Uploaded image for project: 'HBase'
  1. HBase
  2. HBASE-15085

IllegalStateException was thrown when scanning on bulkloaded HFiles



    • Bug
    • Status: Closed
    • Critical
    • Resolution: Fixed
    • 0.98.12, 1.1.2
    • 1.2.0, 1.3.0, 1.0.3, 1.1.3, 0.98.17, 2.0.0
    • None
    • HBase-0.98.12 & Hadoop-2.6.0 & JDK1.7
      HBase-1.1.2 & Hadoop-2.6.0 & JDK1.7

    • Reviewed


      IllegalStateException was thrown when we scanned from an HFile which was bulk loaded several minutes ago, as shown below:

      2015-12-16 22:20:54,456 ERROR com.taobao.kart.coprocessor.server.KartCoprocessor: icbu_ae_ws_product,/0055,1450275490479.6a6a700f465ad074287fed720c950f7c. batchNotify exception
      java.lang.IllegalStateException: EncodedScanner works only on encoded data blocks
              at org.apache.hadoop.hbase.io.hfile.HFileReaderV2$EncodedScannerV2.updateCurrentBlock(HFileReaderV2.java:1042)
              at org.apache.hadoop.hbase.io.hfile.HFileReaderV2$EncodedScannerV2.seekTo(HFileReaderV2.java:1093)
              at org.apache.hadoop.hbase.regionserver.StoreFileScanner.seekAtOrAfter(StoreFileScanner.java:244)
              at org.apache.hadoop.hbase.regionserver.StoreFileScanner.seek(StoreFileScanner.java:152)
              at org.apache.hadoop.hbase.regionserver.StoreScanner.seekScanners(StoreScanner.java:329)
              at org.apache.hadoop.hbase.regionserver.StoreScanner.<init>(StoreScanner.java:188)
              at org.apache.hadoop.hbase.regionserver.HStore.getScanner(HStore.java:1879)
              at org.apache.hadoop.hbase.regionserver.HRegion$RegionScannerImpl.<init>(HRegion.java:4068)
              at org.apache.hadoop.hbase.regionserver.HRegion.instantiateRegionScanner(HRegion.java:2029)
              at org.apache.hadoop.hbase.regionserver.HRegion.getScanner(HRegion.java:2015)
              at org.apache.hadoop.hbase.regionserver.HRegion.getScanner(HRegion.java:1992)

      I used 'hbase hfile' command to analyse the meta and block info of the hfile, finding that even through the DATA_BLOCK_ENCODING was 'DIFF' in FileInfo, the actual data blocks was written without any encoding algorithms(BlockType was 'DATA', not 'ENCODED_DATA'):

          BULKLOAD_SOURCE_TASK = attempt_1442077249005_606706_r_000012_0
          BULKLOAD_TIMESTAMP = \x00\x00\x01R\x12$\x13\x12
      DataBlock Header:
      HFileBlock [ fileOffset=0 headerSize()=33 blockType=DATA onDiskSizeWithoutHeader=65591 uncompressedSizeWithoutHeader=65571 prevBlockOffset=-1 isUseHBaseChecksum()=true checksumType=CRC32 bytesPerChecksum=16384 onDiskDataSizeWithHeader=65604 getOnDiskSizeWithHeader()=65624 totalChecksumBytes()=20 isUnpacked()=true buf=[ java.nio.HeapByteBuffer[pos=0 lim=65624 cap=65657], array().length=65657, arrayOffset()=0 ] dataBeginsWith=\x00\x00\x003\x00\x00\x00\x0A\x00\x10/0008:1000000008\x01dprod fileContext=HFileContext [ usesHBaseChecksum=true checksumType=CRC32 bytesPerChecksum=16384 blocksize=65536 encoding=NONE includesMvcc=true includesTags=false compressAlgo=NONE compressTags=false cryptoContext=[ cipher=NONE keyHash=NONE ] ] ]

      The data block encoding in file info was not consistent with the one in data block, which means there must be something wrong with the bulkload process.
      After debugging on each step of bulkload, I found that LoadIncrementalHFiles had a bug when loading hfile into a splitted region.

         * Copy half of an HFile into a new HFile.
        private static void copyHFileHalf(
            Configuration conf, Path inFile, Path outFile, Reference reference,
            HColumnDescriptor familyDescriptor)
        throws IOException {
          FileSystem fs = inFile.getFileSystem(conf);
          CacheConfig cacheConf = new CacheConfig(conf);
          HalfStoreFileReader halfReader = null;
          StoreFile.Writer halfWriter = null;
          try {
            halfReader = new HalfStoreFileReader(fs, inFile, cacheConf, reference, conf);
            Map<byte[], byte[]> fileInfo = halfReader.loadFileInfo();
            int blocksize = familyDescriptor.getBlocksize();
            Algorithm compression = familyDescriptor.getCompression();
            BloomType bloomFilterType = familyDescriptor.getBloomFilterType();
      // use CF's DATA_BLOCK_ENCODING to initialize HFile writer
            HFileContext hFileContext = new HFileContextBuilder()
            halfWriter = new StoreFile.WriterBuilder(conf, cacheConf,
            HFileScanner scanner = halfReader.getScanner(false, false, false);
            do {
              KeyValue kv = KeyValueUtil.ensureKeyValue(scanner.getKeyValue());
            } while (scanner.next());
      // force encoding setting with the original HFile's file info
            for (Map.Entry<byte[],byte[]> entry : fileInfo.entrySet()) {
              if (shouldCopyHFileMetaKey(entry.getKey())) {
                halfWriter.appendFileInfo(entry.getKey(), entry.getValue());
          } finally {
            if (halfWriter != null) halfWriter.close();
            if (halfReader != null) halfReader.close(cacheConf.shouldEvictOnClose());

      As shown above, when an HFile which has a DIFF encoding is bulkloaded into a splitted region whose CF's DATA_BLOCK_ENCODING is NONE, the two new HFiles would have inconsistent encodings.

      Besides, it would be OK if splitting region's DATA_BLOCK_ENCODING is DIFF and bulk loaded HFile has NONE, because the initial bulkloaded HFile would not write the encoding info into its meta (NoOpDataBlockEncoder.saveMetadata() is empty), and It then would not rewrite encoding in two generated Files in copyHFileHalf(). Two new HFiles' meta info would be consistent with their block headers, which would all be DIFF. So, no Exception would be thrown when scanning these files.


        1. HBASE-15085-0.98-v1.patch
          6 kB
          Victor Xu
        2. HBASE-15085-0.98-v2.patch
          7 kB
          Victor Xu
        3. HBASE-15085-0.98-v3.patch
          7 kB
          Victor Xu
        4. HBASE-15085-0.98-v4.patch
          7 kB
          Victor Xu
        5. HBASE-15085-0.98-v4.patch
          7 kB
          Victor Xu
        6. HBASE-15085-0.98-v5.patch
          7 kB
          Victor Xu
        7. HBASE-15085-branch-1.0-v1.patch
          7 kB
          Victor Xu
        8. HBASE-15085-branch-1.0-v2.patch
          7 kB
          Victor Xu
        9. HBASE-15085-branch-1.1-v1.patch
          7 kB
          Victor Xu
        10. HBASE-15085-branch-1.1-v2.patch
          7 kB
          Victor Xu
        11. HBASE-15085-branch-1.2-v1.patch
          7 kB
          Victor Xu
        12. HBASE-15085-branch-1.2-v2.patch
          7 kB
          Victor Xu
        13. HBASE-15085-v1.patch
          7 kB
          Victor Xu
        14. HBASE-15085-v2.patch
          7 kB
          Victor Xu
        15. HBASE-15085-v3.patch
          7 kB
          Victor Xu
        16. HBASE-15085-v4.patch
          7 kB
          Victor Xu
        17. HBASE-15085-v4.patch
          7 kB
          Victor Xu
        18. HBASE-15085-v4.patch
          7 kB
        19. HBASE-15085-v4.patch
          7 kB
          Victor Xu
        20. HBASE-15085-v5.patch
          7 kB
          Victor Xu



            victorunique Victor Xu
            victorunique Victor Xu
            0 Vote for this issue
            11 Start watching this issue