name="line.240"></a> +<span class="sourceLineNo">241</span> }<a name="line.241"></a> +<span class="sourceLineNo">242</span><a name="line.242"></a> +<span class="sourceLineNo">243</span> @VisibleForTesting<a name="line.243"></a> +<span class="sourceLineNo">244</span> public WALProcedureStore(final Configuration conf, final Path walDir, final Path walArchiveDir,<a name="line.244"></a> +<span class="sourceLineNo">245</span> final LeaseRecovery leaseRecovery) throws IOException {<a name="line.245"></a> +<span class="sourceLineNo">246</span> this.conf = conf;<a name="line.246"></a> +<span class="sourceLineNo">247</span> this.leaseRecovery = leaseRecovery;<a name="line.247"></a> +<span class="sourceLineNo">248</span> this.walDir = walDir;<a name="line.248"></a> +<span class="sourceLineNo">249</span> this.walArchiveDir = walArchiveDir;<a name="line.249"></a> +<span class="sourceLineNo">250</span> this.fs = walDir.getFileSystem(conf);<a name="line.250"></a> +<span class="sourceLineNo">251</span> this.enforceStreamCapability = conf.getBoolean(CommonFSUtils.UNSAFE_STREAM_CAPABILITY_ENFORCE, true);<a name="line.251"></a> +<span class="sourceLineNo">252</span><a name="line.252"></a> +<span class="sourceLineNo">253</span> // Create the log directory for the procedure store<a name="line.253"></a> +<span class="sourceLineNo">254</span> if (!fs.exists(walDir)) {<a name="line.254"></a> +<span class="sourceLineNo">255</span> if (!fs.mkdirs(walDir)) {<a name="line.255"></a> +<span class="sourceLineNo">256</span> throw new IOException("Unable to mkdir " + walDir);<a name="line.256"></a> +<span class="sourceLineNo">257</span> }<a name="line.257"></a> +<span class="sourceLineNo">258</span> }<a name="line.258"></a> +<span class="sourceLineNo">259</span> // Now that it exists, set the log policy<a name="line.259"></a> +<span class="sourceLineNo">260</span> String storagePolicy =<a name="line.260"></a> +<span class="sourceLineNo">261</span> conf.get(HConstants.WAL_STORAGE_POLICY, HConstants.DEFAULT_WAL_STORAGE_POLICY);<a name="line.261"></a> +<span class="sourceLineNo">262</span> CommonFSUtils.setStoragePolicy(fs, walDir, storagePolicy);<a name="line.262"></a> +<span class="sourceLineNo">263</span><a name="line.263"></a> +<span class="sourceLineNo">264</span> // Create archive dir up front. Rename won't work w/o it up on HDFS.<a name="line.264"></a> +<span class="sourceLineNo">265</span> if (this.walArchiveDir != null && !this.fs.exists(this.walArchiveDir)) {<a name="line.265"></a> +<span class="sourceLineNo">266</span> if (this.fs.mkdirs(this.walArchiveDir)) {<a name="line.266"></a> +<span class="sourceLineNo">267</span> LOG.debug("Created Procedure Store WAL archive dir {}", this.walArchiveDir);<a name="line.267"></a> +<span class="sourceLineNo">268</span> } else {<a name="line.268"></a> +<span class="sourceLineNo">269</span> LOG.warn("Failed create of {}", this.walArchiveDir);<a name="line.269"></a> +<span class="sourceLineNo">270</span> }<a name="line.270"></a> +<span class="sourceLineNo">271</span> }<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> @Override<a name="line.274"></a> +<span class="sourceLineNo">275</span> public void start(int numSlots) throws IOException {<a name="line.275"></a> +<span class="sourceLineNo">276</span> if (!setRunning(true)) {<a name="line.276"></a> +<span class="sourceLineNo">277</span> return;<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> // Init buffer slots<a name="line.280"></a> +<span class="sourceLineNo">281</span> loading.set(true);<a name="line.281"></a> +<span class="sourceLineNo">282</span> runningProcCount = numSlots;<a name="line.282"></a> +<span class="sourceLineNo">283</span> syncMaxSlot = numSlots;<a name="line.283"></a> +<span class="sourceLineNo">284</span> slots = new ByteSlot[numSlots];<a name="line.284"></a> +<span class="sourceLineNo">285</span> slotsCache = new LinkedTransferQueue<>();<a name="line.285"></a> +<span class="sourceLineNo">286</span> while (slotsCache.size() < numSlots) {<a name="line.286"></a> +<span class="sourceLineNo">287</span> slotsCache.offer(new ByteSlot());<a name="line.287"></a> +<span class="sourceLineNo">288</span> }<a name="line.288"></a> +<span class="sourceLineNo">289</span><a name="line.289"></a> +<span class="sourceLineNo">290</span> // Tunings<a name="line.290"></a> +<span class="sourceLineNo">291</span> walCountWarnThreshold =<a name="line.291"></a> +<span class="sourceLineNo">292</span> conf.getInt(WAL_COUNT_WARN_THRESHOLD_CONF_KEY, DEFAULT_WAL_COUNT_WARN_THRESHOLD);<a name="line.292"></a> +<span class="sourceLineNo">293</span> maxRetriesBeforeRoll =<a name="line.293"></a> +<span class="sourceLineNo">294</span> conf.getInt(MAX_RETRIES_BEFORE_ROLL_CONF_KEY, DEFAULT_MAX_RETRIES_BEFORE_ROLL);<a name="line.294"></a> +<span class="sourceLineNo">295</span> maxSyncFailureRoll = conf.getInt(MAX_SYNC_FAILURE_ROLL_CONF_KEY, DEFAULT_MAX_SYNC_FAILURE_ROLL);<a name="line.295"></a> +<span class="sourceLineNo">296</span> waitBeforeRoll = conf.getInt(WAIT_BEFORE_ROLL_CONF_KEY, DEFAULT_WAIT_BEFORE_ROLL);<a name="line.296"></a> +<span class="sourceLineNo">297</span> rollRetries = conf.getInt(ROLL_RETRIES_CONF_KEY, DEFAULT_ROLL_RETRIES);<a name="line.297"></a> +<span class="sourceLineNo">298</span> rollThreshold = conf.getLong(ROLL_THRESHOLD_CONF_KEY, DEFAULT_ROLL_THRESHOLD);<a name="line.298"></a> +<span class="sourceLineNo">299</span> periodicRollMsec = conf.getInt(PERIODIC_ROLL_CONF_KEY, DEFAULT_PERIODIC_ROLL);<a name="line.299"></a> +<span class="sourceLineNo">300</span> syncWaitMsec = conf.getInt(SYNC_WAIT_MSEC_CONF_KEY, DEFAULT_SYNC_WAIT_MSEC);<a name="line.300"></a> +<span class="sourceLineNo">301</span> useHsync = conf.getBoolean(USE_HSYNC_CONF_KEY, DEFAULT_USE_HSYNC);<a name="line.301"></a> <span class="sourceLineNo">302</span><a name="line.302"></a> -<span class="sourceLineNo">303</span> // Init sync thread<a name="line.303"></a> -<span class="sourceLineNo">304</span> syncThread = new Thread("WALProcedureStoreSyncThread") {<a name="line.304"></a> -<span class="sourceLineNo">305</span> @Override<a name="line.305"></a> -<span class="sourceLineNo">306</span> public void run() {<a name="line.306"></a> -<span class="sourceLineNo">307</span> try {<a name="line.307"></a> -<span class="sourceLineNo">308</span> syncLoop();<a name="line.308"></a> -<span class="sourceLineNo">309</span> } catch (Throwable e) {<a name="line.309"></a> -<span class="sourceLineNo">310</span> LOG.error("Got an exception from the sync-loop", e);<a name="line.310"></a> -<span class="sourceLineNo">311</span> if (!isSyncAborted()) {<a name="line.311"></a> -<span class="sourceLineNo">312</span> sendAbortProcessSignal();<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> }<a name="line.315"></a> -<span class="sourceLineNo">316</span> };<a name="line.316"></a> -<span class="sourceLineNo">317</span> syncThread.start();<a name="line.317"></a> -<span class="sourceLineNo">318</span> }<a name="line.318"></a> -<span class="sourceLineNo">319</span><a name="line.319"></a> -<span class="sourceLineNo">320</span> @Override<a name="line.320"></a> -<span class="sourceLineNo">321</span> public void stop(final boolean abort) {<a name="line.321"></a> -<span class="sourceLineNo">322</span> if (!setRunning(false)) {<a name="line.322"></a> -<span class="sourceLineNo">323</span> return;<a name="line.323"></a> -<span class="sourceLineNo">324</span> }<a name="line.324"></a> -<span class="sourceLineNo">325</span><a name="line.325"></a> -<span class="sourceLineNo">326</span> LOG.info("Stopping the WAL Procedure Store, isAbort=" + abort +<a name="line.326"></a> -<span class="sourceLineNo">327</span> (isSyncAborted() ? " (self aborting)" : ""));<a name="line.327"></a> -<span class="sourceLineNo">328</span> sendStopSignal();<a name="line.328"></a> -<span class="sourceLineNo">329</span> if (!isSyncAborted()) {<a name="line.329"></a> -<span class="sourceLineNo">330</span> try {<a name="line.330"></a> -<span class="sourceLineNo">331</span> while (syncThread.isAlive()) {<a name="line.331"></a> -<span class="sourceLineNo">332</span> sendStopSignal();<a name="line.332"></a> -<span class="sourceLineNo">333</span> syncThread.join(250);<a name="line.333"></a> -<span class="sourceLineNo">334</span> }<a name="line.334"></a> -<span class="sourceLineNo">335</span> } catch (InterruptedException e) {<a name="line.335"></a> -<span class="sourceLineNo">336</span> LOG.warn("join interrupted", e);<a name="line.336"></a> -<span class="sourceLineNo">337</span> Thread.currentThread().interrupt();<a name="line.337"></a> -<span class="sourceLineNo">338</span> }<a name="line.338"></a> -<span class="sourceLineNo">339</span> }<a name="line.339"></a> -<span class="sourceLineNo">340</span><a name="line.340"></a> -<span class="sourceLineNo">341</span> // Close the writer<a name="line.341"></a> -<span class="sourceLineNo">342</span> closeCurrentLogStream();<a name="line.342"></a> -<span class="sourceLineNo">343</span><a name="line.343"></a> -<span class="sourceLineNo">344</span> // Close the old logs<a name="line.344"></a> -<span class="sourceLineNo">345</span> // they should be already closed, this is just in case the load fails<a name="line.345"></a> -<span class="sourceLineNo">346</span> // and we call start() and then stop()<a name="line.346"></a> -<span class="sourceLineNo">347</span> for (ProcedureWALFile log: logs) {<a name="line.347"></a> -<span class="sourceLineNo">348</span> log.close();<a name="line.348"></a> -<span class="sourceLineNo">349</span> }<a name="line.349"></a> -<span class="sourceLineNo">350</span> logs.clear();<a name="line.350"></a> -<span class="sourceLineNo">351</span> loading.set(true);<a name="line.351"></a> -<span class="sourceLineNo">352</span> }<a name="line.352"></a> -<span class="sourceLineNo">353</span><a name="line.353"></a> -<span class="sourceLineNo">354</span> private void sendStopSignal() {<a name="line.354"></a> -<span class="sourceLineNo">355</span> if (lock.tryLock()) {<a name="line.355"></a> -<span class="sourceLineNo">356</span> try {<a name="line.356"></a> -<span class="sourceLineNo">357</span> waitCond.signalAll();<a name="line.357"></a> -<span class="sourceLineNo">358</span> syncCond.signalAll();<a name="line.358"></a> -<span class="sourceLineNo">359</span> } finally {<a name="line.359"></a> -<span class="sourceLineNo">360</span> lock.unlock();<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><a name="line.364"></a> -<span class="sourceLineNo">365</span> @Override<a name="line.365"></a> -<span class="sourceLineNo">366</span> public int getNumThreads() {<a name="line.366"></a> -<span class="sourceLineNo">367</span> return slots == null ? 0 : slots.length;<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> @Override<a name="line.370"></a> -<span class="sourceLineNo">371</span> public int setRunningProcedureCount(final int count) {<a name="line.371"></a> -<span class="sourceLineNo">372</span> this.runningProcCount = count > 0 ? Math.min(count, slots.length) : slots.length;<a name="line.372"></a> -<span class="sourceLineNo">373</span> return this.runningProcCount;<a name="line.373"></a> -<span class="sourceLineNo">374</span> }<a name="line.374"></a> -<span class="sourceLineNo">375</span><a name="line.375"></a> -<span class="sourceLineNo">376</span> public ProcedureStoreTracker getStoreTracker() {<a name="line.376"></a> -<span class="sourceLineNo">377</span> return storeTracker;<a name="line.377"></a> +<span class="sourceLineNo">303</span> // WebUI<a name="line.303"></a> +<span class="sourceLineNo">304</span> syncMetricsQueue = new CircularFifoQueue<>(<a name="line.304"></a> +<span class="sourceLineNo">305</span> conf.getInt(STORE_WAL_SYNC_STATS_COUNT, DEFAULT_SYNC_STATS_COUNT));<a name="line.305"></a> +<span class="sourceLineNo">306</span><a name="line.306"></a> +<span class="sourceLineNo">307</span> // Init sync thread<a name="line.307"></a> +<span class="sourceLineNo">308</span> syncThread = new Thread("WALProcedureStoreSyncThread") {<a name="line.308"></a> +<span class="sourceLineNo">309</span> @Override<a name="line.309"></a> +<span class="sourceLineNo">310</span> public void run() {<a name="line.310"></a> +<span class="sourceLineNo">311</span> try {<a name="line.311"></a> +<span class="sourceLineNo">312</span> syncLoop();<a name="line.312"></a> +<span class="sourceLineNo">313</span> } catch (Throwable e) {<a name="line.313"></a> +<span class="sourceLineNo">314</span> LOG.error("Got an exception from the sync-loop", e);<a name="line.314"></a> +<span class="sourceLineNo">315</span> if (!isSyncAborted()) {<a name="line.315"></a> +<span class="sourceLineNo">316</span> sendAbortProcessSignal();<a name="line.316"></a> +<span class="sourceLineNo">317</span> }<a name="line.317"></a> +<span class="sourceLineNo">318</span> }<a name="line.318"></a> +<span class="sourceLineNo">319</span> }<a name="line.319"></a> +<span class="sourceLineNo">320</span> };<a name="line.320"></a> +<span class="sourceLineNo">321</span> syncThread.start();<a name="line.321"></a> +<span class="sourceLineNo">322</span> }<a name="line.322"></a> +<span class="sourceLineNo">323</span><a name="line.323"></a> +<span class="sourceLineNo">324</span> @Override<a name="line.324"></a> +<span class="sourceLineNo">325</span> public void stop(final boolean abort) {<a name="line.325"></a> +<span class="sourceLineNo">326</span> if (!setRunning(false)) {<a name="line.326"></a> +<span class="sourceLineNo">327</span> return;<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> LOG.info("Stopping the WAL Procedure Store, isAbort=" + abort +<a name="line.330"></a> +<span class="sourceLineNo">331</span> (isSyncAborted() ? " (self aborting)" : ""));<a name="line.331"></a> +<span class="sourceLineNo">332</span> sendStopSignal();<a name="line.332"></a> +<span class="sourceLineNo">333</span> if (!isSyncAborted()) {<a name="line.333"></a> +<span class="sourceLineNo">334</span> try {<a name="line.334"></a> +<span class="sourceLineNo">335</span> while (syncThread.isAlive()) {<a name="line.335"></a> +<span class="sourceLineNo">336</span> sendStopSignal();<a name="line.336"></a> +<span class="sourceLineNo">337</span> syncThread.join(250);<a name="line.337"></a> +<span class="sourceLineNo">338</span> }<a name="line.338"></a> +<span class="sourceLineNo">339</span> } catch (InterruptedException e) {<a name="line.339"></a> +<span class="sourceLineNo">340</span> LOG.warn("join interrupted", e);<a name="line.340"></a> +<span class="sourceLineNo">341</span> Thread.currentThread().interrupt();<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> // Close the writer<a name="line.345"></a> +<span class="sourceLineNo">346</span> closeCurrentLogStream();<a name="line.346"></a> +<span class="sourceLineNo">347</span><a name="line.347"></a> +<span class="sourceLineNo">348</span> // Close the old logs<a name="line.348"></a> +<span class="sourceLineNo">349</span> // they should be already closed, this is just in case the load fails<a name="line.349"></a> +<span class="sourceLineNo">350</span> // and we call start() and then stop()<a name="line.350"></a> +<span class="sourceLineNo">351</span> for (ProcedureWALFile log: logs) {<a name="line.351"></a> +<span class="sourceLineNo">352</span> log.close();<a name="line.352"></a> +<span class="sourceLineNo">353</span> }<a name="line.353"></a> +<span class="sourceLineNo">354</span> logs.clear();<a name="line.354"></a> +<span class="sourceLineNo">355</span> loading.set(true);<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 void sendStopSignal() {<a name="line.358"></a> +<span class="sourceLineNo">359</span> if (lock.tryLock()) {<a name="line.359"></a> +<span class="sourceLineNo">360</span> try {<a name="line.360"></a> +<span class="sourceLineNo">361</span> waitCond.signalAll();<a name="line.361"></a> +<span class="sourceLineNo">362</span> syncCond.signalAll();<a name="line.362"></a> +<span class="sourceLineNo">363</span> } finally {<a name="line.363"></a> +<span class="sourceLineNo">364</span> lock.unlock();<a name="line.364"></a> +<span class="sourceLineNo">365</span> }<a name="line.365"></a> +<span class="sourceLineNo">366</span> }<a name="line.366"></a> +<span class="sourceLineNo">367</span> }<a name="line.367"></a> +<span class="sourceLineNo">368</span><a name="line.368"></a> +<span class="sourceLineNo">369</span> @Override<a name="line.369"></a> +<span class="sourceLineNo">370</span> public int getNumThreads() {<a name="line.370"></a> +<span class="sourceLineNo">371</span> return slots == null ? 0 : slots.length;<a name="line.371"></a> +<span class="sourceLineNo">372</span> }<a name="line.372"></a> +<span class="sourceLineNo">373</span><a name="line.373"></a> +<span class="sourceLineNo">374</span> @Override<a name="line.374"></a> +<span class="sourceLineNo">375</span> public int setRunningProcedureCount(final int count) {<a name="line.375"></a> +<span class="sourceLineNo">376</span> this.runningProcCount = count > 0 ? Math.min(count, slots.length) : slots.length;<a name="line.376"></a> +<span class="sourceLineNo">377</span> return this.runningProcCount;<a name="line.377"></a> <span class="sourceLineNo">378</span> }<a name="line.378"></a> <span class="sourceLineNo">379</span><a name="line.379"></a> -<span class="sourceLineNo">380</span> public ArrayList<ProcedureWALFile> getActiveLogs() {<a name="line.380"></a> -<span class="sourceLineNo">381</span> lock.lock();<a name="line.381"></a> -<span class="sourceLineNo">382</span> try {<a name="line.382"></a> -<span class="sourceLineNo">383</span> return new ArrayList<>(logs);<a name="line.383"></a> -<span class="sourceLineNo">384</span> } finally {<a name="line.384"></a> -<span class="sourceLineNo">385</span> lock.unlock();<a name="line.385"></a> -<span class="sourceLineNo">386</span> }<a name="line.386"></a> -<span class="sourceLineNo">387</span> }<a name="line.387"></a> -<span class="sourceLineNo">388</span><a name="line.388"></a> -<span class="sourceLineNo">389</span> public Set<ProcedureWALFile> getCorruptedLogs() {<a name="line.389"></a> -<span class="sourceLineNo">390</span> return corruptedLogs;<a name="line.390"></a> +<span class="sourceLineNo">380</span> public ProcedureStoreTracker getStoreTracker() {<a name="line.380"></a> +<span class="sourceLineNo">381</span> return storeTracker;<a name="line.381"></a> +<span class="sourceLineNo">382</span> }<a name="line.382"></a> +<span class="sourceLineNo">383</span><a name="line.383"></a> +<span class="sourceLineNo">384</span> public ArrayList<ProcedureWALFile> getActiveLogs() {<a name="line.384"></a> +<span class="sourceLineNo">385</span> lock.lock();<a name="line.385"></a> +<span class="sourceLineNo">386</span> try {<a name="line.386"></a> +<span class="sourceLineNo">387</span> return new ArrayList<>(logs);<a name="line.387"></a> +<span class="sourceLineNo">388</span> } finally {<a name="line.388"></a> +<span class="sourceLineNo">389</span> lock.unlock();<a name="line.389"></a> +<span class="sourceLineNo">390</span> }<a name="line.390"></a> <span class="sourceLineNo">391</span> }<a name="line.391"></a> <span class="sourceLineNo">392</span><a name="line.392"></a> -<span class="sourceLineNo">393</span> @Override<a name="line.393"></a> -<span class="sourceLineNo">394</span> public void recoverLease() throws IOException {<a name="line.394"></a> -<span class="sourceLineNo">395</span> lock.lock();<a name="line.395"></a> -<span class="sourceLineNo">396</span> try {<a name="line.396"></a> -<span class="sourceLineNo">397</span> LOG.trace("Starting WAL Procedure Store lease recovery");<a name="line.397"></a> -<span class="sourceLineNo">398</span> boolean afterFirstAttempt = false;<a name="line.398"></a> -<span class="sourceLineNo">399</span> while (isRunning()) {<a name="line.399"></a> -<span class="sourceLineNo">400</span> // Don't sleep before first attempt<a name="line.400"></a> -<span class="sourceLineNo">401</span> if (afterFirstAttempt) {<a name="line.401"></a> -<span class="sourceLineNo">402</span> LOG.trace("Sleep {} ms after first lease recovery attempt.",<a name="line.402"></a> -<span class="sourceLineNo">403</span> waitBeforeRoll);<a name="line.403"></a> -<span class="sourceLineNo">404</span> Threads.sleepWithoutInterrupt(waitBeforeRoll);<a name="line.404"></a> -<span class="sourceLineNo">405</span> } else {<a name="line.405"></a> -<span class="sourceLineNo">406</span> afterFirstAttempt = true;<a name="line.406"></a> -<span class="sourceLineNo">407</span> }<a name="line.407"></a> -<span class="sourceLineNo">408</span> FileStatus[] oldLogs = getLogFiles();<a name="line.408"></a> -<span class="sourceLineNo">409</span> // Get Log-MaxID and recover lease on old logs<a name="line.409"></a> -<span class="sourceLineNo">410</span> try {<a name="line.410"></a> -<span class="sourceLineNo">411</span> flushLogId = initOldLogs(oldLogs);<a name="line.411"></a> -<span class="sourceLineNo">412</span> } catch (FileNotFoundException e) {<a name="line.412"></a> -<span class="sourceLineNo">413</span> LOG.warn("Someone else is active and deleted logs. retrying.", e);<a name="line.413"></a> -<span class="sourceLineNo">414</span> continue;<a name="line.414"></a> -<span class="sourceLineNo">415</span> }<a name="line.415"></a> -<span class="sourceLineNo">416</span><a name="line.416"></a> -<span class="sourceLineNo">417</span> // Create new state-log<a name="line.417"></a> -<span class="sourceLineNo">418</span> if (!rollWriter(flushLogId + 1)) {<a name="line.418"></a> -<span class="sourceLineNo">419</span> // someone else has already created this log<a name="line.419"></a> -<span class="sourceLineNo">420</span> LOG.debug("Someone else has already created log {}. Retrying.", flushLogId);<a name="line.420"></a> -<span class="sourceLineNo">421</span> continue;<a name="line.421"></a> -<span class="sourceLineNo">422</span> }<a name="line.422"></a> -<span class="sourceLineNo">423</span><a name="line.423"></a> -<span class="sourceLineNo">424</span> // We have the lease on the log<a name="line.424"></a> -<span class="sourceLineNo">425</span> oldLogs = getLogFiles();<a name="line.425"></a> -<span class="sourceLineNo">426</span> if (getMaxLogId(oldLogs) > flushLogId) {<a name="line.426"></a> -<span class="sourceLineNo">427</span> LOG.debug("Someone else created new logs. Expected maxLogId < {}", flushLogId);<a name="line.427"></a> -<span class="sourceLineNo">428</span> logs.getLast().removeFile(this.walArchiveDir);<a name="line.428"></a> -<span class="sourceLineNo">429</span> continue;<a name="line.429"></a> -<span class="sourceLineNo">430</span> }<a name="line.430"></a> -<span class="sourceLineNo">431</span><a name="line.431"></a> -<span class="sourceLineNo">432</span> LOG.trace("Lease acquired for flushLogId={}", flushLogId);<a name="line.432"></a> -<span class="sourceLineNo">433</span> break;<a name="line.433"></a> -<span class="sourceLineNo">434</span> }<a name="line.434"></a> -<span class="sourceLineNo">435</span> } finally {<a name="line.435"></a> -<span class="sourceLineNo">436</span> lock.unlock();<a name="line.436"></a> -<span class="sourceLineNo">437</span> }<a name="line.437"></a> -<span class="sourceLineNo">438</span> }<a name="line.438"></a> -<span class="sourceLineNo">439</span><a name="line.439"></a> -<span class="sourceLineNo">440</span> @Override<a name="line.440"></a> -<span class="sourceLineNo">441</span> public void load(ProcedureLoader loader) throws IOException {<a name="line.441"></a> -<span class="sourceLineNo">442</span> lock.lock();<a name="line.442"></a> -<span class="sourceLineNo">443</span> try {<a name="line.443"></a> -<span class="sourceLineNo">444</span> if (logs.isEmpty()) {<a name="line.444"></a> -<span class="sourceLineNo">445</span> throw new RuntimeException("recoverLease() must be called before loading data");<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> // Nothing to do, If we have only the current log.<a name="line.448"></a> -<span class="sourceLineNo">449</span> if (logs.size() == 1) {<a name="line.449"></a> -<span class="sourceLineNo">450</span> LOG.trace("No state logs to replay.");<a name="line.450"></a> -<span class="sourceLineNo">451</span> loader.setMaxProcId(0);<a name="line.451"></a> -<span class="sourceLineNo">452</span> return;<a name="line.452"></a> -<span class="sourceLineNo">453</span> }<a name="line.453"></a> -<span class="sourceLineNo">454</span><a name="line.454"></a> -<span class="sourceLineNo">455</span> // Load the old logs<a name="line.455"></a> -<span class="sourceLineNo">456</span> Iterator<ProcedureWALFile> it = logs.descendingIterator();<a name="line.456"></a> -<span class="sourceLineNo">457</span> it.next(); // Skip the current log<a name="line.457"></a> +<span class="sourceLineNo">393</span> public Set<ProcedureWALFile> getCorruptedLogs() {<a name="line.393"></a> +<span class="sourceLineNo">394</span> return corruptedLogs;<a name="line.394"></a> +<span class="sourceLineNo">395</span> }<a name="line.395"></a> +<span class="sourceLineNo">396</span><a name="line.396"></a> +<span class="sourceLineNo">397</span> @Override<a name="line.397"></a> +<span class="sourceLineNo">398</span> public void recoverLease() throws IOException {<a name="line.398"></a> +<span class="sourceLineNo">399</span> lock.lock();<a name="line.399"></a> +<span class="sourceLineNo">400</span> try {<a name="line.400"></a> +<span class="sourceLineNo">401</span> LOG.trace("Starting WAL Procedure Store lease recovery");<a name="line.401"></a> +<span class="sourceLineNo">402</span> boolean afterFirstAttempt = false;<a name="line.402"></a> +<span class="sourceLineNo">403</span> while (isRunning()) {<a name="line.403"></a> +<span class="sourceLineNo">404</span> // Don't sleep before first attempt<a name="line.404"></a> +<span class="sourceLineNo">405</span> if (afterFirstAttempt) {<a name="line.405"></a> +<span class="sourceLineNo">406</span> LOG.trace("Sleep {} ms after first lease recovery attempt.",<a name="line.406"></a> +<span class="sourceLineNo">407</span> waitBeforeRoll);<a name="line.407"></a> +<span class="sourceLineNo">408</span> Threads.sleepWithoutInterrupt(waitBeforeRoll);<a name="line.408"></a> +<span class="sourceLineNo">409</span> } else {<a name="line.409"></a> +<span class="sourceLineNo">410</span> afterFirstAttempt = true;<a name="line.410"></a> +<span class="sourceLineNo">411</span> }<a name="line.411"></a> +<span class="sourceLineNo">412</span> FileStatus[] oldLogs = getLogFiles();<a name="line.412"></a> +<span class="sourceLineNo">413</span> // Get Log-MaxID and recover lease on old logs<a name="line.413"></a> +<span class="sourceLineNo">414</span> try {<a name="line.414"></a> +<span class="sourceLineNo">415</span> flushLogId = initOldLogs(oldLogs);<a name="line.415"></a> +<span class="sourceLineNo">416</span> } catch (FileNotFoundException e) {<a name="line.416"></a> +<span class="sourceLineNo">417</span> LOG.warn("Someone else is active and deleted logs. retrying.", e);<a name="line.417"></a> +<span class="sourceLineNo">418</span> continue;<a name="line.418"></a> +<span class="sourceLineNo">419</span> }<a name="line.419"></a> +<span class="sourceLineNo">420</span><a name="line.420"></a> +<span class="sourceLineNo">421</span> // Create new state-log<a name="line.421"></a> +<span class="sourceLineNo">422</span> if (!rollWriter(flushLogId + 1)) {<a name="line.422"></a> +<span class="sourceLineNo">423</span> // someone else has already created this log<a name="line.423"></a> +<span class="sourceLineNo">424</span> LOG.debug("Someone else has already created log {}. Retrying.", flushLogId);<a name="line.424"></a> +<span class="sourceLineNo">425</span> continue;<a name="line.425"></a> +<span class="sourceLineNo">426</span> }<a name="line.426"></a> +<span class="sourceLineNo">427</span><a name="line.427"></a> +<span class="sourceLineNo">428</span> // We have the lease on the log<a name="line.428"></a> +<span class="sourceLineNo">429</span> oldLogs = getLogFiles();<a name="line.429"></a> +<span class="sourceLineNo">430</span> if (getMaxLogId(oldLogs) > flushLogId) {<a name="line.430"></a> +<span class="sourceLineNo">431</span> LOG.debug("Someone else created new logs. Expected maxLogId < {}", flushLogId);<a name="line.431"></a> +<span class="sourceLineNo">432</span> logs.getLast().removeFile(this.walArchiveDir);<a name="line.432"></a> +<span class="sourceLineNo">433</span> continue;<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> LOG.trace("Lease acquired for flushLogId={}", flushLogId);<a name="line.436"></a> +<span class="sourceLineNo">437</span> break;<a name="line.437"></a> +<span class="sourceLineNo">438</span> }<a name="line.438"></a> +<span class="sourceLineNo">439</span> } finally {<a name="line.439"></a> +<span class="sourceLineNo">440</span> lock.unlock();<a name="line.440"></a> +<span class="sourceLineNo">441</span> }<a name="line.441"></a> +<span class="sourceLineNo">442</span> }<a name="line.442"></a> +<span class="sourceLineNo">443</span><a name="line.443"></a> +<span class="sourceLineNo">444</span> @Override<a name="line.444"></a> +<span class="sourceLineNo">445</span> public void load(ProcedureLoader loader) throws IOException {<a name="line.445"></a> +<span class="sourceLineNo">446</span> lock.lock();<a name="line.446"></a> +<span class="sourceLineNo">447</span> try {<a name="line.447"></a> +<span class="sourceLineNo">448</span> if (logs.isEmpty()) {<a name="line.448"></a> +<span class="sourceLineNo">449</span> throw new RuntimeException("recoverLease() must be called before loading data");<a name="line.449"></a> +<span class="sourceLineNo">450</span> }<a name="line.450"></a> +<span class="sourceLineNo">451</span><a name="line.451"></a> +<span class="sourceLineNo">452</span> // Nothing to do, If we have only the current log.<a name="line.452"></a> +<span class="sourceLineNo">453</span> if (logs.size() == 1) {<a name="line.453"></a> +<span class="sourceLineNo">454</span> LOG.trace("No state logs to replay.");<a name="line.454"></a> +<span class="sourceLineNo">455</span> loader.setMaxProcId(0);<a name="line.455"></a> +<span class="sourceLineNo">456</span> return;<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> ProcedureWALFormat.load(it, storeTracker, new ProcedureWALFormat.Loader() {<a name="line.459"></a> -<span class="sourceLineNo">460</span><a name="line.460"></a> -<span class="sourceLineNo">461</span> @Override<a name="line.461"></a> -<span class="sourceLineNo">462</span> public void setMaxProcId(long maxProcId) {<a name="line.462"></a> -<span class="sourceLineNo">463</span> loader.setMaxProcId(maxProcId);<a name="line.463"></a> -<span class="sourceLineNo">464</span> }<a name="line.464"></a> -<span class="sourceLineNo">465</span><a name="line.465"></a> -<span class="sourceLineNo">466</span> @Override<a name="line.466"></a> -<span class="sourceLineNo">467</span> public void load(ProcedureIterator procIter) throws IOException {<a name="line.467"></a> -<span class="sourceLineNo">468</span> loader.load(procIter);<a name="line.468"></a> -<span class="sourceLineNo">469</span> }<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 void handleCorrupted(ProcedureIterator procIter) throws IOException {<a name="line.472"></a> -<span class="sourceLineNo">473</span> loader.handleCorrupted(procIter);<a name="line.473"></a> -<span class="sourceLineNo">474</span> }<a name="line.474"></a> -<span class="sourceLineNo">475</span><a name="line.475"></a> -<span class="sourceLineNo">476</span> @Override<a name="line.476"></a> -<span class="sourceLineNo">477</span> public void markCorruptedWAL(ProcedureWALFile log, IOException e) {<a name="line.477"></a> -<span class="sourceLineNo">478</span> if (corruptedLogs == null) {<a name="line.478"></a> -<span class="sourceLineNo">479</span> corruptedLogs = new HashSet<>();<a name="line.479"></a> -<span class="sourceLineNo">480</span> }<a name="line.480"></a> -<span class="sourceLineNo">481</span> corruptedLogs.add(log);<a name="line.481"></a> -<span class="sourceLineNo">482</span> // TODO: sideline corrupted log<a name="line.482"></a> -<span class="sourceLineNo">483</span> }<a name="line.483"></a> -<span class="sourceLineNo">484</span> });<a name="line.484"></a> -<span class="sourceLineNo">485</span> } finally {<a name="line.485"></a> -<span class="sourceLineNo">486</span> try {<a name="line.486"></a> -<span class="sourceLineNo">487</span> // try to cleanup inactive wals and complete the operation<a name="line.487"></a> -<span class="sourceLineNo">488</span> buildHoldingCleanupTracker();<a name="line.488"></a> -<span class="sourceLineNo">489</span> tryCleanupLogsOnLoad();<a name="line.489"></a> -<span class="sourceLineNo">490</span> loading.set(false);<a name="line.490"></a> -<span class="sourceLineNo">491</span> } finally {<a name="line.491"></a> -<span class="sourceLineNo">492</span> lock.unlock();<a name="line.492"></a> -<span class="sourceLineNo">493</span> }<a name="line.493"></a> -<span class="sourceLineNo">494</span> }<a name="line.494"></a> -<span class="sourceLineNo">495</span> }<a name="line.495"></a> -<span class="sourceLineNo">496</span><a name="line.496"></a> -<span class="sourceLineNo">497</span> private void tryCleanupLogsOnLoad() {<a name="line.497"></a> -<span class="sourceLineNo">498</span> // nothing to cleanup.<a name="line.498"></a> -<span class="sourceLineNo">499</span> if (logs.size() <= 1) return;<a name="line.499"></a> +<span class="sourceLineNo">459</span> // Load the old logs<a name="line.459"></a> +<span class="sourceLineNo">460</span> Iterator<ProcedureWALFile> it = logs.descendingIterator();<a name="line.460"></a> +<span class="sourceLineNo">461</span> it.next(); // Skip the current log<a name="line.461"></a> +<span class="sourceLineNo">462</span><a name="line.462"></a> +<span class="sourceLineNo">463</span> ProcedureWALFormat.load(it, storeTracker, new ProcedureWALFormat.Loader() {<a name="line.463"></a> +<span class="sourceLineNo">464</span><a name="line.464"></a> +<span class="sourceLineNo">465</span> @Override<a name="line.465"></a> +<span class="sourceLineNo">466</span> public void setMaxProcId(long maxProcId) {<a name="line.466"></a> +<span class="sourceLineNo">467</span> loader.setMaxProcId(maxProcId);<a name="line.467"></a> +<span class="sourceLineNo">468</span> }<a name="line.468"></a> +<span class="sourceLineNo">469</span><a name="line.469"></a> +<span class="sourceLineNo">470</span> @Override<a name="line.470"></a> +<span class="sourceLineNo">471</span> public void load(ProcedureIterator procIter) throws IOException {<a name="line.471"></a> +<span class="sourceLineNo">472</span> loader.load(procIter);<a name="line.472"></a> +<span class="sourceLineNo">473</span> }<a name="line.473"></a> +<span class="sourceLineNo">474</span><a name="line.474"></a> +<span class="sourceLineNo">475</span> @Override<a name="line.475"></a> +<span class="sourceLineNo">476</span> public void handleCorrupted(ProcedureIterator procIter) throws IOException {<a name="line.476"></a> +<span class="sourceLineNo">477</span> loader.handleCorrupted(procIter);<a name="line.477"></a> +<span class="sourceLineNo">478</span> }<a name="line.478"></a> +<span class="sourceLineNo">479</span><a name="line.479"></a> +<span class="sourceLineNo">480</span> @Override<a name="line.480"></a> +<span class="sourceLineNo">481</span> public void markCorruptedWAL(ProcedureWALFile log, IOException e) {<a name="line.481"></a> +<span class="sourceLineNo">482</span> if (corruptedLogs == null) {<a name="line.482"></a> +<span class="sourceLineNo">483</span> corruptedLogs = new HashSet<>();<a name="line.483"></a> +<span class="sourceLineNo">484</span> }<a name="line.484"></a> +<span class="sourceLineNo">485</span> corruptedLogs.add(log);<a name="line.485"></a> +<span class="sourceLineNo">486</span> // TODO: sideline corrupted log<a name="line.486"></a> +<span class="sourceLineNo">487</span> }<a name="line.487"></a> +<span class="sourceLineNo">488</span> });<a name="line.488"></a> +<span class="sourceLineNo">489</span> } finally {<a name="line.489"></a> +<span class="sourceLineNo">490</span> try {<a name="line.490"></a> +<span class="sourceLineNo">491</span> // try to cleanup inactive wals and complete the operation<a name="line.491"></a> +<span class="sourceLineNo">492</span> buildHoldingCleanupTracker();<a name="line.492"></a> +<span class="sourceLineNo">493</span> tryCleanupLogsOnLoad();<a name="line.493"></a> +<span class="sourceLineNo">494</span> loading.set(false);<a name="line.494"></a> +<span class="sourceLineNo">495</span> } finally {<a name="line.495"></a> +<span class="sourceLineNo">496</span> lock.unlock();<a name="line.496"></a> +<span class="sourceLineNo">497</span> }<a name="line.497"></a> +<span class="sourceLineNo">498</span> }<a name="line.498"></a> +<span class="sourceLineNo">499</span> }<a name="line.499"></a> <span class="sourceLineNo">500</span><a name="line.500"></a> -<span class="sourceLineNo">501</span> // the config says to not cleanup wals on load.<a name="line.501"></a> -<span class="sourceLineNo">502</span> if (!conf.getBoolean(EXEC_WAL_CLEANUP_ON_LOAD_CONF_KEY,<a name="line.502"></a> -<span class="sourceLineNo">503</span> DEFAULT_EXEC_WAL_CLEANUP_ON_LOAD_CONF_KEY)) {<a name="line.503"></a> -<span class="sourceLineNo">504</span> LOG.debug("WALs cleanup on load is not enabled: " + getActiveLogs());<a name="line.504"></a> -<span class="sourceLineNo">505</span> return;<a name="line.505"></a> -<span class="sourceLineNo">506</span> }<a name="line.506"></a> -<span class="sourceLineNo">507</span><a name="line.507"></a> -<span class="sourceLineNo">508</span> try {<a name="line.508"></a> -<span class="sourceLineNo">509</span> periodicRoll();<a name="line.509"></a> -<span class="sourceLineNo">510</span> } catch (IOException e) {<a name="line.510"></a> -<span class="sourceLineNo">511</span> LOG.warn("Unable to cleanup logs on load: " + e.getMessage(), e);<a name="line.511"></a> +<span class="sourceLineNo">501</span> private void tryCleanupLogsOnLoad() {<a name="line.501"></a> +<span class="sourceLineNo">502</span> // nothing to cleanup.<a name="line.502"></a> +<span class="sourceLineNo">503</span> if (logs.size() <= 1) {<a name="line.503"></a> +<span class="sourceLineNo">504</span> return;<a name="line.504"></a> +<span class="sourceLineNo">505</span> }<a name="line.505"></a> +<span class="sourceLineNo">506</span><a name="line.506"></a> +<span class="sourceLineNo">507</span> // the config says to not cleanup wals on load.<a name="lin
<TRUNCATED>