You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@hbase.apache.org by st...@apache.org on 2016/11/15 05:46:14 UTC

[23/35] hbase-site git commit: Published site at 9250bf809155ebe93fd6ae8a0485b22c744fdf70.

http://git-wip-us.apache.org/repos/asf/hbase-site/blob/f17356a7/devapidocs/src-html/org/apache/hadoop/hbase/mapreduce/LoadIncrementalHFiles.html
----------------------------------------------------------------------
diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/mapreduce/LoadIncrementalHFiles.html b/devapidocs/src-html/org/apache/hadoop/hbase/mapreduce/LoadIncrementalHFiles.html
index 6d25806..55fa666 100644
--- a/devapidocs/src-html/org/apache/hadoop/hbase/mapreduce/LoadIncrementalHFiles.html
+++ b/devapidocs/src-html/org/apache/hadoop/hbase/mapreduce/LoadIncrementalHFiles.html
@@ -370,12 +370,12 @@
 <span class="sourceLineNo">362</span>   * @param regionLocator region locator<a name="line.362"></a>
 <span class="sourceLineNo">363</span>   * @param silence true to ignore unmatched column families<a name="line.363"></a>
 <span class="sourceLineNo">364</span>   * @param copyFile always copy hfiles if true<a name="line.364"></a>
-<span class="sourceLineNo">365</span>   * @return List of filenames which were not found<a name="line.365"></a>
+<span class="sourceLineNo">365</span>   * @return Map of LoadQueueItem to region<a name="line.365"></a>
 <span class="sourceLineNo">366</span>   * @throws TableNotFoundException if table does not yet exist<a name="line.366"></a>
 <span class="sourceLineNo">367</span>   */<a name="line.367"></a>
-<span class="sourceLineNo">368</span>  public List&lt;String&gt; doBulkLoad(Map&lt;byte[], List&lt;Path&gt;&gt; map, final Admin admin, Table table,<a name="line.368"></a>
-<span class="sourceLineNo">369</span>          RegionLocator regionLocator, boolean silence, boolean copyFile)<a name="line.369"></a>
-<span class="sourceLineNo">370</span>              throws TableNotFoundException, IOException {<a name="line.370"></a>
+<span class="sourceLineNo">368</span>  public Map&lt;LoadQueueItem, ByteBuffer&gt; doBulkLoad(Map&lt;byte[], List&lt;Path&gt;&gt; map, final Admin admin,<a name="line.368"></a>
+<span class="sourceLineNo">369</span>      Table table, RegionLocator regionLocator, boolean silence, boolean copyFile)<a name="line.369"></a>
+<span class="sourceLineNo">370</span>          throws TableNotFoundException, IOException {<a name="line.370"></a>
 <span class="sourceLineNo">371</span>    if (!admin.isTableAvailable(regionLocator.getName())) {<a name="line.371"></a>
 <span class="sourceLineNo">372</span>      throw new TableNotFoundException("Table " + table.getName() + " is not currently available.");<a name="line.372"></a>
 <span class="sourceLineNo">373</span>    }<a name="line.373"></a>
@@ -457,8 +457,8 @@
 <span class="sourceLineNo">449</span>    }<a name="line.449"></a>
 <span class="sourceLineNo">450</span>  }<a name="line.450"></a>
 <span class="sourceLineNo">451</span><a name="line.451"></a>
-<span class="sourceLineNo">452</span>  List&lt;String&gt; performBulkLoad(final Admin admin, Table table, RegionLocator regionLocator,<a name="line.452"></a>
-<span class="sourceLineNo">453</span>      Deque&lt;LoadQueueItem&gt; queue, ExecutorService pool,<a name="line.453"></a>
+<span class="sourceLineNo">452</span>  Map&lt;LoadQueueItem, ByteBuffer&gt; performBulkLoad(final Admin admin, Table table,<a name="line.452"></a>
+<span class="sourceLineNo">453</span>      RegionLocator regionLocator, Deque&lt;LoadQueueItem&gt; queue, ExecutorService pool,<a name="line.453"></a>
 <span class="sourceLineNo">454</span>      SecureBulkLoadClient secureClient, boolean copyFile) throws IOException {<a name="line.454"></a>
 <span class="sourceLineNo">455</span>    int count = 0;<a name="line.455"></a>
 <span class="sourceLineNo">456</span><a name="line.456"></a>
@@ -472,802 +472,815 @@
 <span class="sourceLineNo">464</span>    // fs is the source filesystem<a name="line.464"></a>
 <span class="sourceLineNo">465</span>    fsDelegationToken.acquireDelegationToken(fs);<a name="line.465"></a>
 <span class="sourceLineNo">466</span>    bulkToken = secureClient.prepareBulkLoad(admin.getConnection());<a name="line.466"></a>
-<span class="sourceLineNo">467</span>    Pair&lt;Multimap&lt;ByteBuffer, LoadQueueItem&gt;, List&lt;String&gt;&gt; pair = null;<a name="line.467"></a>
+<span class="sourceLineNo">467</span>    Pair&lt;Multimap&lt;ByteBuffer, LoadQueueItem&gt;, Set&lt;String&gt;&gt; pair = null;<a name="line.467"></a>
 <span class="sourceLineNo">468</span><a name="line.468"></a>
-<span class="sourceLineNo">469</span>    // Assumes that region splits can happen while this occurs.<a name="line.469"></a>
-<span class="sourceLineNo">470</span>    while (!queue.isEmpty()) {<a name="line.470"></a>
-<span class="sourceLineNo">471</span>      // need to reload split keys each iteration.<a name="line.471"></a>
-<span class="sourceLineNo">472</span>      final Pair&lt;byte[][], byte[][]&gt; startEndKeys = regionLocator.getStartEndKeys();<a name="line.472"></a>
-<span class="sourceLineNo">473</span>      if (count != 0) {<a name="line.473"></a>
-<span class="sourceLineNo">474</span>        LOG.info("Split occured while grouping HFiles, retry attempt " +<a name="line.474"></a>
-<span class="sourceLineNo">475</span>            + count + " with " + queue.size() + " files remaining to group or split");<a name="line.475"></a>
-<span class="sourceLineNo">476</span>      }<a name="line.476"></a>
-<span class="sourceLineNo">477</span><a name="line.477"></a>
-<span class="sourceLineNo">478</span>      int maxRetries = getConf().getInt(HConstants.BULKLOAD_MAX_RETRIES_NUMBER, 10);<a name="line.478"></a>
-<span class="sourceLineNo">479</span>      maxRetries = Math.max(maxRetries, startEndKeys.getFirst().length + 1);<a name="line.479"></a>
-<span class="sourceLineNo">480</span>      if (maxRetries != 0 &amp;&amp; count &gt;= maxRetries) {<a name="line.480"></a>
-<span class="sourceLineNo">481</span>        throw new IOException("Retry attempted " + count +<a name="line.481"></a>
-<span class="sourceLineNo">482</span>            " times without completing, bailing out");<a name="line.482"></a>
-<span class="sourceLineNo">483</span>      }<a name="line.483"></a>
-<span class="sourceLineNo">484</span>      count++;<a name="line.484"></a>
-<span class="sourceLineNo">485</span><a name="line.485"></a>
-<span class="sourceLineNo">486</span>      // Using ByteBuffer for byte[] equality semantics<a name="line.486"></a>
-<span class="sourceLineNo">487</span>      pair = groupOrSplitPhase(table, pool, queue, startEndKeys);<a name="line.487"></a>
-<span class="sourceLineNo">488</span>      Multimap&lt;ByteBuffer, LoadQueueItem&gt; regionGroups = pair.getFirst();<a name="line.488"></a>
-<span class="sourceLineNo">489</span><a name="line.489"></a>
-<span class="sourceLineNo">490</span>      if (!checkHFilesCountPerRegionPerFamily(regionGroups)) {<a name="line.490"></a>
-<span class="sourceLineNo">491</span>        // Error is logged inside checkHFilesCountPerRegionPerFamily.<a name="line.491"></a>
-<span class="sourceLineNo">492</span>        throw new IOException("Trying to load more than " + maxFilesPerRegionPerFamily<a name="line.492"></a>
-<span class="sourceLineNo">493</span>            + " hfiles to one family of one region");<a name="line.493"></a>
-<span class="sourceLineNo">494</span>      }<a name="line.494"></a>
-<span class="sourceLineNo">495</span><a name="line.495"></a>
-<span class="sourceLineNo">496</span>      bulkLoadPhase(table, admin.getConnection(), pool, queue, regionGroups, copyFile);<a name="line.496"></a>
-<span class="sourceLineNo">497</span><a name="line.497"></a>
-<span class="sourceLineNo">498</span>      // NOTE: The next iteration's split / group could happen in parallel to<a name="line.498"></a>
-<span class="sourceLineNo">499</span>      // atomic bulkloads assuming that there are splits and no merges, and<a name="line.499"></a>
-<span class="sourceLineNo">500</span>      // that we can atomically pull out the groups we want to retry.<a name="line.500"></a>
-<span class="sourceLineNo">501</span>    }<a name="line.501"></a>
-<span class="sourceLineNo">502</span><a name="line.502"></a>
-<span class="sourceLineNo">503</span>    if (!queue.isEmpty()) {<a name="line.503"></a>
-<span class="sourceLineNo">504</span>      throw new RuntimeException("Bulk load aborted with some files not yet loaded."<a name="line.504"></a>
-<span class="sourceLineNo">505</span>        + "Please check log for more details.");<a name="line.505"></a>
-<span class="sourceLineNo">506</span>    }<a name="line.506"></a>
-<span class="sourceLineNo">507</span>    if (pair == null) return null;<a name="line.507"></a>
-<span class="sourceLineNo">508</span>    return pair.getSecond();<a name="line.508"></a>
-<span class="sourceLineNo">509</span>  }<a name="line.509"></a>
-<span class="sourceLineNo">510</span><a name="line.510"></a>
-<span class="sourceLineNo">511</span>  /**<a name="line.511"></a>
-<span class="sourceLineNo">512</span>   * Prepare a collection of {@link LoadQueueItem} from list of source hfiles contained in the<a name="line.512"></a>
-<span class="sourceLineNo">513</span>   * passed directory and validates whether the prepared queue has all the valid table column<a name="line.513"></a>
-<span class="sourceLineNo">514</span>   * families in it.<a name="line.514"></a>
-<span class="sourceLineNo">515</span>   * @param hfilesDir directory containing list of hfiles to be loaded into the table<a name="line.515"></a>
-<span class="sourceLineNo">516</span>   * @param table table to which hfiles should be loaded<a name="line.516"></a>
-<span class="sourceLineNo">517</span>   * @param queue queue which needs to be loaded into the table<a name="line.517"></a>
-<span class="sourceLineNo">518</span>   * @param validateHFile if true hfiles will be validated for its format<a name="line.518"></a>
-<span class="sourceLineNo">519</span>   * @throws IOException If any I/O or network error occurred<a name="line.519"></a>
-<span class="sourceLineNo">520</span>   */<a name="line.520"></a>
-<span class="sourceLineNo">521</span>  public void prepareHFileQueue(Path hfilesDir, Table table, Deque&lt;LoadQueueItem&gt; queue,<a name="line.521"></a>
-<span class="sourceLineNo">522</span>      boolean validateHFile) throws IOException {<a name="line.522"></a>
-<span class="sourceLineNo">523</span>    prepareHFileQueue(hfilesDir, table, queue, validateHFile, false);<a name="line.523"></a>
-<span class="sourceLineNo">524</span>  }<a name="line.524"></a>
-<span class="sourceLineNo">525</span><a name="line.525"></a>
-<span class="sourceLineNo">526</span>  /**<a name="line.526"></a>
-<span class="sourceLineNo">527</span>   * Prepare a collection of {@link LoadQueueItem} from list of source hfiles contained in the<a name="line.527"></a>
-<span class="sourceLineNo">528</span>   * passed directory and validates whether the prepared queue has all the valid table column<a name="line.528"></a>
-<span class="sourceLineNo">529</span>   * families in it.<a name="line.529"></a>
-<span class="sourceLineNo">530</span>   * @param hfilesDir directory containing list of hfiles to be loaded into the table<a name="line.530"></a>
-<span class="sourceLineNo">531</span>   * @param table table to which hfiles should be loaded<a name="line.531"></a>
-<span class="sourceLineNo">532</span>   * @param queue queue which needs to be loaded into the table<a name="line.532"></a>
-<span class="sourceLineNo">533</span>   * @param validateHFile if true hfiles will be validated for its format<a name="line.533"></a>
-<span class="sourceLineNo">534</span>   * @param silence  true to ignore unmatched column families<a name="line.534"></a>
-<span class="sourceLineNo">535</span>   * @throws IOException If any I/O or network error occurred<a name="line.535"></a>
-<span class="sourceLineNo">536</span>   */<a name="line.536"></a>
-<span class="sourceLineNo">537</span>  public void prepareHFileQueue(Path hfilesDir, Table table,<a name="line.537"></a>
-<span class="sourceLineNo">538</span>      Deque&lt;LoadQueueItem&gt; queue, boolean validateHFile, boolean silence) throws IOException {<a name="line.538"></a>
-<span class="sourceLineNo">539</span>    discoverLoadQueue(queue, hfilesDir, validateHFile);<a name="line.539"></a>
-<span class="sourceLineNo">540</span>    validateFamiliesInHFiles(table, queue, silence);<a name="line.540"></a>
-<span class="sourceLineNo">541</span>  }<a name="line.541"></a>
-<span class="sourceLineNo">542</span><a name="line.542"></a>
-<span class="sourceLineNo">543</span>  /**<a name="line.543"></a>
-<span class="sourceLineNo">544</span>   * Prepare a collection of {@link LoadQueueItem} from list of source hfiles contained in the<a name="line.544"></a>
-<span class="sourceLineNo">545</span>   * passed directory and validates whether the prepared queue has all the valid table column<a name="line.545"></a>
-<span class="sourceLineNo">546</span>   * families in it.<a name="line.546"></a>
-<span class="sourceLineNo">547</span>   * @param map map of family to List of hfiles<a name="line.547"></a>
-<span class="sourceLineNo">548</span>   * @param table table to which hfiles should be loaded<a name="line.548"></a>
-<span class="sourceLineNo">549</span>   * @param queue queue which needs to be loaded into the table<a name="line.549"></a>
-<span class="sourceLineNo">550</span>   * @param silence  true to ignore unmatched column families<a name="line.550"></a>
-<span class="sourceLineNo">551</span>   * @throws IOException If any I/O or network error occurred<a name="line.551"></a>
-<span class="sourceLineNo">552</span>   */<a name="line.552"></a>
-<span class="sourceLineNo">553</span>  public void prepareHFileQueue(Map&lt;byte[], List&lt;Path&gt;&gt; map, Table table,<a name="line.553"></a>
-<span class="sourceLineNo">554</span>      Deque&lt;LoadQueueItem&gt; queue, boolean silence) throws IOException {<a name="line.554"></a>
-<span class="sourceLineNo">555</span>    populateLoadQueue(queue, map);<a name="line.555"></a>
-<span class="sourceLineNo">556</span>    validateFamiliesInHFiles(table, queue, silence);<a name="line.556"></a>
-<span class="sourceLineNo">557</span>  }<a name="line.557"></a>
-<span class="sourceLineNo">558</span><a name="line.558"></a>
-<span class="sourceLineNo">559</span>  // Initialize a thread pool<a name="line.559"></a>
-<span class="sourceLineNo">560</span>  private ExecutorService createExecutorService() {<a name="line.560"></a>
-<span class="sourceLineNo">561</span>    ThreadFactoryBuilder builder = new ThreadFactoryBuilder();<a name="line.561"></a>
-<span class="sourceLineNo">562</span>    builder.setNameFormat("LoadIncrementalHFiles-%1$d");<a name="line.562"></a>
-<span class="sourceLineNo">563</span>    ExecutorService pool = new ThreadPoolExecutor(nrThreads, nrThreads, 60, TimeUnit.SECONDS,<a name="line.563"></a>
-<span class="sourceLineNo">564</span>        new LinkedBlockingQueue&lt;Runnable&gt;(), builder.build());<a name="line.564"></a>
-<span class="sourceLineNo">565</span>    ((ThreadPoolExecutor) pool).allowCoreThreadTimeOut(true);<a name="line.565"></a>
-<span class="sourceLineNo">566</span>    return pool;<a name="line.566"></a>
-<span class="sourceLineNo">567</span>  }<a name="line.567"></a>
-<span class="sourceLineNo">568</span><a name="line.568"></a>
-<span class="sourceLineNo">569</span>  /**<a name="line.569"></a>
-<span class="sourceLineNo">570</span>   * Checks whether there is any invalid family name in HFiles to be bulk loaded.<a name="line.570"></a>
-<span class="sourceLineNo">571</span>   */<a name="line.571"></a>
-<span class="sourceLineNo">572</span>  private void validateFamiliesInHFiles(Table table, Deque&lt;LoadQueueItem&gt; queue, boolean silence)<a name="line.572"></a>
-<span class="sourceLineNo">573</span>      throws IOException {<a name="line.573"></a>
-<span class="sourceLineNo">574</span>    Collection&lt;HColumnDescriptor&gt; families = table.getTableDescriptor().getFamilies();<a name="line.574"></a>
-<span class="sourceLineNo">575</span>    List&lt;String&gt; familyNames = new ArrayList&lt;&gt;(families.size());<a name="line.575"></a>
-<span class="sourceLineNo">576</span>    for (HColumnDescriptor family : families) {<a name="line.576"></a>
-<span class="sourceLineNo">577</span>      familyNames.add(family.getNameAsString());<a name="line.577"></a>
-<span class="sourceLineNo">578</span>    }<a name="line.578"></a>
-<span class="sourceLineNo">579</span>    Iterator&lt;LoadQueueItem&gt; queueIter = queue.iterator();<a name="line.579"></a>
-<span class="sourceLineNo">580</span>    while (queueIter.hasNext()) {<a name="line.580"></a>
-<span class="sourceLineNo">581</span>      LoadQueueItem lqi = queueIter.next();<a name="line.581"></a>
-<span class="sourceLineNo">582</span>      String familyNameInHFile = Bytes.toString(lqi.family);<a name="line.582"></a>
-<span class="sourceLineNo">583</span>      if (!familyNames.contains(familyNameInHFile)) {<a name="line.583"></a>
-<span class="sourceLineNo">584</span>        unmatchedFamilies.add(familyNameInHFile);<a name="line.584"></a>
-<span class="sourceLineNo">585</span>      }<a name="line.585"></a>
-<span class="sourceLineNo">586</span>    }<a name="line.586"></a>
-<span class="sourceLineNo">587</span>    if (unmatchedFamilies.size() &gt; 0) {<a name="line.587"></a>
-<span class="sourceLineNo">588</span>      String msg =<a name="line.588"></a>
-<span class="sourceLineNo">589</span>          "Unmatched family names found: unmatched family names in HFiles to be bulkloaded: "<a name="line.589"></a>
-<span class="sourceLineNo">590</span>              + unmatchedFamilies + "; valid family names of table " + table.getName() + " are: "<a name="line.590"></a>
-<span class="sourceLineNo">591</span>              + familyNames;<a name="line.591"></a>
-<span class="sourceLineNo">592</span>      LOG.error(msg);<a name="line.592"></a>
-<span class="sourceLineNo">593</span>      if (!silence) throw new IOException(msg);<a name="line.593"></a>
-<span class="sourceLineNo">594</span>    }<a name="line.594"></a>
-<span class="sourceLineNo">595</span>  }<a name="line.595"></a>
-<span class="sourceLineNo">596</span><a name="line.596"></a>
-<span class="sourceLineNo">597</span>  /**<a name="line.597"></a>
-<span class="sourceLineNo">598</span>   * Used by the replication sink to load the hfiles from the source cluster. It does the following,<a name="line.598"></a>
-<span class="sourceLineNo">599</span>   * &lt;ol&gt;<a name="line.599"></a>
-<span class="sourceLineNo">600</span>   * &lt;li&gt;LoadIncrementalHFiles#groupOrSplitPhase(Table, ExecutorService, Deque, Pair)}&lt;/li&gt;<a name="line.600"></a>
-<span class="sourceLineNo">601</span>   * &lt;li&gt;LoadIncrementalHFiles#bulkLoadPhase(Table, Connection, ExecutorService, Deque, Multimap)<a name="line.601"></a>
-<span class="sourceLineNo">602</span>   * &lt;/li&gt;<a name="line.602"></a>
-<span class="sourceLineNo">603</span>   * &lt;/ol&gt;<a name="line.603"></a>
-<span class="sourceLineNo">604</span>   * @param table Table to which these hfiles should be loaded to<a name="line.604"></a>
-<span class="sourceLineNo">605</span>   * @param conn Connection to use<a name="line.605"></a>
-<span class="sourceLineNo">606</span>   * @param queue {@link LoadQueueItem} has hfiles yet to be loaded<a name="line.606"></a>
-<span class="sourceLineNo">607</span>   * @param startEndKeys starting and ending row keys of the region<a name="line.607"></a>
-<span class="sourceLineNo">608</span>   */<a name="line.608"></a>
-<span class="sourceLineNo">609</span>  public void loadHFileQueue(final Table table, final Connection conn, Deque&lt;LoadQueueItem&gt; queue,<a name="line.609"></a>
-<span class="sourceLineNo">610</span>      Pair&lt;byte[][], byte[][]&gt; startEndKeys) throws IOException {<a name="line.610"></a>
-<span class="sourceLineNo">611</span>    loadHFileQueue(table, conn, queue, startEndKeys, false);<a name="line.611"></a>
-<span class="sourceLineNo">612</span>  }<a name="line.612"></a>
-<span class="sourceLineNo">613</span><a name="line.613"></a>
-<span class="sourceLineNo">614</span>  /**<a name="line.614"></a>
-<span class="sourceLineNo">615</span>   * Used by the replication sink to load the hfiles from the source cluster. It does the following,<a name="line.615"></a>
-<span class="sourceLineNo">616</span>   * &lt;ol&gt;<a name="line.616"></a>
-<span class="sourceLineNo">617</span>   * &lt;li&gt;LoadIncrementalHFiles#groupOrSplitPhase(Table, ExecutorService, Deque, Pair)}&lt;/li&gt;<a name="line.617"></a>
-<span class="sourceLineNo">618</span>   * &lt;li&gt;LoadIncrementalHFiles#bulkLoadPhase(Table, Connection, ExecutorService, Deque, Multimap)<a name="line.618"></a>
-<span class="sourceLineNo">619</span>   * &lt;/li&gt;<a name="line.619"></a>
-<span class="sourceLineNo">620</span>   * &lt;/ol&gt;<a name="line.620"></a>
-<span class="sourceLineNo">621</span>   * @param table Table to which these hfiles should be loaded to<a name="line.621"></a>
-<span class="sourceLineNo">622</span>   * @param conn Connection to use<a name="line.622"></a>
-<span class="sourceLineNo">623</span>   * @param queue {@link LoadQueueItem} has hfiles yet to be loaded<a name="line.623"></a>
-<span class="sourceLineNo">624</span>   * @param startEndKeys starting and ending row keys of the region<a name="line.624"></a>
-<span class="sourceLineNo">625</span>   */<a name="line.625"></a>
-<span class="sourceLineNo">626</span>  public void loadHFileQueue(final Table table, final Connection conn, Deque&lt;LoadQueueItem&gt; queue,<a name="line.626"></a>
-<span class="sourceLineNo">627</span>      Pair&lt;byte[][], byte[][]&gt; startEndKeys, boolean copyFile) throws IOException {<a name="line.627"></a>
-<span class="sourceLineNo">628</span>    ExecutorService pool = null;<a name="line.628"></a>
-<span class="sourceLineNo">629</span>    try {<a name="line.629"></a>
-<span class="sourceLineNo">630</span>      pool = createExecutorService();<a name="line.630"></a>
-<span class="sourceLineNo">631</span>      Multimap&lt;ByteBuffer, LoadQueueItem&gt; regionGroups =<a name="line.631"></a>
-<span class="sourceLineNo">632</span>          groupOrSplitPhase(table, pool, queue, startEndKeys).getFirst();<a name="line.632"></a>
-<span class="sourceLineNo">633</span>      bulkLoadPhase(table, conn, pool, queue, regionGroups, copyFile);<a name="line.633"></a>
-<span class="sourceLineNo">634</span>    } finally {<a name="line.634"></a>
-<span class="sourceLineNo">635</span>      if (pool != null) {<a name="line.635"></a>
-<span class="sourceLineNo">636</span>        pool.shutdown();<a name="line.636"></a>
-<span class="sourceLineNo">637</span>      }<a name="line.637"></a>
-<span class="sourceLineNo">638</span>    }<a name="line.638"></a>
-<span class="sourceLineNo">639</span>  }<a name="line.639"></a>
-<span class="sourceLineNo">640</span><a name="line.640"></a>
-<span class="sourceLineNo">641</span>  /**<a name="line.641"></a>
-<span class="sourceLineNo">642</span>   * This takes the LQI's grouped by likely regions and attempts to bulk load<a name="line.642"></a>
-<span class="sourceLineNo">643</span>   * them.  Any failures are re-queued for another pass with the<a name="line.643"></a>
-<span class="sourceLineNo">644</span>   * groupOrSplitPhase.<a name="line.644"></a>
-<span class="sourceLineNo">645</span>   */<a name="line.645"></a>
-<span class="sourceLineNo">646</span>  protected void bulkLoadPhase(final Table table, final Connection conn,<a name="line.646"></a>
-<span class="sourceLineNo">647</span>      ExecutorService pool, Deque&lt;LoadQueueItem&gt; queue,<a name="line.647"></a>
-<span class="sourceLineNo">648</span>      final Multimap&lt;ByteBuffer, LoadQueueItem&gt; regionGroups, boolean copyFile) throws IOException {<a name="line.648"></a>
-<span class="sourceLineNo">649</span>    // atomically bulk load the groups.<a name="line.649"></a>
-<span class="sourceLineNo">650</span>    Set&lt;Future&lt;List&lt;LoadQueueItem&gt;&gt;&gt; loadingFutures = new HashSet&lt;&gt;();<a name="line.650"></a>
-<span class="sourceLineNo">651</span>    for (Entry&lt;ByteBuffer, ? extends Collection&lt;LoadQueueItem&gt;&gt; e: regionGroups.asMap().entrySet()){<a name="line.651"></a>
-<span class="sourceLineNo">652</span>      final byte[] first = e.getKey().array();<a name="line.652"></a>
-<span class="sourceLineNo">653</span>      final Collection&lt;LoadQueueItem&gt; lqis =  e.getValue();<a name="line.653"></a>
-<span class="sourceLineNo">654</span><a name="line.654"></a>
-<span class="sourceLineNo">655</span>      final Callable&lt;List&lt;LoadQueueItem&gt;&gt; call = new Callable&lt;List&lt;LoadQueueItem&gt;&gt;() {<a name="line.655"></a>
-<span class="sourceLineNo">656</span>        @Override<a name="line.656"></a>
-<span class="sourceLineNo">657</span>        public List&lt;LoadQueueItem&gt; call() throws Exception {<a name="line.657"></a>
-<span class="sourceLineNo">658</span>          List&lt;LoadQueueItem&gt; toRetry =<a name="line.658"></a>
-<span class="sourceLineNo">659</span>              tryAtomicRegionLoad(conn, table.getName(), first, lqis, copyFile);<a name="line.659"></a>
-<span class="sourceLineNo">660</span>          return toRetry;<a name="line.660"></a>
-<span class="sourceLineNo">661</span>        }<a name="line.661"></a>
-<span class="sourceLineNo">662</span>      };<a name="line.662"></a>
-<span class="sourceLineNo">663</span>      loadingFutures.add(pool.submit(call));<a name="line.663"></a>
-<span class="sourceLineNo">664</span>    }<a name="line.664"></a>
-<span class="sourceLineNo">665</span><a name="line.665"></a>
-<span class="sourceLineNo">666</span>    // get all the results.<a name="line.666"></a>
-<span class="sourceLineNo">667</span>    for (Future&lt;List&lt;LoadQueueItem&gt;&gt; future : loadingFutures) {<a name="line.667"></a>
-<span class="sourceLineNo">668</span>      try {<a name="line.668"></a>
-<span class="sourceLineNo">669</span>        List&lt;LoadQueueItem&gt; toRetry = future.get();<a name="line.669"></a>
-<span class="sourceLineNo">670</span><a name="line.670"></a>
-<span class="sourceLineNo">671</span>        // LQIs that are requeued to be regrouped.<a name="line.671"></a>
-<span class="sourceLineNo">672</span>        queue.addAll(toRetry);<a name="line.672"></a>
-<span class="sourceLineNo">673</span><a name="line.673"></a>
-<span class="sourceLineNo">674</span>      } catch (ExecutionException e1) {<a name="line.674"></a>
-<span class="sourceLineNo">675</span>        Throwable t = e1.getCause();<a name="line.675"></a>
-<span class="sourceLineNo">676</span>        if (t instanceof IOException) {<a name="line.676"></a>
-<span class="sourceLineNo">677</span>          // At this point something unrecoverable has happened.<a name="line.677"></a>
-<span class="sourceLineNo">678</span>          // TODO Implement bulk load recovery<a name="line.678"></a>
-<span class="sourceLineNo">679</span>          throw new IOException("BulkLoad encountered an unrecoverable problem", t);<a name="line.679"></a>
-<span class="sourceLineNo">680</span>        }<a name="line.680"></a>
-<span class="sourceLineNo">681</span>        LOG.error("Unexpected execution exception during bulk load", e1);<a name="line.681"></a>
-<span class="sourceLineNo">682</span>        throw new IllegalStateException(t);<a name="line.682"></a>
-<span class="sourceLineNo">683</span>      } catch (InterruptedException e1) {<a name="line.683"></a>
-<span class="sourceLineNo">684</span>        LOG.error("Unexpected interrupted exception during bulk load", e1);<a name="line.684"></a>
-<span class="sourceLineNo">685</span>        throw (InterruptedIOException)new InterruptedIOException().initCause(e1);<a name="line.685"></a>
-<span class="sourceLineNo">686</span>      }<a name="line.686"></a>
-<span class="sourceLineNo">687</span>    }<a name="line.687"></a>
-<span class="sourceLineNo">688</span>  }<a name="line.688"></a>
-<span class="sourceLineNo">689</span><a name="line.689"></a>
-<span class="sourceLineNo">690</span>  private boolean checkHFilesCountPerRegionPerFamily(<a name="line.690"></a>
-<span class="sourceLineNo">691</span>      final Multimap&lt;ByteBuffer, LoadQueueItem&gt; regionGroups) {<a name="line.691"></a>
-<span class="sourceLineNo">692</span>    for (Entry&lt;ByteBuffer,<a name="line.692"></a>
-<span class="sourceLineNo">693</span>      ? extends Collection&lt;LoadQueueItem&gt;&gt; e: regionGroups.asMap().entrySet()) {<a name="line.693"></a>
-<span class="sourceLineNo">694</span>      final Collection&lt;LoadQueueItem&gt; lqis =  e.getValue();<a name="line.694"></a>
-<span class="sourceLineNo">695</span>      HashMap&lt;byte[], MutableInt&gt; filesMap = new HashMap&lt;&gt;();<a name="line.695"></a>
-<span class="sourceLineNo">696</span>      for (LoadQueueItem lqi: lqis) {<a name="line.696"></a>
-<span class="sourceLineNo">697</span>        MutableInt count = filesMap.get(lqi.family);<a name="line.697"></a>
-<span class="sourceLineNo">698</span>        if (count == null) {<a name="line.698"></a>
-<span class="sourceLineNo">699</span>          count = new MutableInt();<a name="line.699"></a>
-<span class="sourceLineNo">700</span>          filesMap.put(lqi.family, count);<a name="line.700"></a>
-<span class="sourceLineNo">701</span>        }<a name="line.701"></a>
-<span class="sourceLineNo">702</span>        count.increment();<a name="line.702"></a>
-<span class="sourceLineNo">703</span>        if (count.intValue() &gt; maxFilesPerRegionPerFamily) {<a name="line.703"></a>
-<span class="sourceLineNo">704</span>          LOG.error("Trying to load more than " + maxFilesPerRegionPerFamily<a name="line.704"></a>
-<span class="sourceLineNo">705</span>            + " hfiles to family " + Bytes.toStringBinary(lqi.family)<a name="line.705"></a>
-<span class="sourceLineNo">706</span>            + " of region with start key "<a name="line.706"></a>
-<span class="sourceLineNo">707</span>            + Bytes.toStringBinary(e.getKey()));<a name="line.707"></a>
-<span class="sourceLineNo">708</span>          return false;<a name="line.708"></a>
-<span class="sourceLineNo">709</span>        }<a name="line.709"></a>
-<span class="sourceLineNo">710</span>      }<a name="line.710"></a>
-<span class="sourceLineNo">711</span>    }<a name="line.711"></a>
-<span class="sourceLineNo">712</span>    return true;<a name="line.712"></a>
-<span class="sourceLineNo">713</span>  }<a name="line.713"></a>
-<span class="sourceLineNo">714</span><a name="line.714"></a>
-<span class="sourceLineNo">715</span>  /**<a name="line.715"></a>
-<span class="sourceLineNo">716</span>   * @param table the table to load into<a name="line.716"></a>
-<span class="sourceLineNo">717</span>   * @param pool the ExecutorService<a name="line.717"></a>
-<span class="sourceLineNo">718</span>   * @param queue the queue for LoadQueueItem<a name="line.718"></a>
-<span class="sourceLineNo">719</span>   * @param startEndKeys start and end keys<a name="line.719"></a>
-<span class="sourceLineNo">720</span>   * @return A map that groups LQI by likely bulk load region targets and List of missing hfiles.<a name="line.720"></a>
-<span class="sourceLineNo">721</span>   */<a name="line.721"></a>
-<span class="sourceLineNo">722</span>  private Pair&lt;Multimap&lt;ByteBuffer, LoadQueueItem&gt;, List&lt;String&gt;&gt; groupOrSplitPhase(<a name="line.722"></a>
-<span class="sourceLineNo">723</span>      final Table table, ExecutorService pool, Deque&lt;LoadQueueItem&gt; queue,<a name="line.723"></a>
-<span class="sourceLineNo">724</span>      final Pair&lt;byte[][], byte[][]&gt; startEndKeys) throws IOException {<a name="line.724"></a>
-<span class="sourceLineNo">725</span>    // &lt;region start key, LQI&gt; need synchronized only within this scope of this<a name="line.725"></a>
-<span class="sourceLineNo">726</span>    // phase because of the puts that happen in futures.<a name="line.726"></a>
-<span class="sourceLineNo">727</span>    Multimap&lt;ByteBuffer, LoadQueueItem&gt; rgs = HashMultimap.create();<a name="line.727"></a>
-<span class="sourceLineNo">728</span>    final Multimap&lt;ByteBuffer, LoadQueueItem&gt; regionGroups = Multimaps.synchronizedMultimap(rgs);<a name="line.728"></a>
-<span class="sourceLineNo">729</span>    List&lt;String&gt; missingHFiles = new ArrayList&lt;&gt;();<a name="line.729"></a>
-<span class="sourceLineNo">730</span>    Pair&lt;Multimap&lt;ByteBuffer, LoadQueueItem&gt;, List&lt;String&gt;&gt; pair = new Pair&lt;&gt;(regionGroups,<a name="line.730"></a>
-<span class="sourceLineNo">731</span>        missingHFiles);<a name="line.731"></a>
-<span class="sourceLineNo">732</span><a name="line.732"></a>
-<span class="sourceLineNo">733</span>    // drain LQIs and figure out bulk load groups<a name="line.733"></a>
-<span class="sourceLineNo">734</span>    Set&lt;Future&lt;Pair&lt;List&lt;LoadQueueItem&gt;, String&gt;&gt;&gt; splittingFutures = new HashSet&lt;&gt;();<a name="line.734"></a>
-<span class="sourceLineNo">735</span>    while (!queue.isEmpty()) {<a name="line.735"></a>
-<span class="sourceLineNo">736</span>      final LoadQueueItem item = queue.remove();<a name="line.736"></a>
-<span class="sourceLineNo">737</span><a name="line.737"></a>
-<span class="sourceLineNo">738</span>      final Callable&lt;Pair&lt;List&lt;LoadQueueItem&gt;, String&gt;&gt; call =<a name="line.738"></a>
-<span class="sourceLineNo">739</span>          new Callable&lt;Pair&lt;List&lt;LoadQueueItem&gt;, String&gt;&gt;() {<a name="line.739"></a>
-<span class="sourceLineNo">740</span>        @Override<a name="line.740"></a>
-<span class="sourceLineNo">741</span>        public Pair&lt;List&lt;LoadQueueItem&gt;, String&gt; call() throws Exception {<a name="line.741"></a>
-<span class="sourceLineNo">742</span>          Pair&lt;List&lt;LoadQueueItem&gt;, String&gt; splits = groupOrSplit(regionGroups, item, table,<a name="line.742"></a>
-<span class="sourceLineNo">743</span>              startEndKeys);<a name="line.743"></a>
-<span class="sourceLineNo">744</span>          return splits;<a name="line.744"></a>
-<span class="sourceLineNo">745</span>        }<a name="line.745"></a>
-<span class="sourceLineNo">746</span>      };<a name="line.746"></a>
-<span class="sourceLineNo">747</span>      splittingFutures.add(pool.submit(call));<a name="line.747"></a>
-<span class="sourceLineNo">748</span>    }<a name="line.748"></a>
-<span class="sourceLineNo">749</span>    // get all the results.  All grouping and splitting must finish before<a name="line.749"></a>
-<span class="sourceLineNo">750</span>    // we can attempt the atomic loads.<a name="line.750"></a>
-<span class="sourceLineNo">751</span>    for (Future&lt;Pair&lt;List&lt;LoadQueueItem&gt;, String&gt;&gt; lqis : splittingFutures) {<a name="line.751"></a>
-<span class="sourceLineNo">752</span>      try {<a name="line.752"></a>
-<span class="sourceLineNo">753</span>        Pair&lt;List&lt;LoadQueueItem&gt;, String&gt; splits = lqis.get();<a name="line.753"></a>
-<span class="sourceLineNo">754</span>        if (splits != null) {<a name="line.754"></a>
-<span class="sourceLineNo">755</span>          if (splits.getFirst() != null) {<a name="line.755"></a>
-<span class="sourceLineNo">756</span>            queue.addAll(splits.getFirst());<a name="line.756"></a>
-<span class="sourceLineNo">757</span>          } else {<a name="line.757"></a>
-<span class="sourceLineNo">758</span>            missingHFiles.add(splits.getSecond());<a name="line.758"></a>
-<span class="sourceLineNo">759</span>          }<a name="line.759"></a>
-<span class="sourceLineNo">760</span>        }<a name="line.760"></a>
-<span class="sourceLineNo">761</span>      } catch (ExecutionException e1) {<a name="line.761"></a>
-<span class="sourceLineNo">762</span>        Throwable t = e1.getCause();<a name="line.762"></a>
-<span class="sourceLineNo">763</span>        if (t instanceof IOException) {<a name="line.763"></a>
-<span class="sourceLineNo">764</span>          LOG.error("IOException during splitting", e1);<a name="line.764"></a>
-<span class="sourceLineNo">765</span>          throw (IOException)t; // would have been thrown if not parallelized,<a name="line.765"></a>
-<span class="sourceLineNo">766</span>        }<a name="line.766"></a>
-<span class="sourceLineNo">767</span>        LOG.error("Unexpected execution exception during splitting", e1);<a name="line.767"></a>
-<span class="sourceLineNo">768</span>        throw new IllegalStateException(t);<a name="line.768"></a>
-<span class="sourceLineNo">769</span>      } catch (InterruptedException e1) {<a name="line.769"></a>
-<span class="sourceLineNo">770</span>        LOG.error("Unexpected interrupted exception during splitting", e1);<a name="line.770"></a>
-<span class="sourceLineNo">771</span>        throw (InterruptedIOException)new InterruptedIOException().initCause(e1);<a name="line.771"></a>
-<span class="sourceLineNo">772</span>      }<a name="line.772"></a>
-<span class="sourceLineNo">773</span>    }<a name="line.773"></a>
-<span class="sourceLineNo">774</span>    return pair;<a name="line.774"></a>
-<span class="sourceLineNo">775</span>  }<a name="line.775"></a>
-<span class="sourceLineNo">776</span><a name="line.776"></a>
-<span class="sourceLineNo">777</span>  // unique file name for the table<a name="line.777"></a>
-<span class="sourceLineNo">778</span>  private String getUniqueName() {<a name="line.778"></a>
-<span class="sourceLineNo">779</span>    return UUID.randomUUID().toString().replaceAll("-", "");<a name="line.779"></a>
-<span class="sourceLineNo">780</span>  }<a name="line.780"></a>
-<span class="sourceLineNo">781</span><a name="line.781"></a>
-<span class="sourceLineNo">782</span>  protected List&lt;LoadQueueItem&gt; splitStoreFile(final LoadQueueItem item,<a name="line.782"></a>
-<span class="sourceLineNo">783</span>      final Table table, byte[] startKey,<a name="line.783"></a>
-<span class="sourceLineNo">784</span>      byte[] splitKey) throws IOException {<a name="line.784"></a>
-<span class="sourceLineNo">785</span>    final Path hfilePath = item.hfilePath;<a name="line.785"></a>
-<span class="sourceLineNo">786</span><a name="line.786"></a>
-<span class="sourceLineNo">787</span>    Path tmpDir = item.hfilePath.getParent();<a name="line.787"></a>
-<span class="sourceLineNo">788</span>    if (!tmpDir.getName().equals(TMP_DIR)) {<a name="line.788"></a>
-<span class="sourceLineNo">789</span>      tmpDir = new Path(tmpDir, TMP_DIR);<a name="line.789"></a>
-<span class="sourceLineNo">790</span>    }<a name="line.790"></a>
-<span class="sourceLineNo">791</span><a name="line.791"></a>
-<span class="sourceLineNo">792</span>    LOG.info("HFile at " + hfilePath + " no longer fits inside a single " +<a name="line.792"></a>
-<span class="sourceLineNo">793</span>      "region. Splitting...");<a name="line.793"></a>
-<span class="sourceLineNo">794</span><a name="line.794"></a>
-<span class="sourceLineNo">795</span>    String uniqueName = getUniqueName();<a name="line.795"></a>
-<span class="sourceLineNo">796</span>    HColumnDescriptor familyDesc = table.getTableDescriptor().getFamily(item.family);<a name="line.796"></a>
-<span class="sourceLineNo">797</span><a name="line.797"></a>
-<span class="sourceLineNo">798</span>    Path botOut = new Path(tmpDir, uniqueName + ".bottom");<a name="line.798"></a>
-<span class="sourceLineNo">799</span>    Path topOut = new Path(tmpDir, uniqueName + ".top");<a name="line.799"></a>
-<span class="sourceLineNo">800</span>    splitStoreFile(getConf(), hfilePath, familyDesc, splitKey, botOut, topOut);<a name="line.800"></a>
-<span class="sourceLineNo">801</span><a name="line.801"></a>
-<span class="sourceLineNo">802</span>    FileSystem fs = tmpDir.getFileSystem(getConf());<a name="line.802"></a>
-<span class="sourceLineNo">803</span>    fs.setPermission(tmpDir, FsPermission.valueOf("-rwxrwxrwx"));<a name="line.803"></a>
-<span class="sourceLineNo">804</span>    fs.setPermission(botOut, FsPermission.valueOf("-rwxrwxrwx"));<a name="line.804"></a>
-<span class="sourceLineNo">805</span>    fs.setPermission(topOut, FsPermission.valueOf("-rwxrwxrwx"));<a name="line.805"></a>
+<span class="sourceLineNo">469</span>    Map&lt;LoadQueueItem, ByteBuffer&gt; item2RegionMap = new HashMap&lt;&gt;();<a name="line.469"></a>
+<span class="sourceLineNo">470</span>    // Assumes that region splits can happen while this occurs.<a name="line.470"></a>
+<span class="sourceLineNo">471</span>    while (!queue.isEmpty()) {<a name="line.471"></a>
+<span class="sourceLineNo">472</span>      // need to reload split keys each iteration.<a name="line.472"></a>
+<span class="sourceLineNo">473</span>      final Pair&lt;byte[][], byte[][]&gt; startEndKeys = regionLocator.getStartEndKeys();<a name="line.473"></a>
+<span class="sourceLineNo">474</span>      if (count != 0) {<a name="line.474"></a>
+<span class="sourceLineNo">475</span>        LOG.info("Split occured while grouping HFiles, retry attempt " +<a name="line.475"></a>
+<span class="sourceLineNo">476</span>            + count + " with " + queue.size() + " files remaining to group or split");<a name="line.476"></a>
+<span class="sourceLineNo">477</span>      }<a name="line.477"></a>
+<span class="sourceLineNo">478</span><a name="line.478"></a>
+<span class="sourceLineNo">479</span>      int maxRetries = getConf().getInt(HConstants.BULKLOAD_MAX_RETRIES_NUMBER, 10);<a name="line.479"></a>
+<span class="sourceLineNo">480</span>      maxRetries = Math.max(maxRetries, startEndKeys.getFirst().length + 1);<a name="line.480"></a>
+<span class="sourceLineNo">481</span>      if (maxRetries != 0 &amp;&amp; count &gt;= maxRetries) {<a name="line.481"></a>
+<span class="sourceLineNo">482</span>        throw new IOException("Retry attempted " + count +<a name="line.482"></a>
+<span class="sourceLineNo">483</span>            " times without completing, bailing out");<a name="line.483"></a>
+<span class="sourceLineNo">484</span>      }<a name="line.484"></a>
+<span class="sourceLineNo">485</span>      count++;<a name="line.485"></a>
+<span class="sourceLineNo">486</span><a name="line.486"></a>
+<span class="sourceLineNo">487</span>      // Using ByteBuffer for byte[] equality semantics<a name="line.487"></a>
+<span class="sourceLineNo">488</span>      pair = groupOrSplitPhase(table, pool, queue, startEndKeys);<a name="line.488"></a>
+<span class="sourceLineNo">489</span>      Multimap&lt;ByteBuffer, LoadQueueItem&gt; regionGroups = pair.getFirst();<a name="line.489"></a>
+<span class="sourceLineNo">490</span><a name="line.490"></a>
+<span class="sourceLineNo">491</span>      if (!checkHFilesCountPerRegionPerFamily(regionGroups)) {<a name="line.491"></a>
+<span class="sourceLineNo">492</span>        // Error is logged inside checkHFilesCountPerRegionPerFamily.<a name="line.492"></a>
+<span class="sourceLineNo">493</span>        throw new IOException("Trying to load more than " + maxFilesPerRegionPerFamily<a name="line.493"></a>
+<span class="sourceLineNo">494</span>            + " hfiles to one family of one region");<a name="line.494"></a>
+<span class="sourceLineNo">495</span>      }<a name="line.495"></a>
+<span class="sourceLineNo">496</span><a name="line.496"></a>
+<span class="sourceLineNo">497</span>      bulkLoadPhase(table, admin.getConnection(), pool, queue, regionGroups, copyFile,<a name="line.497"></a>
+<span class="sourceLineNo">498</span>          item2RegionMap);<a name="line.498"></a>
+<span class="sourceLineNo">499</span><a name="line.499"></a>
+<span class="sourceLineNo">500</span>      // NOTE: The next iteration's split / group could happen in parallel to<a name="line.500"></a>
+<span class="sourceLineNo">501</span>      // atomic bulkloads assuming that there are splits and no merges, and<a name="line.501"></a>
+<span class="sourceLineNo">502</span>      // that we can atomically pull out the groups we want to retry.<a name="line.502"></a>
+<span class="sourceLineNo">503</span>    }<a name="line.503"></a>
+<span class="sourceLineNo">504</span><a name="line.504"></a>
+<span class="sourceLineNo">505</span>    if (!queue.isEmpty()) {<a name="line.505"></a>
+<span class="sourceLineNo">506</span>      throw new RuntimeException("Bulk load aborted with some files not yet loaded."<a name="line.506"></a>
+<span class="sourceLineNo">507</span>        + "Please check log for more details.");<a name="line.507"></a>
+<span class="sourceLineNo">508</span>    }<a name="line.508"></a>
+<span class="sourceLineNo">509</span>    return item2RegionMap;<a name="line.509"></a>
+<span class="sourceLineNo">510</span>  }<a name="line.510"></a>
+<span class="sourceLineNo">511</span><a name="line.511"></a>
+<span class="sourceLineNo">512</span>  /**<a name="line.512"></a>
+<span class="sourceLineNo">513</span>   * Prepare a collection of {@link LoadQueueItem} from list of source hfiles contained in the<a name="line.513"></a>
+<span class="sourceLineNo">514</span>   * passed directory and validates whether the prepared queue has all the valid table column<a name="line.514"></a>
+<span class="sourceLineNo">515</span>   * families in it.<a name="line.515"></a>
+<span class="sourceLineNo">516</span>   * @param hfilesDir directory containing list of hfiles to be loaded into the table<a name="line.516"></a>
+<span class="sourceLineNo">517</span>   * @param table table to which hfiles should be loaded<a name="line.517"></a>
+<span class="sourceLineNo">518</span>   * @param queue queue which needs to be loaded into the table<a name="line.518"></a>
+<span class="sourceLineNo">519</span>   * @param validateHFile if true hfiles will be validated for its format<a name="line.519"></a>
+<span class="sourceLineNo">520</span>   * @throws IOException If any I/O or network error occurred<a name="line.520"></a>
+<span class="sourceLineNo">521</span>   */<a name="line.521"></a>
+<span class="sourceLineNo">522</span>  public void prepareHFileQueue(Path hfilesDir, Table table, Deque&lt;LoadQueueItem&gt; queue,<a name="line.522"></a>
+<span class="sourceLineNo">523</span>      boolean validateHFile) throws IOException {<a name="line.523"></a>
+<span class="sourceLineNo">524</span>    prepareHFileQueue(hfilesDir, table, queue, validateHFile, false);<a name="line.524"></a>
+<span class="sourceLineNo">525</span>  }<a name="line.525"></a>
+<span class="sourceLineNo">526</span><a name="line.526"></a>
+<span class="sourceLineNo">527</span>  /**<a name="line.527"></a>
+<span class="sourceLineNo">528</span>   * Prepare a collection of {@link LoadQueueItem} from list of source hfiles contained in the<a name="line.528"></a>
+<span class="sourceLineNo">529</span>   * passed directory and validates whether the prepared queue has all the valid table column<a name="line.529"></a>
+<span class="sourceLineNo">530</span>   * families in it.<a name="line.530"></a>
+<span class="sourceLineNo">531</span>   * @param hfilesDir directory containing list of hfiles to be loaded into the table<a name="line.531"></a>
+<span class="sourceLineNo">532</span>   * @param table table to which hfiles should be loaded<a name="line.532"></a>
+<span class="sourceLineNo">533</span>   * @param queue queue which needs to be loaded into the table<a name="line.533"></a>
+<span class="sourceLineNo">534</span>   * @param validateHFile if true hfiles will be validated for its format<a name="line.534"></a>
+<span class="sourceLineNo">535</span>   * @param silence  true to ignore unmatched column families<a name="line.535"></a>
+<span class="sourceLineNo">536</span>   * @throws IOException If any I/O or network error occurred<a name="line.536"></a>
+<span class="sourceLineNo">537</span>   */<a name="line.537"></a>
+<span class="sourceLineNo">538</span>  public void prepareHFileQueue(Path hfilesDir, Table table,<a name="line.538"></a>
+<span class="sourceLineNo">539</span>      Deque&lt;LoadQueueItem&gt; queue, boolean validateHFile, boolean silence) throws IOException {<a name="line.539"></a>
+<span class="sourceLineNo">540</span>    discoverLoadQueue(queue, hfilesDir, validateHFile);<a name="line.540"></a>
+<span class="sourceLineNo">541</span>    validateFamiliesInHFiles(table, queue, silence);<a name="line.541"></a>
+<span class="sourceLineNo">542</span>  }<a name="line.542"></a>
+<span class="sourceLineNo">543</span><a name="line.543"></a>
+<span class="sourceLineNo">544</span>  /**<a name="line.544"></a>
+<span class="sourceLineNo">545</span>   * Prepare a collection of {@link LoadQueueItem} from list of source hfiles contained in the<a name="line.545"></a>
+<span class="sourceLineNo">546</span>   * passed directory and validates whether the prepared queue has all the valid table column<a name="line.546"></a>
+<span class="sourceLineNo">547</span>   * families in it.<a name="line.547"></a>
+<span class="sourceLineNo">548</span>   * @param map map of family to List of hfiles<a name="line.548"></a>
+<span class="sourceLineNo">549</span>   * @param table table to which hfiles should be loaded<a name="line.549"></a>
+<span class="sourceLineNo">550</span>   * @param queue queue which needs to be loaded into the table<a name="line.550"></a>
+<span class="sourceLineNo">551</span>   * @param silence  true to ignore unmatched column families<a name="line.551"></a>
+<span class="sourceLineNo">552</span>   * @throws IOException If any I/O or network error occurred<a name="line.552"></a>
+<span class="sourceLineNo">553</span>   */<a name="line.553"></a>
+<span class="sourceLineNo">554</span>  public void prepareHFileQueue(Map&lt;byte[], List&lt;Path&gt;&gt; map, Table table,<a name="line.554"></a>
+<span class="sourceLineNo">555</span>      Deque&lt;LoadQueueItem&gt; queue, boolean silence) throws IOException {<a name="line.555"></a>
+<span class="sourceLineNo">556</span>    populateLoadQueue(queue, map);<a name="line.556"></a>
+<span class="sourceLineNo">557</span>    validateFamiliesInHFiles(table, queue, silence);<a name="line.557"></a>
+<span class="sourceLineNo">558</span>  }<a name="line.558"></a>
+<span class="sourceLineNo">559</span><a name="line.559"></a>
+<span class="sourceLineNo">560</span>  // Initialize a thread pool<a name="line.560"></a>
+<span class="sourceLineNo">561</span>  private ExecutorService createExecutorService() {<a name="line.561"></a>
+<span class="sourceLineNo">562</span>    ThreadFactoryBuilder builder = new ThreadFactoryBuilder();<a name="line.562"></a>
+<span class="sourceLineNo">563</span>    builder.setNameFormat("LoadIncrementalHFiles-%1$d");<a name="line.563"></a>
+<span class="sourceLineNo">564</span>    ExecutorService pool = new ThreadPoolExecutor(nrThreads, nrThreads, 60, TimeUnit.SECONDS,<a name="line.564"></a>
+<span class="sourceLineNo">565</span>        new LinkedBlockingQueue&lt;Runnable&gt;(), builder.build());<a name="line.565"></a>
+<span class="sourceLineNo">566</span>    ((ThreadPoolExecutor) pool).allowCoreThreadTimeOut(true);<a name="line.566"></a>
+<span class="sourceLineNo">567</span>    return pool;<a name="line.567"></a>
+<span class="sourceLineNo">568</span>  }<a name="line.568"></a>
+<span class="sourceLineNo">569</span><a name="line.569"></a>
+<span class="sourceLineNo">570</span>  /**<a name="line.570"></a>
+<span class="sourceLineNo">571</span>   * Checks whether there is any invalid family name in HFiles to be bulk loaded.<a name="line.571"></a>
+<span class="sourceLineNo">572</span>   */<a name="line.572"></a>
+<span class="sourceLineNo">573</span>  private void validateFamiliesInHFiles(Table table, Deque&lt;LoadQueueItem&gt; queue, boolean silence)<a name="line.573"></a>
+<span class="sourceLineNo">574</span>      throws IOException {<a name="line.574"></a>
+<span class="sourceLineNo">575</span>    Collection&lt;HColumnDescriptor&gt; families = table.getTableDescriptor().getFamilies();<a name="line.575"></a>
+<span class="sourceLineNo">576</span>    List&lt;String&gt; familyNames = new ArrayList&lt;&gt;(families.size());<a name="line.576"></a>
+<span class="sourceLineNo">577</span>    for (HColumnDescriptor family : families) {<a name="line.577"></a>
+<span class="sourceLineNo">578</span>      familyNames.add(family.getNameAsString());<a name="line.578"></a>
+<span class="sourceLineNo">579</span>    }<a name="line.579"></a>
+<span class="sourceLineNo">580</span>    Iterator&lt;LoadQueueItem&gt; queueIter = queue.iterator();<a name="line.580"></a>
+<span class="sourceLineNo">581</span>    while (queueIter.hasNext()) {<a name="line.581"></a>
+<span class="sourceLineNo">582</span>      LoadQueueItem lqi = queueIter.next();<a name="line.582"></a>
+<span class="sourceLineNo">583</span>      String familyNameInHFile = Bytes.toString(lqi.family);<a name="line.583"></a>
+<span class="sourceLineNo">584</span>      if (!familyNames.contains(familyNameInHFile)) {<a name="line.584"></a>
+<span class="sourceLineNo">585</span>        unmatchedFamilies.add(familyNameInHFile);<a name="line.585"></a>
+<span class="sourceLineNo">586</span>      }<a name="line.586"></a>
+<span class="sourceLineNo">587</span>    }<a name="line.587"></a>
+<span class="sourceLineNo">588</span>    if (unmatchedFamilies.size() &gt; 0) {<a name="line.588"></a>
+<span class="sourceLineNo">589</span>      String msg =<a name="line.589"></a>
+<span class="sourceLineNo">590</span>          "Unmatched family names found: unmatched family names in HFiles to be bulkloaded: "<a name="line.590"></a>
+<span class="sourceLineNo">591</span>              + unmatchedFamilies + "; valid family names of table " + table.getName() + " are: "<a name="line.591"></a>
+<span class="sourceLineNo">592</span>              + familyNames;<a name="line.592"></a>
+<span class="sourceLineNo">593</span>      LOG.error(msg);<a name="line.593"></a>
+<span class="sourceLineNo">594</span>      if (!silence) throw new IOException(msg);<a name="line.594"></a>
+<span class="sourceLineNo">595</span>    }<a name="line.595"></a>
+<span class="sourceLineNo">596</span>  }<a name="line.596"></a>
+<span class="sourceLineNo">597</span><a name="line.597"></a>
+<span class="sourceLineNo">598</span>  /**<a name="line.598"></a>
+<span class="sourceLineNo">599</span>   * Used by the replication sink to load the hfiles from the source cluster. It does the following,<a name="line.599"></a>
+<span class="sourceLineNo">600</span>   * &lt;ol&gt;<a name="line.600"></a>
+<span class="sourceLineNo">601</span>   * &lt;li&gt;LoadIncrementalHFiles#groupOrSplitPhase(Table, ExecutorService, Deque, Pair)}&lt;/li&gt;<a name="line.601"></a>
+<span class="sourceLineNo">602</span>   * &lt;li&gt;LoadIncrementalHFiles#bulkLoadPhase(Table, Connection, ExecutorService, Deque, Multimap)<a name="line.602"></a>
+<span class="sourceLineNo">603</span>   * &lt;/li&gt;<a name="line.603"></a>
+<span class="sourceLineNo">604</span>   * &lt;/ol&gt;<a name="line.604"></a>
+<span class="sourceLineNo">605</span>   * @param table Table to which these hfiles should be loaded to<a name="line.605"></a>
+<span class="sourceLineNo">606</span>   * @param conn Connection to use<a name="line.606"></a>
+<span class="sourceLineNo">607</span>   * @param queue {@link LoadQueueItem} has hfiles yet to be loaded<a name="line.607"></a>
+<span class="sourceLineNo">608</span>   * @param startEndKeys starting and ending row keys of the region<a name="line.608"></a>
+<span class="sourceLineNo">609</span>   */<a name="line.609"></a>
+<span class="sourceLineNo">610</span>  public void loadHFileQueue(final Table table, final Connection conn, Deque&lt;LoadQueueItem&gt; queue,<a name="line.610"></a>
+<span class="sourceLineNo">611</span>      Pair&lt;byte[][], byte[][]&gt; startEndKeys) throws IOException {<a name="line.611"></a>
+<span class="sourceLineNo">612</span>    loadHFileQueue(table, conn, queue, startEndKeys, false);<a name="line.612"></a>
+<span class="sourceLineNo">613</span>  }<a name="line.613"></a>
+<span class="sourceLineNo">614</span><a name="line.614"></a>
+<span class="sourceLineNo">615</span>  /**<a name="line.615"></a>
+<span class="sourceLineNo">616</span>   * Used by the replication sink to load the hfiles from the source cluster. It does the following,<a name="line.616"></a>
+<span class="sourceLineNo">617</span>   * &lt;ol&gt;<a name="line.617"></a>
+<span class="sourceLineNo">618</span>   * &lt;li&gt;LoadIncrementalHFiles#groupOrSplitPhase(Table, ExecutorService, Deque, Pair)}&lt;/li&gt;<a name="line.618"></a>
+<span class="sourceLineNo">619</span>   * &lt;li&gt;LoadIncrementalHFiles#bulkLoadPhase(Table, Connection, ExecutorService, Deque, Multimap)<a name="line.619"></a>
+<span class="sourceLineNo">620</span>   * &lt;/li&gt;<a name="line.620"></a>
+<span class="sourceLineNo">621</span>   * &lt;/ol&gt;<a name="line.621"></a>
+<span class="sourceLineNo">622</span>   * @param table Table to which these hfiles should be loaded to<a name="line.622"></a>
+<span class="sourceLineNo">623</span>   * @param conn Connection to use<a name="line.623"></a>
+<span class="sourceLineNo">624</span>   * @param queue {@link LoadQueueItem} has hfiles yet to be loaded<a name="line.624"></a>
+<span class="sourceLineNo">625</span>   * @param startEndKeys starting and ending row keys of the region<a name="line.625"></a>
+<span class="sourceLineNo">626</span>   */<a name="line.626"></a>
+<span class="sourceLineNo">627</span>  public void loadHFileQueue(final Table table, final Connection conn, Deque&lt;LoadQueueItem&gt; queue,<a name="line.627"></a>
+<span class="sourceLineNo">628</span>      Pair&lt;byte[][], byte[][]&gt; startEndKeys, boolean copyFile) throws IOException {<a name="line.628"></a>
+<span class="sourceLineNo">629</span>    ExecutorService pool = null;<a name="line.629"></a>
+<span class="sourceLineNo">630</span>    try {<a name="line.630"></a>
+<span class="sourceLineNo">631</span>      pool = createExecutorService();<a name="line.631"></a>
+<span class="sourceLineNo">632</span>      Multimap&lt;ByteBuffer, LoadQueueItem&gt; regionGroups =<a name="line.632"></a>
+<span class="sourceLineNo">633</span>          groupOrSplitPhase(table, pool, queue, startEndKeys).getFirst();<a name="line.633"></a>
+<span class="sourceLineNo">634</span>      bulkLoadPhase(table, conn, pool, queue, regionGroups, copyFile, null);<a name="line.634"></a>
+<span class="sourceLineNo">635</span>    } finally {<a name="line.635"></a>
+<span class="sourceLineNo">636</span>      if (pool != null) {<a name="line.636"></a>
+<span class="sourceLineNo">637</span>        pool.shutdown();<a name="line.637"></a>
+<span class="sourceLineNo">638</span>      }<a name="line.638"></a>
+<span class="sourceLineNo">639</span>    }<a name="line.639"></a>
+<span class="sourceLineNo">640</span>  }<a name="line.640"></a>
+<span class="sourceLineNo">641</span><a name="line.641"></a>
+<span class="sourceLineNo">642</span>  /**<a name="line.642"></a>
+<span class="sourceLineNo">643</span>   * This takes the LQI's grouped by likely regions and attempts to bulk load<a name="line.643"></a>
+<span class="sourceLineNo">644</span>   * them.  Any failures are re-queued for another pass with the<a name="line.644"></a>
+<span class="sourceLineNo">645</span>   * groupOrSplitPhase.<a name="line.645"></a>
+<span class="sourceLineNo">646</span>   */<a name="line.646"></a>
+<span class="sourceLineNo">647</span>  protected void bulkLoadPhase(final Table table, final Connection conn,<a name="line.647"></a>
+<span class="sourceLineNo">648</span>      ExecutorService pool, Deque&lt;LoadQueueItem&gt; queue,<a name="line.648"></a>
+<span class="sourceLineNo">649</span>      final Multimap&lt;ByteBuffer, LoadQueueItem&gt; regionGroups, boolean copyFile,<a name="line.649"></a>
+<span class="sourceLineNo">650</span>      Map&lt;LoadQueueItem, ByteBuffer&gt; item2RegionMap) throws IOException {<a name="line.650"></a>
+<span class="sourceLineNo">651</span>    // atomically bulk load the groups.<a name="line.651"></a>
+<span class="sourceLineNo">652</span>    Set&lt;Future&lt;List&lt;LoadQueueItem&gt;&gt;&gt; loadingFutures = new HashSet&lt;&gt;();<a name="line.652"></a>
+<span class="sourceLineNo">653</span>    for (Entry&lt;ByteBuffer, ? extends Collection&lt;LoadQueueItem&gt;&gt; e: regionGroups.asMap().entrySet()){<a name="line.653"></a>
+<span class="sourceLineNo">654</span>      final byte[] first = e.getKey().array();<a name="line.654"></a>
+<span class="sourceLineNo">655</span>      final Collection&lt;LoadQueueItem&gt; lqis =  e.getValue();<a name="line.655"></a>
+<span class="sourceLineNo">656</span><a name="line.656"></a>
+<span class="sourceLineNo">657</span>      final Callable&lt;List&lt;LoadQueueItem&gt;&gt; call = new Callable&lt;List&lt;LoadQueueItem&gt;&gt;() {<a name="line.657"></a>
+<span class="sourceLineNo">658</span>        @Override<a name="line.658"></a>
+<span class="sourceLineNo">659</span>        public List&lt;LoadQueueItem&gt; call() throws Exception {<a name="line.659"></a>
+<span class="sourceLineNo">660</span>          List&lt;LoadQueueItem&gt; toRetry =<a name="line.660"></a>
+<span class="sourceLineNo">661</span>              tryAtomicRegionLoad(conn, table.getName(), first, lqis, copyFile);<a name="line.661"></a>
+<span class="sourceLineNo">662</span>          return toRetry;<a name="line.662"></a>
+<span class="sourceLineNo">663</span>        }<a name="line.663"></a>
+<span class="sourceLineNo">664</span>      };<a name="line.664"></a>
+<span class="sourceLineNo">665</span>      if (item2RegionMap != null) {<a name="line.665"></a>
+<span class="sourceLineNo">666</span>        for (LoadQueueItem lqi : lqis) {<a name="line.666"></a>
+<span class="sourceLineNo">667</span>          item2RegionMap.put(lqi, e.getKey());<a name="line.667"></a>
+<span class="sourceLineNo">668</span>        }<a name="line.668"></a>
+<span class="sourceLineNo">669</span>      }<a name="line.669"></a>
+<span class="sourceLineNo">670</span>      loadingFutures.add(pool.submit(call));<a name="line.670"></a>
+<span class="sourceLineNo">671</span>    }<a name="line.671"></a>
+<span class="sourceLineNo">672</span><a name="line.672"></a>
+<span class="sourceLineNo">673</span>    // get all the results.<a name="line.673"></a>
+<span class="sourceLineNo">674</span>    for (Future&lt;List&lt;LoadQueueItem&gt;&gt; future : loadingFutures) {<a name="line.674"></a>
+<span class="sourceLineNo">675</span>      try {<a name="line.675"></a>
+<span class="sourceLineNo">676</span>        List&lt;LoadQueueItem&gt; toRetry = future.get();<a name="line.676"></a>
+<span class="sourceLineNo">677</span><a name="line.677"></a>
+<span class="sourceLineNo">678</span>        if (item2RegionMap != null) {<a name="line.678"></a>
+<span class="sourceLineNo">679</span>          for (LoadQueueItem lqi : toRetry) {<a name="line.679"></a>
+<span class="sourceLineNo">680</span>            item2RegionMap.remove(lqi);<a name="line.680"></a>
+<span class="sourceLineNo">681</span>          }<a name="line.681"></a>
+<span class="sourceLineNo">682</span>        }<a name="line.682"></a>
+<span class="sourceLineNo">683</span>        // LQIs that are requeued to be regrouped.<a name="line.683"></a>
+<span class="sourceLineNo">684</span>        queue.addAll(toRetry);<a name="line.684"></a>
+<span class="sourceLineNo">685</span><a name="line.685"></a>
+<span class="sourceLineNo">686</span>      } catch (ExecutionException e1) {<a name="line.686"></a>
+<span class="sourceLineNo">687</span>        Throwable t = e1.getCause();<a name="line.687"></a>
+<span class="sourceLineNo">688</span>        if (t instanceof IOException) {<a name="line.688"></a>
+<span class="sourceLineNo">689</span>          // At this point something unrecoverable has happened.<a name="line.689"></a>
+<span class="sourceLineNo">690</span>          // TODO Implement bulk load recovery<a name="line.690"></a>
+<span class="sourceLineNo">691</span>          throw new IOException("BulkLoad encountered an unrecoverable problem", t);<a name="line.691"></a>
+<span class="sourceLineNo">692</span>        }<a name="line.692"></a>
+<span class="sourceLineNo">693</span>        LOG.error("Unexpected execution exception during bulk load", e1);<a name="line.693"></a>
+<span class="sourceLineNo">694</span>        throw new IllegalStateException(t);<a name="line.694"></a>
+<span class="sourceLineNo">695</span>      } catch (InterruptedException e1) {<a name="line.695"></a>
+<span class="sourceLineNo">696</span>        LOG.error("Unexpected interrupted exception during bulk load", e1);<a name="line.696"></a>
+<span class="sourceLineNo">697</span>        throw (InterruptedIOException)new InterruptedIOException().initCause(e1);<a name="line.697"></a>
+<span class="sourceLineNo">698</span>      }<a name="line.698"></a>
+<span class="sourceLineNo">699</span>    }<a name="line.699"></a>
+<span class="sourceLineNo">700</span>  }<a name="line.700"></a>
+<span class="sourceLineNo">701</span><a name="line.701"></a>
+<span class="sourceLineNo">702</span>  private boolean checkHFilesCountPerRegionPerFamily(<a name="line.702"></a>
+<span class="sourceLineNo">703</span>      final Multimap&lt;ByteBuffer, LoadQueueItem&gt; regionGroups) {<a name="line.703"></a>
+<span class="sourceLineNo">704</span>    for (Entry&lt;ByteBuffer,<a name="line.704"></a>
+<span class="sourceLineNo">705</span>      ? extends Collection&lt;LoadQueueItem&gt;&gt; e: regionGroups.asMap().entrySet()) {<a name="line.705"></a>
+<span class="sourceLineNo">706</span>      final Collection&lt;LoadQueueItem&gt; lqis =  e.getValue();<a name="line.706"></a>
+<span class="sourceLineNo">707</span>      HashMap&lt;byte[], MutableInt&gt; filesMap = new HashMap&lt;&gt;();<a name="line.707"></a>
+<span class="sourceLineNo">708</span>      for (LoadQueueItem lqi: lqis) {<a name="line.708"></a>
+<span class="sourceLineNo">709</span>        MutableInt count = filesMap.get(lqi.family);<a name="line.709"></a>
+<span class="sourceLineNo">710</span>        if (count == null) {<a name="line.710"></a>
+<span class="sourceLineNo">711</span>          count = new MutableInt();<a name="line.711"></a>
+<span class="sourceLineNo">712</span>          filesMap.put(lqi.family, count);<a name="line.712"></a>
+<span class="sourceLineNo">713</span>        }<a name="line.713"></a>
+<span class="sourceLineNo">714</span>        count.increment();<a name="line.714"></a>
+<span class="sourceLineNo">715</span>        if (count.intValue() &gt; maxFilesPerRegionPerFamily) {<a name="line.715"></a>
+<span class="sourceLineNo">716</span>          LOG.error("Trying to load more than " + maxFilesPerRegionPerFamily<a name="line.716"></a>
+<span class="sourceLineNo">717</span>            + " hfiles to family " + Bytes.toStringBinary(lqi.family)<a name="line.717"></a>
+<span class="sourceLineNo">718</span>            + " of region with start key "<a name="line.718"></a>
+<span class="sourceLineNo">719</span>            + Bytes.toStringBinary(e.getKey()));<a name="line.719"></a>
+<span class="sourceLineNo">720</span>          return false;<a name="line.720"></a>
+<span class="sourceLineNo">721</span>        }<a name="line.721"></a>
+<span class="sourceLineNo">722</span>      }<a name="line.722"></a>
+<span class="sourceLineNo">723</span>    }<a name="line.723"></a>
+<span class="sourceLineNo">724</span>    return true;<a name="line.724"></a>
+<span class="sourceLineNo">725</span>  }<a name="line.725"></a>
+<span class="sourceLineNo">726</span><a name="line.726"></a>
+<span class="sourceLineNo">727</span>  /**<a name="line.727"></a>
+<span class="sourceLineNo">728</span>   * @param table the table to load into<a name="line.728"></a>
+<span class="sourceLineNo">729</span>   * @param pool the ExecutorService<a name="line.729"></a>
+<span class="sourceLineNo">730</span>   * @param queue the queue for LoadQueueItem<a name="line.730"></a>
+<span class="sourceLineNo">731</span>   * @param startEndKeys start and end keys<a name="line.731"></a>
+<span class="sourceLineNo">732</span>   * @return A map that groups LQI by likely bulk load region targets and Set of missing hfiles.<a name="line.732"></a>
+<span class="sourceLineNo">733</span>   */<a name="line.733"></a>
+<span class="sourceLineNo">734</span>  private Pair&lt;Multimap&lt;ByteBuffer, LoadQueueItem&gt;, Set&lt;String&gt;&gt; groupOrSplitPhase(<a name="line.734"></a>
+<span class="sourceLineNo">735</span>      final Table table, ExecutorService pool, Deque&lt;LoadQueueItem&gt; queue,<a name="line.735"></a>
+<span class="sourceLineNo">736</span>      final Pair&lt;byte[][], byte[][]&gt; startEndKeys) throws IOException {<a name="line.736"></a>
+<span class="sourceLineNo">737</span>    // &lt;region start key, LQI&gt; need synchronized only within this scope of this<a name="line.737"></a>
+<span class="sourceLineNo">738</span>    // phase because of the puts that happen in futures.<a name="line.738"></a>
+<span class="sourceLineNo">739</span>    Multimap&lt;ByteBuffer, LoadQueueItem&gt; rgs = HashMultimap.create();<a name="line.739"></a>
+<span class="sourceLineNo">740</span>    final Multimap&lt;ByteBuffer, LoadQueueItem&gt; regionGroups = Multimaps.synchronizedMultimap(rgs);<a name="line.740"></a>
+<span class="sourceLineNo">741</span>    Set&lt;String&gt; missingHFiles = new HashSet&lt;&gt;();<a name="line.741"></a>
+<span class="sourceLineNo">742</span>    Pair&lt;Multimap&lt;ByteBuffer, LoadQueueItem&gt;, Set&lt;String&gt;&gt; pair = new Pair&lt;&gt;(regionGroups,<a name="line.742"></a>
+<span class="sourceLineNo">743</span>        missingHFiles);<a name="line.743"></a>
+<span class="sourceLineNo">744</span><a name="line.744"></a>
+<span class="sourceLineNo">745</span>    // drain LQIs and figure out bulk load groups<a name="line.745"></a>
+<span class="sourceLineNo">746</span>    Set&lt;Future&lt;Pair&lt;List&lt;LoadQueueItem&gt;, String&gt;&gt;&gt; splittingFutures = new HashSet&lt;&gt;();<a name="line.746"></a>
+<span class="sourceLineNo">747</span>    while (!queue.isEmpty()) {<a name="line.747"></a>
+<span class="sourceLineNo">748</span>      final LoadQueueItem item = queue.remove();<a name="line.748"></a>
+<span class="sourceLineNo">749</span><a name="line.749"></a>
+<span class="sourceLineNo">750</span>      final Callable&lt;Pair&lt;List&lt;LoadQueueItem&gt;, String&gt;&gt; call =<a name="line.750"></a>
+<span class="sourceLineNo">751</span>          new Callable&lt;Pair&lt;List&lt;LoadQueueItem&gt;, String&gt;&gt;() {<a name="line.751"></a>
+<span class="sourceLineNo">752</span>        @Override<a name="line.752"></a>
+<span class="sourceLineNo">753</span>        public Pair&lt;List&lt;LoadQueueItem&gt;, String&gt; call() throws Exception {<a name="line.753"></a>
+<span class="sourceLineNo">754</span>          Pair&lt;List&lt;LoadQueueItem&gt;, String&gt; splits = groupOrSplit(regionGroups, item, table,<a name="line.754"></a>
+<span class="sourceLineNo">755</span>              startEndKeys);<a name="line.755"></a>
+<span class="sourceLineNo">756</span>          return splits;<a name="line.756"></a>
+<span class="sourceLineNo">757</span>        }<a name="line.757"></a>
+<span class="sourceLineNo">758</span>      };<a name="line.758"></a>
+<span class="sourceLineNo">759</span>      splittingFutures.add(pool.submit(call));<a name="line.759"></a>
+<span class="sourceLineNo">760</span>    }<a name="line.760"></a>
+<span class="sourceLineNo">761</span>    // get all the results.  All grouping and splitting must finish before<a name="line.761"></a>
+<span class="sourceLineNo">762</span>    // we can attempt the atomic loads.<a name="line.762"></a>
+<span class="sourceLineNo">763</span>    for (Future&lt;Pair&lt;List&lt;LoadQueueItem&gt;, String&gt;&gt; lqis : splittingFutures) {<a name="line.763"></a>
+<span class="sourceLineNo">764</span>      try {<a name="line.764"></a>
+<span class="sourceLineNo">765</span>        Pair&lt;List&lt;LoadQueueItem&gt;, String&gt; splits = lqis.get();<a name="line.765"></a>
+<span class="sourceLineNo">766</span>        if (splits != null) {<a name="line.766"></a>
+<span class="sourceLineNo">767</span>          if (splits.getFirst() != null) {<a name="line.767"></a>
+<span class="sourceLineNo">768</span>            queue.addAll(splits.getFirst());<a name="line.768"></a>
+<span class="sourceLineNo">769</span>          } else {<a name="line.769"></a>
+<span class="sourceLineNo">770</span>            missingHFiles.add(splits.getSecond());<a name="line.770"></a>
+<span class="sourceLineNo">771</span>          }<a name="line.771"></a>
+<span class="sourceLineNo">772</span>        }<a name="line.772"></a>
+<span class="sourceLineNo">773</span>      } catch (ExecutionException e1) {<a name="line.773"></a>
+<span class="sourceLineNo">774</span>        Throwable t = e1.getCause();<a name="line.774"></a>
+<span class="sourceLineNo">775</span>        if (t instanceof IOException) {<a name="line.775"></a>
+<span class="sourceLineNo">776</span>          LOG.error("IOException during splitting", e1);<a name="line.776"></a>
+<span class="sourceLineNo">777</span>          throw (IOException)t; // would have been thrown if not parallelized,<a name="line.777"></a>
+<span class="sourceLineNo">778</span>        }<a name="line.778"></a>
+<span class="sourceLineNo">779</span>        LOG.error("Unexpected execution exception during splitting", e1);<a name="line.779"></a>
+<span class="sourceLineNo">780</span>        throw new IllegalStateException(t);<a name="line.780"></a>
+<span class="sourceLineNo">781</span>      } catch (InterruptedException e1) {<a name="line.781"></a>
+<span class="sourceLineNo">782</span>        LOG.error("Unexpected interrupted exception during splitting", e1);<a name="line.782"></a>
+<span class="sourceLineNo">783</span>        throw (InterruptedIOException)new InterruptedIOException().initCause(e1);<a name="line.783"></a>
+<span class="sourceLineNo">784</span>      }<a name="line.784"></a>
+<span class="sourceLineNo">785</span>    }<a name="line.785"></a>
+<span class="sourceLineNo">786</span>    return pair;<a name="line.786"></a>
+<span class="sourceLineNo">787</span>  }<a name="line.787"></a>
+<span class="sourceLineNo">788</span><a name="line.788"></a>
+<span class="sourceLineNo">789</span>  // unique file name for the table<a name="line.789"></a>
+<span class="sourceLineNo">790</span>  private String getUniqueName() {<a name="line.790"></a>
+<span class="sourceLineNo">791</span>    return UUID.randomUUID().toString().replaceAll("-", "");<a name="line.791"></a>
+<span class="sourceLineNo">792</span>  }<a name="line.792"></a>
+<span class="sourceLineNo">793</span><a name="line.793"></a>
+<span class="sourceLineNo">794</span>  protected List&lt;LoadQueueItem&gt; splitStoreFile(final LoadQueueItem item,<a name="line.794"></a>
+<span class="sourceLineNo">795</span>      final Table table, byte[] startKey,<a name="line.795"></a>
+<span class="sourceLineNo">796</span>      byte[] splitKey) throws IOException {<a name="line.796"></a>
+<span class="sourceLineNo">797</span>    final Path hfilePath = item.hfilePath;<a name="line.797"></a>
+<span class="sourceLineNo">798</span><a name="line.798"></a>
+<span class="sourceLineNo">799</span>    Path tmpDir = item.hfilePath.getParent();<a name="line.799"></a>
+<span class="sourceLineNo">800</span>    if (!tmpDir.getName().equals(TMP_DIR)) {<a name="line.800"></a>
+<span class="sourceLineNo">801</span>      tmpDir = new Path(tmpDir, TMP_DIR);<a name="line.801"></a>
+<span class="sourceLineNo">802</span>    }<a name="line.802"></a>
+<span class="sourceLineNo">803</span><a name="line.803"></a>
+<span class="sourceLineNo">804</span>    LOG.info("HFile at " + hfilePath + " no longer fits inside a single " +<a name="line.804"></a>
+<span class="sourceLineNo">805</span>      "region. Splitting...");<a name="line.805"></a>
 <span class="sourceLineNo">806</span><a name="line.806"></a>
-<span class="sourceLineNo">807</span>    // Add these back at the *front* of the queue, so there's a lower<a name="line.807"></a>
-<span class="sourceLineNo">808</span>    // chance that the region will just split again before we get there.<a name="line.808"></a>
-<span class="sourceLineNo">809</span>    List&lt;LoadQueueItem&gt; lqis = new ArrayList&lt;LoadQueueItem&gt;(2);<a name="line.809"></a>
-<span class="sourceLineNo">810</span>    lqis.add(new LoadQueueItem(item.family, botOut));<a name="line.810"></a>
-<span class="sourceLineNo">811</span>    lqis.add(new LoadQueueItem(item.family, topOut));<a name="line.811"></a>
-<span class="sourceLineNo">812</span><a name="line.812"></a>
-<span class="sourceLineNo">813</span>    // If the current item is already the result of previous splits,<a name="line.813"></a>
-<span class="sourceLineNo">814</span>    // we don't need it anymore. Clean up to save space.<a name="line.814"></a>
-<span class="sourceLineNo">815</span>    // It is not part of the original input files.<a name="line.815"></a>
-<span class="sourceLineNo">816</span>    try {<a name="line.816"></a>
-<span class="sourceLineNo">817</span>      tmpDir = item.hfilePath.getParent();<a name="line.817"></a>
-<span class="sourceLineNo">818</span>      if (tmpDir.getName().equals(TMP_DIR)) {<a name="line.818"></a>
-<span class="sourceLineNo">819</span>        fs.delete(item.hfilePath, false);<a name="line.819"></a>
-<span class="sourceLineNo">820</span>      }<a name="line.820"></a>
-<span class="sourceLineNo">821</span>    } catch (IOException e) {<a name="line.821"></a>
-<span class="sourceLineNo">822</span>      LOG.warn("Unable to delete temporary split file " + item.hfilePath);<a name="line.822"></a>
-<span class="sourceLineNo">823</span>    }<a name="line.823"></a>
-<span class="sourceLineNo">824</span>    LOG.info("Successfully split into new HFiles " + botOut + " and " + topOut);<a name="line.824"></a>
-<span class="sourceLineNo">825</span>    return lqis;<a name="line.825"></a>
-<span class="sourceLineNo">826</span>  }<a name="line.826"></a>
-<span class="sourceLineNo">827</span><a name="line.827"></a>
-<span class="sourceLineNo">828</span>  /**<a name="line.828"></a>
-<span class="sourceLineNo">829</span>   * Attempt to assign the given load queue item into its target region group.<a name="line.829"></a>
-<span class="sourceLineNo">830</span>   * If the hfile boundary no longer fits into a region, physically splits<a name="line.830"></a>
-<span class="sourceLineNo">831</span>   * the hfile such that the new bottom half will fit and returns the list of<a name="line.831"></a>
-<span class="sourceLineNo">832</span>   * LQI's corresponding to the resultant hfiles.<a name="line.832"></a>
-<span class="sourceLineNo">833</span>   *<a name="line.833"></a>
-<span class="sourceLineNo">834</span>   * protected for testing<a name="line.834"></a>
-<span class="sourceLineNo">835</span>   * @throws IOException if an IO failure is encountered<a name="line.835"></a>
-<span class="sourceLineNo">836</span>   */<a name="line.836"></a>
-<span class="sourceLineNo">837</span>  protected Pair&lt;List&lt;LoadQueueItem&gt;, String&gt; groupOrSplit(<a name="line.837"></a>
-<span class="sourceLineNo">838</span>      Multimap&lt;ByteBuffer, LoadQueueItem&gt; regionGroups, final LoadQueueItem item, final Table table,<a name="line.838"></a>
-<span class="sourceLineNo">839</span>      final Pair&lt;byte[][], byte[][]&gt; startEndKeys) throws IOException {<a name="line.839"></a>
-<span class="sourceLineNo">840</span>    final Path hfilePath = item.hfilePath;<a name="line.840"></a>
-<span class="sourceLineNo">841</span>    // fs is the source filesystem<a name="line.841"></a>
-<span class="sourceLineNo">842</span>    if (fs == null) {<a name="line.842"></a>
-<span class="sourceLineNo">843</span>      fs = hfilePath.getFileSystem(getConf());<a name="line.843"></a>
-<span class="sourceLineNo">844</span>    }<a name="line.844"></a>
-<span class="sourceLineNo">845</span>    HFile.Reader hfr = null;<a name="line.845"></a>
-<span class="sourceLineNo">846</span>    try {<a name="line.846"></a>
-<span class="sourceLineNo">847</span>      hfr = HFile.createReader(fs, hfilePath,<a name="line.847"></a>
-<span class="sourceLineNo">848</span>          new CacheConfig(getConf()), getConf());<a name="line.848"></a>
-<span class="sourceLineNo">849</span>    } catch (FileNotFoundException fnfe) {<a name="line.849"></a>
-<span class="sourceLineNo">850</span>      LOG.debug("encountered", fnfe);<a name="line.850"></a>
-<span class="sourceLineNo">851</span>      return new Pair&lt;&gt;(null, hfilePath.getName());<a name="line.851"></a>
-<span class="sourceLineNo">852</span>    }<a name="line.852"></a>
-<span class="sourceLineNo">853</span>    final byte[] first, last;<a name="line.853"></a>
-<span class="sourceLineNo">854</span>    try {<a name="line.854"></a>
-<span class="sourceLineNo">855</span>      hfr.loadFileInfo();<a name="line.855"></a>
-<span class="sourceLineNo">856</span>      first = hfr.getFirstRowKey();<a name="line.856"></a>
-<span class="sourceLineNo">857</span>      last = hfr.getLastRowKey();<a name="line.857"></a>
-<span class="sourceLineNo">858</span>    }  finally {<a name="line.858"></a>
-<span class="sourceLineNo">859</span>      hfr.close();<a name="line.859"></a>
-<span class="sourceLineNo">860</span>    }<a name="line.860"></a>
-<span class="sourceLineNo">861</span><a name="line.861"></a>
-<span class="sourceLineNo">862</span>    LOG.info("Trying to load hfile=" + hfilePath +<a name="line.862"></a>
-<span class="sourceLineNo">863</span>        " first=" + Bytes.toStringBinary(first) +<a name="line.863"></a>
-<span class="sourceLineNo">864</span>        " last="  + Bytes.toStringBinary(last));<a name="line.864"></a>
-<span class="sourceLineNo">865</span>    if (first == null || last == null) {<a name="line.865"></a>
-<span class="sourceLineNo">866</span>      assert first == null &amp;&amp; last == null;<a name="line.866"></a>
-<span class="sourceLineNo">867</span>      // TODO what if this is due to a bad HFile?<a name="line.867"></a>
-<span class="sourceLineNo">868</span>      LOG.info("hfile " + hfilePath + " has no entries, skipping");<a name="line.868"></a>
-<span class="sourceLineNo">869</span>      return null;<a name="line.869"></a>
-<span class="sourceLineNo">870</span>    }<a name="line.870"></a>
-<span class="sourceLineNo">871</span>    if (Bytes.compareTo(first, last) &gt; 0) {<a name="line.871"></a>
-<span class="sourceLineNo">872</span>      throw new IllegalArgumentException(<a name="line.872"></a>
-<span class="sourceLineNo">873</span>      "Invalid range: " + Bytes.toStringBinary(first) +<a name="line.873"></a>
-<span class="sourceLineNo">874</span>      " &gt; " + Bytes.toStringBinary(last));<a name="line.874"></a>
-<span class="sourceLineNo">875</span>    }<a name="line.875"></a>
-<span class="sourceLineNo">876</span>    int idx = Arrays.binarySearch(startEndKeys.getFirst(), first,<a name="line.876"></a>
-<span class="sourceLineNo">877</span>        Bytes.BYTES_COMPARATOR);<a name="line.877"></a>
-<span class="sourceLineNo">878</span>    if (idx &lt; 0) {<a name="line.878"></a>
-<span class="sourceLineNo">879</span>      // not on boundary, returns -(insertion index).  Calculate region it<a name="line.879"></a>
-<span class="sourceLineNo">880</span>      // would be in.<a name="line.880"></a>
-<span class="sourceLineNo">881</span>      idx = -(idx + 1) - 1;<a name="line.881"></a>
+<span class="sourceLineNo">807</span>    String uniqueName = getUniqueName();<a name="line.807"></a>
+<span class="sourceLineNo">808</span>    HColumnDescriptor familyDesc = table.getTableDescriptor().getFamily(item.family);<a name="line.808"></a>
+<span class="sourceLineNo">809</span><a name="line.809"></a>
+<span class="sourceLineNo">810</span>    Path botOut = new Path(tmpDir, uniqueName + ".bottom");<a name="line.810"></a>
+<span class="sourceLineNo">811</span>    Path topOut = new Path(tmpDir, uniqueName + ".top");<a name="line.811"></a>
+<span class="sourceLineNo">812</span>    splitStoreFile(getConf(), hfilePath, familyDesc, splitKey, botOut, topOut);<a name="line.812"></a>
+<span class="sourceLineNo">813</span><a name="line.813"></a>
+<span class="sourceLineNo">814</span>    FileSystem fs = tmpDir.getFileSystem(getConf());<a name="line.814"></a>
+<span class="sourceLineNo">815</span>    fs.setPermission(tmpDir, FsPermission.valueOf("-rwxrwxrwx"));<a name="line.815"></a>
+<span class="sourceLineNo">816</span>    fs.setPermission(botOut, FsPermission.valueOf("-rwxrwxrwx"));<a name="line.816"></a>
+<span class="sourceLineNo">817</span>    fs.setPermission(topOut, FsPermission.valueOf("-rwxrwxrwx"));<a name="line.817"></a>
+<span class="sourceLineNo">818</span><a name="line.818"></a>
+<span class="sourceLineNo">819</span>    // Add these back at the *front* of the queue, so there's a lower<a name="line.819"></a>
+<span class="sourceLineNo">820</span>    // chance that the region will just split again before we get there.<a name="line.820"></a>
+<span class="sourceLineNo">821</span>    List&lt;LoadQueueItem&gt; lqis = new ArrayList&lt;LoadQueueItem&gt;(2);<a name="line.821"></a>
+<span class="sourceLineNo">822</span>    lqis.add(new LoadQueueItem(item.family, botOut));<a name="line.822"></a>
+<span class="sourceLineNo">823</span>    lqis.add(new LoadQueueItem(item.family, topOut));<a name="line.823"></a>
+<span class="sourceLineNo">824</span><a name="line.824"></a>
+<span class="sourceLineNo">825</span>    // If the current item is already the result of previous splits,<a name="line.825"></a>
+<span class="sourceLineNo">826</span>    // we don't need it anymore. Clean up to save space.<a name="line.826"></a>
+<span class="sourceLineNo">827</span>    // It is not part of the original input files.<a name="line.827"></a>
+<span class="sourceLineNo">828</span>    try {<a name="line.828"></a>
+<span class="sourceLineNo">829</span>      tmpDir = item.hfilePath.getParent();<a name="line.829"></a>
+<span class="sourceLineNo">830</span>      if (tmpDir.getName().equals(TMP_DIR)) {<a name="line.830"></a>
+<span class="sourceLineNo">831</span>        fs.delete(item.hfilePath, false);<a name="line.831"></a>
+<span class="sourceLineNo">832</span>      }<a name="line.832"></a>
+<span class="sourceLineNo">833</span>    } catch (IOException e) {<a name="line.833"></a>
+<span class="sourceLineNo">834</span>      LOG.warn("Unable to delete temporary split file " + item.hfilePath);<a name="line.834"></a>
+<span class="sourceLineNo">835</span>    }<a name="line.835"></a>
+<span class="sourceLineNo">836</span>    LOG.info("Successfully split into new HFiles " + botOut + " and " + topOut);<a name="line.836"></a>
+<span class="sourceLineNo">837</span>    return lqis;<a name="line.837"></a>
+<span class="sourceLineNo">838</span>  }<a name="line.838"></a>
+<span class="sourceLineNo">839</span><a name="line.839"></a>
+<span class="sourceLineNo">840</span>  /**<a name="line.840"></a>
+<span class="sourceLineNo">841</span>   * Attempt to assign the given load queue item into its target region group.<a name="line.841"></a>
+<span class="sourceLineNo">842</span>   * If the hfile boundary no longer fits into a region, physically splits<a name="line.842"></a>
+<span class="sourceLineNo">843</span>   * the hfile such that the new bottom half will fit and returns the list of<a name="line.843"></a>
+<span class="sourceLineNo">844</span>   * LQI's corresponding to the resultant hfiles.<a name="line.844"></a>
+<span class="sourceLineNo">845</span>   *<a name="line.845"></a>
+<span class="sourceLineNo">846</span>   * protected for testing<a name="line.846"></a>
+<span class="sourceLineNo">847</span>   * @throws IOException if an IO failure is encountered<a name="line.847"></a>
+<span class="sourceLineNo">848</span>   */<a name="line.848"></a>
+<span class="sourceLineNo">849</span>  protected Pair&lt;List&lt;LoadQueueItem&gt;, String&gt; groupOrSplit(<a name="line.849"></a>
+<span class="sourceLineNo">850</span>      Multimap&lt;ByteBuffer, LoadQueueItem&gt; regionGroups, final LoadQueueItem item, final Table table,<a name="line.850"></a>
+<span class="sourceLineNo">851</span>      final Pair&lt;byte[][], byte[][]&gt; startEndKeys) throws IOException {<a name="line.851"></a>
+<span class="sourceLineNo">852</span>    final Path hfilePath = item.hfilePath;<a name="line.852"></a>
+<span class="sourceLineNo">853</span>    // fs is the source filesystem<a name="line.853"></a>
+<span class="sourceLineNo">854</span>    if (fs == null) {<a name="line.854"></a>
+<span class="sourceLineNo">855</span>      fs = hfilePath.getFileSystem(getConf());<a name="line.855"></a>
+<span class="sourceLineNo">856</span>    }<a name="line.856"></a>
+<span class="sourceLineNo">857</span>    HFile.Reader hfr = null;<a name="line.857"></a>
+<span class="sourceLineNo">858</span>    try {<a name="line.858"></a>
+<span class="sourceLineNo">859</span>      hfr = HFile.createReader(fs, hfilePath,<a name="line.859"></a>
+<span class="sourceLineNo">860</span>          new CacheConfig(getConf()), getConf());<a name="line.860"></a>
+<span class="sourceLineNo">861</span>    } catch (FileNotFoundException fnfe) {<a name="line.861"></a>
+<span class="sourceLineNo">862</span>      LOG.debug("encountered", fnfe);<a name="line.862"></a>
+<span class="sourceLineNo">863</span>      return new Pair&lt;&gt;(null, hfilePath.getName());<a name="line.863"></a>
+<span class="sourceLineNo">864</span>    }<a name="line.864"></a>
+<span class="sourceLineNo">865</span>    final byte[] first, last;<a name="line.865"></a>
+<span class="sourceLineNo">866</span>    try {<a name="li

<TRUNCATED>