HBASE-2052 Upper bound of outstanding WALs can be overrun

git-svn-id: https://svn.apache.org/repos/asf/hadoop/hbase/trunk@895331 13f79535-47bb-0310-9956-ffa450edef68

CHANGES.txt

@@ -278,6 +278,7 @@ Release 0.21.0 - Unreleased
    HBASE-2036  Use Configuration instead of HBaseConfiguration (Enis Soztutar via Stack)
    HBASE-2085  StringBuffer -> StringBuilder - conversion of references as necessary
                (Kay Kay via Stack)
+   HBASE-2052  Upper bound of outstanding WALs can be overrun
 
   NEW FEATURES
    HBASE-1901  "General" partitioner for "hbase-48" bulk (behind the api, write

src/java/org/apache/hadoop/hbase/regionserver/LogRoller.java

@@ -82,9 +82,9 @@ class LogRoller extends Thread implements LogRollListener {
       rollLock.lock(); // FindBugs UL_UNRELEASED_LOCK_EXCEPTION_PATH
       try {
         this.lastrolltime = now;
-        byte [] regionToFlush = server.getLog().rollWriter();
+        byte [][] regionsToFlush = server.getLog().rollWriter();
-        if (regionToFlush != null) {
+        if (regionsToFlush != null) {
-          scheduleFlush(regionToFlush);
+          for (byte [] r: regionsToFlush) scheduleFlush(r);
         }
       } catch (FailedLogCloseException e) {
         LOG.fatal("Forcing server shutdown", e);

src/java/org/apache/hadoop/hbase/regionserver/wal/HLog.java

@@ -154,7 +154,7 @@ public class HLog implements HConstants, Syncable {
     Collections.synchronizedSortedMap(new TreeMap<Long, Path>());
 
   /*
-   * Map of region to last sequence/edit id. 
+   * Map of regions to first sequence/edit id in their memstore.
    */
   private final ConcurrentSkipListMap<byte [], Long> lastSeqWritten =
     new ConcurrentSkipListMap<byte [], Long>(Bytes.BYTES_COMPARATOR);
@@ -327,21 +327,21 @@ public class HLog implements HConstants, Syncable {
    * cacheFlushLock and then completeCacheFlush could be called which would wait
    * for the lock on this and consequently never release the cacheFlushLock
    *
-   * @return If lots of logs, flush the returned region so next time through
+   * @return If lots of logs, flush the returned regions so next time through
    * we can clean logs. Returns null if nothing to flush.
    * @throws org.apache.hadoop.hbase.regionserver.wal.FailedLogCloseException
    * @throws IOException
    */
-  public byte [] rollWriter() throws FailedLogCloseException, IOException {
+  public byte [][] rollWriter() throws FailedLogCloseException, IOException {
     // Return if nothing to flush.
     if (this.writer != null && this.numEntries.get() <= 0) {
       return null;
     }
-    byte [] regionToFlush = null;
+    byte [][] regionsToFlush = null;
     this.cacheFlushLock.lock();
     try {
       if (closed) {
-        return regionToFlush;
+        return regionsToFlush;
       }
       synchronized (updateLock) {
         // Clean up current writer.
@@ -367,7 +367,7 @@ public class HLog implements HConstants, Syncable {
             }
             this.outputfiles.clear();
           } else {
-            regionToFlush = cleanOldLogs();
+            regionsToFlush = cleanOldLogs();
           }
         }
         this.numEntries.set(0);
@@ -376,7 +376,7 @@ public class HLog implements HConstants, Syncable {
     } finally {
       this.cacheFlushLock.unlock();
     }
-    return regionToFlush;
+    return regionsToFlush;
   }
 
   /**
@@ -433,8 +433,7 @@ public class HLog implements HConstants, Syncable {
    * we can clean logs. Returns null if nothing to flush.
    * @throws IOException
    */
-  private byte [] cleanOldLogs() throws IOException {
+  private byte [][] cleanOldLogs() throws IOException {
-    byte [] regionToFlush = null;
     Long oldestOutstandingSeqNum = getOldestOutstandingSeqNum();
     // Get the set of all log files whose final ID is older than or
     // equal to the oldest pending region operation
@@ -442,29 +441,60 @@ public class HLog implements HConstants, Syncable {
       new TreeSet<Long>(this.outputfiles.headMap(
         (Long.valueOf(oldestOutstandingSeqNum.longValue() + 1L))).keySet());
     // Now remove old log files (if any)
-    byte [] oldestRegion = null;
+    int logsToRemove = sequenceNumbers.size();
+    if (logsToRemove > 0) {
       if (LOG.isDebugEnabled()) {
-      // Find region associated with oldest key -- helps debugging.
+        // Find associated region; helps debugging.
-      oldestRegion = getOldestRegion(oldestOutstandingSeqNum);
+        byte [] oldestRegion = getOldestRegion(oldestOutstandingSeqNum);
-      LOG.debug("Found " + sequenceNumbers.size() + " hlogs to remove " +
+        LOG.debug("Found " + logsToRemove + " hlogs to remove " +
           " out of total " + this.outputfiles.size() + "; " +
           "oldest outstanding seqnum is " + oldestOutstandingSeqNum +
-        " from region " + Bytes.toStringBinary(oldestRegion));
+          " from region " + Bytes.toString(oldestRegion));
       }
-    if (sequenceNumbers.size() > 0) {
       for (Long seq : sequenceNumbers) {
         deleteLogFile(this.outputfiles.remove(seq), seq);
       }
     }
-    int countOfLogs = this.outputfiles.size() - sequenceNumbers.size();
+
-    if (countOfLogs > this.maxLogs) {
+    // If too many log files, figure which regions we need to flush.
-      regionToFlush = oldestRegion != null?
+    byte [][] regions = null;
-        oldestRegion: getOldestRegion(oldestOutstandingSeqNum);
+    int logCount = this.outputfiles.size() - logsToRemove;
-      LOG.info("Too many hlogs: logs=" + countOfLogs + ", maxlogs=" +
+    if (logCount > this.maxLogs && this.outputfiles != null &&
-        this.maxLogs + "; forcing flush of region with oldest edits: " +
+        this.outputfiles.size() > 0) {
-        Bytes.toStringBinary(regionToFlush));
+      regions = findMemstoresWithEditsOlderThan(this.outputfiles.firstKey(),
+        this.lastSeqWritten);
+      StringBuilder sb = new StringBuilder();
+      for (int i = 0; i < regions.length; i++) {
+        if (i > 0) sb.append(", ");
+        sb.append(Bytes.toStringBinary(regions[i]));
       }
-    return regionToFlush;
+      LOG.info("Too many hlogs: logs=" + logCount + ", maxlogs=" +
+        this.maxLogs + "; forcing flush of " + regions.length + " regions(s): " +
+        sb.toString());
+    }
+    return regions;
+  }
+
+  /**
+   * Return regions (memstores) that have edits that are less than the passed
+   * <code>oldestWALseqid</code>.
+   * @param oldestWALseqid
+   * @param regionsToSeqids
+   * @return All regions whose seqid is < than <code>oldestWALseqid</code> (Not
+   * necessarily in order).  Null if no regions found.
+   */
+  static byte [][] findMemstoresWithEditsOlderThan(final long oldestWALseqid,
+      final Map<byte [], Long> regionsToSeqids) {
+    //  This method is static so it can be unit tested the easier.
+    List<byte []> regions = null;
+    for (Map.Entry<byte [], Long> e: regionsToSeqids.entrySet()) {
+      if (e.getValue().longValue() < oldestWALseqid) {
+        if (regions == null) regions = new ArrayList<byte []>();
+        regions.add(e.getKey());
+      }
+    }
+    return regions == null?
+      null: regions.toArray(new byte [][] {HConstants.EMPTY_BYTE_ARRAY});
   }
 
   /*
@@ -614,7 +644,8 @@ public class HLog implements HConstants, Syncable {
       long seqNum = obtainSeqNum();
       logKey.setLogSeqNum(seqNum);
       // The 'lastSeqWritten' map holds the sequence number of the oldest
-      // write for each region. When the cache is flushed, the entry for the
+      // write for each region (i.e. the first edit added to the particular
+      // memstore). When the cache is flushed, the entry for the
       // region being flushed is removed if the sequence number of the flush
       // is greater than or equal to the value in lastSeqWritten.
       this.lastSeqWritten.putIfAbsent(regionName, Long.valueOf(seqNum));
@@ -661,7 +692,8 @@ public class HLog implements HConstants, Syncable {
     long seqNum [] = obtainSeqNum(edits.size());
     synchronized (this.updateLock) {
       // The 'lastSeqWritten' map holds the sequence number of the oldest
-      // write for each region. When the cache is flushed, the entry for the
+      // write for each region (i.e. the first edit added to the particular
+      // memstore). . When the cache is flushed, the entry for the
       // region being flushed is removed if the sequence number of the flush
       // is greater than or equal to the value in lastSeqWritten.
       this.lastSeqWritten.putIfAbsent(regionName, Long.valueOf(seqNum[0]));

src/test/org/apache/hadoop/hbase/regionserver/wal/TestHLog.java

@@ -21,7 +21,9 @@ package org.apache.hadoop.hbase.regionserver.wal;
 
 import java.io.IOException;
 import java.util.ArrayList;
+import java.util.HashMap;
 import java.util.List;
+import java.util.Map;
 
 import org.apache.hadoop.fs.FSDataInputStream;
 import org.apache.hadoop.fs.FSDataOutputStream;
@@ -184,6 +186,31 @@ public class TestHLog extends HBaseTestCase implements HConstants {
     reader.close();
   }
 
+  /**
+   * Test the findMemstoresWithEditsOlderThan method.
+   * @throws IOException
+   */
+  public void testFindMemstoresWithEditsOlderThan() throws IOException {
+    Map<byte [], Long> regionsToSeqids = new HashMap<byte [], Long>();
+    for (int i = 0; i < 10; i++) {
+      Long l = new Long(i);
+      regionsToSeqids.put(l.toString().getBytes(), l);
+    }
+    byte [][] regions =
+      HLog.findMemstoresWithEditsOlderThan(1, regionsToSeqids);
+    assertEquals(1, regions.length);
+    assertTrue(Bytes.equals(regions[0], "0".getBytes()));
+    regions = HLog.findMemstoresWithEditsOlderThan(3, regionsToSeqids);
+    int count = 3;
+    assertEquals(count, regions.length);
+    // Regions returned are not ordered.
+    for (int i = 0; i < count; i++) {
+      assertTrue(Bytes.equals(regions[i], "0".getBytes()) ||
+        Bytes.equals(regions[i], "1".getBytes()) ||
+        Bytes.equals(regions[i], "2".getBytes()));
+    }
+  }
+ 
   private void verifySplits(List<Path> splits, final int howmany)
   throws IOException {
     assertEquals(howmany, splits.size());