View Javadoc

1   /**
2    *
3    * Licensed to the Apache Software Foundation (ASF) under one
4    * or more contributor license agreements.  See the NOTICE file
5    * distributed with this work for additional information
6    * regarding copyright ownership.  The ASF licenses this file
7    * to you under the Apache License, Version 2.0 (the
8    * "License"); you may not use this file except in compliance
9    * with the License.  You may obtain a copy of the License at
10   *
11   *     http://www.apache.org/licenses/LICENSE-2.0
12   *
13   * Unless required by applicable law or agreed to in writing, software
14   * distributed under the License is distributed on an "AS IS" BASIS,
15   * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
16   * See the License for the specific language governing permissions and
17   * limitations under the License.
18   */
19  package org.apache.hadoop.hbase.regionserver.wal;
20  
21  import java.io.EOFException;
22  import java.io.FileNotFoundException;
23  import java.io.IOException;
24  import java.io.InterruptedIOException;
25  import java.text.ParseException;
26  import java.util.ArrayList;
27  import java.util.Collections;
28  import java.util.HashSet;
29  import java.util.LinkedList;
30  import java.util.List;
31  import java.util.Map;
32  import java.util.Set;
33  import java.util.TreeMap;
34  import java.util.TreeSet;
35  import java.util.UUID;
36  import java.util.concurrent.Callable;
37  import java.util.concurrent.CompletionService;
38  import java.util.concurrent.ConcurrentHashMap;
39  import java.util.concurrent.ExecutionException;
40  import java.util.concurrent.ExecutorCompletionService;
41  import java.util.concurrent.Future;
42  import java.util.concurrent.ThreadFactory;
43  import java.util.concurrent.ThreadPoolExecutor;
44  import java.util.concurrent.TimeUnit;
45  import java.util.concurrent.atomic.AtomicBoolean;
46  import java.util.concurrent.atomic.AtomicLong;
47  import java.util.concurrent.atomic.AtomicReference;
48  
49  import org.apache.commons.logging.Log;
50  import org.apache.commons.logging.LogFactory;
51  import org.apache.hadoop.hbase.classification.InterfaceAudience;
52  import org.apache.hadoop.conf.Configuration;
53  import org.apache.hadoop.fs.FileStatus;
54  import org.apache.hadoop.fs.FileSystem;
55  import org.apache.hadoop.fs.Path;
56  import org.apache.hadoop.hbase.Cell;
57  import org.apache.hadoop.hbase.CellScanner;
58  import org.apache.hadoop.hbase.CellUtil;
59  import org.apache.hadoop.hbase.HBaseConfiguration;
60  import org.apache.hadoop.hbase.HConstants;
61  import org.apache.hadoop.hbase.HRegionInfo;
62  import org.apache.hadoop.hbase.HRegionLocation;
63  import org.apache.hadoop.hbase.KeyValue;
64  import org.apache.hadoop.hbase.KeyValueUtil;
65  import org.apache.hadoop.hbase.RemoteExceptionHandler;
66  import org.apache.hadoop.hbase.ServerName;
67  import org.apache.hadoop.hbase.TableName;
68  import org.apache.hadoop.hbase.TableNotFoundException;
69  import org.apache.hadoop.hbase.Tag;
70  import org.apache.hadoop.hbase.TagType;
71  import org.apache.hadoop.hbase.client.ConnectionUtils;
72  import org.apache.hadoop.hbase.client.Delete;
73  import org.apache.hadoop.hbase.client.HConnection;
74  import org.apache.hadoop.hbase.client.HConnectionManager;
75  import org.apache.hadoop.hbase.client.Mutation;
76  import org.apache.hadoop.hbase.client.Put;
77  import org.apache.hadoop.hbase.exceptions.RegionOpeningException;
78  import org.apache.hadoop.hbase.io.HeapSize;
79  import org.apache.hadoop.hbase.master.SplitLogManager;
80  import org.apache.hadoop.hbase.monitoring.MonitoredTask;
81  import org.apache.hadoop.hbase.monitoring.TaskMonitor;
82  import org.apache.hadoop.hbase.protobuf.ProtobufUtil;
83  import org.apache.hadoop.hbase.protobuf.RequestConverter;
84  import org.apache.hadoop.hbase.protobuf.generated.AdminProtos.AdminService.BlockingInterface;
85  import org.apache.hadoop.hbase.protobuf.generated.AdminProtos.GetRegionInfoRequest;
86  import org.apache.hadoop.hbase.protobuf.generated.AdminProtos.GetRegionInfoResponse;
87  import org.apache.hadoop.hbase.protobuf.generated.AdminProtos.WALEntry;
88  import org.apache.hadoop.hbase.protobuf.generated.ClientProtos.MutationProto.MutationType;
89  import org.apache.hadoop.hbase.protobuf.generated.HBaseProtos;
90  import org.apache.hadoop.hbase.protobuf.generated.WALProtos.CompactionDescriptor;
91  import org.apache.hadoop.hbase.protobuf.generated.WALProtos.WALKey;
92  import org.apache.hadoop.hbase.protobuf.generated.ZooKeeperProtos.RegionStoreSequenceIds;
93  import org.apache.hadoop.hbase.protobuf.generated.ZooKeeperProtos.SplitLogTask.RecoveryMode;
94  import org.apache.hadoop.hbase.protobuf.generated.ZooKeeperProtos.StoreSequenceId;
95  import org.apache.hadoop.hbase.regionserver.HRegion;
96  import org.apache.hadoop.hbase.regionserver.LastSequenceId;
97  import org.apache.hadoop.hbase.regionserver.wal.HLog.Entry;
98  import org.apache.hadoop.hbase.regionserver.wal.HLog.Reader;
99  import org.apache.hadoop.hbase.regionserver.wal.HLog.Writer;
100 import org.apache.hadoop.hbase.util.Bytes;
101 import org.apache.hadoop.hbase.util.CancelableProgressable;
102 import org.apache.hadoop.hbase.util.ClassSize;
103 import org.apache.hadoop.hbase.util.EnvironmentEdgeManager;
104 import org.apache.hadoop.hbase.util.FSUtils;
105 import org.apache.hadoop.hbase.util.Pair;
106 import org.apache.hadoop.hbase.util.Threads;
107 import org.apache.hadoop.hbase.zookeeper.ZKSplitLog;
108 import org.apache.hadoop.hbase.zookeeper.ZKTable;
109 import org.apache.hadoop.hbase.zookeeper.ZooKeeperWatcher;
110 import org.apache.hadoop.io.MultipleIOException;
111 import org.apache.zookeeper.KeeperException;
112 
113 import com.google.common.base.Preconditions;
114 import com.google.common.collect.Lists;
115 import com.google.protobuf.ServiceException;
116 
117 /**
118  * This class is responsible for splitting up a bunch of regionserver commit log
119  * files that are no longer being written to, into new files, one per region for
120  * region to replay on startup. Delete the old log files when finished.
121  */
122 @InterfaceAudience.Private
123 public class HLogSplitter {
124   static final Log LOG = LogFactory.getLog(HLogSplitter.class);
125 
126   // Parameters for split process
127   protected final Path rootDir;
128   protected final FileSystem fs;
129   protected final Configuration conf;
130 
131   // Major subcomponents of the split process.
132   // These are separated into inner classes to make testing easier.
133   OutputSink outputSink;
134   EntryBuffers entryBuffers;
135 
136   private Set<TableName> disablingOrDisabledTables =
137       new HashSet<TableName>();
138   private ZooKeeperWatcher watcher;
139 
140   // If an exception is thrown by one of the other threads, it will be
141   // stored here.
142   protected AtomicReference<Throwable> thrown = new AtomicReference<Throwable>();
143 
144   // Wait/notify for when data has been produced by the reader thread,
145   // consumed by the reader thread, or an exception occurred
146   final Object dataAvailable = new Object();
147 
148   private MonitoredTask status;
149 
150   // For checking the latest flushed sequence id
151   protected final LastSequenceId sequenceIdChecker;
152 
153   protected boolean distributedLogReplay;
154 
155   // Map encodedRegionName -> lastFlushedSequenceId
156   protected Map<String, Long> lastFlushedSequenceIds = new ConcurrentHashMap<String, Long>();
157 
158   // Map encodedRegionName -> maxSeqIdInStores
159   protected Map<String, Map<byte[], Long>> regionMaxSeqIdInStores =
160       new ConcurrentHashMap<String, Map<byte[], Long>>();
161 
162   // Failed region server that the wal file being split belongs to
163   protected String failedServerName = "";
164 
165   // Number of writer threads
166   private final int numWriterThreads;
167 
168   // Min batch size when replay WAL edits
169   private final int minBatchSize;
170 
171   HLogSplitter(Configuration conf, Path rootDir,
172       FileSystem fs, LastSequenceId idChecker, ZooKeeperWatcher zkw, RecoveryMode mode) {
173     this.conf = HBaseConfiguration.create(conf);
174     String codecClassName = conf
175         .get(WALCellCodec.WAL_CELL_CODEC_CLASS_KEY, WALCellCodec.class.getName());
176     this.conf.set(HConstants.RPC_CODEC_CONF_KEY, codecClassName);
177     this.rootDir = rootDir;
178     this.fs = fs;
179     this.sequenceIdChecker = idChecker;
180     this.watcher = zkw;
181 
182     entryBuffers = new EntryBuffers(
183         this.conf.getInt("hbase.regionserver.hlog.splitlog.buffersize",
184             128*1024*1024));
185 
186     // a larger minBatchSize may slow down recovery because replay writer has to wait for
187     // enough edits before replaying them
188     this.minBatchSize = this.conf.getInt("hbase.regionserver.wal.logreplay.batch.size", 64);
189     this.distributedLogReplay = (RecoveryMode.LOG_REPLAY == mode);
190 
191     this.numWriterThreads = this.conf.getInt("hbase.regionserver.hlog.splitlog.writer.threads", 3);
192     if (zkw != null && this.distributedLogReplay) {
193       outputSink = new LogReplayOutputSink(numWriterThreads);
194     } else {
195       if (this.distributedLogReplay) {
196         LOG.info("ZooKeeperWatcher is passed in as NULL so disable distrubitedLogRepaly.");
197       }
198       this.distributedLogReplay = false;
199       outputSink = new LogRecoveredEditsOutputSink(numWriterThreads);
200     }
201 
202   }
203 
204   /**
205    * Splits a HLog file into region's recovered-edits directory.
206    * This is the main entry point for distributed log splitting from SplitLogWorker.
207    * <p>
208    * If the log file has N regions then N recovered.edits files will be produced.
209    * <p>
210    * @param rootDir
211    * @param logfile
212    * @param fs
213    * @param conf
214    * @param reporter
215    * @param idChecker
216    * @param zkw ZooKeeperWatcher if it's null, we will back to the old-style log splitting where we
217    *          dump out recoved.edits files for regions to replay on.
218    * @return false if it is interrupted by the progress-able.
219    * @throws IOException
220    */
221   public static boolean splitLogFile(Path rootDir, FileStatus logfile, FileSystem fs,
222       Configuration conf, CancelableProgressable reporter, LastSequenceId idChecker,
223       ZooKeeperWatcher zkw, RecoveryMode mode) throws IOException {
224     HLogSplitter s = new HLogSplitter(conf, rootDir, fs, idChecker, zkw, mode);
225     return s.splitLogFile(logfile, reporter);
226   }
227 
228   // A wrapper to split one log folder using the method used by distributed
229   // log splitting. Used by tools and unit tests. It should be package private.
230   // It is public only because TestWALObserver is in a different package,
231   // which uses this method to to log splitting.
232   public static List<Path> split(Path rootDir, Path logDir, Path oldLogDir,
233       FileSystem fs, Configuration conf) throws IOException {
234     FileStatus[] logfiles = fs.listStatus(logDir);
235     List<Path> splits = new ArrayList<Path>();
236     if (logfiles != null && logfiles.length > 0) {
237       for (FileStatus logfile: logfiles) {
238         HLogSplitter s = new HLogSplitter(conf, rootDir, fs, null, null, 
239           RecoveryMode.LOG_SPLITTING);
240         if (s.splitLogFile(logfile, null)) {
241           finishSplitLogFile(rootDir, oldLogDir, logfile.getPath(), conf);
242           if (s.outputSink.splits != null) {
243             splits.addAll(s.outputSink.splits);
244           }
245         }
246       }
247     }
248     if (!fs.delete(logDir, true)) {
249       throw new IOException("Unable to delete src dir: " + logDir);
250     }
251     return splits;
252   }
253 
254   // The real log splitter. It just splits one log file.
255   boolean splitLogFile(FileStatus logfile,
256       CancelableProgressable reporter) throws IOException {
257     boolean isCorrupted = false;
258     Preconditions.checkState(status == null);
259     boolean skipErrors = conf.getBoolean("hbase.hlog.split.skip.errors",
260       HLog.SPLIT_SKIP_ERRORS_DEFAULT);
261     int interval = conf.getInt("hbase.splitlog.report.interval.loglines", 1024);
262     Path logPath = logfile.getPath();
263     boolean outputSinkStarted = false;
264     boolean progress_failed = false;
265     int editsCount = 0;
266     int editsSkipped = 0;
267     Reader in = null;
268 
269     status =
270         TaskMonitor.get().createStatus(
271           "Splitting log file " + logfile.getPath() + "into a temporary staging area.");
272     try {
273       long logLength = logfile.getLen();
274       LOG.info("Splitting hlog: " + logPath + ", length=" + logLength);
275       LOG.info("DistributedLogReplay = " + this.distributedLogReplay);
276       status.setStatus("Opening log file");
277       if (reporter != null && !reporter.progress()) {
278         progress_failed = true;
279         return false;
280       }
281       try {
282         in = getReader(fs, logfile, conf, skipErrors, reporter);
283       } catch (CorruptedLogFileException e) {
284         LOG.warn("Could not get reader, corrupted log file " + logPath, e);
285         ZKSplitLog.markCorrupted(rootDir, logfile.getPath().getName(), fs);
286         isCorrupted = true;
287       }
288       if (in == null) {
289         LOG.warn("Nothing to split in log file " + logPath);
290         return true;
291       }
292       if(watcher != null) {
293         try {
294           disablingOrDisabledTables = ZKTable.getDisabledOrDisablingTables(watcher);
295         } catch (KeeperException e) {
296           throw new IOException("Can't get disabling/disabled tables", e);
297         }
298       }
299       int numOpenedFilesBeforeReporting = conf.getInt("hbase.splitlog.report.openedfiles", 3);
300       int numOpenedFilesLastCheck = 0;
301       outputSink.setReporter(reporter);
302       outputSink.startWriterThreads();
303       outputSinkStarted = true;
304       Entry entry;
305       Long lastFlushedSequenceId = -1L;
306       ServerName serverName = HLogUtil.getServerNameFromHLogDirectoryName(logPath);
307       failedServerName = (serverName == null) ? "" : serverName.getServerName();
308       while ((entry = getNextLogLine(in, logPath, skipErrors)) != null) {
309         byte[] region = entry.getKey().getEncodedRegionName();
310         String key = Bytes.toString(region);
311         lastFlushedSequenceId = lastFlushedSequenceIds.get(key);
312         if (lastFlushedSequenceId == null) {
313           if (this.distributedLogReplay) {
314             RegionStoreSequenceIds ids =
315                 SplitLogManager.getRegionFlushedSequenceId(this.watcher, failedServerName, key);
316             if (ids != null) {
317               lastFlushedSequenceId = ids.getLastFlushedSequenceId();
318             }
319           } else if (sequenceIdChecker != null) {
320             lastFlushedSequenceId = sequenceIdChecker.getLastSequenceId(region);
321           }
322           if (lastFlushedSequenceId == null) {
323             lastFlushedSequenceId = -1L;
324           }
325           lastFlushedSequenceIds.put(key, lastFlushedSequenceId);
326         }
327         if (lastFlushedSequenceId >= entry.getKey().getLogSeqNum()) {
328           editsSkipped++;
329           continue;
330         }
331         entryBuffers.appendEntry(entry);
332         editsCount++;
333         int moreWritersFromLastCheck = this.getNumOpenWriters() - numOpenedFilesLastCheck;
334         // If sufficient edits have passed, check if we should report progress.
335         if (editsCount % interval == 0
336             || moreWritersFromLastCheck > numOpenedFilesBeforeReporting) {
337           numOpenedFilesLastCheck = this.getNumOpenWriters();
338           String countsStr = (editsCount - (editsSkipped + outputSink.getSkippedEdits()))
339               + " edits, skipped " + editsSkipped + " edits.";
340           status.setStatus("Split " + countsStr);
341           if (reporter != null && !reporter.progress()) {
342             progress_failed = true;
343             return false;
344           }
345         }
346       }
347     } catch (InterruptedException ie) {
348       IOException iie = new InterruptedIOException();
349       iie.initCause(ie);
350       throw iie;
351     } catch (CorruptedLogFileException e) {
352       LOG.warn("Could not parse, corrupted log file " + logPath, e);
353       ZKSplitLog.markCorrupted(rootDir, logfile.getPath().getName(), fs);
354       isCorrupted = true;
355     } catch (IOException e) {
356       e = RemoteExceptionHandler.checkIOException(e);
357       throw e;
358     } finally {
359       LOG.debug("Finishing writing output logs and closing down.");
360       try {
361         if (null != in) {
362           in.close();
363         }
364       } catch (IOException exception) {
365         LOG.warn("Could not close wal reader: " + exception.getMessage());
366         LOG.debug("exception details", exception);
367       }
368       try {
369         if (outputSinkStarted) {
370           // Set progress_failed to true as the immediate following statement will reset its value
371           // when finishWritingAndClose() throws exception, progress_failed has the right value
372           progress_failed = true;
373           progress_failed = outputSink.finishWritingAndClose() == null;
374         }
375       } finally {
376         String msg =
377             "Processed " + editsCount + " edits across " + outputSink.getNumberOfRecoveredRegions()
378                 + " regions; log file=" + logPath + " is corrupted = " + isCorrupted
379                 + " progress failed = " + progress_failed;
380         LOG.info(msg);
381         status.markComplete(msg);
382       }
383     }
384     return !progress_failed;
385   }
386 
387   /**
388    * Completes the work done by splitLogFile by archiving logs
389    * <p>
390    * It is invoked by SplitLogManager once it knows that one of the
391    * SplitLogWorkers have completed the splitLogFile() part. If the master
392    * crashes then this function might get called multiple times.
393    * <p>
394    * @param logfile
395    * @param conf
396    * @throws IOException
397    */
398   public static void finishSplitLogFile(String logfile,
399       Configuration conf)  throws IOException {
400     Path rootdir = FSUtils.getRootDir(conf);
401     Path oldLogDir = new Path(rootdir, HConstants.HREGION_OLDLOGDIR_NAME);
402     Path logPath;
403     if (FSUtils.isStartingWithPath(rootdir, logfile)) {
404       logPath = new Path(logfile);
405     } else {
406       logPath = new Path(rootdir, logfile);
407     }
408     finishSplitLogFile(rootdir, oldLogDir, logPath, conf);
409   }
410 
411   static void finishSplitLogFile(Path rootdir, Path oldLogDir,
412       Path logPath, Configuration conf) throws IOException {
413     List<Path> processedLogs = new ArrayList<Path>();
414     List<Path> corruptedLogs = new ArrayList<Path>();
415     FileSystem fs;
416     fs = rootdir.getFileSystem(conf);
417     if (ZKSplitLog.isCorrupted(rootdir, logPath.getName(), fs)) {
418       corruptedLogs.add(logPath);
419     } else {
420       processedLogs.add(logPath);
421     }
422     archiveLogs(corruptedLogs, processedLogs, oldLogDir, fs, conf);
423     Path stagingDir = ZKSplitLog.getSplitLogDir(rootdir, logPath.getName());
424     fs.delete(stagingDir, true);
425   }
426 
427   /**
428    * Moves processed logs to a oldLogDir after successful processing Moves
429    * corrupted logs (any log that couldn't be successfully parsed to corruptDir
430    * (.corrupt) for later investigation
431    *
432    * @param corruptedLogs
433    * @param processedLogs
434    * @param oldLogDir
435    * @param fs
436    * @param conf
437    * @throws IOException
438    */
439   private static void archiveLogs(
440       final List<Path> corruptedLogs,
441       final List<Path> processedLogs, final Path oldLogDir,
442       final FileSystem fs, final Configuration conf) throws IOException {
443     final Path corruptDir = new Path(FSUtils.getRootDir(conf), conf.get(
444         "hbase.regionserver.hlog.splitlog.corrupt.dir",  HConstants.CORRUPT_DIR_NAME));
445 
446     if (!fs.mkdirs(corruptDir)) {
447       LOG.info("Unable to mkdir " + corruptDir);
448     }
449     fs.mkdirs(oldLogDir);
450 
451     // this method can get restarted or called multiple times for archiving
452     // the same log files.
453     for (Path corrupted : corruptedLogs) {
454       Path p = new Path(corruptDir, corrupted.getName());
455       if (fs.exists(corrupted)) {
456         if (!fs.rename(corrupted, p)) {
457           LOG.warn("Unable to move corrupted log " + corrupted + " to " + p);
458         } else {
459           LOG.warn("Moved corrupted log " + corrupted + " to " + p);
460         }
461       }
462     }
463 
464     for (Path p : processedLogs) {
465       Path newPath = FSHLog.getHLogArchivePath(oldLogDir, p);
466       if (fs.exists(p)) {
467         if (!FSUtils.renameAndSetModifyTime(fs, p, newPath)) {
468           LOG.warn("Unable to move  " + p + " to " + newPath);
469         } else {
470           LOG.debug("Archived processed log " + p + " to " + newPath);
471         }
472       }
473     }
474   }
475 
476   /**
477    * Path to a file under RECOVERED_EDITS_DIR directory of the region found in
478    * <code>logEntry</code> named for the sequenceid in the passed
479    * <code>logEntry</code>: e.g. /hbase/some_table/2323432434/recovered.edits/2332.
480    * This method also ensures existence of RECOVERED_EDITS_DIR under the region
481    * creating it if necessary.
482    * @param fs
483    * @param logEntry
484    * @param rootDir HBase root dir.
485    * @return Path to file into which to dump split log edits.
486    * @throws IOException
487    */
488   @SuppressWarnings("deprecation")
489   static Path getRegionSplitEditsPath(final FileSystem fs,
490       final Entry logEntry, final Path rootDir, boolean isCreate)
491   throws IOException {
492     Path tableDir = FSUtils.getTableDir(rootDir, logEntry.getKey().getTablename());
493     String encodedRegionName = Bytes.toString(logEntry.getKey().getEncodedRegionName());
494     Path regiondir = HRegion.getRegionDir(tableDir, encodedRegionName);
495     Path dir = HLogUtil.getRegionDirRecoveredEditsDir(regiondir);
496 
497     if (!fs.exists(regiondir)) {
498       LOG.info("This region's directory doesn't exist: "
499           + regiondir.toString() + ". It is very likely that it was" +
500           " already split so it's safe to discard those edits.");
501       return null;
502     }
503     if (fs.exists(dir) && fs.isFile(dir)) {
504       Path tmp = new Path("/tmp");
505       if (!fs.exists(tmp)) {
506         fs.mkdirs(tmp);
507       }
508       tmp = new Path(tmp,
509         HConstants.RECOVERED_EDITS_DIR + "_" + encodedRegionName);
510       LOG.warn("Found existing old file: " + dir + ". It could be some "
511         + "leftover of an old installation. It should be a folder instead. "
512         + "So moving it to " + tmp);
513       if (!fs.rename(dir, tmp)) {
514         LOG.warn("Failed to sideline old file " + dir);
515       }
516     }
517 
518     if (isCreate && !fs.exists(dir)) {
519       if (!fs.mkdirs(dir)) LOG.warn("mkdir failed on " + dir);
520     }
521     // Append file name ends with RECOVERED_LOG_TMPFILE_SUFFIX to ensure
522     // region's replayRecoveredEdits will not delete it
523     String fileName = formatRecoveredEditsFileName(logEntry.getKey().getLogSeqNum());
524     fileName = getTmpRecoveredEditsFileName(fileName);
525     return new Path(dir, fileName);
526   }
527 
528   static String getTmpRecoveredEditsFileName(String fileName) {
529     return fileName + HLog.RECOVERED_LOG_TMPFILE_SUFFIX;
530   }
531 
532   /**
533    * Get the completed recovered edits file path, renaming it to be by last edit
534    * in the file from its first edit. Then we could use the name to skip
535    * recovered edits when doing {@link HRegion#replayRecoveredEditsIfAny}.
536    * @param srcPath
537    * @param maximumEditLogSeqNum
538    * @return dstPath take file's last edit log seq num as the name
539    */
540   static Path getCompletedRecoveredEditsFilePath(Path srcPath,
541       Long maximumEditLogSeqNum) {
542     String fileName = formatRecoveredEditsFileName(maximumEditLogSeqNum);
543     return new Path(srcPath.getParent(), fileName);
544   }
545 
546   static String formatRecoveredEditsFileName(final long seqid) {
547     return String.format("%019d", seqid);
548   }
549 
550   /**
551    * Create a new {@link Reader} for reading logs to split.
552    *
553    * @param fs
554    * @param file
555    * @param conf
556    * @return A new Reader instance
557    * @throws IOException
558    * @throws CorruptedLogFileException
559    */
560   protected Reader getReader(FileSystem fs, FileStatus file, Configuration conf,
561       boolean skipErrors, CancelableProgressable reporter)
562       throws IOException, CorruptedLogFileException {
563     Path path = file.getPath();
564     long length = file.getLen();
565     Reader in;
566 
567     // Check for possibly empty file. With appends, currently Hadoop reports a
568     // zero length even if the file has been sync'd. Revisit if HDFS-376 or
569     // HDFS-878 is committed.
570     if (length <= 0) {
571       LOG.warn("File " + path + " might be still open, length is 0");
572     }
573 
574     try {
575       FSUtils.getInstance(fs, conf).recoverFileLease(fs, path, conf, reporter);
576       try {
577         in = getReader(fs, path, conf, reporter);
578       } catch (EOFException e) {
579         if (length <= 0) {
580           // TODO should we ignore an empty, not-last log file if skip.errors
581           // is false? Either way, the caller should decide what to do. E.g.
582           // ignore if this is the last log in sequence.
583           // TODO is this scenario still possible if the log has been
584           // recovered (i.e. closed)
585           LOG.warn("Could not open " + path + " for reading. File is empty", e);
586           return null;
587         } else {
588           // EOFException being ignored
589           return null;
590         }
591       }
592     } catch (IOException e) {
593       if (e instanceof FileNotFoundException) {
594         // A wal file may not exist anymore. Nothing can be recovered so move on
595         LOG.warn("File " + path + " doesn't exist anymore.", e);
596         return null;
597       }
598       if (!skipErrors || e instanceof InterruptedIOException) {
599         throw e; // Don't mark the file corrupted if interrupted, or not skipErrors
600       }
601       CorruptedLogFileException t =
602         new CorruptedLogFileException("skipErrors=true Could not open hlog " +
603             path + " ignoring");
604       t.initCause(e);
605       throw t;
606     }
607     return in;
608   }
609 
610   static private Entry getNextLogLine(Reader in, Path path, boolean skipErrors)
611   throws CorruptedLogFileException, IOException {
612     try {
613       return in.next();
614     } catch (EOFException eof) {
615       // truncated files are expected if a RS crashes (see HBASE-2643)
616       LOG.info("EOF from hlog " + path + ".  continuing");
617       return null;
618     } catch (IOException e) {
619       // If the IOE resulted from bad file format,
620       // then this problem is idempotent and retrying won't help
621       if (e.getCause() != null &&
622           (e.getCause() instanceof ParseException ||
623            e.getCause() instanceof org.apache.hadoop.fs.ChecksumException)) {
624         LOG.warn("Parse exception " + e.getCause().toString() + " from hlog "
625            + path + ".  continuing");
626         return null;
627       }
628       if (!skipErrors) {
629         throw e;
630       }
631       CorruptedLogFileException t =
632         new CorruptedLogFileException("skipErrors=true Ignoring exception" +
633             " while parsing hlog " + path + ". Marking as corrupted");
634       t.initCause(e);
635       throw t;
636     }
637   }
638 
639   private void writerThreadError(Throwable t) {
640     thrown.compareAndSet(null, t);
641   }
642 
643   /**
644    * Check for errors in the writer threads. If any is found, rethrow it.
645    */
646   private void checkForErrors() throws IOException {
647     Throwable thrown = this.thrown.get();
648     if (thrown == null) return;
649     if (thrown instanceof IOException) {
650       throw new IOException(thrown);
651     } else {
652       throw new RuntimeException(thrown);
653     }
654   }
655   /**
656    * Create a new {@link Writer} for writing log splits.
657    */
658   protected Writer createWriter(FileSystem fs, Path logfile, Configuration conf)
659       throws IOException {
660     return HLogFactory.createRecoveredEditsWriter(fs, logfile, conf);
661   }
662 
663   /**
664    * Create a new {@link Reader} for reading logs to split.
665    */
666   protected Reader getReader(FileSystem fs, Path curLogFile,
667       Configuration conf, CancelableProgressable reporter) throws IOException {
668     return HLogFactory.createReader(fs, curLogFile, conf, reporter);
669   }
670 
671   /**
672    * Get current open writers
673    */
674   private int getNumOpenWriters() {
675     int result = 0;
676     if (this.outputSink != null) {
677       result += this.outputSink.getNumOpenWriters();
678     }
679     return result;
680   }
681 
682   /**
683    * Class which accumulates edits and separates them into a buffer per region
684    * while simultaneously accounting RAM usage. Blocks if the RAM usage crosses
685    * a predefined threshold.
686    *
687    * Writer threads then pull region-specific buffers from this class.
688    */
689   class EntryBuffers {
690     Map<byte[], RegionEntryBuffer> buffers =
691       new TreeMap<byte[], RegionEntryBuffer>(Bytes.BYTES_COMPARATOR);
692 
693     /* Track which regions are currently in the middle of writing. We don't allow
694        an IO thread to pick up bytes from a region if we're already writing
695        data for that region in a different IO thread. */
696     Set<byte[]> currentlyWriting = new TreeSet<byte[]>(Bytes.BYTES_COMPARATOR);
697 
698     long totalBuffered = 0;
699     long maxHeapUsage;
700 
701     EntryBuffers(long maxHeapUsage) {
702       this.maxHeapUsage = maxHeapUsage;
703     }
704 
705     /**
706      * Append a log entry into the corresponding region buffer.
707      * Blocks if the total heap usage has crossed the specified threshold.
708      *
709      * @throws InterruptedException
710      * @throws IOException
711      */
712     void appendEntry(Entry entry) throws InterruptedException, IOException {
713       HLogKey key = entry.getKey();
714 
715       RegionEntryBuffer buffer;
716       long incrHeap;
717       synchronized (this) {
718         buffer = buffers.get(key.getEncodedRegionName());
719         if (buffer == null) {
720           buffer = new RegionEntryBuffer(key.getTablename(), key.getEncodedRegionName());
721           buffers.put(key.getEncodedRegionName(), buffer);
722         }
723         incrHeap= buffer.appendEntry(entry);
724       }
725 
726       // If we crossed the chunk threshold, wait for more space to be available
727       synchronized (dataAvailable) {
728         totalBuffered += incrHeap;
729         while (totalBuffered > maxHeapUsage && thrown.get() == null) {
730           LOG.debug("Used " + totalBuffered + " bytes of buffered edits, waiting for IO threads...");
731           dataAvailable.wait(2000);
732         }
733         dataAvailable.notifyAll();
734       }
735       checkForErrors();
736     }
737 
738     /**
739      * @return RegionEntryBuffer a buffer of edits to be written or replayed.
740      */
741     synchronized RegionEntryBuffer getChunkToWrite() {
742       long biggestSize = 0;
743       byte[] biggestBufferKey = null;
744 
745       for (Map.Entry<byte[], RegionEntryBuffer> entry : buffers.entrySet()) {
746         long size = entry.getValue().heapSize();
747         if (size > biggestSize && (!currentlyWriting.contains(entry.getKey()))) {
748           biggestSize = size;
749           biggestBufferKey = entry.getKey();
750         }
751       }
752       if (biggestBufferKey == null) {
753         return null;
754       }
755 
756       RegionEntryBuffer buffer = buffers.remove(biggestBufferKey);
757       currentlyWriting.add(biggestBufferKey);
758       return buffer;
759     }
760 
761     void doneWriting(RegionEntryBuffer buffer) {
762       synchronized (this) {
763         boolean removed = currentlyWriting.remove(buffer.encodedRegionName);
764         assert removed;
765       }
766       long size = buffer.heapSize();
767 
768       synchronized (dataAvailable) {
769         totalBuffered -= size;
770         // We may unblock writers
771         dataAvailable.notifyAll();
772       }
773     }
774 
775     synchronized boolean isRegionCurrentlyWriting(byte[] region) {
776       return currentlyWriting.contains(region);
777     }
778   }
779 
780   /**
781    * A buffer of some number of edits for a given region.
782    * This accumulates edits and also provides a memory optimization in order to
783    * share a single byte array instance for the table and region name.
784    * Also tracks memory usage of the accumulated edits.
785    */
786   static class RegionEntryBuffer implements HeapSize {
787     long heapInBuffer = 0;
788     List<Entry> entryBuffer;
789     TableName tableName;
790     byte[] encodedRegionName;
791 
792     RegionEntryBuffer(TableName tableName, byte[] region) {
793       this.tableName = tableName;
794       this.encodedRegionName = region;
795       this.entryBuffer = new LinkedList<Entry>();
796     }
797 
798     long appendEntry(Entry entry) {
799       internify(entry);
800       entryBuffer.add(entry);
801       long incrHeap = entry.getEdit().heapSize() +
802         ClassSize.align(2 * ClassSize.REFERENCE) + // HLogKey pointers
803         0; // TODO linkedlist entry
804       heapInBuffer += incrHeap;
805       return incrHeap;
806     }
807 
808     private void internify(Entry entry) {
809       HLogKey k = entry.getKey();
810       k.internTableName(this.tableName);
811       k.internEncodedRegionName(this.encodedRegionName);
812     }
813 
814     public long heapSize() {
815       return heapInBuffer;
816     }
817   }
818 
819   class WriterThread extends Thread {
820     private volatile boolean shouldStop = false;
821     private OutputSink outputSink = null;
822 
823     WriterThread(OutputSink sink, int i) {
824       super(Thread.currentThread().getName() + "-Writer-" + i);
825       outputSink = sink;
826     }
827 
828     public void run()  {
829       try {
830         doRun();
831       } catch (Throwable t) {
832         LOG.error("Exiting thread", t);
833         writerThreadError(t);
834       }
835     }
836 
837     private void doRun() throws IOException {
838       LOG.debug("Writer thread " + this + ": starting");
839       while (true) {
840         RegionEntryBuffer buffer = entryBuffers.getChunkToWrite();
841         if (buffer == null) {
842           // No data currently available, wait on some more to show up
843           synchronized (dataAvailable) {
844             if (shouldStop && !this.outputSink.flush()) {
845               return;
846             }
847             try {
848               dataAvailable.wait(500);
849             } catch (InterruptedException ie) {
850               if (!shouldStop) {
851                 throw new RuntimeException(ie);
852               }
853             }
854           }
855           continue;
856         }
857 
858         assert buffer != null;
859         try {
860           writeBuffer(buffer);
861         } finally {
862           entryBuffers.doneWriting(buffer);
863         }
864       }
865     }
866 
867     private void writeBuffer(RegionEntryBuffer buffer) throws IOException {
868       outputSink.append(buffer);
869     }
870 
871     void finish() {
872       synchronized (dataAvailable) {
873         shouldStop = true;
874         dataAvailable.notifyAll();
875       }
876     }
877   }
878 
879   /**
880    * The following class is an abstraction class to provide a common interface to support both
881    * existing recovered edits file sink and region server WAL edits replay sink
882    */
883    abstract class OutputSink {
884 
885     protected Map<byte[], SinkWriter> writers = Collections
886         .synchronizedMap(new TreeMap<byte[], SinkWriter>(Bytes.BYTES_COMPARATOR));;
887 
888     protected final Map<byte[], Long> regionMaximumEditLogSeqNum = Collections
889         .synchronizedMap(new TreeMap<byte[], Long>(Bytes.BYTES_COMPARATOR));
890 
891     protected final List<WriterThread> writerThreads = Lists.newArrayList();
892 
893     /* Set of regions which we've decided should not output edits */
894     protected final Set<byte[]> blacklistedRegions = Collections
895         .synchronizedSet(new TreeSet<byte[]>(Bytes.BYTES_COMPARATOR));
896 
897     protected boolean closeAndCleanCompleted = false;
898 
899     protected boolean writersClosed = false;
900 
901     protected final int numThreads;
902 
903     protected CancelableProgressable reporter = null;
904 
905     protected AtomicLong skippedEdits = new AtomicLong();
906 
907     protected List<Path> splits = null;
908 
909     public OutputSink(int numWriters) {
910       numThreads = numWriters;
911     }
912 
913     void setReporter(CancelableProgressable reporter) {
914       this.reporter = reporter;
915     }
916 
917     /**
918      * Start the threads that will pump data from the entryBuffers to the output files.
919      */
920     synchronized void startWriterThreads() {
921       for (int i = 0; i < numThreads; i++) {
922         WriterThread t = new WriterThread(this, i);
923         t.start();
924         writerThreads.add(t);
925       }
926     }
927 
928     /**
929      *
930      * Update region's maximum edit log SeqNum.
931      */
932     void updateRegionMaximumEditLogSeqNum(Entry entry) {
933       synchronized (regionMaximumEditLogSeqNum) {
934         Long currentMaxSeqNum = regionMaximumEditLogSeqNum.get(entry.getKey()
935             .getEncodedRegionName());
936         if (currentMaxSeqNum == null || entry.getKey().getLogSeqNum() > currentMaxSeqNum) {
937           regionMaximumEditLogSeqNum.put(entry.getKey().getEncodedRegionName(), entry.getKey()
938               .getLogSeqNum());
939         }
940       }
941     }
942 
943     Long getRegionMaximumEditLogSeqNum(byte[] region) {
944       return regionMaximumEditLogSeqNum.get(region);
945     }
946 
947     /**
948      * @return the number of currently opened writers
949      */
950     int getNumOpenWriters() {
951       return this.writers.size();
952     }
953 
954     long getSkippedEdits() {
955       return this.skippedEdits.get();
956     }
957 
958     /**
959      * Wait for writer threads to dump all info to the sink
960      * @return true when there is no error
961      * @throws IOException
962      */
963     protected boolean finishWriting() throws IOException {
964       LOG.info("Waiting for split writer threads to finish");
965       boolean progress_failed = false;
966       for (WriterThread t : writerThreads) {
967         t.finish();
968       }
969       for (WriterThread t : writerThreads) {
970         if (!progress_failed && reporter != null && !reporter.progress()) {
971           progress_failed = true;
972         }
973         try {
974           t.join();
975         } catch (InterruptedException ie) {
976           IOException iie = new InterruptedIOException();
977           iie.initCause(ie);
978           throw iie;
979         }
980       }
981       checkForErrors();
982       LOG.info("Split writers finished");
983       return (!progress_failed);
984     }
985 
986     abstract List<Path> finishWritingAndClose() throws IOException;
987 
988     /**
989      * @return a map from encoded region ID to the number of edits written out for that region.
990      */
991     abstract Map<byte[], Long> getOutputCounts();
992 
993     /**
994      * @return number of regions we've recovered
995      */
996     abstract int getNumberOfRecoveredRegions();
997 
998     /**
999      * @param buffer A WAL Edit Entry
1000      * @throws IOException
1001      */
1002     abstract void append(RegionEntryBuffer buffer) throws IOException;
1003 
1004     /**
1005      * WriterThread call this function to help flush internal remaining edits in buffer before close
1006      * @return true when underlying sink has something to flush
1007      */
1008     protected boolean flush() throws IOException {
1009       return false;
1010     }
1011   }
1012 
1013   /**
1014    * Class that manages the output streams from the log splitting process.
1015    */
1016   class LogRecoveredEditsOutputSink extends OutputSink {
1017 
1018     public LogRecoveredEditsOutputSink(int numWriters) {
1019       // More threads could potentially write faster at the expense
1020       // of causing more disk seeks as the logs are split.
1021       // 3. After a certain setting (probably around 3) the
1022       // process will be bound on the reader in the current
1023       // implementation anyway.
1024       super(numWriters);
1025     }
1026 
1027     /**
1028      * @return null if failed to report progress
1029      * @throws IOException
1030      */
1031     @Override
1032     List<Path> finishWritingAndClose() throws IOException {
1033       boolean isSuccessful = false;
1034       List<Path> result = null;
1035       try {
1036         isSuccessful = finishWriting();
1037       } finally {
1038         result = close();
1039         List<IOException> thrown = closeLogWriters(null);
1040         if (thrown != null && !thrown.isEmpty()) {
1041           throw MultipleIOException.createIOException(thrown);
1042         }
1043       }
1044       if (isSuccessful) {
1045         splits = result;
1046       }
1047       return splits;
1048     }
1049 
1050     /**
1051      * Close all of the output streams.
1052      * @return the list of paths written.
1053      */
1054     private List<Path> close() throws IOException {
1055       Preconditions.checkState(!closeAndCleanCompleted);
1056 
1057       final List<Path> paths = new ArrayList<Path>();
1058       final List<IOException> thrown = Lists.newArrayList();
1059       ThreadPoolExecutor closeThreadPool = Threads.getBoundedCachedThreadPool(numThreads, 30L,
1060         TimeUnit.SECONDS, new ThreadFactory() {
1061           private int count = 1;
1062 
1063           public Thread newThread(Runnable r) {
1064             Thread t = new Thread(r, "split-log-closeStream-" + count++);
1065             return t;
1066           }
1067         });
1068       CompletionService<Void> completionService =
1069         new ExecutorCompletionService<Void>(closeThreadPool);
1070       for (final Map.Entry<byte[], ? extends SinkWriter> writersEntry : writers.entrySet()) {
1071         LOG.debug("Submitting close of " + ((WriterAndPath)writersEntry.getValue()).p);
1072         completionService.submit(new Callable<Void>() {
1073           public Void call() throws Exception {
1074             WriterAndPath wap = (WriterAndPath) writersEntry.getValue();
1075             LOG.debug("Closing " + wap.p);
1076             try {
1077               wap.w.close();
1078             } catch (IOException ioe) {
1079               LOG.error("Couldn't close log at " + wap.p, ioe);
1080               thrown.add(ioe);
1081               return null;
1082             }
1083             LOG.info("Closed wap " + wap.p + " (wrote " + wap.editsWritten + " edits in "
1084                 + (wap.nanosSpent / 1000 / 1000) + "ms)");
1085 
1086             if (wap.editsWritten == 0) {
1087               // just remove the empty recovered.edits file
1088               if (fs.exists(wap.p) && !fs.delete(wap.p, false)) {
1089                 LOG.warn("Failed deleting empty " + wap.p);
1090                 throw new IOException("Failed deleting empty  " + wap.p);
1091               }
1092               return null;
1093             }
1094 
1095             Path dst = getCompletedRecoveredEditsFilePath(wap.p,
1096               regionMaximumEditLogSeqNum.get(writersEntry.getKey()));
1097             try {
1098               if (!dst.equals(wap.p) && fs.exists(dst)) {
1099                 LOG.warn("Found existing old edits file. It could be the "
1100                     + "result of a previous failed split attempt. Deleting " + dst + ", length="
1101                     + fs.getFileStatus(dst).getLen());
1102                 if (!fs.delete(dst, false)) {
1103                   LOG.warn("Failed deleting of old " + dst);
1104                   throw new IOException("Failed deleting of old " + dst);
1105                 }
1106               }
1107               // Skip the unit tests which create a splitter that reads and
1108               // writes the data without touching disk.
1109               // TestHLogSplit#testThreading is an example.
1110               if (fs.exists(wap.p)) {
1111                 if (!fs.rename(wap.p, dst)) {
1112                   throw new IOException("Failed renaming " + wap.p + " to " + dst);
1113                 }
1114                 LOG.debug("Rename " + wap.p + " to " + dst);
1115               }
1116             } catch (IOException ioe) {
1117               LOG.error("Couldn't rename " + wap.p + " to " + dst, ioe);
1118               thrown.add(ioe);
1119               return null;
1120             }
1121             paths.add(dst);
1122             return null;
1123           }
1124         });
1125       }
1126 
1127       boolean progress_failed = false;
1128       try {
1129         for (int i = 0, n = this.writers.size(); i < n; i++) {
1130           Future<Void> future = completionService.take();
1131           future.get();
1132           if (!progress_failed && reporter != null && !reporter.progress()) {
1133             progress_failed = true;
1134           }
1135         }
1136       } catch (InterruptedException e) {
1137         IOException iie = new InterruptedIOException();
1138         iie.initCause(e);
1139         throw iie;
1140       } catch (ExecutionException e) {
1141         throw new IOException(e.getCause());
1142       } finally {
1143         closeThreadPool.shutdownNow();
1144       }
1145 
1146       if (!thrown.isEmpty()) {
1147         throw MultipleIOException.createIOException(thrown);
1148       }
1149       writersClosed = true;
1150       closeAndCleanCompleted = true;
1151       if (progress_failed) {
1152         return null;
1153       }
1154       return paths;
1155     }
1156 
1157     private List<IOException> closeLogWriters(List<IOException> thrown) throws IOException {
1158       if (writersClosed) {
1159         return thrown;
1160       }
1161 
1162       if (thrown == null) {
1163         thrown = Lists.newArrayList();
1164       }
1165       try {
1166         for (WriterThread t : writerThreads) {
1167           while (t.isAlive()) {
1168             t.shouldStop = true;
1169             t.interrupt();
1170             try {
1171               t.join(10);
1172             } catch (InterruptedException e) {
1173               IOException iie = new InterruptedIOException();
1174               iie.initCause(e);
1175               throw iie;
1176             }
1177           }
1178         }
1179       } finally {
1180         synchronized (writers) {
1181           WriterAndPath wap = null;
1182           for (SinkWriter tmpWAP : writers.values()) {
1183             try {
1184               wap = (WriterAndPath) tmpWAP;
1185               wap.w.close();
1186             } catch (IOException ioe) {
1187               LOG.error("Couldn't close log at " + wap.p, ioe);
1188               thrown.add(ioe);
1189               continue;
1190             }
1191             LOG.info("Closed log " + wap.p + " (wrote " + wap.editsWritten + " edits in "
1192                 + (wap.nanosSpent / 1000 / 1000) + "ms)");
1193           }
1194         }
1195         writersClosed = true;
1196       }
1197 
1198       return thrown;
1199     }
1200 
1201     /**
1202      * Get a writer and path for a log starting at the given entry. This function is threadsafe so
1203      * long as multiple threads are always acting on different regions.
1204      * @return null if this region shouldn't output any logs
1205      */
1206     private WriterAndPath getWriterAndPath(Entry entry) throws IOException {
1207       byte region[] = entry.getKey().getEncodedRegionName();
1208       WriterAndPath ret = (WriterAndPath) writers.get(region);
1209       if (ret != null) {
1210         return ret;
1211       }
1212       // If we already decided that this region doesn't get any output
1213       // we don't need to check again.
1214       if (blacklistedRegions.contains(region)) {
1215         return null;
1216       }
1217       ret = createWAP(region, entry, rootDir, fs, conf);
1218       if (ret == null) {
1219         blacklistedRegions.add(region);
1220         return null;
1221       }
1222       writers.put(region, ret);
1223       return ret;
1224     }
1225 
1226     private WriterAndPath createWAP(byte[] region, Entry entry, Path rootdir, FileSystem fs,
1227         Configuration conf) throws IOException {
1228       Path regionedits = getRegionSplitEditsPath(fs, entry, rootdir, true);
1229       if (regionedits == null) {
1230         return null;
1231       }
1232       if (fs.exists(regionedits)) {
1233         LOG.warn("Found old edits file. It could be the "
1234             + "result of a previous failed split attempt. Deleting " + regionedits + ", length="
1235             + fs.getFileStatus(regionedits).getLen());
1236         if (!fs.delete(regionedits, false)) {
1237           LOG.warn("Failed delete of old " + regionedits);
1238         }
1239       }
1240       Writer w = createWriter(fs, regionedits, conf);
1241       LOG.debug("Creating writer path=" + regionedits + " region=" + Bytes.toStringBinary(region));
1242       return (new WriterAndPath(regionedits, w));
1243     }
1244 
1245     void append(RegionEntryBuffer buffer) throws IOException {
1246       List<Entry> entries = buffer.entryBuffer;
1247       if (entries.isEmpty()) {
1248         LOG.warn("got an empty buffer, skipping");
1249         return;
1250       }
1251 
1252       WriterAndPath wap = null;
1253 
1254       long startTime = System.nanoTime();
1255       try {
1256         int editsCount = 0;
1257 
1258         for (Entry logEntry : entries) {
1259           if (wap == null) {
1260             wap = getWriterAndPath(logEntry);
1261             if (wap == null) {
1262               // getWriterAndPath decided we don't need to write these edits
1263               return;
1264             }
1265           }
1266           wap.w.append(logEntry);
1267           this.updateRegionMaximumEditLogSeqNum(logEntry);
1268           editsCount++;
1269         }
1270         // Pass along summary statistics
1271         wap.incrementEdits(editsCount);
1272         wap.incrementNanoTime(System.nanoTime() - startTime);
1273       } catch (IOException e) {
1274         e = RemoteExceptionHandler.checkIOException(e);
1275         LOG.fatal(" Got while writing log entry to log", e);
1276         throw e;
1277       }
1278     }
1279 
1280     /**
1281      * @return a map from encoded region ID to the number of edits written out for that region.
1282      */
1283     Map<byte[], Long> getOutputCounts() {
1284       TreeMap<byte[], Long> ret = new TreeMap<byte[], Long>(Bytes.BYTES_COMPARATOR);
1285       synchronized (writers) {
1286         for (Map.Entry<byte[], ? extends SinkWriter> entry : writers.entrySet()) {
1287           ret.put(entry.getKey(), entry.getValue().editsWritten);
1288         }
1289       }
1290       return ret;
1291     }
1292 
1293     @Override
1294     int getNumberOfRecoveredRegions() {
1295       return writers.size();
1296     }
1297   }
1298 
1299   /**
1300    * Class wraps the actual writer which writes data out and related statistics
1301    */
1302   private abstract static class SinkWriter {
1303     /* Count of edits written to this path */
1304     long editsWritten = 0;
1305     /* Number of nanos spent writing to this log */
1306     long nanosSpent = 0;
1307 
1308     void incrementEdits(int edits) {
1309       editsWritten += edits;
1310     }
1311 
1312     void incrementNanoTime(long nanos) {
1313       nanosSpent += nanos;
1314     }
1315   }
1316 
1317   /**
1318    * Private data structure that wraps a Writer and its Path, also collecting statistics about the
1319    * data written to this output.
1320    */
1321   private final static class WriterAndPath extends SinkWriter {
1322     final Path p;
1323     final Writer w;
1324 
1325     WriterAndPath(final Path p, final Writer w) {
1326       this.p = p;
1327       this.w = w;
1328     }
1329   }
1330 
1331   /**
1332    * Class that manages to replay edits from WAL files directly to assigned fail over region servers
1333    */
1334   class LogReplayOutputSink extends OutputSink {
1335     private static final double BUFFER_THRESHOLD = 0.35;
1336     private static final String KEY_DELIMITER = "#";
1337 
1338     private long waitRegionOnlineTimeOut;
1339     private final Set<String> recoveredRegions = Collections.synchronizedSet(new HashSet<String>());
1340     private final Map<String, RegionServerWriter> writers =
1341         new ConcurrentHashMap<String, RegionServerWriter>();
1342     // online encoded region name -> region location map
1343     private final Map<String, HRegionLocation> onlineRegions =
1344         new ConcurrentHashMap<String, HRegionLocation>();
1345 
1346     private Map<TableName, HConnection> tableNameToHConnectionMap = Collections
1347         .synchronizedMap(new TreeMap<TableName, HConnection>());
1348     /**
1349      * Map key -> value layout
1350      * <servername>:<table name> -> Queue<Row>
1351      */
1352     private Map<String, List<Pair<HRegionLocation, HLog.Entry>>> serverToBufferQueueMap =
1353         new ConcurrentHashMap<String, List<Pair<HRegionLocation, HLog.Entry>>>();
1354     private List<Throwable> thrown = new ArrayList<Throwable>();
1355 
1356     // The following sink is used in distrubitedLogReplay mode for entries of regions in a disabling
1357     // table. It's a limitation of distributedLogReplay. Because log replay needs a region is
1358     // assigned and online before it can replay wal edits while regions of disabling/disabled table
1359     // won't be assigned by AM. We can retire this code after HBASE-8234.
1360     private LogRecoveredEditsOutputSink logRecoveredEditsOutputSink;
1361     private boolean hasEditsInDisablingOrDisabledTables = false;
1362 
1363     public LogReplayOutputSink(int numWriters) {
1364       super(numWriters);
1365       this.waitRegionOnlineTimeOut = conf.getInt("hbase.splitlog.manager.timeout",
1366         SplitLogManager.DEFAULT_TIMEOUT);
1367       this.logRecoveredEditsOutputSink = new LogRecoveredEditsOutputSink(numWriters);
1368       this.logRecoveredEditsOutputSink.setReporter(reporter);
1369     }
1370 
1371     void append(RegionEntryBuffer buffer) throws IOException {
1372       List<Entry> entries = buffer.entryBuffer;
1373       if (entries.isEmpty()) {
1374         LOG.warn("got an empty buffer, skipping");
1375         return;
1376       }
1377 
1378       // check if current region in a disabling or disabled table
1379       if (disablingOrDisabledTables.contains(buffer.tableName)) {
1380         // need fall back to old way
1381         logRecoveredEditsOutputSink.append(buffer);
1382         hasEditsInDisablingOrDisabledTables = true;
1383         // store regions we have recovered so far
1384         addToRecoveredRegions(Bytes.toString(buffer.encodedRegionName));
1385         return;
1386       }
1387 
1388       // group entries by region servers
1389       groupEditsByServer(entries);
1390 
1391       // process workitems
1392       String maxLocKey = null;
1393       int maxSize = 0;
1394       List<Pair<HRegionLocation, HLog.Entry>> maxQueue = null;
1395       synchronized (this.serverToBufferQueueMap) {
1396         for (String key : this.serverToBufferQueueMap.keySet()) {
1397           List<Pair<HRegionLocation, HLog.Entry>> curQueue = this.serverToBufferQueueMap.get(key);
1398           if (curQueue.size() > maxSize) {
1399             maxSize = curQueue.size();
1400             maxQueue = curQueue;
1401             maxLocKey = key;
1402           }
1403         }
1404         if (maxSize < minBatchSize
1405             && entryBuffers.totalBuffered < BUFFER_THRESHOLD * entryBuffers.maxHeapUsage) {
1406           // buffer more to process
1407           return;
1408         } else if (maxSize > 0) {
1409           this.serverToBufferQueueMap.remove(maxLocKey);
1410         }
1411       }
1412 
1413       if (maxSize > 0) {
1414         processWorkItems(maxLocKey, maxQueue);
1415       }
1416     }
1417 
1418     private void addToRecoveredRegions(String encodedRegionName) {
1419       if (!recoveredRegions.contains(encodedRegionName)) {
1420         recoveredRegions.add(encodedRegionName);
1421       }
1422     }
1423 
1424     /**
1425      * Helper function to group WALEntries to individual region servers
1426      * @throws IOException
1427      */
1428     private void groupEditsByServer(List<Entry> entries) throws IOException {
1429       Set<TableName> nonExistentTables = null;
1430       Long cachedLastFlushedSequenceId = -1l;
1431       for (HLog.Entry entry : entries) {
1432         WALEdit edit = entry.getEdit();
1433         TableName table = entry.getKey().getTablename();
1434         // clear scopes which isn't needed for recovery
1435         entry.getKey().setScopes(null);
1436         String encodeRegionNameStr = Bytes.toString(entry.getKey().getEncodedRegionName());
1437         // skip edits of non-existent tables
1438         if (nonExistentTables != null && nonExistentTables.contains(table)) {
1439           this.skippedEdits.incrementAndGet();
1440           continue;
1441         }
1442 
1443         Map<byte[], Long> maxStoreSequenceIds = null;
1444         boolean needSkip = false;
1445         HRegionLocation loc = null;
1446         String locKey = null;
1447         List<KeyValue> kvs = edit.getKeyValues();
1448         List<KeyValue> skippedKVs = new ArrayList<KeyValue>();
1449         HConnection hconn = this.getConnectionByTableName(table);
1450 
1451         for (KeyValue kv : kvs) {
1452           byte[] row = kv.getRow();
1453           byte[] family = kv.getFamily();
1454           boolean isCompactionEntry = false;
1455 	
1456           if (kv.matchingFamily(WALEdit.METAFAMILY)) {
1457             CompactionDescriptor compaction = WALEdit.getCompaction(kv);
1458             if (compaction != null && compaction.hasRegionName()) {
1459               try {
1460                 byte[][] regionName = HRegionInfo.parseRegionName(compaction.getRegionName()
1461                   .toByteArray());
1462                 row = regionName[1]; // startKey of the region
1463                 family = compaction.getFamilyName().toByteArray();
1464                 isCompactionEntry = true;
1465               } catch (Exception ex) {
1466                 LOG.warn("Unexpected exception received, ignoring " + ex);
1467                 skippedKVs.add(kv);
1468                 continue;
1469               }
1470             } else {
1471               skippedKVs.add(kv);
1472               continue;
1473             }
1474           }
1475 
1476           try {
1477             loc =
1478                 locateRegionAndRefreshLastFlushedSequenceId(hconn, table, row,
1479                   encodeRegionNameStr);
1480             // skip replaying the compaction if the region is gone
1481             if (isCompactionEntry && !encodeRegionNameStr.equalsIgnoreCase(
1482               loc.getRegionInfo().getEncodedName())) {
1483               LOG.info("Not replaying a compaction marker for an older region: "
1484                   + encodeRegionNameStr);
1485               needSkip = true;
1486             }
1487           } catch (TableNotFoundException ex) {
1488             // table has been deleted so skip edits of the table
1489             LOG.info("Table " + table + " doesn't exist. Skip log replay for region "
1490                 + encodeRegionNameStr);
1491             lastFlushedSequenceIds.put(encodeRegionNameStr, Long.MAX_VALUE);
1492             if (nonExistentTables == null) {
1493               nonExistentTables = new TreeSet<TableName>();
1494             }
1495             nonExistentTables.add(table);
1496             this.skippedEdits.incrementAndGet();
1497             needSkip = true;
1498             break;
1499           }
1500 
1501           cachedLastFlushedSequenceId =
1502               lastFlushedSequenceIds.get(loc.getRegionInfo().getEncodedName());
1503           if (cachedLastFlushedSequenceId != null
1504               && cachedLastFlushedSequenceId >= entry.getKey().getLogSeqNum()) {
1505             // skip the whole HLog entry
1506             this.skippedEdits.incrementAndGet();
1507             needSkip = true;
1508             break;
1509           } else {
1510             if (maxStoreSequenceIds == null) {
1511               maxStoreSequenceIds =
1512                   regionMaxSeqIdInStores.get(loc.getRegionInfo().getEncodedName());
1513             }
1514             if (maxStoreSequenceIds != null) {
1515               Long maxStoreSeqId = maxStoreSequenceIds.get(family);
1516               if (maxStoreSeqId == null || maxStoreSeqId >= entry.getKey().getLogSeqNum()) {
1517                 // skip current kv if column family doesn't exist anymore or already flushed
1518                 skippedKVs.add(kv);
1519                 continue;
1520               }
1521             }
1522           }
1523         }
1524 
1525         // skip the edit
1526         if (loc == null || needSkip) continue;
1527 
1528         if (!skippedKVs.isEmpty()) {
1529           kvs.removeAll(skippedKVs);
1530         }
1531 
1532         synchronized (serverToBufferQueueMap) {
1533           locKey = loc.getHostnamePort() + KEY_DELIMITER + table;
1534           List<Pair<HRegionLocation, HLog.Entry>> queue = serverToBufferQueueMap.get(locKey);
1535           if (queue == null) {
1536             queue =
1537                 Collections.synchronizedList(new ArrayList<Pair<HRegionLocation, HLog.Entry>>());
1538             serverToBufferQueueMap.put(locKey, queue);
1539           }
1540           queue.add(new Pair<HRegionLocation, HLog.Entry>(loc, entry));
1541         }
1542         // store regions we have recovered so far
1543         addToRecoveredRegions(loc.getRegionInfo().getEncodedName());
1544       }
1545     }
1546 
1547     /**
1548      * Locate destination region based on table name & row. This function also makes sure the
1549      * destination region is online for replay.
1550      * @throws IOException
1551      */
1552     private HRegionLocation locateRegionAndRefreshLastFlushedSequenceId(HConnection hconn,
1553         TableName table, byte[] row, String originalEncodedRegionName) throws IOException {
1554       // fetch location from cache
1555       HRegionLocation loc = onlineRegions.get(originalEncodedRegionName);
1556       if(loc != null) return loc;
1557       // fetch location from hbase:meta directly without using cache to avoid hit old dead server
1558       loc = hconn.getRegionLocation(table, row, true);
1559       if (loc == null) {
1560         throw new IOException("Can't locate location for row:" + Bytes.toString(row)
1561             + " of table:" + table);
1562       }
1563       // check if current row moves to a different region due to region merge/split
1564       if (!originalEncodedRegionName.equalsIgnoreCase(loc.getRegionInfo().getEncodedName())) {
1565         // originalEncodedRegionName should have already flushed
1566         lastFlushedSequenceIds.put(originalEncodedRegionName, Long.MAX_VALUE);
1567         HRegionLocation tmpLoc = onlineRegions.get(loc.getRegionInfo().getEncodedName());
1568         if (tmpLoc != null) return tmpLoc;
1569       }
1570 
1571       Long lastFlushedSequenceId = -1l;
1572       AtomicBoolean isRecovering = new AtomicBoolean(true);
1573       loc = waitUntilRegionOnline(loc, row, this.waitRegionOnlineTimeOut, isRecovering);
1574       if (!isRecovering.get()) {
1575         // region isn't in recovering at all because WAL file may contain a region that has
1576         // been moved to somewhere before hosting RS fails
1577         lastFlushedSequenceIds.put(loc.getRegionInfo().getEncodedName(), Long.MAX_VALUE);
1578         LOG.info("logReplay skip region: " + loc.getRegionInfo().getEncodedName()
1579             + " because it's not in recovering.");
1580       } else {
1581         Long cachedLastFlushedSequenceId =
1582             lastFlushedSequenceIds.get(loc.getRegionInfo().getEncodedName());
1583 
1584         // retrieve last flushed sequence Id from ZK. Because region postOpenDeployTasks will
1585         // update the value for the region
1586         RegionStoreSequenceIds ids =
1587             SplitLogManager.getRegionFlushedSequenceId(watcher, failedServerName, loc
1588                 .getRegionInfo().getEncodedName());
1589         if (ids != null) {
1590           lastFlushedSequenceId = ids.getLastFlushedSequenceId();
1591           Map<byte[], Long> storeIds = new TreeMap<byte[], Long>(Bytes.BYTES_COMPARATOR);
1592           List<StoreSequenceId> maxSeqIdInStores = ids.getStoreSequenceIdList();
1593           for (StoreSequenceId id : maxSeqIdInStores) {
1594             storeIds.put(id.getFamilyName().toByteArray(), id.getSequenceId());
1595           }
1596           regionMaxSeqIdInStores.put(loc.getRegionInfo().getEncodedName(), storeIds);
1597         }
1598 
1599         if (cachedLastFlushedSequenceId == null
1600             || lastFlushedSequenceId > cachedLastFlushedSequenceId) {
1601           lastFlushedSequenceIds.put(loc.getRegionInfo().getEncodedName(), lastFlushedSequenceId);
1602         }
1603       }
1604 
1605       onlineRegions.put(loc.getRegionInfo().getEncodedName(), loc);
1606       return loc;
1607     }
1608 
1609     private void processWorkItems(String key, List<Pair<HRegionLocation, HLog.Entry>> actions)
1610         throws IOException {
1611       RegionServerWriter rsw = null;
1612 
1613       long startTime = System.nanoTime();
1614       try {
1615         rsw = getRegionServerWriter(key);
1616         rsw.sink.replayEntries(actions);
1617 
1618         // Pass along summary statistics
1619         rsw.incrementEdits(actions.size());
1620         rsw.incrementNanoTime(System.nanoTime() - startTime);
1621       } catch (IOException e) {
1622         e = RemoteExceptionHandler.checkIOException(e);
1623         LOG.fatal(" Got while writing log entry to log", e);
1624         throw e;
1625       }
1626     }
1627 
1628     /**
1629      * Wait until region is online on the destination region server
1630      * @param loc
1631      * @param row
1632      * @param timeout How long to wait
1633      * @param isRecovering Recovering state of the region interested on destination region server.
1634      * @return True when region is online on the destination region server
1635      * @throws InterruptedException
1636      */
1637     private HRegionLocation waitUntilRegionOnline(HRegionLocation loc, byte[] row,
1638         final long timeout, AtomicBoolean isRecovering)
1639         throws IOException {
1640       final long endTime = EnvironmentEdgeManager.currentTimeMillis() + timeout;
1641       final long pause = conf.getLong(HConstants.HBASE_CLIENT_PAUSE,
1642         HConstants.DEFAULT_HBASE_CLIENT_PAUSE);
1643       boolean reloadLocation = false;
1644       TableName tableName = loc.getRegionInfo().getTable();
1645       int tries = 0;
1646       Throwable cause = null;
1647       while (endTime > EnvironmentEdgeManager.currentTimeMillis()) {
1648         try {
1649           // Try and get regioninfo from the hosting server.
1650           HConnection hconn = getConnectionByTableName(tableName);
1651           if(reloadLocation) {
1652             loc = hconn.getRegionLocation(tableName, row, true);
1653           }
1654           BlockingInterface remoteSvr = hconn.getAdmin(loc.getServerName());
1655           HRegionInfo region = loc.getRegionInfo();
1656           try {
1657             GetRegionInfoRequest request =
1658                 RequestConverter.buildGetRegionInfoRequest(region.getRegionName());
1659             GetRegionInfoResponse response = remoteSvr.getRegionInfo(null, request);
1660             if (HRegionInfo.convert(response.getRegionInfo()) != null) {
1661               isRecovering.set((response.hasIsRecovering()) ? response.getIsRecovering() : true);
1662               return loc;
1663             }
1664           } catch (ServiceException se) {
1665             throw ProtobufUtil.getRemoteException(se);
1666           }
1667         } catch (IOException e) {
1668           cause = e.getCause();
1669           if(!(cause instanceof RegionOpeningException)) {
1670             reloadLocation = true;
1671           }
1672         }
1673         long expectedSleep = ConnectionUtils.getPauseTime(pause, tries);
1674         try {
1675           Thread.sleep(expectedSleep);
1676         } catch (InterruptedException e) {
1677           throw new IOException("Interrupted when waiting region " +
1678               loc.getRegionInfo().getEncodedName() + " online.", e);
1679         }
1680         tries++;
1681       }
1682 
1683       throw new IOException("Timeout when waiting region " + loc.getRegionInfo().getEncodedName() +
1684         " online for " + timeout + " milliseconds.", cause);
1685     }
1686 
1687     @Override
1688     protected boolean flush() throws IOException {
1689       String curLoc = null;
1690       int curSize = 0;
1691       List<Pair<HRegionLocation, HLog.Entry>> curQueue = null;
1692       synchronized (this.serverToBufferQueueMap) {
1693         for (String locationKey : this.serverToBufferQueueMap.keySet()) {
1694           curQueue = this.serverToBufferQueueMap.get(locationKey);
1695           if (!curQueue.isEmpty()) {
1696             curSize = curQueue.size();
1697             curLoc = locationKey;
1698             break;
1699           }
1700         }
1701         if (curSize > 0) {
1702           this.serverToBufferQueueMap.remove(curLoc);
1703         }
1704       }
1705 
1706       if (curSize > 0) {
1707         this.processWorkItems(curLoc, curQueue);
1708         dataAvailable.notifyAll();
1709         return true;
1710       }
1711       return false;
1712     }
1713 
1714     void addWriterError(Throwable t) {
1715       thrown.add(t);
1716     }
1717 
1718     @Override
1719     List<Path> finishWritingAndClose() throws IOException {
1720       try {
1721         if (!finishWriting()) {
1722           return null;
1723         }
1724         if (hasEditsInDisablingOrDisabledTables) {
1725           splits = logRecoveredEditsOutputSink.finishWritingAndClose();
1726         } else {
1727           splits = new ArrayList<Path>();
1728         }
1729         // returns an empty array in order to keep interface same as old way
1730         return splits;
1731       } finally {
1732         List<IOException> thrown = closeRegionServerWriters();
1733         if (thrown != null && !thrown.isEmpty()) {
1734           throw MultipleIOException.createIOException(thrown);
1735         }
1736       }
1737     }
1738 
1739     @Override
1740     int getNumOpenWriters() {
1741       return this.writers.size() + this.logRecoveredEditsOutputSink.getNumOpenWriters();
1742     }
1743 
1744     private List<IOException> closeRegionServerWriters() throws IOException {
1745       List<IOException> result = null;
1746       if (!writersClosed) {
1747         result = Lists.newArrayList();
1748         try {
1749           for (WriterThread t : writerThreads) {
1750             while (t.isAlive()) {
1751               t.shouldStop = true;
1752               t.interrupt();
1753               try {
1754                 t.join(10);
1755               } catch (InterruptedException e) {
1756                 IOException iie = new InterruptedIOException();
1757                 iie.initCause(e);
1758                 throw iie;
1759               }
1760             }
1761           }
1762         } finally {
1763           synchronized (writers) {
1764             for (String locationKey : writers.keySet()) {
1765               RegionServerWriter tmpW = writers.get(locationKey);
1766               try {
1767                 tmpW.close();
1768               } catch (IOException ioe) {
1769                 LOG.error("Couldn't close writer for region server:" + locationKey, ioe);
1770                 result.add(ioe);
1771               }
1772             }
1773           }
1774 
1775           // close connections
1776           synchronized (this.tableNameToHConnectionMap) {
1777             for (TableName tableName : this.tableNameToHConnectionMap.keySet()) {
1778               HConnection hconn = this.tableNameToHConnectionMap.get(tableName);
1779               try {
1780                 hconn.clearRegionCache();
1781                 hconn.close();
1782               } catch (IOException ioe) {
1783                 result.add(ioe);
1784               }
1785             }
1786           }
1787           writersClosed = true;
1788         }
1789       }
1790       return result;
1791     }
1792 
1793     Map<byte[], Long> getOutputCounts() {
1794       TreeMap<byte[], Long> ret = new TreeMap<byte[], Long>(Bytes.BYTES_COMPARATOR);
1795       synchronized (writers) {
1796         for (Map.Entry<String, RegionServerWriter> entry : writers.entrySet()) {
1797           ret.put(Bytes.toBytes(entry.getKey()), entry.getValue().editsWritten);
1798         }
1799       }
1800       return ret;
1801     }
1802 
1803     @Override
1804     int getNumberOfRecoveredRegions() {
1805       return this.recoveredRegions.size();
1806     }
1807 
1808     /**
1809      * Get a writer and path for a log starting at the given entry. This function is threadsafe so
1810      * long as multiple threads are always acting on different regions.
1811      * @return null if this region shouldn't output any logs
1812      */
1813     private RegionServerWriter getRegionServerWriter(String loc) throws IOException {
1814       RegionServerWriter ret = writers.get(loc);
1815       if (ret != null) {
1816         return ret;
1817       }
1818 
1819       TableName tableName = getTableFromLocationStr(loc);
1820       if(tableName == null){
1821         throw new IOException("Invalid location string:" + loc + " found. Replay aborted.");
1822       }
1823 
1824       HConnection hconn = getConnectionByTableName(tableName);
1825       synchronized (writers) {
1826         ret = writers.get(loc);
1827         if (ret == null) {
1828           ret = new RegionServerWriter(conf, tableName, hconn);
1829           writers.put(loc, ret);
1830         }
1831       }
1832       return ret;
1833     }
1834 
1835     private HConnection getConnectionByTableName(final TableName tableName) throws IOException {
1836       HConnection hconn = this.tableNameToHConnectionMap.get(tableName);
1837       if (hconn == null) {
1838         synchronized (this.tableNameToHConnectionMap) {
1839           hconn = this.tableNameToHConnectionMap.get(tableName);
1840           if (hconn == null) {
1841             hconn = HConnectionManager.getConnection(conf);
1842             this.tableNameToHConnectionMap.put(tableName, hconn);
1843           }
1844         }
1845       }
1846       return hconn;
1847     }
1848     private TableName getTableFromLocationStr(String loc) {
1849       /**
1850        * location key is in format <server name:port>#<table name>
1851        */
1852       String[] splits = loc.split(KEY_DELIMITER);
1853       if (splits.length != 2) {
1854         return null;
1855       }
1856       return TableName.valueOf(splits[1]);
1857     }
1858   }
1859 
1860   /**
1861    * Private data structure that wraps a receiving RS and collecting statistics about the data
1862    * written to this newly assigned RS.
1863    */
1864   private final static class RegionServerWriter extends SinkWriter {
1865     final WALEditsReplaySink sink;
1866 
1867     RegionServerWriter(final Configuration conf, final TableName tableName, final HConnection conn)
1868         throws IOException {
1869       this.sink = new WALEditsReplaySink(conf, tableName, conn);
1870     }
1871 
1872     void close() throws IOException {
1873     }
1874   }
1875 
1876   static class CorruptedLogFileException extends Exception {
1877     private static final long serialVersionUID = 1L;
1878 
1879     CorruptedLogFileException(String s) {
1880       super(s);
1881     }
1882   }
1883 
1884   /** A struct used by getMutationsFromWALEntry */
1885   public static class MutationReplay {
1886     public MutationReplay(MutationType type, Mutation mutation, long nonceGroup, long nonce) {
1887       this.type = type;
1888       this.mutation = mutation;
1889       this.nonceGroup = nonceGroup;
1890       this.nonce = nonce;
1891     }
1892 
1893     public final MutationType type;
1894     public final Mutation mutation;
1895     public final long nonceGroup;
1896     public final long nonce;
1897   }
1898 
1899  /**
1900   * Tag original sequence number for each edit to be replayed
1901   * @param entry
1902   * @param cell
1903   * @return
1904   */
1905   private static Cell tagReplayLogSequenceNumber(WALEntry entry, Cell cell) {
1906     // Tag puts with original sequence number if there is no LOG_REPLAY_TAG yet
1907     boolean needAddRecoveryTag = true;
1908     if (cell.getTagsLengthUnsigned() > 0) {
1909       Tag tmpTag = Tag.getTag(cell.getTagsArray(), cell.getTagsOffset(),
1910           cell.getTagsLengthUnsigned(), TagType.LOG_REPLAY_TAG_TYPE);
1911       if (tmpTag != null) {
1912         // found an existing log replay tag so reuse it
1913         needAddRecoveryTag = false;
1914       }
1915     }
1916     if (needAddRecoveryTag) {
1917       List<Tag> newTags = new ArrayList<Tag>();
1918       Tag replayTag = new Tag(TagType.LOG_REPLAY_TAG_TYPE, Bytes.toBytes(entry.getKey()
1919           .getLogSequenceNumber()));
1920       newTags.add(replayTag);
1921       return KeyValue.cloneAndAddTags(cell, newTags);
1922     }
1923     return cell;
1924   }
1925 
1926   /**
1927    * This function is used to construct mutations from a WALEntry. It also reconstructs HLogKey &
1928    * WALEdit from the passed in WALEntry
1929    * @param entry
1930    * @param cells
1931    * @param logEntry pair of HLogKey and WALEdit instance stores HLogKey and WALEdit instances
1932    *          extracted from the passed in WALEntry.
1933    * @param addLogReplayTag
1934    * @return list of Pair<MutationType, Mutation> to be replayed
1935    * @throws IOException
1936    */
1937   public static List<MutationReplay> getMutationsFromWALEntry(WALEntry entry, CellScanner cells,
1938       Pair<HLogKey, WALEdit> logEntry, boolean addLogReplayTag) throws IOException {
1939 
1940     if (entry == null) {
1941       // return an empty array
1942       return new ArrayList<MutationReplay>();
1943     }
1944 
1945     int count = entry.getAssociatedCellCount();
1946     List<MutationReplay> mutations = new ArrayList<MutationReplay>();
1947     Cell previousCell = null;
1948     Mutation m = null;
1949     HLogKey key = null;
1950     WALEdit val = null;
1951     if (logEntry != null) val = new WALEdit();
1952 
1953     for (int i = 0; i < count; i++) {
1954       // Throw index out of bounds if our cell count is off
1955       if (!cells.advance()) {
1956         throw new ArrayIndexOutOfBoundsException("Expected=" + count + ", index=" + i);
1957       }
1958       Cell cell = cells.current();
1959       if (val != null) val.add(KeyValueUtil.ensureKeyValue(cell));
1960 
1961       boolean isNewRowOrType =
1962           previousCell == null || previousCell.getTypeByte() != cell.getTypeByte()
1963               || !CellUtil.matchingRow(previousCell, cell);
1964       if (isNewRowOrType) {
1965         // Create new mutation
1966         if (CellUtil.isDelete(cell)) {
1967           m = new Delete(cell.getRowArray(), cell.getRowOffset(), cell.getRowLength());
1968           // Deletes don't have nonces.
1969           mutations.add(new MutationReplay(
1970               MutationType.DELETE, m, HConstants.NO_NONCE, HConstants.NO_NONCE));
1971         } else {
1972           m = new Put(cell.getRowArray(), cell.getRowOffset(), cell.getRowLength());
1973           // Puts might come from increment or append, thus we need nonces.
1974           long nonceGroup = entry.getKey().hasNonceGroup()
1975               ? entry.getKey().getNonceGroup() : HConstants.NO_NONCE;
1976           long nonce = entry.getKey().hasNonce() ? entry.getKey().getNonce() : HConstants.NO_NONCE;
1977           mutations.add(new MutationReplay(MutationType.PUT, m, nonceGroup, nonce));
1978         }
1979       }
1980       if (CellUtil.isDelete(cell)) {
1981         ((Delete) m).addDeleteMarker(KeyValueUtil.ensureKeyValue(cell));
1982       } else {
1983         Cell tmpNewCell = cell;
1984         if (addLogReplayTag) {
1985           tmpNewCell = tagReplayLogSequenceNumber(entry, cell);
1986         }
1987         ((Put) m).add(KeyValueUtil.ensureKeyValue(tmpNewCell));
1988       }
1989       previousCell = cell;
1990     }
1991 
1992     // reconstruct HLogKey
1993     if (logEntry != null) {
1994       WALKey walKey = entry.getKey();
1995       List<UUID> clusterIds = new ArrayList<UUID>(walKey.getClusterIdsCount());
1996       for (HBaseProtos.UUID uuid : entry.getKey().getClusterIdsList()) {
1997         clusterIds.add(new UUID(uuid.getMostSigBits(), uuid.getLeastSigBits()));
1998       }
1999       key = new HLogKey(walKey.getEncodedRegionName().toByteArray(), TableName.valueOf(walKey
2000               .getTableName().toByteArray()), walKey.getLogSequenceNumber(), walKey.getWriteTime(),
2001               clusterIds, walKey.getNonceGroup(), walKey.getNonce());
2002       logEntry.setFirst(key);
2003       logEntry.setSecond(val);
2004     }
2005 
2006     return mutations;
2007   }
2008 }