In the user portrait scenario, many tags are usually developed. As a qualifier, each tag needs to be offline after some tags are no longer in use. However, the delete related api provided by hbase can only be used for a single line. It is not easy to clean up all the data of a qualifier. Here we provide an implementation scheme based on coprocessor;

hbase provides the following five hook s for the compact process to embed custom code:

• preCompactSelection
• postCompactSelection
• preCompactScannerOpen
• preCompact
• postCompact

preCompact will be called before reading the data after storeScanner is created, so the idea here is to proxy scanner, create a new scanner to implement its next method, and then process the raw data read.

The code is as follows, referring to the ValueRewritingObserver class in HBase examples module:

public class QualifierDeletingObserver implements RegionObserver, RegionCoprocessor {
  private static final Logger LOG = LoggerFactory.getLogger(QualifierDeletingObserver.class);

  private byte[] qualifierToDelete = null;
  private Bytes.ByteArrayComparator comparator;

  @Override
  public Optional<RegionObserver> getRegionObserver() {
    // Extremely important to be sure that the coprocessor is invoked as a RegionObserver
    return Optional.of(this);
  }

  @Override
  public void start(
      @SuppressWarnings("rawtypes") CoprocessorEnvironment env) throws IOException {
    RegionCoprocessorEnvironment renv = (RegionCoprocessorEnvironment) env;
    qualifierToDelete = Bytes.toBytes(renv.getConfiguration().get("qualifier.to.delete"));
    comparator = new Bytes.ByteArrayComparator();
  }

  @Override
  public InternalScanner preCompact(
      ObserverContext<RegionCoprocessorEnvironment> c, Store store,
      final InternalScanner scanner, ScanType scanType, CompactionLifeCycleTracker tracker,
      CompactionRequest request) {
    InternalScanner modifyingScanner = new InternalScanner() {
      @Override
      public boolean next(List<Cell> result, ScannerContext scannerContext) throws IOException {
        boolean ret = scanner.next(result, scannerContext);
        for (int i = 0; i < result.size(); i++) {
          Cell c = result.get(i);
          byte[] qualifier = CellUtil.cloneQualifier(c);
          if (comparator.compare(qualifier, qualifierToDelete) == 0) {
            result.remove(i);
          }
        }
        return ret;
      }

      @Override
      public void close() throws IOException {
        scanner.close();
      }
    };

    return modifyingScanner;
  }
}

Upload the jar package to hdfs;

The following is a simple test process display;

create 'cp_test','f'

put 'cp_test','rk1','f:q1','123'
put 'cp_test','rk1','f:q2','123'
put 'cp_test','rk2','f:q1','123'
put 'cp_test','rk2','f:q2','123'
put 'cp_test','rk2','f:q3','123'


hbase(main):015:0> scan 'cp_test'
ROW                                    COLUMN+CELL                                                                                                     
 rk1                                   column=f:q1, timestamp=1590567958995, value=123                                                                 
 rk1                                   column=f:q2, timestamp=1590567959023, value=123                                                                 
 rk2                                   column=f:q1, timestamp=1590567959048, value=123                                                                 
 rk2                                   column=f:q2, timestamp=1590567959073, value=123                                                                 
 rk2                                   column=f:q3, timestamp=1590567959842, value=123        
 

alter 'cp_test' \
, METHOD => 'table_att', 'coprocessor'=>'hdfs://xxx.jar|xxx.QualifierDelexxxtingObserver|1024|qualifier.to.delete=q1'

flush 'cp_test'

major_compact 'cp_test'

hbase(main):017:0> scan 'cp_test'
ROW                                    COLUMN+CELL                                                                                                     
 rk1                                   column=f:q2, timestamp=1590567959023, value=123                                                                 
 rk2                                   column=f:q2, timestamp=1590567959073, value=123                                                                 
 rk2                                   column=f:q3, timestamp=1590567959842, value=123