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.mapreduce;
20  
21  import java.io.FileNotFoundException;
22  import java.io.IOException;
23  import java.io.InterruptedIOException;
24  import java.nio.ByteBuffer;
25  import java.util.ArrayList;
26  import java.util.Arrays;
27  import java.util.Collection;
28  import java.util.Deque;
29  import java.util.HashMap;
30  import java.util.HashSet;
31  import java.util.LinkedList;
32  import java.util.List;
33  import java.util.Map;
34  import java.util.Map.Entry;
35  import java.util.Set;
36  import java.util.TreeMap;
37  import java.util.concurrent.Callable;
38  import java.util.concurrent.ExecutionException;
39  import java.util.concurrent.ExecutorService;
40  import java.util.concurrent.Future;
41  import java.util.concurrent.LinkedBlockingQueue;
42  import java.util.concurrent.ThreadPoolExecutor;
43  import java.util.concurrent.TimeUnit;
44  import java.util.concurrent.atomic.AtomicLong;
45  
46  import org.apache.commons.lang.mutable.MutableInt;
47  import org.apache.commons.logging.Log;
48  import org.apache.commons.logging.LogFactory;
49  import org.apache.hadoop.hbase.classification.InterfaceAudience;
50  import org.apache.hadoop.hbase.classification.InterfaceStability;
51  import org.apache.hadoop.conf.Configuration;
52  import org.apache.hadoop.conf.Configured;
53  import org.apache.hadoop.fs.permission.FsPermission;
54  import org.apache.hadoop.fs.FileStatus;
55  import org.apache.hadoop.fs.FileSystem;
56  import org.apache.hadoop.fs.FileUtil;
57  import org.apache.hadoop.fs.Path;
58  import org.apache.hadoop.hbase.HBaseConfiguration;
59  import org.apache.hadoop.hbase.HColumnDescriptor;
60  import org.apache.hadoop.hbase.HConstants;
61  import org.apache.hadoop.hbase.HTableDescriptor;
62  import org.apache.hadoop.hbase.KeyValue;
63  import org.apache.hadoop.hbase.TableName;
64  import org.apache.hadoop.hbase.TableNotFoundException;
65  import org.apache.hadoop.hbase.client.HBaseAdmin;
66  import org.apache.hadoop.hbase.client.HConnection;
67  import org.apache.hadoop.hbase.client.HTable;
68  import org.apache.hadoop.hbase.client.RegionServerCallable;
69  import org.apache.hadoop.hbase.client.RpcRetryingCallerFactory;
70  import org.apache.hadoop.hbase.client.coprocessor.SecureBulkLoadClient;
71  import org.apache.hadoop.hbase.coprocessor.CoprocessorHost;
72  import org.apache.hadoop.hbase.io.HFileLink;
73  import org.apache.hadoop.hbase.io.HalfStoreFileReader;
74  import org.apache.hadoop.hbase.io.Reference;
75  import org.apache.hadoop.hbase.io.compress.Compression.Algorithm;
76  import org.apache.hadoop.hbase.io.hfile.CacheConfig;
77  import org.apache.hadoop.hbase.io.hfile.HFile;
78  import org.apache.hadoop.hbase.io.hfile.HFileContext;
79  import org.apache.hadoop.hbase.io.hfile.HFileContextBuilder;
80  import org.apache.hadoop.hbase.io.hfile.HFileScanner;
81  import org.apache.hadoop.hbase.protobuf.ProtobufUtil;
82  import org.apache.hadoop.hbase.regionserver.BloomType;
83  import org.apache.hadoop.hbase.regionserver.HStore;
84  import org.apache.hadoop.hbase.regionserver.StoreFile;
85  import org.apache.hadoop.hbase.security.UserProvider;
86  import org.apache.hadoop.hbase.security.token.FsDelegationToken;
87  import org.apache.hadoop.hbase.util.Bytes;
88  import org.apache.hadoop.hbase.util.FSHDFSUtils;
89  import org.apache.hadoop.hbase.util.Pair;
90  import org.apache.hadoop.util.Tool;
91  import org.apache.hadoop.util.ToolRunner;
92  
93  import com.google.common.collect.HashMultimap;
94  import com.google.common.collect.Multimap;
95  import com.google.common.collect.Multimaps;
96  import com.google.common.util.concurrent.ThreadFactoryBuilder;
97  
98  import java.util.UUID;
99  
100 /**
101  * Tool to load the output of HFileOutputFormat into an existing table.
102  * @see #usage()
103  */
104 @InterfaceAudience.Public
105 @InterfaceStability.Stable
106 public class LoadIncrementalHFiles extends Configured implements Tool {
107   private static final Log LOG = LogFactory.getLog(LoadIncrementalHFiles.class);
108   private HBaseAdmin hbAdmin;
109 
110   public static final String NAME = "completebulkload";
111   public static final String MAX_FILES_PER_REGION_PER_FAMILY
112     = "hbase.mapreduce.bulkload.max.hfiles.perRegion.perFamily";
113   private static final String ASSIGN_SEQ_IDS = "hbase.mapreduce.bulkload.assign.sequenceNumbers";
114 
115   private int maxFilesPerRegionPerFamily;
116   private boolean assignSeqIds;
117 
118   // Source filesystem
119   private FileSystem fs;
120   // Source delegation token
121   private FsDelegationToken fsDelegationToken;
122   private String bulkToken;
123   private UserProvider userProvider;
124 
125   private LoadIncrementalHFiles() {}
126 
127   public LoadIncrementalHFiles(Configuration conf) throws Exception {
128     super(conf);
129     initialize();
130   }
131 
132   private void initialize() throws Exception {
133     if (hbAdmin == null) {
134       // make a copy, just to be sure we're not overriding someone else's config
135       setConf(HBaseConfiguration.create(getConf()));
136       Configuration conf = getConf();
137       // disable blockcache for tool invocation, see HBASE-10500
138       conf.setFloat(HConstants.HFILE_BLOCK_CACHE_SIZE_KEY, 0);
139       this.hbAdmin = new HBaseAdmin(conf);
140       this.userProvider = UserProvider.instantiate(conf);
141       this.fsDelegationToken = new FsDelegationToken(userProvider, "renewer");
142       assignSeqIds = conf.getBoolean(ASSIGN_SEQ_IDS, true);
143       maxFilesPerRegionPerFamily = conf.getInt(MAX_FILES_PER_REGION_PER_FAMILY, 32);
144     }
145   }
146 
147   private void usage() {
148     System.err.println("usage: " + NAME +
149         " /path/to/hfileoutputformat-output " +
150         "tablename");
151   }
152 
153   /**
154    * Represents an HFile waiting to be loaded. An queue is used
155    * in this class in order to support the case where a region has
156    * split during the process of the load. When this happens,
157    * the HFile is split into two physical parts across the new
158    * region boundary, and each part is added back into the queue.
159    * The import process finishes when the queue is empty.
160    */
161   static class LoadQueueItem {
162     final byte[] family;
163     final Path hfilePath;
164 
165     public LoadQueueItem(byte[] family, Path hfilePath) {
166       this.family = family;
167       this.hfilePath = hfilePath;
168     }
169 
170     public String toString() {
171       return "family:"+ Bytes.toString(family) + " path:" + hfilePath.toString();
172     }
173   }
174 
175   /**
176    * Walk the given directory for all HFiles, and return a Queue
177    * containing all such files.
178    */
179   private void discoverLoadQueue(Deque<LoadQueueItem> ret, Path hfofDir)
180   throws IOException {
181     fs = hfofDir.getFileSystem(getConf());
182 
183     if (!fs.exists(hfofDir)) {
184       throw new FileNotFoundException("HFileOutputFormat dir " +
185           hfofDir + " not found");
186     }
187 
188     FileStatus[] familyDirStatuses = fs.listStatus(hfofDir);
189     if (familyDirStatuses == null) {
190       throw new FileNotFoundException("No families found in " + hfofDir);
191     }
192 
193     for (FileStatus stat : familyDirStatuses) {
194       if (!stat.isDir()) {
195         LOG.warn("Skipping non-directory " + stat.getPath());
196         continue;
197       }
198       Path familyDir = stat.getPath();
199       // Skip _logs, etc
200       if (familyDir.getName().startsWith("_")) continue;
201       byte[] family = familyDir.getName().getBytes();
202       Path[] hfiles = FileUtil.stat2Paths(fs.listStatus(familyDir));
203       for (Path hfile : hfiles) {
204         if (hfile.getName().startsWith("_")) continue;
205         ret.add(new LoadQueueItem(family, hfile));
206       }
207     }
208   }
209 
210   /**
211    * Perform a bulk load of the given directory into the given
212    * pre-existing table.  This method is not threadsafe.
213    *
214    * @param hfofDir the directory that was provided as the output path
215    * of a job using HFileOutputFormat
216    * @param table the table to load into
217    * @throws TableNotFoundException if table does not yet exist
218    */
219   @SuppressWarnings("deprecation")
220   public void doBulkLoad(Path hfofDir, final HTable table)
221     throws TableNotFoundException, IOException
222   {
223     final HConnection conn = table.getConnection();
224 
225     if (!conn.isTableAvailable(table.getName())) {
226       throw new TableNotFoundException("Table " +
227           Bytes.toStringBinary(table.getTableName()) +
228           "is not currently available.");
229     }
230 
231     // initialize thread pools
232     int nrThreads = getConf().getInt("hbase.loadincremental.threads.max",
233       Runtime.getRuntime().availableProcessors());
234     ThreadFactoryBuilder builder = new ThreadFactoryBuilder();
235     builder.setNameFormat("LoadIncrementalHFiles-%1$d");
236     ExecutorService pool = new ThreadPoolExecutor(nrThreads, nrThreads,
237         60, TimeUnit.SECONDS,
238         new LinkedBlockingQueue<Runnable>(),
239         builder.build());
240     ((ThreadPoolExecutor)pool).allowCoreThreadTimeOut(true);
241 
242     // LQI queue does not need to be threadsafe -- all operations on this queue
243     // happen in this thread
244     Deque<LoadQueueItem> queue = new LinkedList<LoadQueueItem>();
245     try {
246       discoverLoadQueue(queue, hfofDir);
247       // check whether there is invalid family name in HFiles to be bulkloaded
248       Collection<HColumnDescriptor> families = table.getTableDescriptor().getFamilies();
249       ArrayList<String> familyNames = new ArrayList<String>();
250       for (HColumnDescriptor family : families) {
251         familyNames.add(family.getNameAsString());
252       }
253       ArrayList<String> unmatchedFamilies = new ArrayList<String>();
254       for (LoadQueueItem lqi : queue) {
255         String familyNameInHFile = Bytes.toString(lqi.family);
256         if (!familyNames.contains(familyNameInHFile)) {
257           unmatchedFamilies.add(familyNameInHFile);
258         }
259       }
260       if (unmatchedFamilies.size() > 0) {
261         String msg =
262             "Unmatched family names found: unmatched family names in HFiles to be bulkloaded: "
263                 + unmatchedFamilies + "; valid family names of table "
264                 + Bytes.toString(table.getTableName()) + " are: " + familyNames;
265         LOG.error(msg);
266         throw new IOException(msg);
267       }
268       int count = 0;
269 
270       if (queue.isEmpty()) {
271         LOG.warn("Bulk load operation did not find any files to load in " +
272             "directory " + hfofDir.toUri() + ".  Does it contain files in " +
273             "subdirectories that correspond to column family names?");
274         return;
275       }
276 
277       //If using secure bulk load, get source delegation token, and
278       //prepare staging directory and token
279       // fs is the source filesystem
280       fsDelegationToken.acquireDelegationToken(fs);
281       if(isSecureBulkLoadEndpointAvailable()) {
282         bulkToken = new SecureBulkLoadClient(table).prepareBulkLoad(table.getName());
283       }
284 
285       // Assumes that region splits can happen while this occurs.
286       while (!queue.isEmpty()) {
287         // need to reload split keys each iteration.
288         final Pair<byte[][], byte[][]> startEndKeys = table.getStartEndKeys();
289         if (count != 0) {
290           LOG.info("Split occured while grouping HFiles, retry attempt " +
291               + count + " with " + queue.size() + " files remaining to group or split");
292         }
293 
294         int maxRetries = getConf().getInt("hbase.bulkload.retries.number", 0);
295         if (maxRetries != 0 && count >= maxRetries) {
296           LOG.error("Retry attempted " + count +  " times without completing, bailing out");
297           return;
298         }
299         count++;
300 
301         // Using ByteBuffer for byte[] equality semantics
302         Multimap<ByteBuffer, LoadQueueItem> regionGroups = groupOrSplitPhase(table,
303             pool, queue, startEndKeys);
304 
305         if (!checkHFilesCountPerRegionPerFamily(regionGroups)) {
306           // Error is logged inside checkHFilesCountPerRegionPerFamily.
307           throw new IOException("Trying to load more than " + maxFilesPerRegionPerFamily
308             + " hfiles to one family of one region");
309         }
310 
311         bulkLoadPhase(table, conn, pool, queue, regionGroups);
312 
313         // NOTE: The next iteration's split / group could happen in parallel to
314         // atomic bulkloads assuming that there are splits and no merges, and
315         // that we can atomically pull out the groups we want to retry.
316       }
317 
318     } finally {
319       fsDelegationToken.releaseDelegationToken();
320       if(bulkToken != null) {
321         new SecureBulkLoadClient(table).cleanupBulkLoad(bulkToken);
322       }
323       pool.shutdown();
324       if (queue != null && !queue.isEmpty()) {
325         StringBuilder err = new StringBuilder();
326         err.append("-------------------------------------------------\n");
327         err.append("Bulk load aborted with some files not yet loaded:\n");
328         err.append("-------------------------------------------------\n");
329         for (LoadQueueItem q : queue) {
330           err.append("  ").append(q.hfilePath).append('\n');
331         }
332         LOG.error(err);
333       }
334     }
335 
336     if (queue != null && !queue.isEmpty()) {
337         throw new RuntimeException("Bulk load aborted with some files not yet loaded."
338           + "Please check log for more details.");
339     }
340   }
341 
342   /**
343    * This takes the LQI's grouped by likely regions and attempts to bulk load
344    * them.  Any failures are re-queued for another pass with the
345    * groupOrSplitPhase.
346    */
347   protected void bulkLoadPhase(final HTable table, final HConnection conn,
348       ExecutorService pool, Deque<LoadQueueItem> queue,
349       final Multimap<ByteBuffer, LoadQueueItem> regionGroups) throws IOException {
350     // atomically bulk load the groups.
351     Set<Future<List<LoadQueueItem>>> loadingFutures = new HashSet<Future<List<LoadQueueItem>>>();
352     for (Entry<ByteBuffer, ? extends Collection<LoadQueueItem>> e: regionGroups.asMap().entrySet()) {
353       final byte[] first = e.getKey().array();
354       final Collection<LoadQueueItem> lqis =  e.getValue();
355 
356       final Callable<List<LoadQueueItem>> call = new Callable<List<LoadQueueItem>>() {
357         public List<LoadQueueItem> call() throws Exception {
358           List<LoadQueueItem> toRetry =
359               tryAtomicRegionLoad(conn, table.getName(), first, lqis);
360           return toRetry;
361         }
362       };
363       loadingFutures.add(pool.submit(call));
364     }
365 
366     // get all the results.
367     for (Future<List<LoadQueueItem>> future : loadingFutures) {
368       try {
369         List<LoadQueueItem> toRetry = future.get();
370 
371         // LQIs that are requeued to be regrouped.
372         queue.addAll(toRetry);
373 
374       } catch (ExecutionException e1) {
375         Throwable t = e1.getCause();
376         if (t instanceof IOException) {
377           // At this point something unrecoverable has happened.
378           // TODO Implement bulk load recovery
379           throw new IOException("BulkLoad encountered an unrecoverable problem", t);
380         }
381         LOG.error("Unexpected execution exception during bulk load", e1);
382         throw new IllegalStateException(t);
383       } catch (InterruptedException e1) {
384         LOG.error("Unexpected interrupted exception during bulk load", e1);
385         throw (InterruptedIOException)new InterruptedIOException().initCause(e1);
386       }
387     }
388   }
389 
390   private boolean checkHFilesCountPerRegionPerFamily(
391       final Multimap<ByteBuffer, LoadQueueItem> regionGroups) {
392     for (Entry<ByteBuffer,
393         ? extends Collection<LoadQueueItem>> e: regionGroups.asMap().entrySet()) {
394       final Collection<LoadQueueItem> lqis =  e.getValue();
395       HashMap<byte[], MutableInt> filesMap = new HashMap<byte[], MutableInt>();
396       for (LoadQueueItem lqi: lqis) {
397         MutableInt count = filesMap.get(lqi.family);
398         if (count == null) {
399           count = new MutableInt();
400           filesMap.put(lqi.family, count);
401         }
402         count.increment();
403         if (count.intValue() > maxFilesPerRegionPerFamily) {
404           LOG.error("Trying to load more than " + maxFilesPerRegionPerFamily
405             + " hfiles to family " + Bytes.toStringBinary(lqi.family)
406             + " of region with start key "
407             + Bytes.toStringBinary(e.getKey()));
408           return false;
409         }
410       }
411     }
412     return true;
413   }
414 
415   /**
416    * @return A Multimap<startkey, LoadQueueItem> that groups LQI by likely
417    * bulk load region targets.
418    */
419   private Multimap<ByteBuffer, LoadQueueItem> groupOrSplitPhase(final HTable table,
420       ExecutorService pool, Deque<LoadQueueItem> queue,
421       final Pair<byte[][], byte[][]> startEndKeys) throws IOException {
422     // <region start key, LQI> need synchronized only within this scope of this
423     // phase because of the puts that happen in futures.
424     Multimap<ByteBuffer, LoadQueueItem> rgs = HashMultimap.create();
425     final Multimap<ByteBuffer, LoadQueueItem> regionGroups = Multimaps.synchronizedMultimap(rgs);
426 
427     // drain LQIs and figure out bulk load groups
428     Set<Future<List<LoadQueueItem>>> splittingFutures = new HashSet<Future<List<LoadQueueItem>>>();
429     while (!queue.isEmpty()) {
430       final LoadQueueItem item = queue.remove();
431 
432       final Callable<List<LoadQueueItem>> call = new Callable<List<LoadQueueItem>>() {
433         public List<LoadQueueItem> call() throws Exception {
434           List<LoadQueueItem> splits = groupOrSplit(regionGroups, item, table, startEndKeys);
435           return splits;
436         }
437       };
438       splittingFutures.add(pool.submit(call));
439     }
440     // get all the results.  All grouping and splitting must finish before
441     // we can attempt the atomic loads.
442     for (Future<List<LoadQueueItem>> lqis : splittingFutures) {
443       try {
444         List<LoadQueueItem> splits = lqis.get();
445         if (splits != null) {
446           queue.addAll(splits);
447         }
448       } catch (ExecutionException e1) {
449         Throwable t = e1.getCause();
450         if (t instanceof IOException) {
451           LOG.error("IOException during splitting", e1);
452           throw (IOException)t; // would have been thrown if not parallelized,
453         }
454         LOG.error("Unexpected execution exception during splitting", e1);
455         throw new IllegalStateException(t);
456       } catch (InterruptedException e1) {
457         LOG.error("Unexpected interrupted exception during splitting", e1);
458         throw (InterruptedIOException)new InterruptedIOException().initCause(e1);
459       }
460     }
461     return regionGroups;
462   }
463 
464   // unique file name for the table
465   private String getUniqueName() {
466     return UUID.randomUUID().toString().replaceAll("-", "");
467   }
468 
469   protected List<LoadQueueItem> splitStoreFile(final LoadQueueItem item,
470       final HTable table, byte[] startKey,
471       byte[] splitKey) throws IOException {
472     final Path hfilePath = item.hfilePath;
473 
474     // We use a '_' prefix which is ignored when walking directory trees
475     // above.
476     final Path tmpDir = new Path(item.hfilePath.getParent(), "_tmp");
477 
478     LOG.info("HFile at " + hfilePath + " no longer fits inside a single " +
479     "region. Splitting...");
480 
481     String uniqueName = getUniqueName();
482     HColumnDescriptor familyDesc = table.getTableDescriptor().getFamily(item.family);
483     Path botOut = new Path(tmpDir, uniqueName + ".bottom");
484     Path topOut = new Path(tmpDir, uniqueName + ".top");
485     splitStoreFile(getConf(), hfilePath, familyDesc, splitKey,
486         botOut, topOut);
487 
488     FileSystem fs = tmpDir.getFileSystem(getConf());
489     fs.setPermission(tmpDir, FsPermission.valueOf("-rwxrwxrwx"));
490     fs.setPermission(botOut, FsPermission.valueOf("-rwxrwxrwx"));
491 
492     // Add these back at the *front* of the queue, so there's a lower
493     // chance that the region will just split again before we get there.
494     List<LoadQueueItem> lqis = new ArrayList<LoadQueueItem>(2);
495     lqis.add(new LoadQueueItem(item.family, botOut));
496     lqis.add(new LoadQueueItem(item.family, topOut));
497 
498     LOG.info("Successfully split into new HFiles " + botOut + " and " + topOut);
499     return lqis;
500   }
501 
502   /**
503    * Attempt to assign the given load queue item into its target region group.
504    * If the hfile boundary no longer fits into a region, physically splits
505    * the hfile such that the new bottom half will fit and returns the list of
506    * LQI's corresponding to the resultant hfiles.
507    *
508    * protected for testing
509    * @throws IOException
510    */
511   protected List<LoadQueueItem> groupOrSplit(Multimap<ByteBuffer, LoadQueueItem> regionGroups,
512       final LoadQueueItem item, final HTable table,
513       final Pair<byte[][], byte[][]> startEndKeys)
514       throws IOException {
515     final Path hfilePath = item.hfilePath;
516     HFile.Reader hfr = HFile.createReader(fs, hfilePath,
517         new CacheConfig(getConf()), getConf());
518     final byte[] first, last;
519     try {
520       hfr.loadFileInfo();
521       first = hfr.getFirstRowKey();
522       last = hfr.getLastRowKey();
523     }  finally {
524       hfr.close();
525     }
526 
527     LOG.info("Trying to load hfile=" + hfilePath +
528         " first=" + Bytes.toStringBinary(first) +
529         " last="  + Bytes.toStringBinary(last));
530     if (first == null || last == null) {
531       assert first == null && last == null;
532       // TODO what if this is due to a bad HFile?
533       LOG.info("hfile " + hfilePath + " has no entries, skipping");
534       return null;
535     }
536     if (Bytes.compareTo(first, last) > 0) {
537       throw new IllegalArgumentException(
538       "Invalid range: " + Bytes.toStringBinary(first) +
539       " > " + Bytes.toStringBinary(last));
540     }
541     int idx = Arrays.binarySearch(startEndKeys.getFirst(), first,
542         Bytes.BYTES_COMPARATOR);
543     if (idx < 0) {
544       // not on boundary, returns -(insertion index).  Calculate region it
545       // would be in.
546       idx = -(idx + 1) - 1;
547     }
548     final int indexForCallable = idx;
549 
550     /**
551      * we can consider there is a region hole in following conditions. 1) if idx < 0,then first
552      * region info is lost. 2) if the endkey of a region is not equal to the startkey of the next
553      * region. 3) if the endkey of the last region is not empty.
554      */
555     if (indexForCallable < 0) {
556       throw new IOException("The first region info for table "
557           + Bytes.toString(table.getTableName())
558           + " cann't be found in hbase:meta.Please use hbck tool to fix it first.");
559     } else if ((indexForCallable == startEndKeys.getFirst().length - 1)
560         && !Bytes.equals(startEndKeys.getSecond()[indexForCallable], HConstants.EMPTY_BYTE_ARRAY)) {
561       throw new IOException("The last region info for table "
562           + Bytes.toString(table.getTableName())
563           + " cann't be found in hbase:meta.Please use hbck tool to fix it first.");
564     } else if (indexForCallable + 1 < startEndKeys.getFirst().length
565         && !(Bytes.compareTo(startEndKeys.getSecond()[indexForCallable],
566           startEndKeys.getFirst()[indexForCallable + 1]) == 0)) {
567       throw new IOException("The endkey of one region for table "
568           + Bytes.toString(table.getTableName())
569           + " is not equal to the startkey of the next region in hbase:meta."
570           + "Please use hbck tool to fix it first.");
571     }
572 
573     boolean lastKeyInRange =
574       Bytes.compareTo(last, startEndKeys.getSecond()[idx]) < 0 ||
575       Bytes.equals(startEndKeys.getSecond()[idx], HConstants.EMPTY_BYTE_ARRAY);
576     if (!lastKeyInRange) {
577       List<LoadQueueItem> lqis = splitStoreFile(item, table,
578           startEndKeys.getFirst()[indexForCallable],
579           startEndKeys.getSecond()[indexForCallable]);
580       return lqis;
581     }
582 
583     // group regions.
584     regionGroups.put(ByteBuffer.wrap(startEndKeys.getFirst()[idx]), item);
585     return null;
586   }
587 
588   /**
589    * @deprecated Use {@link #tryAtomicRegionLoad(HConnection, TableName, byte[], Collection)}
590    */
591   @Deprecated
592   protected List<LoadQueueItem> tryAtomicRegionLoad(final HConnection conn,
593       final byte [] tableName, final byte[] first, Collection<LoadQueueItem> lqis)
594   throws IOException {
595     return tryAtomicRegionLoad(conn, TableName.valueOf(tableName), first, lqis);
596   }
597 
598   /**
599    * Attempts to do an atomic load of many hfiles into a region.  If it fails,
600    * it returns a list of hfiles that need to be retried.  If it is successful
601    * it will return an empty list.
602    *
603    * NOTE: To maintain row atomicity guarantees, region server callable should
604    * succeed atomically and fails atomically.
605    *
606    * Protected for testing.
607    *
608    * @return empty list if success, list of items to retry on recoverable
609    * failure
610    */
611   protected List<LoadQueueItem> tryAtomicRegionLoad(final HConnection conn,
612       final TableName tableName, final byte[] first, Collection<LoadQueueItem> lqis)
613   throws IOException {
614     final List<Pair<byte[], String>> famPaths =
615       new ArrayList<Pair<byte[], String>>(lqis.size());
616     for (LoadQueueItem lqi : lqis) {
617       famPaths.add(Pair.newPair(lqi.family, lqi.hfilePath.toString()));
618     }
619 
620     final RegionServerCallable<Boolean> svrCallable =
621         new RegionServerCallable<Boolean>(conn, tableName, first) {
622       @Override
623       public Boolean call() throws Exception {
624         SecureBulkLoadClient secureClient = null;
625         boolean success = false;
626 
627         try {
628           LOG.debug("Going to connect to server " + getLocation() + " for row "
629               + Bytes.toStringBinary(getRow()) + " with hfile group " + famPaths);
630           byte[] regionName = getLocation().getRegionInfo().getRegionName();
631           if (!isSecureBulkLoadEndpointAvailable()) {
632             success = ProtobufUtil.bulkLoadHFile(getStub(), famPaths, regionName, assignSeqIds);
633           } else {
634             HTable table = new HTable(conn.getConfiguration(), getTableName());
635             secureClient = new SecureBulkLoadClient(table);
636             success = secureClient.bulkLoadHFiles(famPaths, fsDelegationToken.getUserToken(),
637               bulkToken, getLocation().getRegionInfo().getStartKey());
638           }
639           return success;
640         } finally {
641           //Best effort copying of files that might not have been imported
642           //from the staging directory back to original location
643           //in user directory
644           if(secureClient != null && !success) {
645             FileSystem targetFs = FileSystem.get(getConf());
646             // Check to see if the source and target filesystems are the same
647             // If they are the same filesystem, we will try move the files back
648             // because previously we moved them to the staging directory.
649             if (FSHDFSUtils.isSameHdfs(getConf(), fs, targetFs)) {
650               for(Pair<byte[], String> el : famPaths) {
651                 Path hfileStagingPath = null;
652                 Path hfileOrigPath = new Path(el.getSecond());
653                 try {
654                   hfileStagingPath= new Path(secureClient.getStagingPath(bulkToken, el.getFirst()),
655                     hfileOrigPath.getName());
656                   if(targetFs.rename(hfileStagingPath, hfileOrigPath)) {
657                     LOG.debug("Moved back file " + hfileOrigPath + " from " +
658                         hfileStagingPath);
659                   } else if(targetFs.exists(hfileStagingPath)){
660                     LOG.debug("Unable to move back file " + hfileOrigPath + " from " +
661                         hfileStagingPath);
662                   }
663                 } catch(Exception ex) {
664                   LOG.debug("Unable to move back file " + hfileOrigPath + " from " +
665                       hfileStagingPath, ex);
666                 }
667               }
668             }
669           }
670         }
671       }
672     };
673 
674     try {
675       List<LoadQueueItem> toRetry = new ArrayList<LoadQueueItem>();
676       Configuration conf = getConf();
677       boolean success = RpcRetryingCallerFactory.instantiate(conf).<Boolean> newCaller()
678           .callWithRetries(svrCallable);
679       if (!success) {
680         LOG.warn("Attempt to bulk load region containing "
681             + Bytes.toStringBinary(first) + " into table "
682             + tableName  + " with files " + lqis
683             + " failed.  This is recoverable and they will be retried.");
684         toRetry.addAll(lqis); // return lqi's to retry
685       }
686       // success
687       return toRetry;
688     } catch (IOException e) {
689       LOG.error("Encountered unrecoverable error from region server, additional details: "
690           + svrCallable.getExceptionMessageAdditionalDetail(), e);
691       throw e;
692     }
693   }
694   
695   private boolean isSecureBulkLoadEndpointAvailable() {
696     String classes = getConf().get(CoprocessorHost.REGION_COPROCESSOR_CONF_KEY, "");
697     return classes.contains("org.apache.hadoop.hbase.security.access.SecureBulkLoadEndpoint");
698   }
699 
700   /**
701    * Split a storefile into a top and bottom half, maintaining
702    * the metadata, recreating bloom filters, etc.
703    */
704   static void splitStoreFile(
705       Configuration conf, Path inFile,
706       HColumnDescriptor familyDesc, byte[] splitKey,
707       Path bottomOut, Path topOut) throws IOException
708   {
709     // Open reader with no block cache, and not in-memory
710     Reference topReference = Reference.createTopReference(splitKey);
711     Reference bottomReference = Reference.createBottomReference(splitKey);
712 
713     copyHFileHalf(conf, inFile, topOut, topReference, familyDesc);
714     copyHFileHalf(conf, inFile, bottomOut, bottomReference, familyDesc);
715   }
716 
717   /**
718    * Copy half of an HFile into a new HFile.
719    */
720   private static void copyHFileHalf(
721       Configuration conf, Path inFile, Path outFile, Reference reference,
722       HColumnDescriptor familyDescriptor)
723   throws IOException {
724     FileSystem fs = inFile.getFileSystem(conf);
725     CacheConfig cacheConf = new CacheConfig(conf);
726     HalfStoreFileReader halfReader = null;
727     StoreFile.Writer halfWriter = null;
728     try {
729       halfReader = new HalfStoreFileReader(fs, inFile, cacheConf, reference, conf);
730       Map<byte[], byte[]> fileInfo = halfReader.loadFileInfo();
731 
732       int blocksize = familyDescriptor.getBlocksize();
733       Algorithm compression = familyDescriptor.getCompression();
734       BloomType bloomFilterType = familyDescriptor.getBloomFilterType();
735       HFileContext hFileContext = new HFileContextBuilder()
736                                   .withCompression(compression)
737                                   .withChecksumType(HStore.getChecksumType(conf))
738                                   .withBytesPerCheckSum(HStore.getBytesPerChecksum(conf))
739                                   .withBlockSize(blocksize)
740                                   .withDataBlockEncoding(familyDescriptor.getDataBlockEncoding())
741                                   .build();
742       halfWriter = new StoreFile.WriterBuilder(conf, cacheConf,
743           fs)
744               .withFilePath(outFile)
745               .withBloomType(bloomFilterType)
746               .withFileContext(hFileContext)
747               .build();
748       HFileScanner scanner = halfReader.getScanner(false, false, false);
749       scanner.seekTo();
750       do {
751         KeyValue kv = scanner.getKeyValue();
752         halfWriter.append(kv);
753       } while (scanner.next());
754 
755       for (Map.Entry<byte[],byte[]> entry : fileInfo.entrySet()) {
756         if (shouldCopyHFileMetaKey(entry.getKey())) {
757           halfWriter.appendFileInfo(entry.getKey(), entry.getValue());
758         }
759       }
760     } finally {
761       if (halfWriter != null) halfWriter.close();
762       if (halfReader != null) halfReader.close(cacheConf.shouldEvictOnClose());
763     }
764   }
765 
766   private static boolean shouldCopyHFileMetaKey(byte[] key) {
767     return !HFile.isReservedFileInfoKey(key);
768   }
769 
770   private boolean doesTableExist(TableName tableName) throws Exception {
771     return hbAdmin.tableExists(tableName);
772   }
773 
774   /*
775    * Infers region boundaries for a new table.
776    * Parameter:
777    *   bdryMap is a map between keys to an integer belonging to {+1, -1}
778    *     If a key is a start key of a file, then it maps to +1
779    *     If a key is an end key of a file, then it maps to -1
780    * Algo:
781    * 1) Poll on the keys in order:
782    *    a) Keep adding the mapped values to these keys (runningSum)
783    *    b) Each time runningSum reaches 0, add the start Key from when the runningSum had started to a boundary list.
784    * 2) Return the boundary list.
785    */
786   public static byte[][] inferBoundaries(TreeMap<byte[], Integer> bdryMap) {
787     ArrayList<byte[]> keysArray = new ArrayList<byte[]>();
788     int runningValue = 0;
789     byte[] currStartKey = null;
790     boolean firstBoundary = true;
791 
792     for (Map.Entry<byte[], Integer> item: bdryMap.entrySet()) {
793       if (runningValue == 0) currStartKey = item.getKey();
794       runningValue += item.getValue();
795       if (runningValue == 0) {
796         if (!firstBoundary) keysArray.add(currStartKey);
797         firstBoundary = false;
798       }
799     }
800 
801     return keysArray.toArray(new byte[0][0]);
802   }
803 
804   /*
805    * If the table is created for the first time, then "completebulkload" reads the files twice.
806    * More modifications necessary if we want to avoid doing it.
807    */
808   private void createTable(TableName tableName, String dirPath) throws Exception {
809     Path hfofDir = new Path(dirPath);
810     FileSystem fs = hfofDir.getFileSystem(getConf());
811 
812     if (!fs.exists(hfofDir)) {
813       throw new FileNotFoundException("HFileOutputFormat dir " +
814           hfofDir + " not found");
815     }
816 
817     FileStatus[] familyDirStatuses = fs.listStatus(hfofDir);
818     if (familyDirStatuses == null) {
819       throw new FileNotFoundException("No families found in " + hfofDir);
820     }
821 
822     HTableDescriptor htd = new HTableDescriptor(tableName);
823     HColumnDescriptor hcd;
824 
825     // Add column families
826     // Build a set of keys
827     byte[][] keys;
828     TreeMap<byte[], Integer> map = new TreeMap<byte[], Integer>(Bytes.BYTES_COMPARATOR);
829 
830     for (FileStatus stat : familyDirStatuses) {
831       if (!stat.isDir()) {
832         LOG.warn("Skipping non-directory " + stat.getPath());
833         continue;
834       }
835       Path familyDir = stat.getPath();
836       // Skip _logs, etc
837       if (familyDir.getName().startsWith("_")) continue;
838       byte[] family = familyDir.getName().getBytes();
839 
840       hcd = new HColumnDescriptor(family);
841       htd.addFamily(hcd);
842 
843       Path[] hfiles = FileUtil.stat2Paths(fs.listStatus(familyDir));
844       for (Path hfile : hfiles) {
845         if (hfile.getName().startsWith("_")) continue;
846         HFile.Reader reader = HFile.createReader(fs, hfile,
847             new CacheConfig(getConf()), getConf());
848         final byte[] first, last;
849         try {
850           if (hcd.getCompressionType() != reader.getFileContext().getCompression()) {
851             hcd.setCompressionType(reader.getFileContext().getCompression());
852             LOG.info("Setting compression " + hcd.getCompressionType().name() +
853                      " for family " + hcd.toString());
854           }
855           reader.loadFileInfo();
856           first = reader.getFirstRowKey();
857           last =  reader.getLastRowKey();
858 
859           LOG.info("Trying to figure out region boundaries hfile=" + hfile +
860             " first=" + Bytes.toStringBinary(first) +
861             " last="  + Bytes.toStringBinary(last));
862 
863           // To eventually infer start key-end key boundaries
864           Integer value = map.containsKey(first)? map.get(first):0;
865           map.put(first, value+1);
866 
867           value = map.containsKey(last)? map.get(last):0;
868           map.put(last, value-1);
869         }  finally {
870           reader.close();
871         }
872       }
873     }
874 
875     keys = LoadIncrementalHFiles.inferBoundaries(map);
876     this.hbAdmin.createTable(htd,keys);
877 
878     LOG.info("Table "+ tableName +" is available!!");
879   }
880 
881   @Override
882   public int run(String[] args) throws Exception {
883     if (args.length != 2) {
884       usage();
885       return -1;
886     }
887 
888     initialize();
889 
890     String dirPath = args[0];
891     TableName tableName = TableName.valueOf(args[1]);
892 
893     boolean tableExists = this.doesTableExist(tableName);
894     if (!tableExists) this.createTable(tableName,dirPath);
895 
896     Path hfofDir = new Path(dirPath);
897     HTable table = new HTable(getConf(), tableName);
898 
899     doBulkLoad(hfofDir, table);
900     return 0;
901   }
902 
903   public static void main(String[] args) throws Exception {
904     Configuration conf = HBaseConfiguration.create();
905     int ret = ToolRunner.run(conf, new LoadIncrementalHFiles(), args);
906     System.exit(ret);
907   }
908 
909 }