Beispiel #1
0
 @Override
 public int compare(KeyValue o1, KeyValue o2) {
   int d;
   if ((d = Bytes.memcmp(o1.key(), o2.key())) != 0) {
     return d;
   } else if ((d = Bytes.memcmp(o1.family(), o2.family())) != 0) {
     return d;
   } else if ((d = Bytes.memcmp(o1.qualifier(), o2.qualifier())) != 0) {
     return d;
   } else if ((d = Long.signum(o2.timestamp() - o1.timestamp())) != 0) {
     return d;
   } else {
     d = Bytes.memcmp(o1.value(), o2.value());
   }
   return d;
 }
 @Override
 @SuppressWarnings({"rawtypes", "unchecked"})
 public void execute(TridentTuple tuple, TridentCollector collector) {
   List<KeyValue> keyValueList = (List<KeyValue>) tuple.getValue(0);
   for (KeyValue keyValue : keyValueList) {
     Values values = new Values();
     if (this.rowKey) {
       if (this.rowKeyDeserializer != null) {
         values.add(this.rowKeyDeserializer.deserialize(keyValue.key()));
       } else {
         values.add(new String(keyValue.key()));
       }
     }
     if (this.family) {
       if (this.columnFamilyDeserializer != null) {
         values.add(this.columnFamilyDeserializer.deserialize(keyValue.family()));
       } else {
         values.add(new String(keyValue.family()));
       }
     }
     if (this.qualifier) {
       if (this.columnQualifierDeserializer != null) {
         values.add(this.columnFamilyDeserializer.deserialize(keyValue.qualifier()));
       } else {
         values.add(new String(keyValue.qualifier()));
       }
     }
     if (this.value) {
       if (this.valueDeserializer != null) {
         values.add(this.valueDeserializer.deserialize(keyValue.value()));
       } else {
         values.add(new String(keyValue.value()));
       }
     }
     if (this.timestamp) {
       values.add(keyValue.timestamp());
     }
     collector.emit(values);
   }
 }
Beispiel #3
0
 @Test
 public void parseFromColumn() throws Exception {
   final KeyValue column = mock(KeyValue.class);
   when(column.key()).thenReturn(new byte[] {0, 0, 1, 0, 0, 1, 0, 0, 1});
   when(column.value())
       .thenReturn(
           storage.getColumn(
               new byte[] {0, 0, 1, 0, 0, 1, 0, 0, 1},
               NAME_FAMILY,
               "ts_meta".getBytes(MockBase.ASCII())));
   final TSMeta meta = TSMeta.parseFromColumn(tsdb, column, false).joinUninterruptibly();
   assertNotNull(meta);
   assertEquals("000001000001000001", meta.getTSUID());
   assertNull(meta.getMetric());
 }
Beispiel #4
0
  @Test(expected = NoSuchUniqueId.class)
  public void parseFromColumnWithUIDMetaNSU() throws Exception {
    class ErrBack implements Callback<Object, Exception> {
      @Override
      public Object call(Exception e) throws Exception {
        Throwable ex = e;
        while (ex.getClass().equals(DeferredGroupException.class)) {
          ex = ex.getCause();
        }
        throw (Exception) ex;
      }
    }

    final KeyValue column = mock(KeyValue.class);
    when(column.key()).thenReturn(new byte[] {0, 0, 1, 0, 0, 1, 0, 0, 2});
    when(column.value())
        .thenReturn(
            ("{\"tsuid\":\"000001000001000002\",\""
                    + "description\":\"Description\",\"notes\":\"Notes\",\"created\":1328140800,"
                    + "\"custom\":null,\"units\":\"\",\"retention\":42,\"max\":1.0,\"min\":"
                    + "\"NaN\",\"displayName\":\"Display\",\"dataType\":\"Data\"}")
                .getBytes(MockBase.ASCII()));
    TSMeta.parseFromColumn(tsdb, column, true).addErrback(new ErrBack()).joinUninterruptibly();
  }
Beispiel #5
0
 public Deferred<Object> call(final Object arg) {
   return client.delete(new DeleteRequest(table, kv.key(), kv.family(), kv.qualifier()));
 }
Beispiel #6
0
  private static int fsck(
      final TSDB tsdb,
      final HBaseClient client,
      final byte[] table,
      final boolean fix,
      final String[] args)
      throws Exception {

    /** Callback to asynchronously delete a specific {@link KeyValue}. */
    final class DeleteOutOfOrder implements Callback<Deferred<Object>, Object> {

      private final KeyValue kv;

      public DeleteOutOfOrder(final KeyValue kv) {
        this.kv = kv;
      }

      public Deferred<Object> call(final Object arg) {
        return client.delete(new DeleteRequest(table, kv.key(), kv.family(), kv.qualifier()));
      }

      public String toString() {
        return "delete out-of-order data";
      }
    }

    int errors = 0;
    int correctable = 0;

    final short metric_width = width(tsdb, "metrics");
    final short name_width = width(tsdb, "tag_names");
    final short value_width = width(tsdb, "tag_values");

    final ArrayList<Query> queries = new ArrayList<Query>();
    CliQuery.parseCommandLineQuery(args, tsdb, queries, null, null);
    final StringBuilder buf = new StringBuilder();
    for (final Query query : queries) {
      final long start_time = System.nanoTime();
      long ping_start_time = start_time;
      LOG.info("Starting to fsck data covered by " + query);
      long kvcount = 0;
      long rowcount = 0;
      final Bytes.ByteMap<Seen> seen = new Bytes.ByteMap<Seen>();
      final Scanner scanner = Core.getScanner(query);
      ArrayList<ArrayList<KeyValue>> rows;
      while ((rows = scanner.nextRows().joinUninterruptibly()) != null) {
        for (final ArrayList<KeyValue> row : rows) {
          rowcount++;
          // Take a copy of the row-key because we're going to zero-out the
          // timestamp and use that as a key in our `seen' map.
          final byte[] key = row.get(0).key().clone();
          final long base_time = Bytes.getUnsignedInt(key, metric_width);
          for (int i = metric_width; i < metric_width + Const.TIMESTAMP_BYTES; i++) {
            key[i] = 0;
          }
          Seen prev = seen.get(key);
          if (prev == null) {
            prev = new Seen(base_time - 1, row.get(0));
            seen.put(key, prev);
          }
          for (final KeyValue kv : row) {
            kvcount++;
            if (kvcount % 100000 == 0) {
              final long now = System.nanoTime();
              ping_start_time = (now - ping_start_time) / 1000000;
              LOG.info(
                  "... "
                      + kvcount
                      + " KV analyzed in "
                      + ping_start_time
                      + "ms ("
                      + (100000 * 1000 / ping_start_time)
                      + " KVs/s)");
              ping_start_time = now;
            }
            if (kv.qualifier().length != 2) {
              LOG.warn(
                  "Ignoring unsupported KV with a qualifier of "
                      + kv.qualifier().length
                      + " bytes:"
                      + kv);
              continue;
            }
            final short qualifier = Bytes.getShort(kv.qualifier());
            final short delta = (short) ((qualifier & 0xFFFF) >>> FLAG_BITS);
            final long timestamp = base_time + delta;
            byte[] value = kv.value();
            if (value.length > 8) {
              errors++;
              LOG.error("Value more than 8 byte long with a 2-byte" + " qualifier.\n\t" + kv);
            }
            // TODO(tsuna): Don't hardcode 0x8 / 0x3 here.
            if ((qualifier & (0x8 | 0x3)) == (0x8 | 0x3)) { // float | 4 bytes
              // The qualifier says the value is on 4 bytes, and the value is
              // on 8 bytes, then the 4 MSBs must be 0s.  Old versions of the
              // code were doing this.  It's kinda sad.  Some versions had a
              // bug whereby the value would be sign-extended, so we can
              // detect these values and fix them here.
              if (value.length == 8) {
                if (value[0] == -1 && value[1] == -1 && value[2] == -1 && value[3] == -1) {
                  errors++;
                  correctable++;
                  if (fix) {
                    value = value.clone(); // We're going to change it.
                    value[0] = value[1] = value[2] = value[3] = 0;
                    client.put(new PutRequest(table, kv.key(), kv.family(), kv.qualifier(), value));
                  } else {
                    LOG.error(
                        "Floating point value with 0xFF most significant"
                            + " bytes, probably caused by sign extension bug"
                            + " present in revisions [96908436..607256fc].\n"
                            + "\t"
                            + kv);
                  }
                } else if (value[0] != 0 || value[1] != 0 || value[2] != 0 || value[3] != 0) {
                  errors++;
                }
              } else if (value.length != 4) {
                errors++;
                LOG.error(
                    "This floating point value must be encoded either on"
                        + " 4 or 8 bytes, but it's on "
                        + value.length
                        + " bytes.\n\t"
                        + kv);
              }
            }
            if (timestamp <= prev.timestamp()) {
              errors++;
              correctable++;
              if (fix) {
                final byte[] newkey = kv.key().clone();
                // Fix the timestamp in the row key.
                final long new_base_time = (timestamp - (timestamp % Const.MAX_TIMESPAN));
                Bytes.setInt(newkey, (int) new_base_time, metric_width);
                final short newqual =
                    (short) ((timestamp - new_base_time) << FLAG_BITS | (qualifier & FLAGS_MASK));
                final DeleteOutOfOrder delooo = new DeleteOutOfOrder(kv);
                if (timestamp < prev.timestamp()) {
                  client
                      .put(
                          new PutRequest(
                              table, newkey, kv.family(), Bytes.fromShort(newqual), value))
                      // Only delete the offending KV once we're sure that the new
                      // KV has been persisted in HBase.
                      .addCallbackDeferring(delooo);
                } else {
                  // We have two data points at exactly the same timestamp.
                  // This can happen when only the flags differ.  This is
                  // typically caused by one data point being an integer and
                  // the other being a floating point value.  In this case
                  // we just delete the duplicate data point and keep the
                  // first one we saw.
                  delooo.call(null);
                }
              } else {
                buf.setLength(0);
                buf.append(
                        timestamp < prev.timestamp()
                            ? "Out of order data.\n\t"
                            : "Duplicate data point with different flags.\n\t")
                    .append(timestamp)
                    .append(" (")
                    .append(DumpSeries.date(timestamp))
                    .append(") @ ")
                    .append(kv)
                    .append("\n\t");
                DumpSeries.formatKeyValue(buf, tsdb, kv, base_time);
                buf.append("\n\t  was found after\n\t")
                    .append(prev.timestamp)
                    .append(" (")
                    .append(DumpSeries.date(prev.timestamp))
                    .append(") @ ")
                    .append(prev.kv)
                    .append("\n\t");
                DumpSeries.formatKeyValue(
                    buf, tsdb, prev.kv, Bytes.getUnsignedInt(prev.kv.key(), metric_width));
                LOG.error(buf.toString());
              }
            } else {
              prev.setTimestamp(timestamp);
              prev.kv = kv;
            }
          }
        }
      }
      final long timing = (System.nanoTime() - start_time) / 1000000;
      System.out.println(
          kvcount
              + " KVs (in "
              + rowcount
              + " rows) analyzed in "
              + timing
              + "ms (~"
              + (kvcount * 1000 / timing)
              + " KV/s)");
    }

    System.out.println(errors != 0 ? "Found " + errors + " errors." : "No error found.");
    if (!fix && correctable > 0) {
      System.out.println(
          correctable
              + " of these errors are automatically"
              + " correctable, re-run with --fix.\n"
              + "Make sure you understand the errors above and you"
              + " know what you're doing before using --fix.");
    }
    return errors;
  }