View Javadoc

1   /**
2    * Licensed to the Apache Software Foundation (ASF) under one
3    * or more contributor license agreements.  See the NOTICE file
4    * distributed with this work for additional information
5    * regarding copyright ownership.  The ASF licenses this file
6    * to you under the Apache License, Version 2.0 (the
7    * "License"); you may not use this file except in compliance
8    * with the License.  You may obtain a copy of the License at
9    *
10   *     http://www.apache.org/licenses/LICENSE-2.0
11   *
12   * Unless required by applicable law or agreed to in writing, software
13   * distributed under the License is distributed on an "AS IS" BASIS,
14   * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
15   * See the License for the specific language governing permissions and
16   * limitations under the License.
17   */
18  package org.apache.hadoop.hbase.util;
19  
20  import java.io.FileNotFoundException;
21  import java.io.IOException;
22  import java.io.PrintWriter;
23  import java.io.StringWriter;
24  import java.net.URI;
25  import java.util.ArrayList;
26  import java.util.Arrays;
27  import java.util.Collection;
28  import java.util.Collections;
29  import java.util.Comparator;
30  import java.util.HashMap;
31  import java.util.HashSet;
32  import java.util.Iterator;
33  import java.util.List;
34  import java.util.Map;
35  import java.util.Map.Entry;
36  import java.util.Set;
37  import java.util.SortedMap;
38  import java.util.SortedSet;
39  import java.util.TreeMap;
40  import java.util.TreeSet;
41  import java.util.concurrent.Callable;
42  import java.util.concurrent.ConcurrentSkipListMap;
43  import java.util.concurrent.ExecutionException;
44  import java.util.concurrent.ExecutorService;
45  import java.util.concurrent.Future;
46  import java.util.concurrent.ScheduledThreadPoolExecutor;
47  import java.util.concurrent.atomic.AtomicInteger;
48  
49  import org.apache.commons.logging.Log;
50  import org.apache.commons.logging.LogFactory;
51  import org.apache.hadoop.classification.InterfaceAudience;
52  import org.apache.hadoop.classification.InterfaceStability;
53  import org.apache.hadoop.conf.Configuration;
54  import org.apache.hadoop.conf.Configured;
55  import org.apache.hadoop.fs.FileStatus;
56  import org.apache.hadoop.fs.FileSystem;
57  import org.apache.hadoop.fs.Path;
58  import org.apache.hadoop.fs.permission.FsAction;
59  import org.apache.hadoop.hbase.Abortable;
60  import org.apache.hadoop.hbase.Cell;
61  import org.apache.hadoop.hbase.ClusterStatus;
62  import org.apache.hadoop.hbase.HBaseConfiguration;
63  import org.apache.hadoop.hbase.HColumnDescriptor;
64  import org.apache.hadoop.hbase.HConstants;
65  import org.apache.hadoop.hbase.HRegionInfo;
66  import org.apache.hadoop.hbase.HRegionLocation;
67  import org.apache.hadoop.hbase.HTableDescriptor;
68  import org.apache.hadoop.hbase.KeyValue;
69  import org.apache.hadoop.hbase.MasterNotRunningException;
70  import org.apache.hadoop.hbase.ServerName;
71  import org.apache.hadoop.hbase.TableName;
72  import org.apache.hadoop.hbase.ZooKeeperConnectionException;
73  import org.apache.hadoop.hbase.catalog.MetaEditor;
74  import org.apache.hadoop.hbase.client.Delete;
75  import org.apache.hadoop.hbase.client.Get;
76  import org.apache.hadoop.hbase.client.HBaseAdmin;
77  import org.apache.hadoop.hbase.client.HConnectable;
78  import org.apache.hadoop.hbase.client.HConnection;
79  import org.apache.hadoop.hbase.client.HConnectionManager;
80  import org.apache.hadoop.hbase.client.HTable;
81  import org.apache.hadoop.hbase.client.MetaScanner;
82  import org.apache.hadoop.hbase.client.MetaScanner.MetaScannerVisitor;
83  import org.apache.hadoop.hbase.client.MetaScanner.MetaScannerVisitorBase;
84  import org.apache.hadoop.hbase.client.Put;
85  import org.apache.hadoop.hbase.client.Result;
86  import org.apache.hadoop.hbase.client.RowMutations;
87  import org.apache.hadoop.hbase.io.hfile.CacheConfig;
88  import org.apache.hadoop.hbase.io.hfile.HFile;
89  import org.apache.hadoop.hbase.master.MasterFileSystem;
90  import org.apache.hadoop.hbase.master.RegionState;
91  import org.apache.hadoop.hbase.protobuf.ProtobufUtil;
92  import org.apache.hadoop.hbase.protobuf.generated.AdminProtos.AdminService.BlockingInterface;
93  import org.apache.hadoop.hbase.regionserver.HRegion;
94  import org.apache.hadoop.hbase.regionserver.HRegionFileSystem;
95  import org.apache.hadoop.hbase.regionserver.StoreFileInfo;
96  import org.apache.hadoop.hbase.regionserver.wal.HLogUtil;
97  import org.apache.hadoop.hbase.security.UserProvider;
98  import org.apache.hadoop.hbase.util.Bytes.ByteArrayComparator;
99  import org.apache.hadoop.hbase.util.HBaseFsck.ErrorReporter.ERROR_CODE;
100 import org.apache.hadoop.hbase.util.hbck.HFileCorruptionChecker;
101 import org.apache.hadoop.hbase.util.hbck.TableIntegrityErrorHandler;
102 import org.apache.hadoop.hbase.util.hbck.TableIntegrityErrorHandlerImpl;
103 import org.apache.hadoop.hbase.util.hbck.TableLockChecker;
104 import org.apache.hadoop.hbase.zookeeper.MetaRegionTracker;
105 import org.apache.hadoop.hbase.zookeeper.ZKTableReadOnly;
106 import org.apache.hadoop.hbase.zookeeper.ZooKeeperWatcher;
107 import org.apache.hadoop.io.IOUtils;
108 import org.apache.hadoop.security.AccessControlException;
109 import org.apache.hadoop.security.UserGroupInformation;
110 import org.apache.hadoop.util.ReflectionUtils;
111 import org.apache.hadoop.util.Tool;
112 import org.apache.hadoop.util.ToolRunner;
113 import org.apache.zookeeper.KeeperException;
114 
115 import com.google.common.base.Joiner;
116 import com.google.common.base.Preconditions;
117 import com.google.common.collect.Lists;
118 import com.google.common.collect.Multimap;
119 import com.google.common.collect.TreeMultimap;
120 import com.google.protobuf.ServiceException;
121 
122 /**
123  * HBaseFsck (hbck) is a tool for checking and repairing region consistency and
124  * table integrity problems in a corrupted HBase.
125  * <p>
126  * Region consistency checks verify that hbase:meta, region deployment on region
127  * servers and the state of data in HDFS (.regioninfo files) all are in
128  * accordance.
129  * <p>
130  * Table integrity checks verify that all possible row keys resolve to exactly
131  * one region of a table.  This means there are no individual degenerate
132  * or backwards regions; no holes between regions; and that there are no
133  * overlapping regions.
134  * <p>
135  * The general repair strategy works in two phases:
136  * <ol>
137  * <li> Repair Table Integrity on HDFS. (merge or fabricate regions)
138  * <li> Repair Region Consistency with hbase:meta and assignments
139  * </ol>
140  * <p>
141  * For table integrity repairs, the tables' region directories are scanned
142  * for .regioninfo files.  Each table's integrity is then verified.  If there
143  * are any orphan regions (regions with no .regioninfo files) or holes, new
144  * regions are fabricated.  Backwards regions are sidelined as well as empty
145  * degenerate (endkey==startkey) regions.  If there are any overlapping regions,
146  * a new region is created and all data is merged into the new region.
147  * <p>
148  * Table integrity repairs deal solely with HDFS and could potentially be done
149  * offline -- the hbase region servers or master do not need to be running.
150  * This phase can eventually be used to completely reconstruct the hbase:meta table in
151  * an offline fashion.
152  * <p>
153  * Region consistency requires three conditions -- 1) valid .regioninfo file
154  * present in an HDFS region dir,  2) valid row with .regioninfo data in META,
155  * and 3) a region is deployed only at the regionserver that was assigned to
156  * with proper state in the master.
157  * <p>
158  * Region consistency repairs require hbase to be online so that hbck can
159  * contact the HBase master and region servers.  The hbck#connect() method must
160  * first be called successfully.  Much of the region consistency information
161  * is transient and less risky to repair.
162  * <p>
163  * If hbck is run from the command line, there are a handful of arguments that
164  * can be used to limit the kinds of repairs hbck will do.  See the code in
165  * {@link #printUsageAndExit()} for more details.
166  */
167 @InterfaceAudience.Public
168 @InterfaceStability.Evolving
169 public class HBaseFsck extends Configured {
170   public static final long DEFAULT_TIME_LAG = 60000; // default value of 1 minute
171   public static final long DEFAULT_SLEEP_BEFORE_RERUN = 10000;
172   private static final int MAX_NUM_THREADS = 50; // #threads to contact regions
173   private static boolean rsSupportsOffline = true;
174   private static final int DEFAULT_OVERLAPS_TO_SIDELINE = 2;
175   private static final int DEFAULT_MAX_MERGE = 5;
176   private static final String TO_BE_LOADED = "to_be_loaded";
177 
178   /**********************
179    * Internal resources
180    **********************/
181   private static final Log LOG = LogFactory.getLog(HBaseFsck.class.getName());
182   private ClusterStatus status;
183   private HConnection connection;
184   private HBaseAdmin admin;
185   private HTable meta;
186   // threads to do ||izable tasks: retrieve data from regionservers, handle overlapping regions
187   protected ExecutorService executor;
188   private long startMillis = System.currentTimeMillis();
189   private HFileCorruptionChecker hfcc;
190   private int retcode = 0;
191 
192   /***********
193    * Options
194    ***********/
195   private static boolean details = false; // do we display the full report
196   private long timelag = DEFAULT_TIME_LAG; // tables whose modtime is older
197   private boolean fixAssignments = false; // fix assignment errors?
198   private boolean fixMeta = false; // fix meta errors?
199   private boolean checkHdfs = true; // load and check fs consistency?
200   private boolean fixHdfsHoles = false; // fix fs holes?
201   private boolean fixHdfsOverlaps = false; // fix fs overlaps (risky)
202   private boolean fixHdfsOrphans = false; // fix fs holes (missing .regioninfo)
203   private boolean fixTableOrphans = false; // fix fs holes (missing .tableinfo)
204   private boolean fixVersionFile = false; // fix missing hbase.version file in hdfs
205   private boolean fixSplitParents = false; // fix lingering split parents
206   private boolean fixReferenceFiles = false; // fix lingering reference store file
207   private boolean fixEmptyMetaCells = false; // fix (remove) empty REGIONINFO_QUALIFIER rows
208   private boolean fixTableLocks = false; // fix table locks which are expired
209 
210   // limit checking/fixes to listed tables, if empty attempt to check/fix all
211   // hbase:meta are always checked
212   private Set<TableName> tablesIncluded = new HashSet<TableName>();
213   private int maxMerge = DEFAULT_MAX_MERGE; // maximum number of overlapping regions to merge
214   private int maxOverlapsToSideline = DEFAULT_OVERLAPS_TO_SIDELINE; // maximum number of overlapping regions to sideline
215   private boolean sidelineBigOverlaps = false; // sideline overlaps with >maxMerge regions
216   private Path sidelineDir = null;
217 
218   private boolean rerun = false; // if we tried to fix something, rerun hbck
219   private static boolean summary = false; // if we want to print less output
220   private boolean checkMetaOnly = false;
221   private boolean checkRegionBoundaries = false;
222   private boolean ignorePreCheckPermission = false; // if pre-check permission
223 
224   /*********
225    * State
226    *********/
227   final private ErrorReporter errors;
228   int fixes = 0;
229 
230   /**
231    * This map contains the state of all hbck items.  It maps from encoded region
232    * name to HbckInfo structure.  The information contained in HbckInfo is used
233    * to detect and correct consistency (hdfs/meta/deployment) problems.
234    */
235   private TreeMap<String, HbckInfo> regionInfoMap = new TreeMap<String, HbckInfo>();
236   private TreeSet<TableName> disabledTables =
237     new TreeSet<TableName>();
238   // Empty regioninfo qualifiers in hbase:meta
239   private Set<Result> emptyRegionInfoQualifiers = new HashSet<Result>();
240 
241   /**
242    * This map from Tablename -> TableInfo contains the structures necessary to
243    * detect table consistency problems (holes, dupes, overlaps).  It is sorted
244    * to prevent dupes.
245    *
246    * If tablesIncluded is empty, this map contains all tables.
247    * Otherwise, it contains only meta tables and tables in tablesIncluded,
248    * unless checkMetaOnly is specified, in which case, it contains only
249    * the meta table
250    */
251   private SortedMap<TableName, TableInfo> tablesInfo =
252       new ConcurrentSkipListMap<TableName, TableInfo>();
253 
254   /**
255    * When initially looking at HDFS, we attempt to find any orphaned data.
256    */
257   private List<HbckInfo> orphanHdfsDirs = Collections.synchronizedList(new ArrayList<HbckInfo>());
258 
259   private Map<TableName, Set<String>> orphanTableDirs =
260       new HashMap<TableName, Set<String>>();
261 
262   /**
263    * Constructor
264    *
265    * @param conf Configuration object
266    * @throws MasterNotRunningException if the master is not running
267    * @throws ZooKeeperConnectionException if unable to connect to ZooKeeper
268    */
269   public HBaseFsck(Configuration conf) throws MasterNotRunningException,
270       ZooKeeperConnectionException, IOException, ClassNotFoundException {
271     super(conf);
272     // make a copy, just to be sure we're not overriding someone else's config
273     setConf(HBaseConfiguration.create(getConf()));
274     // disable blockcache for tool invocation, see HBASE-10500
275     getConf().setFloat(HConstants.HFILE_BLOCK_CACHE_SIZE_KEY, 0);
276     errors = getErrorReporter(conf);
277 
278     int numThreads = conf.getInt("hbasefsck.numthreads", MAX_NUM_THREADS);
279     executor = new ScheduledThreadPoolExecutor(numThreads, Threads.newDaemonThreadFactory("hbasefsck"));
280   }
281 
282   /**
283    * Constructor
284    *
285    * @param conf
286    *          Configuration object
287    * @throws MasterNotRunningException
288    *           if the master is not running
289    * @throws ZooKeeperConnectionException
290    *           if unable to connect to ZooKeeper
291    */
292   public HBaseFsck(Configuration conf, ExecutorService exec) throws MasterNotRunningException,
293       ZooKeeperConnectionException, IOException, ClassNotFoundException {
294     super(conf);
295     errors = getErrorReporter(getConf());
296     this.executor = exec;
297   }
298 
299   /**
300    * To repair region consistency, one must call connect() in order to repair
301    * online state.
302    */
303   public void connect() throws IOException {
304     admin = new HBaseAdmin(getConf());
305     meta = new HTable(getConf(), TableName.META_TABLE_NAME);
306     status = admin.getClusterStatus();
307     connection = admin.getConnection();
308   }
309 
310   /**
311    * Get deployed regions according to the region servers.
312    */
313   private void loadDeployedRegions() throws IOException, InterruptedException {
314     // From the master, get a list of all known live region servers
315     Collection<ServerName> regionServers = status.getServers();
316     errors.print("Number of live region servers: " + regionServers.size());
317     if (details) {
318       for (ServerName rsinfo: regionServers) {
319         errors.print("  " + rsinfo.getServerName());
320       }
321     }
322 
323     // From the master, get a list of all dead region servers
324     Collection<ServerName> deadRegionServers = status.getDeadServerNames();
325     errors.print("Number of dead region servers: " + deadRegionServers.size());
326     if (details) {
327       for (ServerName name: deadRegionServers) {
328         errors.print("  " + name);
329       }
330     }
331 
332     // Print the current master name and state
333     errors.print("Master: " + status.getMaster());
334 
335     // Print the list of all backup masters
336     Collection<ServerName> backupMasters = status.getBackupMasters();
337     errors.print("Number of backup masters: " + backupMasters.size());
338     if (details) {
339       for (ServerName name: backupMasters) {
340         errors.print("  " + name);
341       }
342     }
343 
344     errors.print("Average load: " + status.getAverageLoad());
345     errors.print("Number of requests: " + status.getRequestsCount());
346     errors.print("Number of regions: " + status.getRegionsCount());
347 
348     Map<String, RegionState> rits = status.getRegionsInTransition();
349     errors.print("Number of regions in transition: " + rits.size());
350     if (details) {
351       for (RegionState state: rits.values()) {
352         errors.print("  " + state.toDescriptiveString());
353       }
354     }
355 
356     // Determine what's deployed
357     processRegionServers(regionServers);
358   }
359 
360   /**
361    * Clear the current state of hbck.
362    */
363   private void clearState() {
364     // Make sure regionInfo is empty before starting
365     fixes = 0;
366     regionInfoMap.clear();
367     emptyRegionInfoQualifiers.clear();
368     disabledTables.clear();
369     errors.clear();
370     tablesInfo.clear();
371     orphanHdfsDirs.clear();
372   }
373 
374   /**
375    * This repair method analyzes hbase data in hdfs and repairs it to satisfy
376    * the table integrity rules.  HBase doesn't need to be online for this
377    * operation to work.
378    */
379   public void offlineHdfsIntegrityRepair() throws IOException, InterruptedException {
380     // Initial pass to fix orphans.
381     if (shouldCheckHdfs() && (shouldFixHdfsOrphans() || shouldFixHdfsHoles()
382         || shouldFixHdfsOverlaps() || shouldFixTableOrphans())) {
383       LOG.info("Loading regioninfos HDFS");
384       // if nothing is happening this should always complete in two iterations.
385       int maxIterations = getConf().getInt("hbase.hbck.integrityrepair.iterations.max", 3);
386       int curIter = 0;
387       do {
388         clearState(); // clears hbck state and reset fixes to 0 and.
389         // repair what's on HDFS
390         restoreHdfsIntegrity();
391         curIter++;// limit the number of iterations.
392       } while (fixes > 0 && curIter <= maxIterations);
393 
394       // Repairs should be done in the first iteration and verification in the second.
395       // If there are more than 2 passes, something funny has happened.
396       if (curIter > 2) {
397         if (curIter == maxIterations) {
398           LOG.warn("Exiting integrity repairs after max " + curIter + " iterations. "
399               + "Tables integrity may not be fully repaired!");
400         } else {
401           LOG.info("Successfully exiting integrity repairs after " + curIter + " iterations");
402         }
403       }
404     }
405   }
406 
407   /**
408    * This repair method requires the cluster to be online since it contacts
409    * region servers and the masters.  It makes each region's state in HDFS, in
410    * hbase:meta, and deployments consistent.
411    *
412    * @return If > 0 , number of errors detected, if < 0 there was an unrecoverable
413    * error.  If 0, we have a clean hbase.
414    */
415   public int onlineConsistencyRepair() throws IOException, KeeperException,
416     InterruptedException {
417     clearState();
418 
419     // get regions according to what is online on each RegionServer
420     loadDeployedRegions();
421     // check whether hbase:meta is deployed and online
422     recordMetaRegion();
423     // Check if hbase:meta is found only once and in the right place
424     if (!checkMetaRegion()) {
425       String errorMsg = "hbase:meta table is not consistent. ";
426       if (shouldFixAssignments()) {
427         errorMsg += "HBCK will try fixing it. Rerun once hbase:meta is back to consistent state.";
428       } else {
429         errorMsg += "Run HBCK with proper fix options to fix hbase:meta inconsistency.";
430       }
431       errors.reportError(errorMsg + " Exiting...");
432       return -2;
433     }
434     // Not going with further consistency check for tables when hbase:meta itself is not consistent.
435     LOG.info("Loading regionsinfo from the hbase:meta table");
436     boolean success = loadMetaEntries();
437     if (!success) return -1;
438 
439     // Empty cells in hbase:meta?
440     reportEmptyMetaCells();
441 
442     // Check if we have to cleanup empty REGIONINFO_QUALIFIER rows from hbase:meta
443     if (shouldFixEmptyMetaCells()) {
444       fixEmptyMetaCells();
445     }
446 
447     // get a list of all tables that have not changed recently.
448     if (!checkMetaOnly) {
449       reportTablesInFlux();
450     }
451 
452     // load regiondirs and regioninfos from HDFS
453     if (shouldCheckHdfs()) {
454       loadHdfsRegionDirs();
455       loadHdfsRegionInfos();
456     }
457 
458     // Get disabled tables from ZooKeeper
459     loadDisabledTables();
460 
461     // fix the orphan tables
462     fixOrphanTables();
463 
464     // Check and fix consistency
465     checkAndFixConsistency();
466 
467     // Check integrity (does not fix)
468     checkIntegrity();
469     return errors.getErrorList().size();
470   }
471 
472   /**
473    * Contacts the master and prints out cluster-wide information
474    * @return 0 on success, non-zero on failure
475    */
476   public int onlineHbck() throws IOException, KeeperException, InterruptedException, ServiceException {
477     // print hbase server version
478     errors.print("Version: " + status.getHBaseVersion());
479     offlineHdfsIntegrityRepair();
480 
481     // turn the balancer off
482     boolean oldBalancer = admin.setBalancerRunning(false, true);
483     try {
484       onlineConsistencyRepair();
485     }
486     finally {
487       admin.setBalancerRunning(oldBalancer, false);
488     }
489 
490     if (checkRegionBoundaries) {
491       checkRegionBoundaries();
492     }
493 
494     offlineReferenceFileRepair();
495 
496     checkAndFixTableLocks();
497 
498     // Print table summary
499     printTableSummary(tablesInfo);
500     return errors.summarize();
501   }
502 
503   public static byte[] keyOnly (byte[] b) {
504     if (b == null)
505       return b;
506     int rowlength = Bytes.toShort(b, 0);
507     byte[] result = new byte[rowlength];
508     System.arraycopy(b, Bytes.SIZEOF_SHORT, result, 0, rowlength);
509     return result;
510   }
511 
512   private static class RegionBoundariesInformation {
513     public byte [] regionName;
514     public byte [] metaFirstKey;
515     public byte [] metaLastKey;
516     public byte [] storesFirstKey;
517     public byte [] storesLastKey;
518     public String toString () {
519       return "regionName=" + Bytes.toStringBinary(regionName) +
520              "\nmetaFirstKey=" + Bytes.toStringBinary(metaFirstKey) +
521              "\nmetaLastKey=" + Bytes.toStringBinary(metaLastKey) +
522              "\nstoresFirstKey=" + Bytes.toStringBinary(storesFirstKey) +
523              "\nstoresLastKey=" + Bytes.toStringBinary(storesLastKey);
524     }
525   }
526 
527   public void checkRegionBoundaries() {
528     try {
529       ByteArrayComparator comparator = new ByteArrayComparator();
530       List<HRegionInfo> regions = MetaScanner.listAllRegions(getConf(), false);
531       final RegionBoundariesInformation currentRegionBoundariesInformation =
532           new RegionBoundariesInformation();
533       Path hbaseRoot = FSUtils.getRootDir(getConf());
534       for (HRegionInfo regionInfo : regions) {
535         Path tableDir = FSUtils.getTableDir(hbaseRoot, regionInfo.getTable());
536         currentRegionBoundariesInformation.regionName = regionInfo.getRegionName();
537         // For each region, get the start and stop key from the META and compare them to the
538         // same information from the Stores.
539         Path path = new Path(tableDir, regionInfo.getEncodedName());
540         FileSystem fs = path.getFileSystem(getConf());
541         FileStatus[] files = fs.listStatus(path);
542         // For all the column families in this region...
543         byte[] storeFirstKey = null;
544         byte[] storeLastKey = null;
545         for (FileStatus file : files) {
546           String fileName = file.getPath().toString();
547           fileName = fileName.substring(fileName.lastIndexOf("/") + 1);
548           if (!fileName.startsWith(".") && !fileName.endsWith("recovered.edits")) {
549             FileStatus[] storeFiles = fs.listStatus(file.getPath());
550             // For all the stores in this column family.
551             for (FileStatus storeFile : storeFiles) {
552               HFile.Reader reader = HFile.createReader(fs, storeFile.getPath(), new CacheConfig(
553                   getConf()));
554               if ((reader.getFirstKey() != null)
555                   && ((storeFirstKey == null) || (comparator.compare(storeFirstKey,
556                       reader.getFirstKey()) > 0))) {
557                 storeFirstKey = reader.getFirstKey();
558               }
559               if ((reader.getLastKey() != null)
560                   && ((storeLastKey == null) || (comparator.compare(storeLastKey,
561                       reader.getLastKey())) < 0)) {
562                 storeLastKey = reader.getLastKey();
563               }
564               reader.close();
565             }
566           }
567         }
568         currentRegionBoundariesInformation.metaFirstKey = regionInfo.getStartKey();
569         currentRegionBoundariesInformation.metaLastKey = regionInfo.getEndKey();
570         currentRegionBoundariesInformation.storesFirstKey = keyOnly(storeFirstKey);
571         currentRegionBoundariesInformation.storesLastKey = keyOnly(storeLastKey);
572         if (currentRegionBoundariesInformation.metaFirstKey.length == 0)
573           currentRegionBoundariesInformation.metaFirstKey = null;
574         if (currentRegionBoundariesInformation.metaLastKey.length == 0)
575           currentRegionBoundariesInformation.metaLastKey = null;
576 
577         // For a region to be correct, we need the META start key to be smaller or equal to the
578         // smallest start key from all the stores, and the start key from the next META entry to
579         // be bigger than the last key from all the current stores. First region start key is null;
580         // Last region end key is null; some regions can be empty and not have any store.
581 
582         boolean valid = true;
583         // Checking start key.
584         if ((currentRegionBoundariesInformation.storesFirstKey != null)
585             && (currentRegionBoundariesInformation.metaFirstKey != null)) {
586           valid = valid
587               && comparator.compare(currentRegionBoundariesInformation.storesFirstKey,
588                 currentRegionBoundariesInformation.metaFirstKey) >= 0;
589         }
590         // Checking stop key.
591         if ((currentRegionBoundariesInformation.storesLastKey != null)
592             && (currentRegionBoundariesInformation.metaLastKey != null)) {
593           valid = valid
594               && comparator.compare(currentRegionBoundariesInformation.storesLastKey,
595                 currentRegionBoundariesInformation.metaLastKey) < 0;
596         }
597         if (!valid) {
598           errors.reportError(ERROR_CODE.BOUNDARIES_ERROR, "Found issues with regions boundaries",
599             tablesInfo.get(regionInfo.getTable()));
600           LOG.warn("Region's boundaries not alligned between stores and META for:");
601           LOG.warn(currentRegionBoundariesInformation);
602         }
603       }
604     } catch (IOException e) {
605       LOG.error(e);
606     }
607   }
608 
609   /**
610    * Iterates through the list of all orphan/invalid regiondirs.
611    */
612   private void adoptHdfsOrphans(Collection<HbckInfo> orphanHdfsDirs) throws IOException {
613     for (HbckInfo hi : orphanHdfsDirs) {
614       LOG.info("Attempting to handle orphan hdfs dir: " + hi.getHdfsRegionDir());
615       adoptHdfsOrphan(hi);
616     }
617   }
618 
619   /**
620    * Orphaned regions are regions without a .regioninfo file in them.  We "adopt"
621    * these orphans by creating a new region, and moving the column families,
622    * recovered edits, HLogs, into the new region dir.  We determine the region
623    * startkey and endkeys by looking at all of the hfiles inside the column
624    * families to identify the min and max keys. The resulting region will
625    * likely violate table integrity but will be dealt with by merging
626    * overlapping regions.
627    */
628   @SuppressWarnings("deprecation")
629   private void adoptHdfsOrphan(HbckInfo hi) throws IOException {
630     Path p = hi.getHdfsRegionDir();
631     FileSystem fs = p.getFileSystem(getConf());
632     FileStatus[] dirs = fs.listStatus(p);
633     if (dirs == null) {
634       LOG.warn("Attempt to adopt ophan hdfs region skipped becuase no files present in " +
635           p + ". This dir could probably be deleted.");
636       return ;
637     }
638 
639     TableName tableName = hi.getTableName();
640     TableInfo tableInfo = tablesInfo.get(tableName);
641     Preconditions.checkNotNull(tableInfo, "Table '" + tableName + "' not present!");
642     HTableDescriptor template = tableInfo.getHTD();
643 
644     // find min and max key values
645     Pair<byte[],byte[]> orphanRegionRange = null;
646     for (FileStatus cf : dirs) {
647       String cfName= cf.getPath().getName();
648       // TODO Figure out what the special dirs are
649       if (cfName.startsWith(".") || cfName.equals(HConstants.SPLIT_LOGDIR_NAME)) continue;
650 
651       FileStatus[] hfiles = fs.listStatus(cf.getPath());
652       for (FileStatus hfile : hfiles) {
653         byte[] start, end;
654         HFile.Reader hf = null;
655         try {
656           CacheConfig cacheConf = new CacheConfig(getConf());
657           hf = HFile.createReader(fs, hfile.getPath(), cacheConf);
658           hf.loadFileInfo();
659           KeyValue startKv = KeyValue.createKeyValueFromKey(hf.getFirstKey());
660           start = startKv.getRow();
661           KeyValue endKv = KeyValue.createKeyValueFromKey(hf.getLastKey());
662           end = endKv.getRow();
663         } catch (IOException ioe) {
664           LOG.warn("Problem reading orphan file " + hfile + ", skipping");
665           continue;
666         } catch (NullPointerException ioe) {
667           LOG.warn("Orphan file " + hfile + " is possibly corrupted HFile, skipping");
668           continue;
669         } finally {
670           if (hf != null) {
671             hf.close();
672           }
673         }
674 
675         // expand the range to include the range of all hfiles
676         if (orphanRegionRange == null) {
677           // first range
678           orphanRegionRange = new Pair<byte[], byte[]>(start, end);
679         } else {
680           // TODO add test
681 
682           // expand range only if the hfile is wider.
683           if (Bytes.compareTo(orphanRegionRange.getFirst(), start) > 0) {
684             orphanRegionRange.setFirst(start);
685           }
686           if (Bytes.compareTo(orphanRegionRange.getSecond(), end) < 0 ) {
687             orphanRegionRange.setSecond(end);
688           }
689         }
690       }
691     }
692     if (orphanRegionRange == null) {
693       LOG.warn("No data in dir " + p + ", sidelining data");
694       fixes++;
695       sidelineRegionDir(fs, hi);
696       return;
697     }
698     LOG.info("Min max keys are : [" + Bytes.toString(orphanRegionRange.getFirst()) + ", " +
699         Bytes.toString(orphanRegionRange.getSecond()) + ")");
700 
701     // create new region on hdfs.  move data into place.
702     HRegionInfo hri = new HRegionInfo(template.getTableName(), orphanRegionRange.getFirst(), orphanRegionRange.getSecond());
703     LOG.info("Creating new region : " + hri);
704     HRegion region = HBaseFsckRepair.createHDFSRegionDir(getConf(), hri, template);
705     Path target = region.getRegionFileSystem().getRegionDir();
706 
707     // rename all the data to new region
708     mergeRegionDirs(target, hi);
709     fixes++;
710   }
711 
712   /**
713    * This method determines if there are table integrity errors in HDFS.  If
714    * there are errors and the appropriate "fix" options are enabled, the method
715    * will first correct orphan regions making them into legit regiondirs, and
716    * then reload to merge potentially overlapping regions.
717    *
718    * @return number of table integrity errors found
719    */
720   private int restoreHdfsIntegrity() throws IOException, InterruptedException {
721     // Determine what's on HDFS
722     LOG.info("Loading HBase regioninfo from HDFS...");
723     loadHdfsRegionDirs(); // populating regioninfo table.
724 
725     int errs = errors.getErrorList().size();
726     // First time just get suggestions.
727     tablesInfo = loadHdfsRegionInfos(); // update tableInfos based on region info in fs.
728     checkHdfsIntegrity(false, false);
729 
730     if (errors.getErrorList().size() == errs) {
731       LOG.info("No integrity errors.  We are done with this phase. Glorious.");
732       return 0;
733     }
734 
735     if (shouldFixHdfsOrphans() && orphanHdfsDirs.size() > 0) {
736       adoptHdfsOrphans(orphanHdfsDirs);
737       // TODO optimize by incrementally adding instead of reloading.
738     }
739 
740     // Make sure there are no holes now.
741     if (shouldFixHdfsHoles()) {
742       clearState(); // this also resets # fixes.
743       loadHdfsRegionDirs();
744       tablesInfo = loadHdfsRegionInfos(); // update tableInfos based on region info in fs.
745       tablesInfo = checkHdfsIntegrity(shouldFixHdfsHoles(), false);
746     }
747 
748     // Now we fix overlaps
749     if (shouldFixHdfsOverlaps()) {
750       // second pass we fix overlaps.
751       clearState(); // this also resets # fixes.
752       loadHdfsRegionDirs();
753       tablesInfo = loadHdfsRegionInfos(); // update tableInfos based on region info in fs.
754       tablesInfo = checkHdfsIntegrity(false, shouldFixHdfsOverlaps());
755     }
756 
757     return errors.getErrorList().size();
758   }
759 
760   /**
761    * Scan all the store file names to find any lingering reference files,
762    * which refer to some none-exiting files. If "fix" option is enabled,
763    * any lingering reference file will be sidelined if found.
764    * <p>
765    * Lingering reference file prevents a region from opening. It has to
766    * be fixed before a cluster can start properly.
767    */
768   private void offlineReferenceFileRepair() throws IOException {
769     Configuration conf = getConf();
770     Path hbaseRoot = FSUtils.getRootDir(conf);
771     FileSystem fs = hbaseRoot.getFileSystem(conf);
772     Map<String, Path> allFiles = FSUtils.getTableStoreFilePathMap(fs, hbaseRoot);
773     for (Path path: allFiles.values()) {
774       boolean isReference = false;
775       try {
776         isReference = StoreFileInfo.isReference(path);
777       } catch (Throwable t) {
778         // Ignore. Some files may not be store files at all.
779         // For example, files under .oldlogs folder in hbase:meta
780         // Warning message is already logged by
781         // StoreFile#isReference.
782       }
783       if (!isReference) continue;
784 
785       Path referredToFile = StoreFileInfo.getReferredToFile(path);
786       if (fs.exists(referredToFile)) continue;  // good, expected
787 
788       // Found a lingering reference file
789       errors.reportError(ERROR_CODE.LINGERING_REFERENCE_HFILE,
790         "Found lingering reference file " + path);
791       if (!shouldFixReferenceFiles()) continue;
792 
793       // Now, trying to fix it since requested
794       boolean success = false;
795       String pathStr = path.toString();
796 
797       // A reference file path should be like
798       // ${hbase.rootdir}/data/namespace/table_name/region_id/family_name/referred_file.region_name
799       // Up 5 directories to get the root folder.
800       // So the file will be sidelined to a similar folder structure.
801       int index = pathStr.lastIndexOf(Path.SEPARATOR_CHAR);
802       for (int i = 0; index > 0 && i < 5; i++) {
803         index = pathStr.lastIndexOf(Path.SEPARATOR_CHAR, index - 1);
804       }
805       if (index > 0) {
806         Path rootDir = getSidelineDir();
807         Path dst = new Path(rootDir, pathStr.substring(index + 1));
808         fs.mkdirs(dst.getParent());
809         LOG.info("Trying to sildeline reference file "
810           + path + " to " + dst);
811         setShouldRerun();
812 
813         success = fs.rename(path, dst);
814       }
815       if (!success) {
816         LOG.error("Failed to sideline reference file " + path);
817       }
818     }
819   }
820 
821   /**
822    * TODO -- need to add tests for this.
823    */
824   private void reportEmptyMetaCells() {
825     errors.print("Number of empty REGIONINFO_QUALIFIER rows in hbase:meta: " +
826       emptyRegionInfoQualifiers.size());
827     if (details) {
828       for (Result r: emptyRegionInfoQualifiers) {
829         errors.print("  " + r);
830       }
831     }
832   }
833 
834   /**
835    * TODO -- need to add tests for this.
836    */
837   private void reportTablesInFlux() {
838     AtomicInteger numSkipped = new AtomicInteger(0);
839     HTableDescriptor[] allTables = getTables(numSkipped);
840     errors.print("Number of Tables: " + allTables.length);
841     if (details) {
842       if (numSkipped.get() > 0) {
843         errors.detail("Number of Tables in flux: " + numSkipped.get());
844       }
845       for (HTableDescriptor td : allTables) {
846         errors.detail("  Table: " + td.getTableName() + "\t" +
847                            (td.isReadOnly() ? "ro" : "rw") + "\t" +
848                             (td.isMetaRegion() ? "META" : "    ") + "\t" +
849                            " families: " + td.getFamilies().size());
850       }
851     }
852   }
853 
854   public ErrorReporter getErrors() {
855     return errors;
856   }
857 
858   /**
859    * Read the .regioninfo file from the file system.  If there is no
860    * .regioninfo, add it to the orphan hdfs region list.
861    */
862   private void loadHdfsRegioninfo(HbckInfo hbi) throws IOException {
863     Path regionDir = hbi.getHdfsRegionDir();
864     if (regionDir == null) {
865       LOG.warn("No HDFS region dir found: " + hbi + " meta=" + hbi.metaEntry);
866       return;
867     }
868 
869     if (hbi.hdfsEntry.hri != null) {
870       // already loaded data
871       return;
872     }
873 
874     FileSystem fs = FileSystem.get(getConf());
875     HRegionInfo hri = HRegionFileSystem.loadRegionInfoFileContent(fs, regionDir);
876     LOG.debug("HRegionInfo read: " + hri.toString());
877     hbi.hdfsEntry.hri = hri;
878   }
879 
880   /**
881    * Exception thrown when a integrity repair operation fails in an
882    * unresolvable way.
883    */
884   public static class RegionRepairException extends IOException {
885     private static final long serialVersionUID = 1L;
886     final IOException ioe;
887     public RegionRepairException(String s, IOException ioe) {
888       super(s);
889       this.ioe = ioe;
890     }
891   }
892 
893   /**
894    * Populate hbi's from regionInfos loaded from file system.
895    */
896   private SortedMap<TableName, TableInfo> loadHdfsRegionInfos()
897       throws IOException, InterruptedException {
898     tablesInfo.clear(); // regenerating the data
899     // generate region split structure
900     Collection<HbckInfo> hbckInfos = regionInfoMap.values();
901 
902     // Parallelized read of .regioninfo files.
903     List<WorkItemHdfsRegionInfo> hbis = new ArrayList<WorkItemHdfsRegionInfo>(hbckInfos.size());
904     List<Future<Void>> hbiFutures;
905 
906     for (HbckInfo hbi : hbckInfos) {
907       WorkItemHdfsRegionInfo work = new WorkItemHdfsRegionInfo(hbi, this, errors);
908       hbis.add(work);
909     }
910 
911     // Submit and wait for completion
912     hbiFutures = executor.invokeAll(hbis);
913 
914     for(int i=0; i<hbiFutures.size(); i++) {
915       WorkItemHdfsRegionInfo work = hbis.get(i);
916       Future<Void> f = hbiFutures.get(i);
917       try {
918         f.get();
919       } catch(ExecutionException e) {
920         LOG.warn("Failed to read .regioninfo file for region " +
921               work.hbi.getRegionNameAsString(), e.getCause());
922       }
923     }
924 
925     Path hbaseRoot = FSUtils.getRootDir(getConf());
926     FileSystem fs = hbaseRoot.getFileSystem(getConf());
927     // serialized table info gathering.
928     for (HbckInfo hbi: hbckInfos) {
929 
930       if (hbi.getHdfsHRI() == null) {
931         // was an orphan
932         continue;
933       }
934 
935 
936       // get table name from hdfs, populate various HBaseFsck tables.
937       TableName tableName = hbi.getTableName();
938       if (tableName == null) {
939         // There was an entry in hbase:meta not in the HDFS?
940         LOG.warn("tableName was null for: " + hbi);
941         continue;
942       }
943 
944       TableInfo modTInfo = tablesInfo.get(tableName);
945       if (modTInfo == null) {
946         // only executed once per table.
947         modTInfo = new TableInfo(tableName);
948         tablesInfo.put(tableName, modTInfo);
949         try {
950           HTableDescriptor htd =
951               FSTableDescriptors.getTableDescriptorFromFs(fs, hbaseRoot, tableName);
952           modTInfo.htds.add(htd);
953         } catch (IOException ioe) {
954           if (!orphanTableDirs.containsKey(tableName)) {
955             LOG.warn("Unable to read .tableinfo from " + hbaseRoot, ioe);
956             //should only report once for each table
957             errors.reportError(ERROR_CODE.NO_TABLEINFO_FILE,
958                 "Unable to read .tableinfo from " + hbaseRoot + "/" + tableName);
959             Set<String> columns = new HashSet<String>();
960             orphanTableDirs.put(tableName, getColumnFamilyList(columns, hbi));
961           }
962         }
963       }
964       if (!hbi.isSkipChecks()) {
965         modTInfo.addRegionInfo(hbi);
966       }
967     }
968 
969     loadTableInfosForTablesWithNoRegion();
970 
971     return tablesInfo;
972   }
973 
974   /**
975    * To get the column family list according to the column family dirs
976    * @param columns
977    * @param hbi
978    * @return a set of column families
979    * @throws IOException
980    */
981   private Set<String> getColumnFamilyList(Set<String> columns, HbckInfo hbi) throws IOException {
982     Path regionDir = hbi.getHdfsRegionDir();
983     FileSystem fs = regionDir.getFileSystem(getConf());
984     FileStatus[] subDirs = fs.listStatus(regionDir, new FSUtils.FamilyDirFilter(fs));
985     for (FileStatus subdir : subDirs) {
986       String columnfamily = subdir.getPath().getName();
987       columns.add(columnfamily);
988     }
989     return columns;
990   }
991 
992   /**
993    * To fabricate a .tableinfo file with following contents<br>
994    * 1. the correct tablename <br>
995    * 2. the correct colfamily list<br>
996    * 3. the default properties for both {@link HTableDescriptor} and {@link HColumnDescriptor}<br>
997    * @throws IOException
998    */
999   private boolean fabricateTableInfo(FSTableDescriptors fstd, TableName tableName,
1000       Set<String> columns) throws IOException {
1001     if (columns ==null || columns.isEmpty()) return false;
1002     HTableDescriptor htd = new HTableDescriptor(tableName);
1003     for (String columnfamimly : columns) {
1004       htd.addFamily(new HColumnDescriptor(columnfamimly));
1005     }
1006     fstd.createTableDescriptor(htd, true);
1007     return true;
1008   }
1009 
1010   /**
1011    * To fix the empty REGIONINFO_QUALIFIER rows from hbase:meta <br>
1012    * @throws IOException
1013    */
1014   public void fixEmptyMetaCells() throws IOException {
1015     if (shouldFixEmptyMetaCells() && !emptyRegionInfoQualifiers.isEmpty()) {
1016       LOG.info("Trying to fix empty REGIONINFO_QUALIFIER hbase:meta rows.");
1017       for (Result region : emptyRegionInfoQualifiers) {
1018         deleteMetaRegion(region.getRow());
1019         errors.getErrorList().remove(ERROR_CODE.EMPTY_META_CELL);
1020       }
1021       emptyRegionInfoQualifiers.clear();
1022     }
1023   }
1024 
1025   /**
1026    * To fix orphan table by creating a .tableinfo file under tableDir <br>
1027    * 1. if TableInfo is cached, to recover the .tableinfo accordingly <br>
1028    * 2. else create a default .tableinfo file with following items<br>
1029    * &nbsp;2.1 the correct tablename <br>
1030    * &nbsp;2.2 the correct colfamily list<br>
1031    * &nbsp;2.3 the default properties for both {@link HTableDescriptor} and {@link HColumnDescriptor}<br>
1032    * @throws IOException
1033    */
1034   public void fixOrphanTables() throws IOException {
1035     if (shouldFixTableOrphans() && !orphanTableDirs.isEmpty()) {
1036 
1037       List<TableName> tmpList = new ArrayList<TableName>();
1038       tmpList.addAll(orphanTableDirs.keySet());
1039       HTableDescriptor[] htds = getHTableDescriptors(tmpList);
1040       Iterator<Entry<TableName, Set<String>>> iter =
1041           orphanTableDirs.entrySet().iterator();
1042       int j = 0;
1043       int numFailedCase = 0;
1044       FSTableDescriptors fstd = new FSTableDescriptors(getConf());
1045       while (iter.hasNext()) {
1046         Entry<TableName, Set<String>> entry =
1047             (Entry<TableName, Set<String>>) iter.next();
1048         TableName tableName = entry.getKey();
1049         LOG.info("Trying to fix orphan table error: " + tableName);
1050         if (j < htds.length) {
1051           if (tableName.equals(htds[j].getTableName())) {
1052             HTableDescriptor htd = htds[j];
1053             LOG.info("fixing orphan table: " + tableName + " from cache");
1054             fstd.createTableDescriptor(htd, true);
1055             j++;
1056             iter.remove();
1057           }
1058         } else {
1059           if (fabricateTableInfo(fstd, tableName, entry.getValue())) {
1060             LOG.warn("fixing orphan table: " + tableName + " with a default .tableinfo file");
1061             LOG.warn("Strongly recommend to modify the HTableDescriptor if necessary for: " + tableName);
1062             iter.remove();
1063           } else {
1064             LOG.error("Unable to create default .tableinfo for " + tableName + " while missing column family information");
1065             numFailedCase++;
1066           }
1067         }
1068         fixes++;
1069       }
1070 
1071       if (orphanTableDirs.isEmpty()) {
1072         // all orphanTableDirs are luckily recovered
1073         // re-run doFsck after recovering the .tableinfo file
1074         setShouldRerun();
1075         LOG.warn("Strongly recommend to re-run manually hfsck after all orphanTableDirs being fixed");
1076       } else if (numFailedCase > 0) {
1077         LOG.error("Failed to fix " + numFailedCase
1078             + " OrphanTables with default .tableinfo files");
1079       }
1080 
1081     }
1082     //cleanup the list
1083     orphanTableDirs.clear();
1084 
1085   }
1086 
1087   /**
1088    * This borrows code from MasterFileSystem.bootstrap()
1089    *
1090    * @return an open hbase:meta HRegion
1091    */
1092   private HRegion createNewMeta() throws IOException {
1093     Path rootdir = FSUtils.getRootDir(getConf());
1094     Configuration c = getConf();
1095     HRegionInfo metaHRI = new HRegionInfo(HRegionInfo.FIRST_META_REGIONINFO);
1096     MasterFileSystem.setInfoFamilyCachingForMeta(false);
1097     HRegion meta = HRegion.createHRegion(metaHRI, rootdir, c,
1098         HTableDescriptor.META_TABLEDESC);
1099     MasterFileSystem.setInfoFamilyCachingForMeta(true);
1100     return meta;
1101   }
1102 
1103   /**
1104    * Generate set of puts to add to new meta.  This expects the tables to be
1105    * clean with no overlaps or holes.  If there are any problems it returns null.
1106    *
1107    * @return An array list of puts to do in bulk, null if tables have problems
1108    */
1109   private ArrayList<Put> generatePuts(
1110       SortedMap<TableName, TableInfo> tablesInfo) throws IOException {
1111     ArrayList<Put> puts = new ArrayList<Put>();
1112     boolean hasProblems = false;
1113     for (Entry<TableName, TableInfo> e : tablesInfo.entrySet()) {
1114       TableName name = e.getKey();
1115 
1116       // skip "hbase:meta"
1117       if (name.compareTo(TableName.META_TABLE_NAME) == 0) {
1118         continue;
1119       }
1120 
1121       TableInfo ti = e.getValue();
1122       for (Entry<byte[], Collection<HbckInfo>> spl : ti.sc.getStarts().asMap()
1123           .entrySet()) {
1124         Collection<HbckInfo> his = spl.getValue();
1125         int sz = his.size();
1126         if (sz != 1) {
1127           // problem
1128           LOG.error("Split starting at " + Bytes.toStringBinary(spl.getKey())
1129               + " had " +  sz + " regions instead of exactly 1." );
1130           hasProblems = true;
1131           continue;
1132         }
1133 
1134         // add the row directly to meta.
1135         HbckInfo hi = his.iterator().next();
1136         HRegionInfo hri = hi.getHdfsHRI(); // hi.metaEntry;
1137         Put p = MetaEditor.makePutFromRegionInfo(hri);
1138         puts.add(p);
1139       }
1140     }
1141     return hasProblems ? null : puts;
1142   }
1143 
1144   /**
1145    * Suggest fixes for each table
1146    */
1147   private void suggestFixes(
1148       SortedMap<TableName, TableInfo> tablesInfo) throws IOException {
1149     for (TableInfo tInfo : tablesInfo.values()) {
1150       TableIntegrityErrorHandler handler = tInfo.new IntegrityFixSuggester(tInfo, errors);
1151       tInfo.checkRegionChain(handler);
1152     }
1153   }
1154 
1155   /**
1156    * Rebuilds meta from information in hdfs/fs.  Depends on configuration
1157    * settings passed into hbck constructor to point to a particular fs/dir.
1158    *
1159    * @param fix flag that determines if method should attempt to fix holes
1160    * @return true if successful, false if attempt failed.
1161    */
1162   public boolean rebuildMeta(boolean fix) throws IOException,
1163       InterruptedException {
1164 
1165     // TODO check to make sure hbase is offline. (or at least the table
1166     // currently being worked on is off line)
1167 
1168     // Determine what's on HDFS
1169     LOG.info("Loading HBase regioninfo from HDFS...");
1170     loadHdfsRegionDirs(); // populating regioninfo table.
1171 
1172     int errs = errors.getErrorList().size();
1173     tablesInfo = loadHdfsRegionInfos(); // update tableInfos based on region info in fs.
1174     checkHdfsIntegrity(false, false);
1175 
1176     // make sure ok.
1177     if (errors.getErrorList().size() != errs) {
1178       // While in error state, iterate until no more fixes possible
1179       while(true) {
1180         fixes = 0;
1181         suggestFixes(tablesInfo);
1182         errors.clear();
1183         loadHdfsRegionInfos(); // update tableInfos based on region info in fs.
1184         checkHdfsIntegrity(shouldFixHdfsHoles(), shouldFixHdfsOverlaps());
1185 
1186         int errCount = errors.getErrorList().size();
1187 
1188         if (fixes == 0) {
1189           if (errCount > 0) {
1190             return false; // failed to fix problems.
1191           } else {
1192             break; // no fixes and no problems? drop out and fix stuff!
1193           }
1194         }
1195       }
1196     }
1197 
1198     // we can rebuild, move old meta out of the way and start
1199     LOG.info("HDFS regioninfo's seems good.  Sidelining old hbase:meta");
1200     Path backupDir = sidelineOldMeta();
1201 
1202     LOG.info("Creating new hbase:meta");
1203     HRegion meta = createNewMeta();
1204 
1205     // populate meta
1206     List<Put> puts = generatePuts(tablesInfo);
1207     if (puts == null) {
1208       LOG.fatal("Problem encountered when creating new hbase:meta entries.  " +
1209         "You may need to restore the previously sidelined hbase:meta");
1210       return false;
1211     }
1212     meta.batchMutate(puts.toArray(new Put[0]));
1213     HRegion.closeHRegion(meta);
1214     LOG.info("Success! hbase:meta table rebuilt.");
1215     LOG.info("Old hbase:meta is moved into " + backupDir);
1216     return true;
1217   }
1218 
1219   private SortedMap<TableName, TableInfo> checkHdfsIntegrity(boolean fixHoles,
1220       boolean fixOverlaps) throws IOException {
1221     LOG.info("Checking HBase region split map from HDFS data...");
1222     for (TableInfo tInfo : tablesInfo.values()) {
1223       TableIntegrityErrorHandler handler;
1224       if (fixHoles || fixOverlaps) {
1225         handler = tInfo.new HDFSIntegrityFixer(tInfo, errors, getConf(),
1226           fixHoles, fixOverlaps);
1227       } else {
1228         handler = tInfo.new IntegrityFixSuggester(tInfo, errors);
1229       }
1230       if (!tInfo.checkRegionChain(handler)) {
1231         // should dump info as well.
1232         errors.report("Found inconsistency in table " + tInfo.getName());
1233       }
1234     }
1235     return tablesInfo;
1236   }
1237 
1238   private Path getSidelineDir() throws IOException {
1239     if (sidelineDir == null) {
1240       Path hbaseDir = FSUtils.getRootDir(getConf());
1241       Path hbckDir = new Path(hbaseDir, HConstants.HBCK_SIDELINEDIR_NAME);
1242       sidelineDir = new Path(hbckDir, hbaseDir.getName() + "-"
1243           + startMillis);
1244     }
1245     return sidelineDir;
1246   }
1247 
1248   /**
1249    * Sideline a region dir (instead of deleting it)
1250    */
1251   Path sidelineRegionDir(FileSystem fs, HbckInfo hi) throws IOException {
1252     return sidelineRegionDir(fs, null, hi);
1253   }
1254 
1255   /**
1256    * Sideline a region dir (instead of deleting it)
1257    *
1258    * @param parentDir if specified, the region will be sidelined to
1259    * folder like .../parentDir/<table name>/<region name>. The purpose
1260    * is to group together similar regions sidelined, for example, those
1261    * regions should be bulk loaded back later on. If null, it is ignored.
1262    */
1263   Path sidelineRegionDir(FileSystem fs,
1264       String parentDir, HbckInfo hi) throws IOException {
1265     TableName tableName = hi.getTableName();
1266     Path regionDir = hi.getHdfsRegionDir();
1267 
1268     if (!fs.exists(regionDir)) {
1269       LOG.warn("No previous " + regionDir + " exists.  Continuing.");
1270       return null;
1271     }
1272 
1273     Path rootDir = getSidelineDir();
1274     if (parentDir != null) {
1275       rootDir = new Path(rootDir, parentDir);
1276     }
1277     Path sidelineTableDir= FSUtils.getTableDir(rootDir, tableName);
1278     Path sidelineRegionDir = new Path(sidelineTableDir, regionDir.getName());
1279     fs.mkdirs(sidelineRegionDir);
1280     boolean success = false;
1281     FileStatus[] cfs =  fs.listStatus(regionDir);
1282     if (cfs == null) {
1283       LOG.info("Region dir is empty: " + regionDir);
1284     } else {
1285       for (FileStatus cf : cfs) {
1286         Path src = cf.getPath();
1287         Path dst =  new Path(sidelineRegionDir, src.getName());
1288         if (fs.isFile(src)) {
1289           // simple file
1290           success = fs.rename(src, dst);
1291           if (!success) {
1292             String msg = "Unable to rename file " + src +  " to " + dst;
1293             LOG.error(msg);
1294             throw new IOException(msg);
1295           }
1296           continue;
1297         }
1298 
1299         // is a directory.
1300         fs.mkdirs(dst);
1301 
1302         LOG.info("Sidelining files from " + src + " into containing region " + dst);
1303         // FileSystem.rename is inconsistent with directories -- if the
1304         // dst (foo/a) exists and is a dir, and the src (foo/b) is a dir,
1305         // it moves the src into the dst dir resulting in (foo/a/b).  If
1306         // the dst does not exist, and the src a dir, src becomes dst. (foo/b)
1307         FileStatus[] hfiles = fs.listStatus(src);
1308         if (hfiles != null && hfiles.length > 0) {
1309           for (FileStatus hfile : hfiles) {
1310             success = fs.rename(hfile.getPath(), dst);
1311             if (!success) {
1312               String msg = "Unable to rename file " + src +  " to " + dst;
1313               LOG.error(msg);
1314               throw new IOException(msg);
1315             }
1316           }
1317         }
1318         LOG.debug("Sideline directory contents:");
1319         debugLsr(sidelineRegionDir);
1320       }
1321     }
1322 
1323     LOG.info("Removing old region dir: " + regionDir);
1324     success = fs.delete(regionDir, true);
1325     if (!success) {
1326       String msg = "Unable to delete dir " + regionDir;
1327       LOG.error(msg);
1328       throw new IOException(msg);
1329     }
1330     return sidelineRegionDir;
1331   }
1332 
1333   /**
1334    * Side line an entire table.
1335    */
1336   void sidelineTable(FileSystem fs, TableName tableName, Path hbaseDir,
1337       Path backupHbaseDir) throws IOException {
1338     Path tableDir = FSUtils.getTableDir(hbaseDir, tableName);
1339     if (fs.exists(tableDir)) {
1340       Path backupTableDir= FSUtils.getTableDir(backupHbaseDir, tableName);
1341       fs.mkdirs(backupTableDir.getParent());
1342       boolean success = fs.rename(tableDir, backupTableDir);
1343       if (!success) {
1344         throw new IOException("Failed to move  " + tableName + " from "
1345             +  tableDir + " to " + backupTableDir);
1346       }
1347     } else {
1348       LOG.info("No previous " + tableName +  " exists.  Continuing.");
1349     }
1350   }
1351 
1352   /**
1353    * @return Path to backup of original directory
1354    */
1355   Path sidelineOldMeta() throws IOException {
1356     // put current hbase:meta aside.
1357     Path hbaseDir = FSUtils.getRootDir(getConf());
1358     FileSystem fs = hbaseDir.getFileSystem(getConf());
1359     Path backupDir = getSidelineDir();
1360     fs.mkdirs(backupDir);
1361     try {
1362       sidelineTable(fs, TableName.META_TABLE_NAME, hbaseDir, backupDir);
1363     } catch (IOException e) {
1364       LOG.fatal("... failed to sideline meta. Currently in inconsistent state.  To restore "
1365       + "try to rename .META. in " + backupDir.getName() + " to "
1366       + hbaseDir.getName() + ".", e);
1367       throw e; // throw original exception
1368     }
1369     return backupDir;
1370   }
1371 
1372   /**
1373    * Load the list of disabled tables in ZK into local set.
1374    * @throws ZooKeeperConnectionException
1375    * @throws IOException
1376    */
1377   private void loadDisabledTables()
1378   throws ZooKeeperConnectionException, IOException {
1379     HConnectionManager.execute(new HConnectable<Void>(getConf()) {
1380       @Override
1381       public Void connect(HConnection connection) throws IOException {
1382         ZooKeeperWatcher zkw = createZooKeeperWatcher();
1383         try {
1384           for (TableName tableName :
1385               ZKTableReadOnly.getDisabledOrDisablingTables(zkw)) {
1386             disabledTables.add(tableName);
1387           }
1388         } catch (KeeperException ke) {
1389           throw new IOException(ke);
1390         } finally {
1391           zkw.close();
1392         }
1393         return null;
1394       }
1395     });
1396   }
1397 
1398   /**
1399    * Check if the specified region's table is disabled.
1400    */
1401   private boolean isTableDisabled(HRegionInfo regionInfo) {
1402     return disabledTables.contains(regionInfo.getTable());
1403   }
1404 
1405   /**
1406    * Scan HDFS for all regions, recording their information into
1407    * regionInfoMap
1408    */
1409   public void loadHdfsRegionDirs() throws IOException, InterruptedException {
1410     Path rootDir = FSUtils.getRootDir(getConf());
1411     FileSystem fs = rootDir.getFileSystem(getConf());
1412 
1413     // list all tables from HDFS
1414     List<FileStatus> tableDirs = Lists.newArrayList();
1415 
1416     boolean foundVersionFile = fs.exists(new Path(rootDir, HConstants.VERSION_FILE_NAME));
1417 
1418     List<Path> paths = FSUtils.getTableDirs(fs, rootDir);
1419     for (Path path : paths) {
1420       TableName tableName = FSUtils.getTableName(path);
1421        if ((!checkMetaOnly &&
1422            isTableIncluded(tableName)) ||
1423            tableName.equals(TableName.META_TABLE_NAME)) {
1424          tableDirs.add(fs.getFileStatus(path));
1425        }
1426     }
1427 
1428     // verify that version file exists
1429     if (!foundVersionFile) {
1430       errors.reportError(ERROR_CODE.NO_VERSION_FILE,
1431           "Version file does not exist in root dir " + rootDir);
1432       if (shouldFixVersionFile()) {
1433         LOG.info("Trying to create a new " + HConstants.VERSION_FILE_NAME
1434             + " file.");
1435         setShouldRerun();
1436         FSUtils.setVersion(fs, rootDir, getConf().getInt(
1437             HConstants.THREAD_WAKE_FREQUENCY, 10 * 1000), getConf().getInt(
1438             HConstants.VERSION_FILE_WRITE_ATTEMPTS,
1439             HConstants.DEFAULT_VERSION_FILE_WRITE_ATTEMPTS));
1440       }
1441     }
1442 
1443     // level 1:  <HBASE_DIR>/*
1444     List<WorkItemHdfsDir> dirs = new ArrayList<WorkItemHdfsDir>(tableDirs.size());
1445     List<Future<Void>> dirsFutures;
1446 
1447     for (FileStatus tableDir : tableDirs) {
1448       LOG.debug("Loading region dirs from " +tableDir.getPath());
1449       dirs.add(new WorkItemHdfsDir(this, fs, errors, tableDir));
1450     }
1451 
1452     // Invoke and wait for Callables to complete
1453     dirsFutures = executor.invokeAll(dirs);
1454 
1455     for(Future<Void> f: dirsFutures) {
1456       try {
1457         f.get();
1458       } catch(ExecutionException e) {
1459         LOG.warn("Could not load region dir " , e.getCause());
1460       }
1461     }
1462   }
1463 
1464   /**
1465    * Record the location of the hbase:meta region as found in ZooKeeper.
1466    */
1467   private boolean recordMetaRegion() throws IOException {
1468     HRegionLocation metaLocation = connection.locateRegion(
1469       TableName.META_TABLE_NAME, HConstants.EMPTY_START_ROW);
1470 
1471     // Check if Meta region is valid and existing
1472     if (metaLocation == null || metaLocation.getRegionInfo() == null ||
1473         metaLocation.getHostname() == null) {
1474       errors.reportError(ERROR_CODE.NULL_META_REGION,
1475         "META region or some of its attributes are null.");
1476       return false;
1477     }
1478     ServerName sn;
1479     try {
1480       sn = getMetaRegionServerName();
1481     } catch (KeeperException e) {
1482       throw new IOException(e);
1483     }
1484     MetaEntry m = new MetaEntry(metaLocation.getRegionInfo(), sn, System.currentTimeMillis());
1485     HbckInfo hbckInfo = regionInfoMap.get(metaLocation.getRegionInfo().getEncodedName());
1486     if (hbckInfo == null) {
1487       regionInfoMap.put(metaLocation.getRegionInfo().getEncodedName(), new HbckInfo(m));
1488     } else {
1489       hbckInfo.metaEntry = m;
1490     }
1491     return true;
1492   }
1493 
1494   private ZooKeeperWatcher createZooKeeperWatcher() throws IOException {
1495     return new ZooKeeperWatcher(getConf(), "hbase Fsck", new Abortable() {
1496       @Override
1497       public void abort(String why, Throwable e) {
1498         LOG.error(why, e);
1499         System.exit(1);
1500       }
1501 
1502       @Override
1503       public boolean isAborted() {
1504         return false;
1505       }
1506 
1507     });
1508   }
1509 
1510   private ServerName getMetaRegionServerName()
1511   throws IOException, KeeperException {
1512     ZooKeeperWatcher zkw = createZooKeeperWatcher();
1513     ServerName sn = null;
1514     try {
1515       sn = MetaRegionTracker.getMetaRegionLocation(zkw);
1516     } finally {
1517       zkw.close();
1518     }
1519     return sn;
1520   }
1521 
1522   /**
1523    * Contacts each regionserver and fetches metadata about regions.
1524    * @param regionServerList - the list of region servers to connect to
1525    * @throws IOException if a remote or network exception occurs
1526    */
1527   void processRegionServers(Collection<ServerName> regionServerList)
1528     throws IOException, InterruptedException {
1529 
1530     List<WorkItemRegion> workItems = new ArrayList<WorkItemRegion>(regionServerList.size());
1531     List<Future<Void>> workFutures;
1532 
1533     // loop to contact each region server in parallel
1534     for (ServerName rsinfo: regionServerList) {
1535       workItems.add(new WorkItemRegion(this, rsinfo, errors, connection));
1536     }
1537 
1538     workFutures = executor.invokeAll(workItems);
1539 
1540     for(int i=0; i<workFutures.size(); i++) {
1541       WorkItemRegion item = workItems.get(i);
1542       Future<Void> f = workFutures.get(i);
1543       try {
1544         f.get();
1545       } catch(ExecutionException e) {
1546         LOG.warn("Could not process regionserver " + item.rsinfo.getHostAndPort(),
1547             e.getCause());
1548       }
1549     }
1550   }
1551 
1552   /**
1553    * Check consistency of all regions that have been found in previous phases.
1554    */
1555   private void checkAndFixConsistency()
1556   throws IOException, KeeperException, InterruptedException {
1557     for (java.util.Map.Entry<String, HbckInfo> e: regionInfoMap.entrySet()) {
1558       checkRegionConsistency(e.getKey(), e.getValue());
1559     }
1560   }
1561 
1562   private void preCheckPermission() throws IOException, AccessControlException {
1563     if (shouldIgnorePreCheckPermission()) {
1564       return;
1565     }
1566 
1567     Path hbaseDir = FSUtils.getRootDir(getConf());
1568     FileSystem fs = hbaseDir.getFileSystem(getConf());
1569     UserProvider userProvider = UserProvider.instantiate(getConf());
1570     UserGroupInformation ugi = userProvider.getCurrent().getUGI();
1571     FileStatus[] files = fs.listStatus(hbaseDir);
1572     for (FileStatus file : files) {
1573       try {
1574         FSUtils.checkAccess(ugi, file, FsAction.WRITE);
1575       } catch (AccessControlException ace) {
1576         LOG.warn("Got AccessControlException when preCheckPermission ", ace);
1577         errors.reportError(ERROR_CODE.WRONG_USAGE, "Current user " + ugi.getUserName()
1578           + " does not have write perms to " + file.getPath()
1579           + ". Please rerun hbck as hdfs user " + file.getOwner());
1580         throw new AccessControlException(ace);
1581       }
1582     }
1583   }
1584 
1585   /**
1586    * Deletes region from meta table
1587    */
1588   private void deleteMetaRegion(HbckInfo hi) throws IOException {
1589     deleteMetaRegion(hi.metaEntry.getRegionName());
1590   }
1591 
1592   /**
1593    * Deletes region from meta table
1594    */
1595   private void deleteMetaRegion(byte[] metaKey) throws IOException {
1596     Delete d = new Delete(metaKey);
1597     meta.delete(d);
1598     meta.flushCommits();
1599     LOG.info("Deleted " + Bytes.toString(metaKey) + " from META" );
1600   }
1601 
1602   /**
1603    * Reset the split parent region info in meta table
1604    */
1605   private void resetSplitParent(HbckInfo hi) throws IOException {
1606     RowMutations mutations = new RowMutations(hi.metaEntry.getRegionName());
1607     Delete d = new Delete(hi.metaEntry.getRegionName());
1608     d.deleteColumn(HConstants.CATALOG_FAMILY, HConstants.SPLITA_QUALIFIER);
1609     d.deleteColumn(HConstants.CATALOG_FAMILY, HConstants.SPLITB_QUALIFIER);
1610     mutations.add(d);
1611 
1612     HRegionInfo hri = new HRegionInfo(hi.metaEntry);
1613     hri.setOffline(false);
1614     hri.setSplit(false);
1615     Put p = MetaEditor.makePutFromRegionInfo(hri);
1616     mutations.add(p);
1617 
1618     meta.mutateRow(mutations);
1619     meta.flushCommits();
1620     LOG.info("Reset split parent " + hi.metaEntry.getRegionNameAsString() + " in META" );
1621   }
1622 
1623   /**
1624    * This backwards-compatibility wrapper for permanently offlining a region
1625    * that should not be alive.  If the region server does not support the
1626    * "offline" method, it will use the closest unassign method instead.  This
1627    * will basically work until one attempts to disable or delete the affected
1628    * table.  The problem has to do with in-memory only master state, so
1629    * restarting the HMaster or failing over to another should fix this.
1630    */
1631   private void offline(byte[] regionName) throws IOException {
1632     String regionString = Bytes.toStringBinary(regionName);
1633     if (!rsSupportsOffline) {
1634       LOG.warn("Using unassign region " + regionString
1635           + " instead of using offline method, you should"
1636           + " restart HMaster after these repairs");
1637       admin.unassign(regionName, true);
1638       return;
1639     }
1640 
1641     // first time we assume the rs's supports #offline.
1642     try {
1643       LOG.info("Offlining region " + regionString);
1644       admin.offline(regionName);
1645     } catch (IOException ioe) {
1646       String notFoundMsg = "java.lang.NoSuchMethodException: " +
1647         "org.apache.hadoop.hbase.master.HMaster.offline([B)";
1648       if (ioe.getMessage().contains(notFoundMsg)) {
1649         LOG.warn("Using unassign region " + regionString
1650             + " instead of using offline method, you should"
1651             + " restart HMaster after these repairs");
1652         rsSupportsOffline = false; // in the future just use unassign
1653         admin.unassign(regionName, true);
1654         return;
1655       }
1656       throw ioe;
1657     }
1658   }
1659 
1660   private void undeployRegions(HbckInfo hi) throws IOException, InterruptedException {
1661     for (OnlineEntry rse : hi.deployedEntries) {
1662       LOG.debug("Undeploy region "  + rse.hri + " from " + rse.hsa);
1663       try {
1664         HBaseFsckRepair.closeRegionSilentlyAndWait(admin, rse.hsa, rse.hri);
1665         offline(rse.hri.getRegionName());
1666       } catch (IOException ioe) {
1667         LOG.warn("Got exception when attempting to offline region "
1668             + Bytes.toString(rse.hri.getRegionName()), ioe);
1669       }
1670     }
1671   }
1672 
1673   /**
1674    * Attempts to undeploy a region from a region server based in information in
1675    * META.  Any operations that modify the file system should make sure that
1676    * its corresponding region is not deployed to prevent data races.
1677    *
1678    * A separate call is required to update the master in-memory region state
1679    * kept in the AssignementManager.  Because disable uses this state instead of
1680    * that found in META, we can't seem to cleanly disable/delete tables that
1681    * have been hbck fixed.  When used on a version of HBase that does not have
1682    * the offline ipc call exposed on the master (<0.90.5, <0.92.0) a master
1683    * restart or failover may be required.
1684    */
1685   private void closeRegion(HbckInfo hi) throws IOException, InterruptedException {
1686     if (hi.metaEntry == null && hi.hdfsEntry == null) {
1687       undeployRegions(hi);
1688       return;
1689     }
1690 
1691     // get assignment info and hregioninfo from meta.
1692     Get get = new Get(hi.getRegionName());
1693     get.addColumn(HConstants.CATALOG_FAMILY, HConstants.REGIONINFO_QUALIFIER);
1694     get.addColumn(HConstants.CATALOG_FAMILY, HConstants.SERVER_QUALIFIER);
1695     get.addColumn(HConstants.CATALOG_FAMILY, HConstants.STARTCODE_QUALIFIER);
1696     Result r = meta.get(get);
1697     ServerName serverName = HRegionInfo.getServerName(r);
1698     if (serverName == null) {
1699       errors.reportError("Unable to close region "
1700           + hi.getRegionNameAsString() +  " because meta does not "
1701           + "have handle to reach it.");
1702       return;
1703     }
1704 
1705     HRegionInfo hri = HRegionInfo.getHRegionInfo(r);
1706     if (hri == null) {
1707       LOG.warn("Unable to close region " + hi.getRegionNameAsString()
1708           + " because hbase:meta had invalid or missing "
1709           + HConstants.CATALOG_FAMILY_STR + ":"
1710           + Bytes.toString(HConstants.REGIONINFO_QUALIFIER)
1711           + " qualifier value.");
1712       return;
1713     }
1714 
1715     // close the region -- close files and remove assignment
1716     HBaseFsckRepair.closeRegionSilentlyAndWait(admin, serverName, hri);
1717   }
1718 
1719   private void tryAssignmentRepair(HbckInfo hbi, String msg) throws IOException,
1720     KeeperException, InterruptedException {
1721     // If we are trying to fix the errors
1722     if (shouldFixAssignments()) {
1723       errors.print(msg);
1724       undeployRegions(hbi);
1725       setShouldRerun();
1726       HRegionInfo hri = hbi.getHdfsHRI();
1727       if (hri == null) {
1728         hri = hbi.metaEntry;
1729       }
1730       HBaseFsckRepair.fixUnassigned(admin, hri);
1731       HBaseFsckRepair.waitUntilAssigned(admin, hri);
1732     }
1733   }
1734 
1735   /**
1736    * Check a single region for consistency and correct deployment.
1737    */
1738   private void checkRegionConsistency(final String key, final HbckInfo hbi)
1739   throws IOException, KeeperException, InterruptedException {
1740     String descriptiveName = hbi.toString();
1741 
1742     boolean inMeta = hbi.metaEntry != null;
1743     // In case not checking HDFS, assume the region is on HDFS
1744     boolean inHdfs = !shouldCheckHdfs() || hbi.getHdfsRegionDir() != null;
1745     boolean hasMetaAssignment = inMeta && hbi.metaEntry.regionServer != null;
1746     boolean isDeployed = !hbi.deployedOn.isEmpty();
1747     boolean isMultiplyDeployed = hbi.deployedOn.size() > 1;
1748     boolean deploymentMatchesMeta =
1749       hasMetaAssignment && isDeployed && !isMultiplyDeployed &&
1750       hbi.metaEntry.regionServer.equals(hbi.deployedOn.get(0));
1751     boolean splitParent =
1752       (hbi.metaEntry == null)? false: hbi.metaEntry.isSplit() && hbi.metaEntry.isOffline();
1753     boolean shouldBeDeployed = inMeta && !isTableDisabled(hbi.metaEntry);
1754     boolean recentlyModified = inHdfs &&
1755       hbi.getModTime() + timelag > System.currentTimeMillis();
1756 
1757     // ========== First the healthy cases =============
1758     if (hbi.containsOnlyHdfsEdits()) {
1759       return;
1760     }
1761     if (inMeta && inHdfs && isDeployed && deploymentMatchesMeta && shouldBeDeployed) {
1762       return;
1763     } else if (inMeta && inHdfs && !shouldBeDeployed && !isDeployed) {
1764       LOG.info("Region " + descriptiveName + " is in META, and in a disabled " +
1765         "tabled that is not deployed");
1766       return;
1767     } else if (recentlyModified) {
1768       LOG.warn("Region " + descriptiveName + " was recently modified -- skipping");
1769       return;
1770     }
1771     // ========== Cases where the region is not in hbase:meta =============
1772     else if (!inMeta && !inHdfs && !isDeployed) {
1773       // We shouldn't have record of this region at all then!
1774       assert false : "Entry for region with no data";
1775     } else if (!inMeta && !inHdfs && isDeployed) {
1776       errors.reportError(ERROR_CODE.NOT_IN_META_HDFS, "Region "
1777           + descriptiveName + ", key=" + key + ", not on HDFS or in hbase:meta but " +
1778           "deployed on " + Joiner.on(", ").join(hbi.deployedOn));
1779       if (shouldFixAssignments()) {
1780         undeployRegions(hbi);
1781       }
1782 
1783     } else if (!inMeta && inHdfs && !isDeployed) {
1784       if (hbi.isMerged()) {
1785         // This region has already been merged, the remaining hdfs file will be
1786         // cleaned by CatalogJanitor later
1787         hbi.setSkipChecks(true);
1788         LOG.info("Region " + descriptiveName
1789             + " got merge recently, its file(s) will be cleaned by CatalogJanitor later");
1790         return;
1791       }
1792       errors.reportError(ERROR_CODE.NOT_IN_META_OR_DEPLOYED, "Region "
1793           + descriptiveName + " on HDFS, but not listed in hbase:meta " +
1794           "or deployed on any region server");
1795       // restore region consistency of an adopted orphan
1796       if (shouldFixMeta()) {
1797         if (!hbi.isHdfsRegioninfoPresent()) {
1798           LOG.error("Region " + hbi.getHdfsHRI() + " could have been repaired"
1799               +  " in table integrity repair phase if -fixHdfsOrphans was" +
1800               " used.");
1801           return;
1802         }
1803 
1804         LOG.info("Patching hbase:meta with .regioninfo: " + hbi.getHdfsHRI());
1805         HBaseFsckRepair.fixMetaHoleOnline(getConf(), hbi.getHdfsHRI());
1806 
1807         tryAssignmentRepair(hbi, "Trying to reassign region...");
1808       }
1809 
1810     } else if (!inMeta && inHdfs && isDeployed) {
1811       errors.reportError(ERROR_CODE.NOT_IN_META, "Region " + descriptiveName
1812           + " not in META, but deployed on " + Joiner.on(", ").join(hbi.deployedOn));
1813       debugLsr(hbi.getHdfsRegionDir());
1814       if (shouldFixMeta()) {
1815         if (!hbi.isHdfsRegioninfoPresent()) {
1816           LOG.error("This should have been repaired in table integrity repair phase");
1817           return;
1818         }
1819 
1820         LOG.info("Patching hbase:meta with with .regioninfo: " + hbi.getHdfsHRI());
1821         HBaseFsckRepair.fixMetaHoleOnline(getConf(), hbi.getHdfsHRI());
1822 
1823         tryAssignmentRepair(hbi, "Trying to fix unassigned region...");
1824       }
1825 
1826     // ========== Cases where the region is in hbase:meta =============
1827     } else if (inMeta && inHdfs && !isDeployed && splitParent) {
1828       // check whether this is an actual error, or just transient state where parent
1829       // is not cleaned
1830       if (hbi.metaEntry.splitA != null && hbi.metaEntry.splitB != null) {
1831         // check that split daughters are there
1832         HbckInfo infoA = this.regionInfoMap.get(hbi.metaEntry.splitA.getEncodedName());
1833         HbckInfo infoB = this.regionInfoMap.get(hbi.metaEntry.splitB.getEncodedName());
1834         if (infoA != null && infoB != null) {
1835           // we already processed or will process daughters. Move on, nothing to see here.
1836           hbi.setSkipChecks(true);
1837           return;
1838         }
1839       }
1840       errors.reportError(ERROR_CODE.LINGERING_SPLIT_PARENT, "Region "
1841           + descriptiveName + " is a split parent in META, in HDFS, "
1842           + "and not deployed on any region server. This could be transient.");
1843       if (shouldFixSplitParents()) {
1844         setShouldRerun();
1845         resetSplitParent(hbi);
1846       }
1847     } else if (inMeta && !inHdfs && !isDeployed) {
1848       errors.reportError(ERROR_CODE.NOT_IN_HDFS_OR_DEPLOYED, "Region "
1849           + descriptiveName + " found in META, but not in HDFS "
1850           + "or deployed on any region server.");
1851       if (shouldFixMeta()) {
1852         deleteMetaRegion(hbi);
1853       }
1854     } else if (inMeta && !inHdfs && isDeployed) {
1855       errors.reportError(ERROR_CODE.NOT_IN_HDFS, "Region " + descriptiveName
1856           + " found in META, but not in HDFS, " +
1857           "and deployed on " + Joiner.on(", ").join(hbi.deployedOn));
1858       // We treat HDFS as ground truth.  Any information in meta is transient
1859       // and equivalent data can be regenerated.  So, lets unassign and remove
1860       // these problems from META.
1861       if (shouldFixAssignments()) {
1862         errors.print("Trying to fix unassigned region...");
1863         closeRegion(hbi);// Close region will cause RS to abort.
1864       }
1865       if (shouldFixMeta()) {
1866         // wait for it to complete
1867         deleteMetaRegion(hbi);
1868       }
1869     } else if (inMeta && inHdfs && !isDeployed && shouldBeDeployed) {
1870       errors.reportError(ERROR_CODE.NOT_DEPLOYED, "Region " + descriptiveName
1871           + " not deployed on any region server.");
1872       tryAssignmentRepair(hbi, "Trying to fix unassigned region...");
1873     } else if (inMeta && inHdfs && isDeployed && !shouldBeDeployed) {
1874       errors.reportError(ERROR_CODE.SHOULD_NOT_BE_DEPLOYED,
1875           "Region " + descriptiveName + " should not be deployed according " +
1876           "to META, but is deployed on " + Joiner.on(", ").join(hbi.deployedOn));
1877       if (shouldFixAssignments()) {
1878         errors.print("Trying to close the region " + descriptiveName);
1879         setShouldRerun();
1880         HBaseFsckRepair.fixMultiAssignment(admin, hbi.metaEntry, hbi.deployedOn);
1881       }
1882     } else if (inMeta && inHdfs && isMultiplyDeployed) {
1883       errors.reportError(ERROR_CODE.MULTI_DEPLOYED, "Region " + descriptiveName
1884           + " is listed in hbase:meta on region server " + hbi.metaEntry.regionServer
1885           + " but is multiply assigned to region servers " +
1886           Joiner.on(", ").join(hbi.deployedOn));
1887       // If we are trying to fix the errors
1888       if (shouldFixAssignments()) {
1889         errors.print("Trying to fix assignment error...");
1890         setShouldRerun();
1891         HBaseFsckRepair.fixMultiAssignment(admin, hbi.metaEntry, hbi.deployedOn);
1892       }
1893     } else if (inMeta && inHdfs && isDeployed && !deploymentMatchesMeta) {
1894       errors.reportError(ERROR_CODE.SERVER_DOES_NOT_MATCH_META, "Region "
1895           + descriptiveName + " listed in hbase:meta on region server " +
1896           hbi.metaEntry.regionServer + " but found on region server " +
1897           hbi.deployedOn.get(0));
1898       // If we are trying to fix the errors
1899       if (shouldFixAssignments()) {
1900         errors.print("Trying to fix assignment error...");
1901         setShouldRerun();
1902         HBaseFsckRepair.fixMultiAssignment(admin, hbi.metaEntry, hbi.deployedOn);
1903         HBaseFsckRepair.waitUntilAssigned(admin, hbi.getHdfsHRI());
1904       }
1905     } else {
1906       errors.reportError(ERROR_CODE.UNKNOWN, "Region " + descriptiveName +
1907           " is in an unforeseen state:" +
1908           " inMeta=" + inMeta +
1909           " inHdfs=" + inHdfs +
1910           " isDeployed=" + isDeployed +
1911           " isMultiplyDeployed=" + isMultiplyDeployed +
1912           " deploymentMatchesMeta=" + deploymentMatchesMeta +
1913           " shouldBeDeployed=" + shouldBeDeployed);
1914     }
1915   }
1916 
1917   /**
1918    * Checks tables integrity. Goes over all regions and scans the tables.
1919    * Collects all the pieces for each table and checks if there are missing,
1920    * repeated or overlapping ones.
1921    * @throws IOException
1922    */
1923   SortedMap<TableName, TableInfo> checkIntegrity() throws IOException {
1924     tablesInfo = new TreeMap<TableName,TableInfo> ();
1925     List<HbckInfo> noHDFSRegionInfos = new ArrayList<HbckInfo>();
1926     LOG.debug("There are " + regionInfoMap.size() + " region info entries");
1927     for (HbckInfo hbi : regionInfoMap.values()) {
1928       // Check only valid, working regions
1929       if (hbi.metaEntry == null) {
1930         // this assumes that consistency check has run loadMetaEntry
1931         noHDFSRegionInfos.add(hbi);
1932         Path p = hbi.getHdfsRegionDir();
1933         if (p == null) {
1934           errors.report("No regioninfo in Meta or HDFS. " + hbi);
1935         }
1936 
1937         // TODO test.
1938         continue;
1939       }
1940       if (hbi.metaEntry.regionServer == null) {
1941         errors.detail("Skipping region because no region server: " + hbi);
1942         continue;
1943       }
1944       if (hbi.metaEntry.isOffline()) {
1945         errors.detail("Skipping region because it is offline: " + hbi);
1946         continue;
1947       }
1948       if (hbi.containsOnlyHdfsEdits()) {
1949         errors.detail("Skipping region because it only contains edits" + hbi);
1950         continue;
1951       }
1952 
1953       // Missing regionDir or over-deployment is checked elsewhere. Include
1954       // these cases in modTInfo, so we can evaluate those regions as part of
1955       // the region chain in META
1956       //if (hbi.foundRegionDir == null) continue;
1957       //if (hbi.deployedOn.size() != 1) continue;
1958       if (hbi.deployedOn.size() == 0) continue;
1959 
1960       // We should be safe here
1961       TableName tableName = hbi.metaEntry.getTable();
1962       TableInfo modTInfo = tablesInfo.get(tableName);
1963       if (modTInfo == null) {
1964         modTInfo = new TableInfo(tableName);
1965       }
1966       for (ServerName server : hbi.deployedOn) {
1967         modTInfo.addServer(server);
1968       }
1969 
1970       if (!hbi.isSkipChecks()) {
1971         modTInfo.addRegionInfo(hbi);
1972       }
1973 
1974       tablesInfo.put(tableName, modTInfo);
1975     }
1976 
1977     loadTableInfosForTablesWithNoRegion();
1978 
1979     for (TableInfo tInfo : tablesInfo.values()) {
1980       TableIntegrityErrorHandler handler = tInfo.new IntegrityFixSuggester(tInfo, errors);
1981       if (!tInfo.checkRegionChain(handler)) {
1982         errors.report("Found inconsistency in table " + tInfo.getName());
1983       }
1984     }
1985     return tablesInfo;
1986   }
1987 
1988   /** Loads table info's for tables that may not have been included, since there are no
1989    * regions reported for the table, but table dir is there in hdfs
1990    */
1991   private void loadTableInfosForTablesWithNoRegion() throws IOException {
1992     Map<String, HTableDescriptor> allTables = new FSTableDescriptors(getConf()).getAll();
1993     for (HTableDescriptor htd : allTables.values()) {
1994       if (checkMetaOnly && !htd.isMetaTable()) {
1995         continue;
1996       }
1997 
1998       TableName tableName = htd.getTableName();
1999       if (isTableIncluded(tableName) && !tablesInfo.containsKey(tableName)) {
2000         TableInfo tableInfo = new TableInfo(tableName);
2001         tableInfo.htds.add(htd);
2002         tablesInfo.put(htd.getTableName(), tableInfo);
2003       }
2004     }
2005   }
2006 
2007   /**
2008    * Merge hdfs data by moving from contained HbckInfo into targetRegionDir.
2009    * @return number of file move fixes done to merge regions.
2010    */
2011   public int mergeRegionDirs(Path targetRegionDir, HbckInfo contained) throws IOException {
2012     int fileMoves = 0;
2013     String thread = Thread.currentThread().getName();
2014     LOG.debug("[" + thread + "] Contained region dir after close and pause");
2015     debugLsr(contained.getHdfsRegionDir());
2016 
2017     // rename the contained into the container.
2018     FileSystem fs = targetRegionDir.getFileSystem(getConf());
2019     FileStatus[] dirs = null;
2020     try { 
2021       dirs = fs.listStatus(contained.getHdfsRegionDir());
2022     } catch (FileNotFoundException fnfe) {
2023       // region we are attempting to merge in is not present!  Since this is a merge, there is
2024       // no harm skipping this region if it does not exist.
2025       if (!fs.exists(contained.getHdfsRegionDir())) {
2026         LOG.warn("[" + thread + "] HDFS region dir " + contained.getHdfsRegionDir() 
2027             + " is missing. Assuming already sidelined or moved.");
2028       } else {
2029         sidelineRegionDir(fs, contained);
2030       }
2031       return fileMoves;
2032     }
2033 
2034     if (dirs == null) {
2035       if (!fs.exists(contained.getHdfsRegionDir())) {
2036         LOG.warn("[" + thread + "] HDFS region dir " + contained.getHdfsRegionDir() 
2037             + " already sidelined.");
2038       } else {
2039         sidelineRegionDir(fs, contained);
2040       }
2041       return fileMoves;
2042     }
2043 
2044     for (FileStatus cf : dirs) {
2045       Path src = cf.getPath();
2046       Path dst =  new Path(targetRegionDir, src.getName());
2047 
2048       if (src.getName().equals(HRegionFileSystem.REGION_INFO_FILE)) {
2049         // do not copy the old .regioninfo file.
2050         continue;
2051       }
2052 
2053       if (src.getName().equals(HConstants.HREGION_OLDLOGDIR_NAME)) {
2054         // do not copy the .oldlogs files
2055         continue;
2056       }
2057 
2058       LOG.info("[" + thread + "] Moving files from " + src + " into containing region " + dst);
2059       // FileSystem.rename is inconsistent with directories -- if the
2060       // dst (foo/a) exists and is a dir, and the src (foo/b) is a dir,
2061       // it moves the src into the dst dir resulting in (foo/a/b).  If
2062       // the dst does not exist, and the src a dir, src becomes dst. (foo/b)
2063       for (FileStatus hfile : fs.listStatus(src)) {
2064         boolean success = fs.rename(hfile.getPath(), dst);
2065         if (success) {
2066           fileMoves++;
2067         }
2068       }
2069       LOG.debug("[" + thread + "] Sideline directory contents:");
2070       debugLsr(targetRegionDir);
2071     }
2072 
2073     // if all success.
2074     sidelineRegionDir(fs, contained);
2075     LOG.info("[" + thread + "] Sidelined region dir "+ contained.getHdfsRegionDir() + " into " +
2076         getSidelineDir());
2077     debugLsr(contained.getHdfsRegionDir());
2078 
2079     return fileMoves;
2080   }
2081 
2082 
2083   static class WorkItemOverlapMerge implements Callable<Void> {
2084     private TableIntegrityErrorHandler handler;
2085     Collection<HbckInfo> overlapgroup;
2086     
2087     WorkItemOverlapMerge(Collection<HbckInfo> overlapgroup, TableIntegrityErrorHandler handler) {
2088       this.handler = handler;
2089       this.overlapgroup = overlapgroup;
2090     }
2091     
2092     @Override
2093     public Void call() throws Exception {
2094       handler.handleOverlapGroup(overlapgroup);
2095       return null;
2096     }
2097   };
2098   
2099   
2100   /**
2101    * Maintain information about a particular table.
2102    */
2103   public class TableInfo {
2104     TableName tableName;
2105     TreeSet <ServerName> deployedOn;
2106 
2107     // backwards regions
2108     final List<HbckInfo> backwards = new ArrayList<HbckInfo>();
2109 
2110     // sidelined big overlapped regions
2111     final Map<Path, HbckInfo> sidelinedRegions = new HashMap<Path, HbckInfo>();
2112 
2113     // region split calculator
2114     final RegionSplitCalculator<HbckInfo> sc = new RegionSplitCalculator<HbckInfo>(cmp);
2115 
2116     // Histogram of different HTableDescriptors found.  Ideally there is only one!
2117     final Set<HTableDescriptor> htds = new HashSet<HTableDescriptor>();
2118 
2119     // key = start split, values = set of splits in problem group
2120     final Multimap<byte[], HbckInfo> overlapGroups =
2121       TreeMultimap.create(RegionSplitCalculator.BYTES_COMPARATOR, cmp);
2122 
2123     TableInfo(TableName name) {
2124       this.tableName = name;
2125       deployedOn = new TreeSet <ServerName>();
2126     }
2127 
2128     /**
2129      * @return descriptor common to all regions.  null if are none or multiple!
2130      */
2131     private HTableDescriptor getHTD() {
2132       if (htds.size() == 1) {
2133         return (HTableDescriptor)htds.toArray()[0];
2134       } else {
2135         LOG.error("None/Multiple table descriptors found for table '"
2136           + tableName + "' regions: " + htds);
2137       }
2138       return null;
2139     }
2140 
2141     public void addRegionInfo(HbckInfo hir) {
2142       if (Bytes.equals(hir.getEndKey(), HConstants.EMPTY_END_ROW)) {
2143         // end key is absolute end key, just add it.
2144         sc.add(hir);
2145         return;
2146       }
2147 
2148       // if not the absolute end key, check for cycle
2149       if (Bytes.compareTo(hir.getStartKey(), hir.getEndKey()) > 0) {
2150         errors.reportError(
2151             ERROR_CODE.REGION_CYCLE,
2152             String.format("The endkey for this region comes before the "
2153                 + "startkey, startkey=%s, endkey=%s",
2154                 Bytes.toStringBinary(hir.getStartKey()),
2155                 Bytes.toStringBinary(hir.getEndKey())), this, hir);
2156         backwards.add(hir);
2157         return;
2158       }
2159 
2160       // main case, add to split calculator
2161       sc.add(hir);
2162     }
2163 
2164     public void addServer(ServerName server) {
2165       this.deployedOn.add(server);
2166     }
2167 
2168     public TableName getName() {
2169       return tableName;
2170     }
2171 
2172     public int getNumRegions() {
2173       return sc.getStarts().size() + backwards.size();
2174     }
2175 
2176     private class IntegrityFixSuggester extends TableIntegrityErrorHandlerImpl {
2177       ErrorReporter errors;
2178 
2179       IntegrityFixSuggester(TableInfo ti, ErrorReporter errors) {
2180         this.errors = errors;
2181         setTableInfo(ti);
2182       }
2183 
2184       @Override
2185       public void handleRegionStartKeyNotEmpty(HbckInfo hi) throws IOException{
2186         errors.reportError(ERROR_CODE.FIRST_REGION_STARTKEY_NOT_EMPTY,
2187             "First region should start with an empty key.  You need to "
2188             + " create a new region and regioninfo in HDFS to plug the hole.",
2189             getTableInfo(), hi);
2190       }
2191 
2192       @Override
2193       public void handleRegionEndKeyNotEmpty(byte[] curEndKey) throws IOException {
2194         errors.reportError(ERROR_CODE.LAST_REGION_ENDKEY_NOT_EMPTY,
2195             "Last region should end with an empty key. You need to "
2196                 + "create a new region and regioninfo in HDFS to plug the hole.", getTableInfo());
2197       }
2198 
2199       @Override
2200       public void handleDegenerateRegion(HbckInfo hi) throws IOException{
2201         errors.reportError(ERROR_CODE.DEGENERATE_REGION,
2202             "Region has the same start and end key.", getTableInfo(), hi);
2203       }
2204 
2205       @Override
2206       public void handleDuplicateStartKeys(HbckInfo r1, HbckInfo r2) throws IOException{
2207         byte[] key = r1.getStartKey();
2208         // dup start key
2209         errors.reportError(ERROR_CODE.DUPE_STARTKEYS,
2210             "Multiple regions have the same startkey: "
2211             + Bytes.toStringBinary(key), getTableInfo(), r1);
2212         errors.reportError(ERROR_CODE.DUPE_STARTKEYS,
2213             "Multiple regions have the same startkey: "
2214             + Bytes.toStringBinary(key), getTableInfo(), r2);
2215       }
2216 
2217       @Override
2218       public void handleOverlapInRegionChain(HbckInfo hi1, HbckInfo hi2) throws IOException{
2219         errors.reportError(ERROR_CODE.OVERLAP_IN_REGION_CHAIN,
2220             "There is an overlap in the region chain.",
2221             getTableInfo(), hi1, hi2);
2222       }
2223 
2224       @Override
2225       public void handleHoleInRegionChain(byte[] holeStart, byte[] holeStop) throws IOException{
2226         errors.reportError(
2227             ERROR_CODE.HOLE_IN_REGION_CHAIN,
2228             "There is a hole in the region chain between "
2229                 + Bytes.toStringBinary(holeStart) + " and "
2230                 + Bytes.toStringBinary(holeStop)
2231                 + ".  You need to create a new .regioninfo and region "
2232                 + "dir in hdfs to plug the hole.");
2233       }
2234     };
2235 
2236     /**
2237      * This handler fixes integrity errors from hdfs information.  There are
2238      * basically three classes of integrity problems 1) holes, 2) overlaps, and
2239      * 3) invalid regions.
2240      *
2241      * This class overrides methods that fix holes and the overlap group case.
2242      * Individual cases of particular overlaps are handled by the general
2243      * overlap group merge repair case.
2244      *
2245      * If hbase is online, this forces regions offline before doing merge
2246      * operations.
2247      */
2248     private class HDFSIntegrityFixer extends IntegrityFixSuggester {
2249       Configuration conf;
2250 
2251       boolean fixOverlaps = true;
2252 
2253       HDFSIntegrityFixer(TableInfo ti, ErrorReporter errors, Configuration conf,
2254           boolean fixHoles, boolean fixOverlaps) {
2255         super(ti, errors);
2256         this.conf = conf;
2257         this.fixOverlaps = fixOverlaps;
2258         // TODO properly use fixHoles
2259       }
2260 
2261       /**
2262        * This is a special case hole -- when the first region of a table is
2263        * missing from META, HBase doesn't acknowledge the existance of the
2264        * table.
2265        */
2266       @Override
2267       public void handleRegionStartKeyNotEmpty(HbckInfo next) throws IOException {
2268         errors.reportError(ERROR_CODE.FIRST_REGION_STARTKEY_NOT_EMPTY,
2269             "First region should start with an empty key.  Creating a new " +
2270             "region and regioninfo in HDFS to plug the hole.",
2271             getTableInfo(), next);
2272         HTableDescriptor htd = getTableInfo().getHTD();
2273         // from special EMPTY_START_ROW to next region's startKey
2274         HRegionInfo newRegion = new HRegionInfo(htd.getTableName(),
2275             HConstants.EMPTY_START_ROW, next.getStartKey());
2276 
2277         // TODO test
2278         HRegion region = HBaseFsckRepair.createHDFSRegionDir(conf, newRegion, htd);
2279         LOG.info("Table region start key was not empty.  Created new empty region: "
2280             + newRegion + " " +region);
2281         fixes++;
2282       }
2283 
2284       @Override
2285       public void handleRegionEndKeyNotEmpty(byte[] curEndKey) throws IOException {
2286         errors.reportError(ERROR_CODE.LAST_REGION_ENDKEY_NOT_EMPTY,
2287             "Last region should end with an empty key.  Creating a new "
2288                 + "region and regioninfo in HDFS to plug the hole.", getTableInfo());
2289         HTableDescriptor htd = getTableInfo().getHTD();
2290         // from curEndKey to EMPTY_START_ROW
2291         HRegionInfo newRegion = new HRegionInfo(htd.getTableName(), curEndKey,
2292             HConstants.EMPTY_START_ROW);
2293 
2294         HRegion region = HBaseFsckRepair.createHDFSRegionDir(conf, newRegion, htd);
2295         LOG.info("Table region end key was not empty.  Created new empty region: " + newRegion
2296             + " " + region);
2297         fixes++;
2298       }
2299 
2300       /**
2301        * There is a hole in the hdfs regions that violates the table integrity
2302        * rules.  Create a new empty region that patches the hole.
2303        */
2304       @Override
2305       public void handleHoleInRegionChain(byte[] holeStartKey, byte[] holeStopKey) throws IOException {
2306         errors.reportError(
2307             ERROR_CODE.HOLE_IN_REGION_CHAIN,
2308             "There is a hole in the region chain between "
2309                 + Bytes.toStringBinary(holeStartKey) + " and "
2310                 + Bytes.toStringBinary(holeStopKey)
2311                 + ".  Creating a new regioninfo and region "
2312                 + "dir in hdfs to plug the hole.");
2313         HTableDescriptor htd = getTableInfo().getHTD();
2314         HRegionInfo newRegion = new HRegionInfo(htd.getTableName(), holeStartKey, holeStopKey);
2315         HRegion region = HBaseFsckRepair.createHDFSRegionDir(conf, newRegion, htd);
2316         LOG.info("Plugged hold by creating new empty region: "+ newRegion + " " +region);
2317         fixes++;
2318       }
2319 
2320       /**
2321        * This takes set of overlapping regions and merges them into a single
2322        * region.  This covers cases like degenerate regions, shared start key,
2323        * general overlaps, duplicate ranges, and partial overlapping regions.
2324        *
2325        * Cases:
2326        * - Clean regions that overlap
2327        * - Only .oldlogs regions (can't find start/stop range, or figure out)
2328        * 
2329        * This is basically threadsafe, except for the fixer increment in mergeOverlaps.
2330        */
2331       @Override
2332       public void handleOverlapGroup(Collection<HbckInfo> overlap)
2333           throws IOException {
2334         Preconditions.checkNotNull(overlap);
2335         Preconditions.checkArgument(overlap.size() >0);
2336 
2337         if (!this.fixOverlaps) {
2338           LOG.warn("Not attempting to repair overlaps.");
2339           return;
2340         }
2341 
2342         if (overlap.size() > maxMerge) {
2343           LOG.warn("Overlap group has " + overlap.size() + " overlapping " +
2344             "regions which is greater than " + maxMerge + ", the max number of regions to merge");
2345           if (sidelineBigOverlaps) {
2346             // we only sideline big overlapped groups that exceeds the max number of regions to merge
2347             sidelineBigOverlaps(overlap);
2348           }
2349           return;
2350         }
2351 
2352         mergeOverlaps(overlap);
2353       }
2354 
2355       void mergeOverlaps(Collection<HbckInfo> overlap)
2356           throws IOException {
2357         String thread = Thread.currentThread().getName();
2358         LOG.info("== [" + thread + "] Merging regions into one region: "
2359           + Joiner.on(",").join(overlap));
2360         // get the min / max range and close all concerned regions
2361         Pair<byte[], byte[]> range = null;
2362         for (HbckInfo hi : overlap) {
2363           if (range == null) {
2364             range = new Pair<byte[], byte[]>(hi.getStartKey(), hi.getEndKey());
2365           } else {
2366             if (RegionSplitCalculator.BYTES_COMPARATOR
2367                 .compare(hi.getStartKey(), range.getFirst()) < 0) {
2368               range.setFirst(hi.getStartKey());
2369             }
2370             if (RegionSplitCalculator.BYTES_COMPARATOR
2371                 .compare(hi.getEndKey(), range.getSecond()) > 0) {
2372               range.setSecond(hi.getEndKey());
2373             }
2374           }
2375           // need to close files so delete can happen.
2376           LOG.debug("[" + thread + "] Closing region before moving data around: " +  hi);
2377           LOG.debug("[" + thread + "] Contained region dir before close");
2378           debugLsr(hi.getHdfsRegionDir());
2379           try {
2380             LOG.info("[" + thread + "] Closing region: " + hi);
2381             closeRegion(hi);
2382           } catch (IOException ioe) {
2383             LOG.warn("[" + thread + "] Was unable to close region " + hi
2384               + ".  Just continuing... ", ioe);
2385           } catch (InterruptedException e) {
2386             LOG.warn("[" + thread + "] Was unable to close region " + hi
2387               + ".  Just continuing... ", e);
2388           }
2389 
2390           try {
2391             LOG.info("[" + thread + "] Offlining region: " + hi);
2392             offline(hi.getRegionName());
2393           } catch (IOException ioe) {
2394             LOG.warn("[" + thread + "] Unable to offline region from master: " + hi
2395               + ".  Just continuing... ", ioe);
2396           }
2397         }
2398 
2399         // create new empty container region.
2400         HTableDescriptor htd = getTableInfo().getHTD();
2401         // from start key to end Key
2402         HRegionInfo newRegion = new HRegionInfo(htd.getTableName(), range.getFirst(),
2403             range.getSecond());
2404         HRegion region = HBaseFsckRepair.createHDFSRegionDir(conf, newRegion, htd);
2405         LOG.info("[" + thread + "] Created new empty container region: " +
2406             newRegion + " to contain regions: " + Joiner.on(",").join(overlap));
2407         debugLsr(region.getRegionFileSystem().getRegionDir());
2408 
2409         // all target regions are closed, should be able to safely cleanup.
2410         boolean didFix= false;
2411         Path target = region.getRegionFileSystem().getRegionDir();
2412         for (HbckInfo contained : overlap) {
2413           LOG.info("[" + thread + "] Merging " + contained  + " into " + target );
2414           int merges = mergeRegionDirs(target, contained);
2415           if (merges > 0) {
2416             didFix = true;
2417           }
2418         }
2419         if (didFix) {
2420           fixes++;
2421         }
2422       }
2423 
2424       /**
2425        * Sideline some regions in a big overlap group so that it
2426        * will have fewer regions, and it is easier to merge them later on.
2427        *
2428        * @param bigOverlap the overlapped group with regions more than maxMerge
2429        * @throws IOException
2430        */
2431       void sidelineBigOverlaps(
2432           Collection<HbckInfo> bigOverlap) throws IOException {
2433         int overlapsToSideline = bigOverlap.size() - maxMerge;
2434         if (overlapsToSideline > maxOverlapsToSideline) {
2435           overlapsToSideline = maxOverlapsToSideline;
2436         }
2437         List<HbckInfo> regionsToSideline =
2438           RegionSplitCalculator.findBigRanges(bigOverlap, overlapsToSideline);
2439         FileSystem fs = FileSystem.get(conf);
2440         for (HbckInfo regionToSideline: regionsToSideline) {
2441           try {
2442             LOG.info("Closing region: " + regionToSideline);
2443             closeRegion(regionToSideline);
2444           } catch (IOException ioe) {
2445             LOG.warn("Was unable to close region " + regionToSideline
2446               + ".  Just continuing... ", ioe);
2447           } catch (InterruptedException e) {
2448             LOG.warn("Was unable to close region " + regionToSideline
2449               + ".  Just continuing... ", e);
2450           }
2451 
2452           try {
2453             LOG.info("Offlining region: " + regionToSideline);
2454             offline(regionToSideline.getRegionName());
2455           } catch (IOException ioe) {
2456             LOG.warn("Unable to offline region from master: " + regionToSideline
2457               + ".  Just continuing... ", ioe);
2458           }
2459 
2460           LOG.info("Before sideline big overlapped region: " + regionToSideline.toString());
2461           Path sidelineRegionDir = sidelineRegionDir(fs, TO_BE_LOADED, regionToSideline);
2462           if (sidelineRegionDir != null) {
2463             sidelinedRegions.put(sidelineRegionDir, regionToSideline);
2464             LOG.info("After sidelined big overlapped region: "
2465               + regionToSideline.getRegionNameAsString()
2466               + " to " + sidelineRegionDir.toString());
2467             fixes++;
2468           }
2469         }
2470       }
2471     }
2472 
2473     /**
2474      * Check the region chain (from META) of this table.  We are looking for
2475      * holes, overlaps, and cycles.
2476      * @return false if there are errors
2477      * @throws IOException
2478      */
2479     public boolean checkRegionChain(TableIntegrityErrorHandler handler) throws IOException {
2480       // When table is disabled no need to check for the region chain. Some of the regions
2481       // accidently if deployed, this below code might report some issues like missing start
2482       // or end regions or region hole in chain and may try to fix which is unwanted.
2483       if (disabledTables.contains(this.tableName)) {
2484         return true;
2485       }
2486       int originalErrorsCount = errors.getErrorList().size();
2487       Multimap<byte[], HbckInfo> regions = sc.calcCoverage();
2488       SortedSet<byte[]> splits = sc.getSplits();
2489 
2490       byte[] prevKey = null;
2491       byte[] problemKey = null;
2492 
2493       if (splits.size() == 0) {
2494         // no region for this table
2495         handler.handleHoleInRegionChain(HConstants.EMPTY_START_ROW, HConstants.EMPTY_END_ROW);
2496       }
2497 
2498       for (byte[] key : splits) {
2499         Collection<HbckInfo> ranges = regions.get(key);
2500         if (prevKey == null && !Bytes.equals(key, HConstants.EMPTY_BYTE_ARRAY)) {
2501           for (HbckInfo rng : ranges) {
2502             handler.handleRegionStartKeyNotEmpty(rng);
2503           }
2504         }
2505 
2506         // check for degenerate ranges
2507         for (HbckInfo rng : ranges) {
2508           // special endkey case converts '' to null
2509           byte[] endKey = rng.getEndKey();
2510           endKey = (endKey.length == 0) ? null : endKey;
2511           if (Bytes.equals(rng.getStartKey(),endKey)) {
2512             handler.handleDegenerateRegion(rng);
2513           }
2514         }
2515 
2516         if (ranges.size() == 1) {
2517           // this split key is ok -- no overlap, not a hole.
2518           if (problemKey != null) {
2519             LOG.warn("reached end of problem group: " + Bytes.toStringBinary(key));
2520           }
2521           problemKey = null; // fell through, no more problem.
2522         } else if (ranges.size() > 1) {
2523           // set the new problem key group name, if already have problem key, just
2524           // keep using it.
2525           if (problemKey == null) {
2526             // only for overlap regions.
2527             LOG.warn("Naming new problem group: " + Bytes.toStringBinary(key));
2528             problemKey = key;
2529           }
2530           overlapGroups.putAll(problemKey, ranges);
2531 
2532           // record errors
2533           ArrayList<HbckInfo> subRange = new ArrayList<HbckInfo>(ranges);
2534           //  this dumb and n^2 but this shouldn't happen often
2535           for (HbckInfo r1 : ranges) {
2536             subRange.remove(r1);
2537             for (HbckInfo r2 : subRange) {
2538               if (Bytes.compareTo(r1.getStartKey(), r2.getStartKey())==0) {
2539                 handler.handleDuplicateStartKeys(r1,r2);
2540               } else {
2541                 // overlap
2542                 handler.handleOverlapInRegionChain(r1, r2);
2543               }
2544             }
2545           }
2546 
2547         } else if (ranges.size() == 0) {
2548           if (problemKey != null) {
2549             LOG.warn("reached end of problem group: " + Bytes.toStringBinary(key));
2550           }
2551           problemKey = null;
2552 
2553           byte[] holeStopKey = sc.getSplits().higher(key);
2554           // if higher key is null we reached the top.
2555           if (holeStopKey != null) {
2556             // hole
2557             handler.handleHoleInRegionChain(key, holeStopKey);
2558           }
2559         }
2560         prevKey = key;
2561       }
2562 
2563       // When the last region of a table is proper and having an empty end key, 'prevKey'
2564       // will be null.
2565       if (prevKey != null) {
2566         handler.handleRegionEndKeyNotEmpty(prevKey);
2567       }
2568 
2569       // TODO fold this into the TableIntegrityHandler
2570       if (getConf().getBoolean("hbasefsck.overlap.merge.parallel", true)) {
2571         LOG.info("Handling overlap merges in parallel. set hbasefsck.overlap.merge.parallel to" +
2572             " false to run serially.");
2573         boolean ok = handleOverlapsParallel(handler, prevKey);
2574         if (!ok) {
2575           return false;
2576         }
2577       } else {
2578         LOG.info("Handling overlap merges serially.  set hbasefsck.overlap.merge.parallel to" +
2579             " true to run in parallel.");
2580         for (Collection<HbckInfo> overlap : overlapGroups.asMap().values()) {
2581           handler.handleOverlapGroup(overlap);
2582         }
2583       }
2584 
2585       if (details) {
2586         // do full region split map dump
2587         errors.print("---- Table '"  +  this.tableName
2588             + "': region split map");
2589         dump(splits, regions);
2590         errors.print("---- Table '"  +  this.tableName
2591             + "': overlap groups");
2592         dumpOverlapProblems(overlapGroups);
2593         errors.print("There are " + overlapGroups.keySet().size()
2594             + " overlap groups with " + overlapGroups.size()
2595             + " overlapping regions");
2596       }
2597       if (!sidelinedRegions.isEmpty()) {
2598         LOG.warn("Sidelined big overlapped regions, please bulk load them!");
2599         errors.print("---- Table '"  +  this.tableName
2600             + "': sidelined big overlapped regions");
2601         dumpSidelinedRegions(sidelinedRegions);
2602       }
2603       return errors.getErrorList().size() == originalErrorsCount;
2604     }
2605 
2606     private boolean handleOverlapsParallel(TableIntegrityErrorHandler handler, byte[] prevKey)
2607         throws IOException {
2608       // we parallelize overlap handler for the case we have lots of groups to fix.  We can
2609       // safely assume each group is independent. 
2610       List<WorkItemOverlapMerge> merges = new ArrayList<WorkItemOverlapMerge>(overlapGroups.size());
2611       List<Future<Void>> rets;
2612       for (Collection<HbckInfo> overlap : overlapGroups.asMap().values()) {
2613         // 
2614         merges.add(new WorkItemOverlapMerge(overlap, handler));
2615       }
2616       try {
2617         rets = executor.invokeAll(merges);
2618       } catch (InterruptedException e) {
2619         e.printStackTrace();
2620         LOG.error("Overlap merges were interrupted", e);
2621         return false;
2622       }
2623       for(int i=0; i<merges.size(); i++) {
2624         WorkItemOverlapMerge work = merges.get(i);
2625         Future<Void> f = rets.get(i);
2626         try {
2627           f.get();
2628         } catch(ExecutionException e) {
2629           LOG.warn("Failed to merge overlap group" + work, e.getCause());
2630         } catch (InterruptedException e) {
2631           LOG.error("Waiting for overlap merges was interrupted", e);
2632           return false;
2633         }
2634       }
2635       return true;
2636     }
2637 
2638     /**
2639      * This dumps data in a visually reasonable way for visual debugging
2640      *
2641      * @param splits
2642      * @param regions
2643      */
2644     void dump(SortedSet<byte[]> splits, Multimap<byte[], HbckInfo> regions) {
2645       // we display this way because the last end key should be displayed as well.
2646       StringBuilder sb = new StringBuilder();
2647       for (byte[] k : splits) {
2648         sb.setLength(0); // clear out existing buffer, if any.
2649         sb.append(Bytes.toStringBinary(k) + ":\t");
2650         for (HbckInfo r : regions.get(k)) {
2651           sb.append("[ "+ r.toString() + ", "
2652               + Bytes.toStringBinary(r.getEndKey())+ "]\t");
2653         }
2654         errors.print(sb.toString());
2655       }
2656     }
2657   }
2658 
2659   public void dumpOverlapProblems(Multimap<byte[], HbckInfo> regions) {
2660     // we display this way because the last end key should be displayed as
2661     // well.
2662     for (byte[] k : regions.keySet()) {
2663       errors.print(Bytes.toStringBinary(k) + ":");
2664       for (HbckInfo r : regions.get(k)) {
2665         errors.print("[ " + r.toString() + ", "
2666             + Bytes.toStringBinary(r.getEndKey()) + "]");
2667       }
2668       errors.print("----");
2669     }
2670   }
2671 
2672   public void dumpSidelinedRegions(Map<Path, HbckInfo> regions) {
2673     for (Map.Entry<Path, HbckInfo> entry: regions.entrySet()) {
2674       TableName tableName = entry.getValue().getTableName();
2675       Path path = entry.getKey();
2676       errors.print("This sidelined region dir should be bulk loaded: "
2677         + path.toString());
2678       errors.print("Bulk load command looks like: "
2679         + "hbase org.apache.hadoop.hbase.mapreduce.LoadIncrementalHFiles "
2680         + path.toUri().getPath() + " "+ tableName);
2681     }
2682   }
2683 
2684   public Multimap<byte[], HbckInfo> getOverlapGroups(
2685       TableName table) {
2686     TableInfo ti = tablesInfo.get(table);
2687     return ti.overlapGroups;
2688   }
2689 
2690   /**
2691    * Return a list of user-space table names whose metadata have not been
2692    * modified in the last few milliseconds specified by timelag
2693    * if any of the REGIONINFO_QUALIFIER, SERVER_QUALIFIER, STARTCODE_QUALIFIER,
2694    * SPLITA_QUALIFIER, SPLITB_QUALIFIER have not changed in the last
2695    * milliseconds specified by timelag, then the table is a candidate to be returned.
2696    * @return tables that have not been modified recently
2697    * @throws IOException if an error is encountered
2698    */
2699   HTableDescriptor[] getTables(AtomicInteger numSkipped) {
2700     List<TableName> tableNames = new ArrayList<TableName>();
2701     long now = System.currentTimeMillis();
2702 
2703     for (HbckInfo hbi : regionInfoMap.values()) {
2704       MetaEntry info = hbi.metaEntry;
2705 
2706       // if the start key is zero, then we have found the first region of a table.
2707       // pick only those tables that were not modified in the last few milliseconds.
2708       if (info != null && info.getStartKey().length == 0 && !info.isMetaRegion()) {
2709         if (info.modTime + timelag < now) {
2710           tableNames.add(info.getTable());
2711         } else {
2712           numSkipped.incrementAndGet(); // one more in-flux table
2713         }
2714       }
2715     }
2716     return getHTableDescriptors(tableNames);
2717   }
2718 
2719   HTableDescriptor[] getHTableDescriptors(List<TableName> tableNames) {
2720     HTableDescriptor[] htd = new HTableDescriptor[0];
2721     try {
2722       LOG.info("getHTableDescriptors == tableNames => " + tableNames);
2723       htd = new HBaseAdmin(getConf()).getTableDescriptorsByTableName(tableNames);
2724     } catch (IOException e) {
2725       LOG.debug("Exception getting table descriptors", e);
2726     }
2727     return htd;
2728   }
2729 
2730   /**
2731    * Gets the entry in regionInfo corresponding to the the given encoded
2732    * region name. If the region has not been seen yet, a new entry is added
2733    * and returned.
2734    */
2735   private synchronized HbckInfo getOrCreateInfo(String name) {
2736     HbckInfo hbi = regionInfoMap.get(name);
2737     if (hbi == null) {
2738       hbi = new HbckInfo(null);
2739       regionInfoMap.put(name, hbi);
2740     }
2741     return hbi;
2742   }
2743 
2744   private void checkAndFixTableLocks() throws IOException {
2745     TableLockChecker checker = new TableLockChecker(createZooKeeperWatcher(), errors);
2746     checker.checkTableLocks();
2747 
2748     if (this.fixTableLocks) {
2749       checker.fixExpiredTableLocks();
2750     }
2751   }
2752 
2753   /**
2754     * Check values in regionInfo for hbase:meta
2755     * Check if zero or more than one regions with hbase:meta are found.
2756     * If there are inconsistencies (i.e. zero or more than one regions
2757     * pretend to be holding the hbase:meta) try to fix that and report an error.
2758     * @throws IOException from HBaseFsckRepair functions
2759    * @throws KeeperException
2760    * @throws InterruptedException
2761     */
2762   boolean checkMetaRegion() throws IOException, KeeperException, InterruptedException {
2763     List<HbckInfo> metaRegions = Lists.newArrayList();
2764     for (HbckInfo value : regionInfoMap.values()) {
2765       if (value.metaEntry != null && value.metaEntry.isMetaRegion()) {
2766         metaRegions.add(value);
2767       }
2768     }
2769 
2770     // There will be always one entry in regionInfoMap corresponding to hbase:meta
2771     // Check the deployed servers. It should be exactly one server.
2772     List<ServerName> servers = new ArrayList<ServerName>();
2773     HbckInfo metaHbckInfo = null;
2774     if (!metaRegions.isEmpty()) {
2775       metaHbckInfo = metaRegions.get(0);
2776       servers = metaHbckInfo.deployedOn;
2777     }
2778     if (servers.size() != 1) {
2779       if (servers.size() == 0) {
2780         errors.reportError(ERROR_CODE.NO_META_REGION, "hbase:meta is not found on any region.");
2781         if (shouldFixAssignments()) {
2782           errors.print("Trying to fix a problem with hbase:meta..");
2783           setShouldRerun();
2784           // try to fix it (treat it as unassigned region)
2785           HBaseFsckRepair.fixUnassigned(admin, HRegionInfo.FIRST_META_REGIONINFO);
2786           HBaseFsckRepair.waitUntilAssigned(admin, HRegionInfo.FIRST_META_REGIONINFO);
2787         }
2788       } else if (servers.size() > 1) {
2789         errors
2790             .reportError(ERROR_CODE.MULTI_META_REGION, "hbase:meta is found on more than one region.");
2791         if (shouldFixAssignments()) {
2792           errors.print("Trying to fix a problem with hbase:meta..");
2793           setShouldRerun();
2794           // try fix it (treat is a dupe assignment)
2795           HBaseFsckRepair.fixMultiAssignment(admin, metaHbckInfo.metaEntry, servers);
2796         }
2797       }
2798       // rerun hbck with hopefully fixed META
2799       return false;
2800     }
2801     // no errors, so continue normally
2802     return true;
2803   }
2804 
2805   /**
2806    * Scan hbase:meta, adding all regions found to the regionInfo map.
2807    * @throws IOException if an error is encountered
2808    */
2809   boolean loadMetaEntries() throws IOException {
2810     MetaScannerVisitor visitor = new MetaScannerVisitorBase() {
2811       int countRecord = 1;
2812 
2813       // comparator to sort KeyValues with latest modtime
2814       final Comparator<Cell> comp = new Comparator<Cell>() {
2815         @Override
2816         public int compare(Cell k1, Cell k2) {
2817           return (int)(k1.getTimestamp() - k2.getTimestamp());
2818         }
2819       };
2820 
2821       @Override
2822       public boolean processRow(Result result) throws IOException {
2823         try {
2824 
2825           // record the latest modification of this META record
2826           long ts =  Collections.max(result.listCells(), comp).getTimestamp();
2827           Pair<HRegionInfo, ServerName> pair = HRegionInfo.getHRegionInfoAndServerName(result);
2828           if (pair == null || pair.getFirst() == null) {
2829             emptyRegionInfoQualifiers.add(result);
2830             errors.reportError(ERROR_CODE.EMPTY_META_CELL,
2831               "Empty REGIONINFO_QUALIFIER found in hbase:meta");
2832             return true;
2833           }
2834           ServerName sn = null;
2835           if (pair.getSecond() != null) {
2836             sn = pair.getSecond();
2837           }
2838           HRegionInfo hri = pair.getFirst();
2839           if (!(isTableIncluded(hri.getTable())
2840               || hri.isMetaRegion())) {
2841             return true;
2842           }
2843           PairOfSameType<HRegionInfo> daughters = HRegionInfo.getDaughterRegions(result);
2844           MetaEntry m = new MetaEntry(hri, sn, ts, daughters.getFirst(), daughters.getSecond());
2845           HbckInfo previous = regionInfoMap.get(hri.getEncodedName());
2846           if (previous == null) {
2847             regionInfoMap.put(hri.getEncodedName(), new HbckInfo(m));
2848           } else if (previous.metaEntry == null) {
2849             previous.metaEntry = m;
2850           } else {
2851             throw new IOException("Two entries in hbase:meta are same " + previous);
2852           }
2853           
2854           PairOfSameType<HRegionInfo> mergeRegions = HRegionInfo.getMergeRegions(result);
2855           for (HRegionInfo mergeRegion : new HRegionInfo[] {
2856               mergeRegions.getFirst(), mergeRegions.getSecond() }) {
2857             if (mergeRegion != null) {
2858               // This region is already been merged
2859               HbckInfo hbInfo = getOrCreateInfo(mergeRegion.getEncodedName());
2860               hbInfo.setMerged(true);
2861             }
2862           }
2863 
2864           // show proof of progress to the user, once for every 100 records.
2865           if (countRecord % 100 == 0) {
2866             errors.progress();
2867           }
2868           countRecord++;
2869           return true;
2870         } catch (RuntimeException e) {
2871           LOG.error("Result=" + result);
2872           throw e;
2873         }
2874       }
2875     };
2876     if (!checkMetaOnly) {
2877       // Scan hbase:meta to pick up user regions
2878       MetaScanner.metaScan(getConf(), visitor);
2879     }
2880 
2881     errors.print("");
2882     return true;
2883   }
2884 
2885   /**
2886    * Stores the regioninfo entries scanned from META
2887    */
2888   static class MetaEntry extends HRegionInfo {
2889     ServerName regionServer;   // server hosting this region
2890     long modTime;          // timestamp of most recent modification metadata
2891     HRegionInfo splitA, splitB; //split daughters
2892 
2893     public MetaEntry(HRegionInfo rinfo, ServerName regionServer, long modTime) {
2894       this(rinfo, regionServer, modTime, null, null);
2895     }
2896 
2897     public MetaEntry(HRegionInfo rinfo, ServerName regionServer, long modTime,
2898         HRegionInfo splitA, HRegionInfo splitB) {
2899       super(rinfo);
2900       this.regionServer = regionServer;
2901       this.modTime = modTime;
2902       this.splitA = splitA;
2903       this.splitB = splitB;
2904     }
2905 
2906     @Override
2907     public boolean equals(Object o) {
2908       boolean superEq = super.equals(o);
2909       if (!superEq) {
2910         return superEq;
2911       }
2912 
2913       MetaEntry me = (MetaEntry) o;
2914       if (!regionServer.equals(me.regionServer)) {
2915         return false;
2916       }
2917       return (modTime == me.modTime);
2918     }
2919 
2920     @Override
2921     public int hashCode() {
2922       int hash = Arrays.hashCode(getRegionName());
2923       hash ^= getRegionId();
2924       hash ^= Arrays.hashCode(getStartKey());
2925       hash ^= Arrays.hashCode(getEndKey());
2926       hash ^= Boolean.valueOf(isOffline()).hashCode();
2927       hash ^= getTable().hashCode();
2928       if (regionServer != null) {
2929         hash ^= regionServer.hashCode();
2930       }
2931       hash ^= modTime;
2932       return hash;
2933     }
2934   }
2935 
2936   /**
2937    * Stores the regioninfo entries from HDFS
2938    */
2939   static class HdfsEntry {
2940     HRegionInfo hri;
2941     Path hdfsRegionDir = null;
2942     long hdfsRegionDirModTime  = 0;
2943     boolean hdfsRegioninfoFilePresent = false;
2944     boolean hdfsOnlyEdits = false;
2945   }
2946 
2947   /**
2948    * Stores the regioninfo retrieved from Online region servers.
2949    */
2950   static class OnlineEntry {
2951     HRegionInfo hri;
2952     ServerName hsa;
2953 
2954     @Override
2955     public String toString() {
2956       return hsa.toString() + ";" + hri.getRegionNameAsString();
2957     }
2958   }
2959 
2960   /**
2961    * Maintain information about a particular region.  It gathers information
2962    * from three places -- HDFS, META, and region servers.
2963    */
2964   public static class HbckInfo implements KeyRange {
2965     private MetaEntry metaEntry = null; // info in META
2966     private HdfsEntry hdfsEntry = null; // info in HDFS
2967     private List<OnlineEntry> deployedEntries = Lists.newArrayList(); // on Region Server
2968     private List<ServerName> deployedOn = Lists.newArrayList(); // info on RS's
2969     private boolean skipChecks = false; // whether to skip further checks to this region info.
2970     private boolean isMerged = false;// whether this region has already been merged into another one
2971 
2972     HbckInfo(MetaEntry metaEntry) {
2973       this.metaEntry = metaEntry;
2974     }
2975 
2976     public synchronized void addServer(HRegionInfo hri, ServerName server) {
2977       OnlineEntry rse = new OnlineEntry() ;
2978       rse.hri = hri;
2979       rse.hsa = server;
2980       this.deployedEntries.add(rse);
2981       this.deployedOn.add(server);
2982     }
2983 
2984     @Override
2985     public synchronized String toString() {
2986       StringBuilder sb = new StringBuilder();
2987       sb.append("{ meta => ");
2988       sb.append((metaEntry != null)? metaEntry.getRegionNameAsString() : "null");
2989       sb.append( ", hdfs => " + getHdfsRegionDir());
2990       sb.append( ", deployed => " + Joiner.on(", ").join(deployedEntries));
2991       sb.append(" }");
2992       return sb.toString();
2993     }
2994 
2995     @Override
2996     public byte[] getStartKey() {
2997       if (this.metaEntry != null) {
2998         return this.metaEntry.getStartKey();
2999       } else if (this.hdfsEntry != null) {
3000         return this.hdfsEntry.hri.getStartKey();
3001       } else {
3002         LOG.error("Entry " + this + " has no meta or hdfs region start key.");
3003         return null;
3004       }
3005     }
3006 
3007     @Override
3008     public byte[] getEndKey() {
3009       if (this.metaEntry != null) {
3010         return this.metaEntry.getEndKey();
3011       } else if (this.hdfsEntry != null) {
3012         return this.hdfsEntry.hri.getEndKey();
3013       } else {
3014         LOG.error("Entry " + this + " has no meta or hdfs region start key.");
3015         return null;
3016       }
3017     }
3018 
3019     public TableName getTableName() {
3020       if (this.metaEntry != null) {
3021         return this.metaEntry.getTable();
3022       } else if (this.hdfsEntry != null) {
3023         // we are only guaranteed to have a path and not an HRI for hdfsEntry,
3024         // so we get the name from the Path
3025         Path tableDir = this.hdfsEntry.hdfsRegionDir.getParent();
3026         return FSUtils.getTableName(tableDir);
3027       } else {
3028         // Currently no code exercises this path, but we could add one for
3029         // getting table name from OnlineEntry
3030         return null;
3031       }
3032     }
3033 
3034     public String getRegionNameAsString() {
3035       if (metaEntry != null) {
3036         return metaEntry.getRegionNameAsString();
3037       } else if (hdfsEntry != null) {
3038         if (hdfsEntry.hri != null) {
3039           return hdfsEntry.hri.getRegionNameAsString();
3040         }
3041       }
3042       return null;
3043     }
3044 
3045     public byte[] getRegionName() {
3046       if (metaEntry != null) {
3047         return metaEntry.getRegionName();
3048       } else if (hdfsEntry != null) {
3049         return hdfsEntry.hri.getRegionName();
3050       } else {
3051         return null;
3052       }
3053     }
3054 
3055     Path getHdfsRegionDir() {
3056       if (hdfsEntry == null) {
3057         return null;
3058       }
3059       return hdfsEntry.hdfsRegionDir;
3060     }
3061 
3062     boolean containsOnlyHdfsEdits() {
3063       if (hdfsEntry == null) {
3064         return false;
3065       }
3066       return hdfsEntry.hdfsOnlyEdits;
3067     }
3068 
3069     boolean isHdfsRegioninfoPresent() {
3070       if (hdfsEntry == null) {
3071         return false;
3072       }
3073       return hdfsEntry.hdfsRegioninfoFilePresent;
3074     }
3075 
3076     long getModTime() {
3077       if (hdfsEntry == null) {
3078         return 0;
3079       }
3080       return hdfsEntry.hdfsRegionDirModTime;
3081     }
3082 
3083     HRegionInfo getHdfsHRI() {
3084       if (hdfsEntry == null) {
3085         return null;
3086       }
3087       return hdfsEntry.hri;
3088     }
3089 
3090     public void setSkipChecks(boolean skipChecks) {
3091       this.skipChecks = skipChecks;
3092     }
3093 
3094     public boolean isSkipChecks() {
3095       return skipChecks;
3096     }
3097 
3098     public void setMerged(boolean isMerged) {
3099       this.isMerged = isMerged;
3100     }
3101 
3102     public boolean isMerged() {
3103       return this.isMerged;
3104     }
3105   }
3106 
3107   final static Comparator<HbckInfo> cmp = new Comparator<HbckInfo>() {
3108     @Override
3109     public int compare(HbckInfo l, HbckInfo r) {
3110       if (l == r) {
3111         // same instance
3112         return 0;
3113       }
3114 
3115       int tableCompare = l.getTableName().compareTo(r.getTableName());
3116       if (tableCompare != 0) {
3117         return tableCompare;
3118       }
3119 
3120       int startComparison = RegionSplitCalculator.BYTES_COMPARATOR.compare(
3121           l.getStartKey(), r.getStartKey());
3122       if (startComparison != 0) {
3123         return startComparison;
3124       }
3125 
3126       // Special case for absolute endkey
3127       byte[] endKey = r.getEndKey();
3128       endKey = (endKey.length == 0) ? null : endKey;
3129       byte[] endKey2 = l.getEndKey();
3130       endKey2 = (endKey2.length == 0) ? null : endKey2;
3131       int endComparison = RegionSplitCalculator.BYTES_COMPARATOR.compare(
3132           endKey2,  endKey);
3133 
3134       if (endComparison != 0) {
3135         return endComparison;
3136       }
3137 
3138       // use regionId as tiebreaker.
3139       // Null is considered after all possible values so make it bigger.
3140       if (l.hdfsEntry == null && r.hdfsEntry == null) {
3141         return 0;
3142       }
3143       if (l.hdfsEntry == null && r.hdfsEntry != null) {
3144         return 1;
3145       }
3146       // l.hdfsEntry must not be null
3147       if (r.hdfsEntry == null) {
3148         return -1;
3149       }
3150       // both l.hdfsEntry and r.hdfsEntry must not be null.
3151       return (int) (l.hdfsEntry.hri.getRegionId()- r.hdfsEntry.hri.getRegionId());
3152     }
3153   };
3154 
3155   /**
3156    * Prints summary of all tables found on the system.
3157    */
3158   private void printTableSummary(SortedMap<TableName, TableInfo> tablesInfo) {
3159     StringBuilder sb = new StringBuilder();
3160     errors.print("Summary:");
3161     for (TableInfo tInfo : tablesInfo.values()) {
3162       if (errors.tableHasErrors(tInfo)) {
3163         errors.print("Table " + tInfo.getName() + " is inconsistent.");
3164       } else {
3165         errors.print("  " + tInfo.getName() + " is okay.");
3166       }
3167       errors.print("    Number of regions: " + tInfo.getNumRegions());
3168       sb.setLength(0); // clear out existing buffer, if any.
3169       sb.append("    Deployed on: ");
3170       for (ServerName server : tInfo.deployedOn) {
3171         sb.append(" " + server.toString());
3172       }
3173       errors.print(sb.toString());
3174     }
3175   }
3176 
3177   static ErrorReporter getErrorReporter(
3178       final Configuration conf) throws ClassNotFoundException {
3179     Class<? extends ErrorReporter> reporter = conf.getClass("hbasefsck.errorreporter", PrintingErrorReporter.class, ErrorReporter.class);
3180     return (ErrorReporter)ReflectionUtils.newInstance(reporter, conf);
3181   }
3182 
3183   public interface ErrorReporter {
3184     enum ERROR_CODE {
3185       UNKNOWN, NO_META_REGION, NULL_META_REGION, NO_VERSION_FILE, NOT_IN_META_HDFS, NOT_IN_META,
3186       NOT_IN_META_OR_DEPLOYED, NOT_IN_HDFS_OR_DEPLOYED, NOT_IN_HDFS, SERVER_DOES_NOT_MATCH_META, NOT_DEPLOYED,
3187       MULTI_DEPLOYED, SHOULD_NOT_BE_DEPLOYED, MULTI_META_REGION, RS_CONNECT_FAILURE,
3188       FIRST_REGION_STARTKEY_NOT_EMPTY, LAST_REGION_ENDKEY_NOT_EMPTY, DUPE_STARTKEYS,
3189       HOLE_IN_REGION_CHAIN, OVERLAP_IN_REGION_CHAIN, REGION_CYCLE, DEGENERATE_REGION,
3190       ORPHAN_HDFS_REGION, LINGERING_SPLIT_PARENT, NO_TABLEINFO_FILE, LINGERING_REFERENCE_HFILE,
3191       WRONG_USAGE, EMPTY_META_CELL, EXPIRED_TABLE_LOCK, BOUNDARIES_ERROR
3192     }
3193     void clear();
3194     void report(String message);
3195     void reportError(String message);
3196     void reportError(ERROR_CODE errorCode, String message);
3197     void reportError(ERROR_CODE errorCode, String message, TableInfo table);
3198     void reportError(ERROR_CODE errorCode, String message, TableInfo table, HbckInfo info);
3199     void reportError(
3200       ERROR_CODE errorCode,
3201       String message,
3202       TableInfo table,
3203       HbckInfo info1,
3204       HbckInfo info2
3205     );
3206     int summarize();
3207     void detail(String details);
3208     ArrayList<ERROR_CODE> getErrorList();
3209     void progress();
3210     void print(String message);
3211     void resetErrors();
3212     boolean tableHasErrors(TableInfo table);
3213   }
3214 
3215   static class PrintingErrorReporter implements ErrorReporter {
3216     public int errorCount = 0;
3217     private int showProgress;
3218 
3219     Set<TableInfo> errorTables = new HashSet<TableInfo>();
3220 
3221     // for use by unit tests to verify which errors were discovered
3222     private ArrayList<ERROR_CODE> errorList = new ArrayList<ERROR_CODE>();
3223 
3224     @Override
3225     public void clear() {
3226       errorTables.clear();
3227       errorList.clear();
3228       errorCount = 0;
3229     }
3230 
3231     @Override
3232     public synchronized void reportError(ERROR_CODE errorCode, String message) {
3233       if (errorCode == ERROR_CODE.WRONG_USAGE) {
3234         System.err.println(message);
3235         return;
3236       }
3237 
3238       errorList.add(errorCode);
3239       if (!summary) {
3240         System.out.println("ERROR: " + message);
3241       }
3242       errorCount++;
3243       showProgress = 0;
3244     }
3245 
3246     @Override
3247     public synchronized void reportError(ERROR_CODE errorCode, String message, TableInfo table) {
3248       errorTables.add(table);
3249       reportError(errorCode, message);
3250     }
3251 
3252     @Override
3253     public synchronized void reportError(ERROR_CODE errorCode, String message, TableInfo table,
3254                                          HbckInfo info) {
3255       errorTables.add(table);
3256       String reference = "(region " + info.getRegionNameAsString() + ")";
3257       reportError(errorCode, reference + " " + message);
3258     }
3259 
3260     @Override
3261     public synchronized void reportError(ERROR_CODE errorCode, String message, TableInfo table,
3262                                          HbckInfo info1, HbckInfo info2) {
3263       errorTables.add(table);
3264       String reference = "(regions " + info1.getRegionNameAsString()
3265           + " and " + info2.getRegionNameAsString() + ")";
3266       reportError(errorCode, reference + " " + message);
3267     }
3268 
3269     @Override
3270     public synchronized void reportError(String message) {
3271       reportError(ERROR_CODE.UNKNOWN, message);
3272     }
3273 
3274     /**
3275      * Report error information, but do not increment the error count.  Intended for cases
3276      * where the actual error would have been reported previously.
3277      * @param message
3278      */
3279     @Override
3280     public synchronized void report(String message) {
3281       if (! summary) {
3282         System.out.println("ERROR: " + message);
3283       }
3284       showProgress = 0;
3285     }
3286 
3287     @Override
3288     public synchronized int summarize() {
3289       System.out.println(Integer.toString(errorCount) +
3290                          " inconsistencies detected.");
3291       if (errorCount == 0) {
3292         System.out.println("Status: OK");
3293         return 0;
3294       } else {
3295         System.out.println("Status: INCONSISTENT");
3296         return -1;
3297       }
3298     }
3299 
3300     @Override
3301     public ArrayList<ERROR_CODE> getErrorList() {
3302       return errorList;
3303     }
3304 
3305     @Override
3306     public synchronized void print(String message) {
3307       if (!summary) {
3308         System.out.println(message);
3309       }
3310     }
3311 
3312     @Override
3313     public boolean tableHasErrors(TableInfo table) {
3314       return errorTables.contains(table);
3315     }
3316 
3317     @Override
3318     public void resetErrors() {
3319       errorCount = 0;
3320     }
3321 
3322     @Override
3323     public synchronized void detail(String message) {
3324       if (details) {
3325         System.out.println(message);
3326       }
3327       showProgress = 0;
3328     }
3329 
3330     @Override
3331     public synchronized void progress() {
3332       if (showProgress++ == 10) {
3333         if (!summary) {
3334           System.out.print(".");
3335         }
3336         showProgress = 0;
3337       }
3338     }
3339   }
3340 
3341   /**
3342    * Contact a region server and get all information from it
3343    */
3344   static class WorkItemRegion implements Callable<Void> {
3345     private HBaseFsck hbck;
3346     private ServerName rsinfo;
3347     private ErrorReporter errors;
3348     private HConnection connection;
3349 
3350     WorkItemRegion(HBaseFsck hbck, ServerName info,
3351                    ErrorReporter errors, HConnection connection) {
3352       this.hbck = hbck;
3353       this.rsinfo = info;
3354       this.errors = errors;
3355       this.connection = connection;
3356     }
3357 
3358     @Override
3359     public synchronized Void call() throws IOException {
3360       errors.progress();
3361       try {
3362         BlockingInterface server = connection.getAdmin(rsinfo);
3363 
3364         // list all online regions from this region server
3365         List<HRegionInfo> regions = ProtobufUtil.getOnlineRegions(server);
3366         regions = filterRegions(regions);
3367 
3368         if (details) {
3369           errors.detail("RegionServer: " + rsinfo.getServerName() +
3370                            " number of regions: " + regions.size());
3371           for (HRegionInfo rinfo: regions) {
3372             errors.detail("  " + rinfo.getRegionNameAsString() +
3373                              " id: " + rinfo.getRegionId() +
3374                              " encoded_name: " + rinfo.getEncodedName() +
3375                              " start: " + Bytes.toStringBinary(rinfo.getStartKey()) +
3376                              " end: " + Bytes.toStringBinary(rinfo.getEndKey()));
3377           }
3378         }
3379 
3380         // check to see if the existence of this region matches the region in META
3381         for (HRegionInfo r:regions) {
3382           HbckInfo hbi = hbck.getOrCreateInfo(r.getEncodedName());
3383           hbi.addServer(r, rsinfo);
3384         }
3385       } catch (IOException e) {          // unable to connect to the region server.
3386         errors.reportError(ERROR_CODE.RS_CONNECT_FAILURE, "RegionServer: " + rsinfo.getServerName() +
3387           " Unable to fetch region information. " + e);
3388         throw e;
3389       }
3390       return null;
3391     }
3392 
3393     private List<HRegionInfo> filterRegions(List<HRegionInfo> regions) {
3394       List<HRegionInfo> ret = Lists.newArrayList();
3395       for (HRegionInfo hri : regions) {
3396         if (hri.isMetaTable() || (!hbck.checkMetaOnly
3397             && hbck.isTableIncluded(hri.getTable()))) {
3398           ret.add(hri);
3399         }
3400       }
3401       return ret;
3402     }
3403   }
3404 
3405   /**
3406    * Contact hdfs and get all information about specified table directory into
3407    * regioninfo list.
3408    */
3409   static class WorkItemHdfsDir implements Callable<Void> {
3410     private HBaseFsck hbck;
3411     private FileStatus tableDir;
3412     private ErrorReporter errors;
3413     private FileSystem fs;
3414 
3415     WorkItemHdfsDir(HBaseFsck hbck, FileSystem fs, ErrorReporter errors,
3416                     FileStatus status) {
3417       this.hbck = hbck;
3418       this.fs = fs;
3419       this.tableDir = status;
3420       this.errors = errors;
3421     }
3422 
3423     @Override
3424     public synchronized Void call() throws IOException {
3425       try {
3426         // level 2: <HBASE_DIR>/<table>/*
3427         FileStatus[] regionDirs = fs.listStatus(tableDir.getPath());
3428         for (FileStatus regionDir : regionDirs) {
3429           String encodedName = regionDir.getPath().getName();
3430           // ignore directories that aren't hexadecimal
3431           if (!encodedName.toLowerCase().matches("[0-9a-f]+")) {
3432             continue;
3433           }
3434 
3435           LOG.debug("Loading region info from hdfs:"+ regionDir.getPath());
3436           HbckInfo hbi = hbck.getOrCreateInfo(encodedName);
3437           HdfsEntry he = new HdfsEntry();
3438           synchronized (hbi) {
3439             if (hbi.getHdfsRegionDir() != null) {
3440               errors.print("Directory " + encodedName + " duplicate??" +
3441                            hbi.getHdfsRegionDir());
3442             }
3443 
3444             he.hdfsRegionDir = regionDir.getPath();
3445             he.hdfsRegionDirModTime = regionDir.getModificationTime();
3446             Path regioninfoFile = new Path(he.hdfsRegionDir, HRegionFileSystem.REGION_INFO_FILE);
3447             he.hdfsRegioninfoFilePresent = fs.exists(regioninfoFile);
3448             // we add to orphan list when we attempt to read .regioninfo
3449 
3450             // Set a flag if this region contains only edits
3451             // This is special case if a region is left after split
3452             he.hdfsOnlyEdits = true;
3453             FileStatus[] subDirs = fs.listStatus(regionDir.getPath());
3454             Path ePath = HLogUtil.getRegionDirRecoveredEditsDir(regionDir.getPath());
3455             for (FileStatus subDir : subDirs) {
3456               String sdName = subDir.getPath().getName();
3457               if (!sdName.startsWith(".") && !sdName.equals(ePath.getName())) {
3458                 he.hdfsOnlyEdits = false;
3459                 break;
3460               }
3461             }
3462             hbi.hdfsEntry = he;
3463           }
3464         }
3465       } catch (IOException e) {
3466         // unable to connect to the region server.
3467         errors.reportError(ERROR_CODE.RS_CONNECT_FAILURE, "Table Directory: "
3468             + tableDir.getPath().getName()
3469             + " Unable to fetch region information. " + e);
3470         throw e;
3471       }
3472       return null;
3473     }
3474   }
3475 
3476   /**
3477    * Contact hdfs and get all information about specified table directory into
3478    * regioninfo list.
3479    */
3480   static class WorkItemHdfsRegionInfo implements Callable<Void> {
3481     private HbckInfo hbi;
3482     private HBaseFsck hbck;
3483     private ErrorReporter errors;
3484 
3485     WorkItemHdfsRegionInfo(HbckInfo hbi, HBaseFsck hbck, ErrorReporter errors) {
3486       this.hbi = hbi;
3487       this.hbck = hbck;
3488       this.errors = errors;
3489     }
3490 
3491     @Override
3492     public synchronized Void call() throws IOException {
3493       // only load entries that haven't been loaded yet.
3494       if (hbi.getHdfsHRI() == null) {
3495         try {
3496           hbck.loadHdfsRegioninfo(hbi);
3497         } catch (IOException ioe) {
3498           String msg = "Orphan region in HDFS: Unable to load .regioninfo from table "
3499               + hbi.getTableName() + " in hdfs dir "
3500               + hbi.getHdfsRegionDir()
3501               + "!  It may be an invalid format or version file.  Treating as "
3502               + "an orphaned regiondir.";
3503           errors.reportError(ERROR_CODE.ORPHAN_HDFS_REGION, msg);
3504           try {
3505             hbck.debugLsr(hbi.getHdfsRegionDir());
3506           } catch (IOException ioe2) {
3507             LOG.error("Unable to read directory " + hbi.getHdfsRegionDir(), ioe2);
3508             throw ioe2;
3509           }
3510           hbck.orphanHdfsDirs.add(hbi);
3511           throw ioe;
3512         }
3513       }
3514       return null;
3515     }
3516   };
3517 
3518   /**
3519    * Display the full report from fsck. This displays all live and dead region
3520    * servers, and all known regions.
3521    */
3522   public static void setDisplayFullReport() {
3523     details = true;
3524   }
3525 
3526   /**
3527    * Set summary mode.
3528    * Print only summary of the tables and status (OK or INCONSISTENT)
3529    */
3530   void setSummary() {
3531     summary = true;
3532   }
3533 
3534   /**
3535    * Set hbase:meta check mode.
3536    * Print only info about hbase:meta table deployment/state
3537    */
3538   void setCheckMetaOnly() {
3539     checkMetaOnly = true;
3540   }
3541 
3542   /**
3543    * Set region boundaries check mode.
3544    */
3545   void setRegionBoundariesCheck() {
3546     checkRegionBoundaries = true;
3547   }
3548 
3549   /**
3550    * Set table locks fix mode.
3551    * Delete table locks held for a long time
3552    */
3553   public void setFixTableLocks(boolean shouldFix) {
3554     fixTableLocks = shouldFix;
3555   }
3556 
3557   /**
3558    * Check if we should rerun fsck again. This checks if we've tried to
3559    * fix something and we should rerun fsck tool again.
3560    * Display the full report from fsck. This displays all live and dead
3561    * region servers, and all known regions.
3562    */
3563   void setShouldRerun() {
3564     rerun = true;
3565   }
3566 
3567   boolean shouldRerun() {
3568     return rerun;
3569   }
3570 
3571   /**
3572    * Fix inconsistencies found by fsck. This should try to fix errors (if any)
3573    * found by fsck utility.
3574    */
3575   public void setFixAssignments(boolean shouldFix) {
3576     fixAssignments = shouldFix;
3577   }
3578 
3579   boolean shouldFixAssignments() {
3580     return fixAssignments;
3581   }
3582 
3583   public void setFixMeta(boolean shouldFix) {
3584     fixMeta = shouldFix;
3585   }
3586 
3587   boolean shouldFixMeta() {
3588     return fixMeta;
3589   }
3590 
3591   public void setFixEmptyMetaCells(boolean shouldFix) {
3592     fixEmptyMetaCells = shouldFix;
3593   }
3594 
3595   boolean shouldFixEmptyMetaCells() {
3596     return fixEmptyMetaCells;
3597   }
3598 
3599   public void setCheckHdfs(boolean checking) {
3600     checkHdfs = checking;
3601   }
3602 
3603   boolean shouldCheckHdfs() {
3604     return checkHdfs;
3605   }
3606 
3607   public void setFixHdfsHoles(boolean shouldFix) {
3608     fixHdfsHoles = shouldFix;
3609   }
3610 
3611   boolean shouldFixHdfsHoles() {
3612     return fixHdfsHoles;
3613   }
3614 
3615   public void setFixTableOrphans(boolean shouldFix) {
3616     fixTableOrphans = shouldFix;
3617   }
3618 
3619   boolean shouldFixTableOrphans() {
3620     return fixTableOrphans;
3621   }
3622 
3623   public void setFixHdfsOverlaps(boolean shouldFix) {
3624     fixHdfsOverlaps = shouldFix;
3625   }
3626 
3627   boolean shouldFixHdfsOverlaps() {
3628     return fixHdfsOverlaps;
3629   }
3630 
3631   public void setFixHdfsOrphans(boolean shouldFix) {
3632     fixHdfsOrphans = shouldFix;
3633   }
3634 
3635   boolean shouldFixHdfsOrphans() {
3636     return fixHdfsOrphans;
3637   }
3638 
3639   public void setFixVersionFile(boolean shouldFix) {
3640     fixVersionFile = shouldFix;
3641   }
3642 
3643   public boolean shouldFixVersionFile() {
3644     return fixVersionFile;
3645   }
3646 
3647   public void setSidelineBigOverlaps(boolean sbo) {
3648     this.sidelineBigOverlaps = sbo;
3649   }
3650 
3651   public boolean shouldSidelineBigOverlaps() {
3652     return sidelineBigOverlaps;
3653   }
3654 
3655   public void setFixSplitParents(boolean shouldFix) {
3656     fixSplitParents = shouldFix;
3657   }
3658 
3659   boolean shouldFixSplitParents() {
3660     return fixSplitParents;
3661   }
3662 
3663   public void setFixReferenceFiles(boolean shouldFix) {
3664     fixReferenceFiles = shouldFix;
3665   }
3666 
3667   boolean shouldFixReferenceFiles() {
3668     return fixReferenceFiles;
3669   }
3670 
3671   public boolean shouldIgnorePreCheckPermission() {
3672     return ignorePreCheckPermission;
3673   }
3674 
3675   public void setIgnorePreCheckPermission(boolean ignorePreCheckPermission) {
3676     this.ignorePreCheckPermission = ignorePreCheckPermission;
3677   }
3678 
3679   /**
3680    * @param mm maximum number of regions to merge into a single region.
3681    */
3682   public void setMaxMerge(int mm) {
3683     this.maxMerge = mm;
3684   }
3685 
3686   public int getMaxMerge() {
3687     return maxMerge;
3688   }
3689 
3690   public void setMaxOverlapsToSideline(int mo) {
3691     this.maxOverlapsToSideline = mo;
3692   }
3693 
3694   public int getMaxOverlapsToSideline() {
3695     return maxOverlapsToSideline;
3696   }
3697 
3698   /**
3699    * Only check/fix tables specified by the list,
3700    * Empty list means all tables are included.
3701    */
3702   boolean isTableIncluded(TableName table) {
3703     return (tablesIncluded.size() == 0) || tablesIncluded.contains(table);
3704   }
3705 
3706   public void includeTable(TableName table) {
3707     tablesIncluded.add(table);
3708   }
3709 
3710   Set<TableName> getIncludedTables() {
3711     return new HashSet<TableName>(tablesIncluded);
3712   }
3713 
3714   /**
3715    * We are interested in only those tables that have not changed their state in
3716    * hbase:meta during the last few seconds specified by hbase.admin.fsck.timelag
3717    * @param seconds - the time in seconds
3718    */
3719   public void setTimeLag(long seconds) {
3720     timelag = seconds * 1000; // convert to milliseconds
3721   }
3722 
3723   /**
3724    *
3725    * @param sidelineDir - HDFS path to sideline data
3726    */
3727   public void setSidelineDir(String sidelineDir) {
3728     this.sidelineDir = new Path(sidelineDir);
3729   }
3730 
3731   protected HFileCorruptionChecker createHFileCorruptionChecker(boolean sidelineCorruptHFiles) throws IOException {
3732     return new HFileCorruptionChecker(getConf(), executor, sidelineCorruptHFiles);
3733   }
3734 
3735   public HFileCorruptionChecker getHFilecorruptionChecker() {
3736     return hfcc;
3737   }
3738 
3739   public void setHFileCorruptionChecker(HFileCorruptionChecker hfcc) {
3740     this.hfcc = hfcc;
3741   }
3742 
3743   public void setRetCode(int code) {
3744     this.retcode = code;
3745   }
3746 
3747   public int getRetCode() {
3748     return retcode;
3749   }
3750 
3751   protected HBaseFsck printUsageAndExit() {
3752     StringWriter sw = new StringWriter(2048);
3753     PrintWriter out = new PrintWriter(sw);
3754     out.println("Usage: fsck [opts] {only tables}");
3755     out.println(" where [opts] are:");
3756     out.println("   -help Display help options (this)");
3757     out.println("   -details Display full report of all regions.");
3758     out.println("   -timelag <timeInSeconds>  Process only regions that " +
3759                        " have not experienced any metadata updates in the last " +
3760                        " <timeInSeconds> seconds.");
3761     out.println("   -sleepBeforeRerun <timeInSeconds> Sleep this many seconds" +
3762         " before checking if the fix worked if run with -fix");
3763     out.println("   -summary Print only summary of the tables and status.");
3764     out.println("   -metaonly Only check the state of the hbase:meta table.");
3765     out.println("   -sidelineDir <hdfs://> HDFS path to backup existing meta.");
3766     out.println("   -boundaries Verify that regions boundaries are the same between META and store files.");
3767 
3768     out.println("");
3769     out.println("  Metadata Repair options: (expert features, use with caution!)");
3770     out.println("   -fix              Try to fix region assignments.  This is for backwards compatiblity");
3771     out.println("   -fixAssignments   Try to fix region assignments.  Replaces the old -fix");
3772     out.println("   -fixMeta          Try to fix meta problems.  This assumes HDFS region info is good.");
3773     out.println("   -noHdfsChecking   Don't load/check region info from HDFS."
3774         + " Assumes hbase:meta region info is good. Won't check/fix any HDFS issue, e.g. hole, orphan, or overlap");
3775     out.println("   -fixHdfsHoles     Try to fix region holes in hdfs.");
3776     out.println("   -fixHdfsOrphans   Try to fix region dirs with no .regioninfo file in hdfs");
3777     out.println("   -fixTableOrphans  Try to fix table dirs with no .tableinfo file in hdfs (online mode only)");
3778     out.println("   -fixHdfsOverlaps  Try to fix region overlaps in hdfs.");
3779     out.println("   -fixVersionFile   Try to fix missing hbase.version file in hdfs.");
3780     out.println("   -maxMerge <n>     When fixing region overlaps, allow at most <n> regions to merge. (n=" + DEFAULT_MAX_MERGE +" by default)");
3781     out.println("   -sidelineBigOverlaps  When fixing region overlaps, allow to sideline big overlaps");
3782     out.println("   -maxOverlapsToSideline <n>  When fixing region overlaps, allow at most <n> regions to sideline per group. (n=" + DEFAULT_OVERLAPS_TO_SIDELINE +" by default)");
3783     out.println("   -fixSplitParents  Try to force offline split parents to be online.");
3784     out.println("   -ignorePreCheckPermission  ignore filesystem permission pre-check");
3785     out.println("   -fixReferenceFiles  Try to offline lingering reference store files");
3786     out.println("   -fixEmptyMetaCells  Try to fix hbase:meta entries not referencing any region"
3787         + " (empty REGIONINFO_QUALIFIER rows)");
3788 
3789     out.println("");
3790     out.println("  Datafile Repair options: (expert features, use with caution!)");
3791     out.println("   -checkCorruptHFiles     Check all Hfiles by opening them to make sure they are valid");
3792     out.println("   -sidelineCorruptHFiles  Quarantine corrupted HFiles.  implies -checkCorruptHFiles");
3793 
3794     out.println("");
3795     out.println("  Metadata Repair shortcuts");
3796     out.println("   -repair           Shortcut for -fixAssignments -fixMeta -fixHdfsHoles " +
3797         "-fixHdfsOrphans -fixHdfsOverlaps -fixVersionFile -sidelineBigOverlaps -fixReferenceFiles -fixTableLocks");
3798     out.println("   -repairHoles      Shortcut for -fixAssignments -fixMeta -fixHdfsHoles");
3799 
3800     out.println("");
3801     out.println("  Table lock options");
3802     out.println("   -fixTableLocks    Deletes table locks held for a long time (hbase.table.lock.expire.ms, 10min by default)");
3803 
3804     out.flush();
3805     errors.reportError(ERROR_CODE.WRONG_USAGE, sw.toString());
3806 
3807     setRetCode(-2);
3808     return this;
3809   }
3810 
3811   /**
3812    * Main program
3813    *
3814    * @param args
3815    * @throws Exception
3816    */
3817   public static void main(String[] args) throws Exception {
3818     // create a fsck object
3819     Configuration conf = HBaseConfiguration.create();
3820     Path hbasedir = FSUtils.getRootDir(conf);
3821     URI defaultFs = hbasedir.getFileSystem(conf).getUri();
3822     FSUtils.setFsDefault(conf, new Path(defaultFs));
3823 
3824     int ret = ToolRunner.run(new HBaseFsckTool(conf), args);
3825     System.exit(ret);
3826   }
3827 
3828   /**
3829    * This is a Tool wrapper that gathers -Dxxx=yyy configuration settings from the command line.
3830    */
3831   static class HBaseFsckTool extends Configured implements Tool {
3832     HBaseFsckTool(Configuration conf) { super(conf); }
3833     @Override
3834     public int run(String[] args) throws Exception {
3835       HBaseFsck hbck = new HBaseFsck(getConf());
3836       hbck.exec(hbck.executor, args);
3837       return hbck.getRetCode();
3838     }
3839   };
3840   
3841 
3842   public HBaseFsck exec(ExecutorService exec, String[] args) throws KeeperException, IOException,
3843     ServiceException, InterruptedException {
3844     long sleepBeforeRerun = DEFAULT_SLEEP_BEFORE_RERUN;
3845 
3846     boolean checkCorruptHFiles = false;
3847     boolean sidelineCorruptHFiles = false;
3848 
3849     // Process command-line args.
3850     for (int i = 0; i < args.length; i++) {
3851       String cmd = args[i];
3852       if (cmd.equals("-help") || cmd.equals("-h")) {
3853         return printUsageAndExit();
3854       } else if (cmd.equals("-details")) {
3855         setDisplayFullReport();
3856       } else if (cmd.equals("-timelag")) {
3857         if (i == args.length - 1) {
3858           errors.reportError(ERROR_CODE.WRONG_USAGE, "HBaseFsck: -timelag needs a value.");
3859           return printUsageAndExit();
3860         }
3861         try {
3862           long timelag = Long.parseLong(args[i+1]);
3863           setTimeLag(timelag);
3864         } catch (NumberFormatException e) {
3865           errors.reportError(ERROR_CODE.WRONG_USAGE, "-timelag needs a numeric value.");
3866           return printUsageAndExit();
3867         }
3868         i++;
3869       } else if (cmd.equals("-sleepBeforeRerun")) {
3870         if (i == args.length - 1) {
3871           errors.reportError(ERROR_CODE.WRONG_USAGE,
3872             "HBaseFsck: -sleepBeforeRerun needs a value.");
3873           return printUsageAndExit();
3874         }
3875         try {
3876           sleepBeforeRerun = Long.parseLong(args[i+1]);
3877         } catch (NumberFormatException e) {
3878           errors.reportError(ERROR_CODE.WRONG_USAGE, "-sleepBeforeRerun needs a numeric value.");
3879           return printUsageAndExit();
3880         }
3881         i++;
3882       } else if (cmd.equals("-sidelineDir")) {
3883         if (i == args.length - 1) {
3884           errors.reportError(ERROR_CODE.WRONG_USAGE, "HBaseFsck: -sidelineDir needs a value.");
3885           return printUsageAndExit();
3886         }
3887         i++;
3888         setSidelineDir(args[i]);
3889       } else if (cmd.equals("-fix")) {
3890         errors.reportError(ERROR_CODE.WRONG_USAGE,
3891           "This option is deprecated, please use  -fixAssignments instead.");
3892         setFixAssignments(true);
3893       } else if (cmd.equals("-fixAssignments")) {
3894         setFixAssignments(true);
3895       } else if (cmd.equals("-fixMeta")) {
3896         setFixMeta(true);
3897       } else if (cmd.equals("-noHdfsChecking")) {
3898         setCheckHdfs(false);
3899       } else if (cmd.equals("-fixHdfsHoles")) {
3900         setFixHdfsHoles(true);
3901       } else if (cmd.equals("-fixHdfsOrphans")) {
3902         setFixHdfsOrphans(true);
3903       } else if (cmd.equals("-fixTableOrphans")) {
3904         setFixTableOrphans(true);
3905       } else if (cmd.equals("-fixHdfsOverlaps")) {
3906         setFixHdfsOverlaps(true);
3907       } else if (cmd.equals("-fixVersionFile")) {
3908         setFixVersionFile(true);
3909       } else if (cmd.equals("-sidelineBigOverlaps")) {
3910         setSidelineBigOverlaps(true);
3911       } else if (cmd.equals("-fixSplitParents")) {
3912         setFixSplitParents(true);
3913       } else if (cmd.equals("-ignorePreCheckPermission")) {
3914         setIgnorePreCheckPermission(true);
3915       } else if (cmd.equals("-checkCorruptHFiles")) {
3916         checkCorruptHFiles = true;
3917       } else if (cmd.equals("-sidelineCorruptHFiles")) {
3918         sidelineCorruptHFiles = true;
3919       } else if (cmd.equals("-fixReferenceFiles")) {
3920         setFixReferenceFiles(true);
3921       } else if (cmd.equals("-fixEmptyMetaCells")) {
3922         setFixEmptyMetaCells(true);
3923       } else if (cmd.equals("-repair")) {
3924         // this attempts to merge overlapping hdfs regions, needs testing
3925         // under load
3926         setFixHdfsHoles(true);
3927         setFixHdfsOrphans(true);
3928         setFixMeta(true);
3929         setFixAssignments(true);
3930         setFixHdfsOverlaps(true);
3931         setFixVersionFile(true);
3932         setSidelineBigOverlaps(true);
3933         setFixSplitParents(false);
3934         setCheckHdfs(true);
3935         setFixReferenceFiles(true);
3936         setFixTableLocks(true);
3937       } else if (cmd.equals("-repairHoles")) {
3938         // this will make all missing hdfs regions available but may lose data
3939         setFixHdfsHoles(true);
3940         setFixHdfsOrphans(false);
3941         setFixMeta(true);
3942         setFixAssignments(true);
3943         setFixHdfsOverlaps(false);
3944         setSidelineBigOverlaps(false);
3945         setFixSplitParents(false);
3946         setCheckHdfs(true);
3947       } else if (cmd.equals("-maxOverlapsToSideline")) {
3948         if (i == args.length - 1) {
3949           errors.reportError(ERROR_CODE.WRONG_USAGE,
3950             "-maxOverlapsToSideline needs a numeric value argument.");
3951           return printUsageAndExit();
3952         }
3953         try {
3954           int maxOverlapsToSideline = Integer.parseInt(args[i+1]);
3955           setMaxOverlapsToSideline(maxOverlapsToSideline);
3956         } catch (NumberFormatException e) {
3957           errors.reportError(ERROR_CODE.WRONG_USAGE,
3958             "-maxOverlapsToSideline needs a numeric value argument.");
3959           return printUsageAndExit();
3960         }
3961         i++;
3962       } else if (cmd.equals("-maxMerge")) {
3963         if (i == args.length - 1) {
3964           errors.reportError(ERROR_CODE.WRONG_USAGE,
3965             "-maxMerge needs a numeric value argument.");
3966           return printUsageAndExit();
3967         }
3968         try {
3969           int maxMerge = Integer.parseInt(args[i+1]);
3970           setMaxMerge(maxMerge);
3971         } catch (NumberFormatException e) {
3972           errors.reportError(ERROR_CODE.WRONG_USAGE,
3973             "-maxMerge needs a numeric value argument.");
3974           return printUsageAndExit();
3975         }
3976         i++;
3977       } else if (cmd.equals("-summary")) {
3978         setSummary();
3979       } else if (cmd.equals("-metaonly")) {
3980         setCheckMetaOnly();
3981       } else if (cmd.equals("-boundaries")) {
3982         setRegionBoundariesCheck();
3983       } else if (cmd.equals("-fixTableLocks")) {
3984         setFixTableLocks(true);
3985       } else if (cmd.startsWith("-")) {
3986         errors.reportError(ERROR_CODE.WRONG_USAGE, "Unrecognized option:" + cmd);
3987         return printUsageAndExit();
3988       } else {
3989         includeTable(TableName.valueOf(cmd));
3990         errors.print("Allow checking/fixes for table: " + cmd);
3991       }
3992     }
3993 
3994     // pre-check current user has FS write permission or not
3995     try {
3996       preCheckPermission();
3997     } catch (AccessControlException ace) {
3998       Runtime.getRuntime().exit(-1);
3999     } catch (IOException ioe) {
4000       Runtime.getRuntime().exit(-1);
4001     }
4002 
4003     // do the real work of hbck
4004     connect();
4005 
4006     try {
4007       // if corrupt file mode is on, first fix them since they may be opened later
4008       if (checkCorruptHFiles || sidelineCorruptHFiles) {
4009         LOG.info("Checking all hfiles for corruption");
4010         HFileCorruptionChecker hfcc = createHFileCorruptionChecker(sidelineCorruptHFiles);
4011         setHFileCorruptionChecker(hfcc); // so we can get result
4012         Collection<TableName> tables = getIncludedTables();
4013         Collection<Path> tableDirs = new ArrayList<Path>();
4014         Path rootdir = FSUtils.getRootDir(getConf());
4015         if (tables.size() > 0) {
4016           for (TableName t : tables) {
4017             tableDirs.add(FSUtils.getTableDir(rootdir, t));
4018           }
4019         } else {
4020           tableDirs = FSUtils.getTableDirs(FSUtils.getCurrentFileSystem(getConf()), rootdir);
4021         }
4022         hfcc.checkTables(tableDirs);
4023         hfcc.report(errors);
4024       }
4025 
4026       // check and fix table integrity, region consistency.
4027       int code = onlineHbck();
4028       setRetCode(code);
4029       // If we have changed the HBase state it is better to run hbck again
4030       // to see if we haven't broken something else in the process.
4031       // We run it only once more because otherwise we can easily fall into
4032       // an infinite loop.
4033       if (shouldRerun()) {
4034         try {
4035           LOG.info("Sleeping " + sleepBeforeRerun + "ms before re-checking after fix...");
4036           Thread.sleep(sleepBeforeRerun);
4037         } catch (InterruptedException ie) {
4038           return this;
4039         }
4040         // Just report
4041         setFixAssignments(false);
4042         setFixMeta(false);
4043         setFixHdfsHoles(false);
4044         setFixHdfsOverlaps(false);
4045         setFixVersionFile(false);
4046         setFixTableOrphans(false);
4047         errors.resetErrors();
4048         code = onlineHbck();
4049         setRetCode(code);
4050       }
4051     } finally {
4052       IOUtils.cleanup(null, connection, meta, admin);
4053     }
4054     return this;
4055   }
4056 
4057   /**
4058    * ls -r for debugging purposes
4059    */
4060   void debugLsr(Path p) throws IOException {
4061     debugLsr(getConf(), p, errors);
4062   }
4063 
4064   /**
4065    * ls -r for debugging purposes
4066    */
4067   public static void debugLsr(Configuration conf,
4068       Path p) throws IOException {
4069     debugLsr(conf, p, new PrintingErrorReporter());
4070   }
4071 
4072   /**
4073    * ls -r for debugging purposes
4074    */
4075   public static void debugLsr(Configuration conf,
4076       Path p, ErrorReporter errors) throws IOException {
4077     if (!LOG.isDebugEnabled() || p == null) {
4078       return;
4079     }
4080     FileSystem fs = p.getFileSystem(conf);
4081 
4082     if (!fs.exists(p)) {
4083       // nothing
4084       return;
4085     }
4086     errors.print(p.toString());
4087 
4088     if (fs.isFile(p)) {
4089       return;
4090     }
4091 
4092     if (fs.getFileStatus(p).isDir()) {
4093       FileStatus[] fss= fs.listStatus(p);
4094       for (FileStatus status : fss) {
4095         debugLsr(conf, status.getPath(), errors);
4096       }
4097     }
4098   }
4099 }