Search in sources :

Example 1 with PRow

use of org.apache.phoenix.schema.PRow in project phoenix by apache.

the class PhoenixIndexBuilder method executeAtomicOp.

@Override
public List<Mutation> executeAtomicOp(Increment inc) throws IOException {
    byte[] opBytes = inc.getAttribute(ATOMIC_OP_ATTRIB);
    if (opBytes == null) {
        // Unexpected
        return null;
    }
    inc.setAttribute(ATOMIC_OP_ATTRIB, null);
    Put put = null;
    Delete delete = null;
    // We cannot neither use the time stamp in the Increment to set the Get time range
    // nor set the Put/Delete time stamp and have this be atomic as HBase does not
    // handle that. Though we disallow using ON DUPLICATE KEY clause when the
    // CURRENT_SCN is set, we still may have a time stamp set as of when the table
    // was resolved on the client side. We need to ignore this as well due to limitations
    // in HBase, but this isn't too bad as the time will be very close the the current
    // time anyway.
    long ts = HConstants.LATEST_TIMESTAMP;
    byte[] rowKey = inc.getRow();
    final Get get = new Get(rowKey);
    if (isDupKeyIgnore(opBytes)) {
        get.setFilter(new FirstKeyOnlyFilter());
        Result result = this.env.getRegion().get(get);
        return result.isEmpty() ? convertIncrementToPutInSingletonList(inc) : Collections.<Mutation>emptyList();
    }
    ByteArrayInputStream stream = new ByteArrayInputStream(opBytes);
    DataInputStream input = new DataInputStream(stream);
    boolean skipFirstOp = input.readBoolean();
    short repeat = input.readShort();
    final int[] estimatedSizeHolder = { 0 };
    List<Pair<PTable, List<Expression>>> operations = Lists.newArrayListWithExpectedSize(3);
    while (true) {
        ExpressionVisitor<Void> visitor = new StatelessTraverseAllExpressionVisitor<Void>() {

            @Override
            public Void visit(KeyValueColumnExpression expression) {
                get.addColumn(expression.getColumnFamily(), expression.getColumnQualifier());
                estimatedSizeHolder[0]++;
                return null;
            }
        };
        try {
            int nExpressions = WritableUtils.readVInt(input);
            List<Expression> expressions = Lists.newArrayListWithExpectedSize(nExpressions);
            for (int i = 0; i < nExpressions; i++) {
                Expression expression = ExpressionType.values()[WritableUtils.readVInt(input)].newInstance();
                expression.readFields(input);
                expressions.add(expression);
                expression.accept(visitor);
            }
            PTableProtos.PTable tableProto = PTableProtos.PTable.parseDelimitedFrom(input);
            PTable table = PTableImpl.createFromProto(tableProto);
            operations.add(new Pair<>(table, expressions));
        } catch (EOFException e) {
            break;
        }
    }
    int estimatedSize = estimatedSizeHolder[0];
    if (get.getFamilyMap().isEmpty()) {
        get.setFilter(new FirstKeyOnlyFilter());
    }
    MultiKeyValueTuple tuple;
    List<Cell> flattenedCells = null;
    List<Cell> cells = ((HRegion) this.env.getRegion()).get(get, false);
    if (cells.isEmpty()) {
        if (skipFirstOp) {
            if (operations.size() <= 1 && repeat <= 1) {
                return convertIncrementToPutInSingletonList(inc);
            }
            // Skip first operation (if first wasn't ON DUPLICATE KEY IGNORE)
            repeat--;
        }
        // Base current state off of new row
        flattenedCells = flattenCells(inc, estimatedSize);
        tuple = new MultiKeyValueTuple(flattenedCells);
    } else {
        // Base current state off of existing row
        tuple = new MultiKeyValueTuple(cells);
    }
    ImmutableBytesWritable ptr = new ImmutableBytesWritable();
    for (int opIndex = 0; opIndex < operations.size(); opIndex++) {
        Pair<PTable, List<Expression>> operation = operations.get(opIndex);
        PTable table = operation.getFirst();
        List<Expression> expressions = operation.getSecond();
        for (int j = 0; j < repeat; j++) {
            // repeater loop
            ptr.set(rowKey);
            // executed, not when the outer loop is exited. Hence we do it here, at the top of the loop.
            if (flattenedCells != null) {
                Collections.sort(flattenedCells, KeyValue.COMPARATOR);
            }
            PRow row = table.newRow(GenericKeyValueBuilder.INSTANCE, ts, ptr, false);
            for (int i = 0; i < expressions.size(); i++) {
                Expression expression = expressions.get(i);
                ptr.set(ByteUtil.EMPTY_BYTE_ARRAY);
                expression.evaluate(tuple, ptr);
                PColumn column = table.getColumns().get(i + 1);
                Object value = expression.getDataType().toObject(ptr, column.getSortOrder());
                // same type.
                if (!column.getDataType().isSizeCompatible(ptr, value, column.getDataType(), expression.getSortOrder(), expression.getMaxLength(), expression.getScale(), column.getMaxLength(), column.getScale())) {
                    throw new DataExceedsCapacityException(column.getDataType(), column.getMaxLength(), column.getScale());
                }
                column.getDataType().coerceBytes(ptr, value, expression.getDataType(), expression.getMaxLength(), expression.getScale(), expression.getSortOrder(), column.getMaxLength(), column.getScale(), column.getSortOrder(), table.rowKeyOrderOptimizable());
                byte[] bytes = ByteUtil.copyKeyBytesIfNecessary(ptr);
                row.setValue(column, bytes);
            }
            flattenedCells = Lists.newArrayListWithExpectedSize(estimatedSize);
            List<Mutation> mutations = row.toRowMutations();
            for (Mutation source : mutations) {
                flattenCells(source, flattenedCells);
            }
            tuple.setKeyValues(flattenedCells);
        }
        // Repeat only applies to first statement
        repeat = 1;
    }
    List<Mutation> mutations = Lists.newArrayListWithExpectedSize(2);
    for (int i = 0; i < tuple.size(); i++) {
        Cell cell = tuple.getValue(i);
        if (Type.codeToType(cell.getTypeByte()) == Type.Put) {
            if (put == null) {
                put = new Put(rowKey);
                transferAttributes(inc, put);
                mutations.add(put);
            }
            put.add(cell);
        } else {
            if (delete == null) {
                delete = new Delete(rowKey);
                transferAttributes(inc, delete);
                mutations.add(delete);
            }
            delete.addDeleteMarker(cell);
        }
    }
    return mutations;
}
Also used : Delete(org.apache.hadoop.hbase.client.Delete) PTable(org.apache.phoenix.schema.PTable) Result(org.apache.hadoop.hbase.client.Result) PRow(org.apache.phoenix.schema.PRow) PColumn(org.apache.phoenix.schema.PColumn) StatelessTraverseAllExpressionVisitor(org.apache.phoenix.expression.visitor.StatelessTraverseAllExpressionVisitor) EOFException(java.io.EOFException) List(java.util.List) KeyValueColumnExpression(org.apache.phoenix.expression.KeyValueColumnExpression) Cell(org.apache.hadoop.hbase.Cell) Pair(org.apache.hadoop.hbase.util.Pair) ImmutableBytesWritable(org.apache.hadoop.hbase.io.ImmutableBytesWritable) FirstKeyOnlyFilter(org.apache.hadoop.hbase.filter.FirstKeyOnlyFilter) DataInputStream(java.io.DataInputStream) Put(org.apache.hadoop.hbase.client.Put) PTableProtos(org.apache.phoenix.coprocessor.generated.PTableProtos) HRegion(org.apache.hadoop.hbase.regionserver.HRegion) DataExceedsCapacityException(org.apache.phoenix.exception.DataExceedsCapacityException) ByteArrayInputStream(java.io.ByteArrayInputStream) KeyValueColumnExpression(org.apache.phoenix.expression.KeyValueColumnExpression) Expression(org.apache.phoenix.expression.Expression) Get(org.apache.hadoop.hbase.client.Get) MultiKeyValueTuple(org.apache.phoenix.schema.tuple.MultiKeyValueTuple) Mutation(org.apache.hadoop.hbase.client.Mutation)

Example 2 with PRow

use of org.apache.phoenix.schema.PRow in project phoenix by apache.

the class IndexTestUtil method generateIndexData.

public static List<Mutation> generateIndexData(PTable indexTable, PTable dataTable, Mutation dataMutation, ImmutableBytesWritable ptr, KeyValueBuilder builder) throws SQLException {
    byte[] dataRowKey = dataMutation.getRow();
    RowKeySchema dataRowKeySchema = dataTable.getRowKeySchema();
    List<PColumn> dataPKColumns = dataTable.getPKColumns();
    int i = 0;
    int indexOffset = 0;
    Boolean hasValue;
    // Skip salt column
    int maxOffset = dataRowKey.length;
    dataRowKeySchema.iterator(dataRowKey, ptr, dataTable.getBucketNum() == null ? i : ++i);
    List<PColumn> indexPKColumns = indexTable.getPKColumns();
    List<PColumn> indexColumns = indexTable.getColumns();
    int nIndexColumns = indexPKColumns.size();
    int maxIndexValues = indexColumns.size() - nIndexColumns - indexOffset;
    BitSet indexValuesSet = new BitSet(maxIndexValues);
    byte[][] indexValues = new byte[indexColumns.size() - indexOffset][];
    while ((hasValue = dataRowKeySchema.next(ptr, i, maxOffset)) != null) {
        if (hasValue) {
            PColumn dataColumn = dataPKColumns.get(i);
            PColumn indexColumn = indexTable.getColumnForColumnName(IndexUtil.getIndexColumnName(dataColumn));
            coerceDataValueToIndexValue(dataColumn, indexColumn, ptr);
            indexValues[indexColumn.getPosition() - indexOffset] = ptr.copyBytes();
        }
        i++;
    }
    PRow row;
    long ts = MetaDataUtil.getClientTimeStamp(dataMutation);
    if (dataMutation instanceof Delete && dataMutation.getFamilyCellMap().values().isEmpty()) {
        indexTable.newKey(ptr, indexValues);
        row = indexTable.newRow(builder, ts, ptr, false);
        row.delete();
    } else {
        // If no column families in table, then nothing to look for 
        if (!dataTable.getColumnFamilies().isEmpty()) {
            for (Map.Entry<byte[], List<Cell>> entry : dataMutation.getFamilyCellMap().entrySet()) {
                PColumnFamily family = dataTable.getColumnFamily(entry.getKey());
                for (Cell kv : entry.getValue()) {
                    @SuppressWarnings("deprecation") byte[] cq = kv.getQualifier();
                    byte[] emptyKVQualifier = EncodedColumnsUtil.getEmptyKeyValueInfo(dataTable).getFirst();
                    if (Bytes.compareTo(emptyKVQualifier, cq) != 0) {
                        try {
                            PColumn dataColumn = family.getPColumnForColumnQualifier(cq);
                            PColumn indexColumn = indexTable.getColumnForColumnName(IndexUtil.getIndexColumnName(family.getName().getString(), dataColumn.getName().getString()));
                            ptr.set(kv.getValueArray(), kv.getValueOffset(), kv.getValueLength());
                            coerceDataValueToIndexValue(dataColumn, indexColumn, ptr);
                            indexValues[indexPKColumns.indexOf(indexColumn) - indexOffset] = ptr.copyBytes();
                            if (!SchemaUtil.isPKColumn(indexColumn)) {
                                indexValuesSet.set(indexColumn.getPosition() - nIndexColumns - indexOffset);
                            }
                        } catch (ColumnNotFoundException e) {
                        // Ignore as this means that the data column isn't in the index
                        }
                    }
                }
            }
        }
        indexTable.newKey(ptr, indexValues);
        row = indexTable.newRow(builder, ts, ptr, false);
        int pos = 0;
        while ((pos = indexValuesSet.nextSetBit(pos)) >= 0) {
            int index = nIndexColumns + indexOffset + pos++;
            PColumn indexColumn = indexColumns.get(index);
            row.setValue(indexColumn, indexValues[index]);
        }
    }
    return row.toRowMutations();
}
Also used : Delete(org.apache.hadoop.hbase.client.Delete) BitSet(java.util.BitSet) RowKeySchema(org.apache.phoenix.schema.RowKeySchema) PColumnFamily(org.apache.phoenix.schema.PColumnFamily) PRow(org.apache.phoenix.schema.PRow) PColumn(org.apache.phoenix.schema.PColumn) ColumnNotFoundException(org.apache.phoenix.schema.ColumnNotFoundException) List(java.util.List) Map(java.util.Map) Cell(org.apache.hadoop.hbase.Cell)

Example 3 with PRow

use of org.apache.phoenix.schema.PRow in project phoenix by apache.

the class UngroupedAggregateRegionObserver method doPostScannerOpen.

@Override
protected RegionScanner doPostScannerOpen(final ObserverContext<RegionCoprocessorEnvironment> c, final Scan scan, final RegionScanner s) throws IOException, SQLException {
    RegionCoprocessorEnvironment env = c.getEnvironment();
    Region region = env.getRegion();
    long ts = scan.getTimeRange().getMax();
    boolean localIndexScan = ScanUtil.isLocalIndex(scan);
    if (ScanUtil.isAnalyzeTable(scan)) {
        byte[] gp_width_bytes = scan.getAttribute(BaseScannerRegionObserver.GUIDEPOST_WIDTH_BYTES);
        byte[] gp_per_region_bytes = scan.getAttribute(BaseScannerRegionObserver.GUIDEPOST_PER_REGION);
        // Let this throw, as this scan is being done for the sole purpose of collecting stats
        StatisticsCollector statsCollector = StatisticsCollectorFactory.createStatisticsCollector(env, region.getRegionInfo().getTable().getNameAsString(), ts, gp_width_bytes, gp_per_region_bytes);
        return collectStats(s, statsCollector, region, scan, env.getConfiguration());
    } else if (ScanUtil.isIndexRebuild(scan)) {
        return rebuildIndices(s, region, scan, env.getConfiguration());
    }
    int offsetToBe = 0;
    if (localIndexScan) {
        /*
             * For local indexes, we need to set an offset on row key expressions to skip
             * the region start key.
             */
        offsetToBe = region.getRegionInfo().getStartKey().length != 0 ? region.getRegionInfo().getStartKey().length : region.getRegionInfo().getEndKey().length;
        ScanUtil.setRowKeyOffset(scan, offsetToBe);
    }
    final int offset = offsetToBe;
    PTable projectedTable = null;
    PTable writeToTable = null;
    byte[][] values = null;
    byte[] descRowKeyTableBytes = scan.getAttribute(UPGRADE_DESC_ROW_KEY);
    boolean isDescRowKeyOrderUpgrade = descRowKeyTableBytes != null;
    if (isDescRowKeyOrderUpgrade) {
        logger.debug("Upgrading row key for " + region.getRegionInfo().getTable().getNameAsString());
        projectedTable = deserializeTable(descRowKeyTableBytes);
        try {
            writeToTable = PTableImpl.makePTable(projectedTable, true);
        } catch (SQLException e) {
            // Impossible
            ServerUtil.throwIOException("Upgrade failed", e);
        }
        values = new byte[projectedTable.getPKColumns().size()][];
    }
    boolean useProto = false;
    byte[] localIndexBytes = scan.getAttribute(LOCAL_INDEX_BUILD_PROTO);
    useProto = localIndexBytes != null;
    if (localIndexBytes == null) {
        localIndexBytes = scan.getAttribute(LOCAL_INDEX_BUILD);
    }
    List<IndexMaintainer> indexMaintainers = localIndexBytes == null ? null : IndexMaintainer.deserialize(localIndexBytes, useProto);
    MutationList indexMutations = localIndexBytes == null ? new MutationList() : new MutationList(1024);
    RegionScanner theScanner = s;
    boolean replayMutations = scan.getAttribute(BaseScannerRegionObserver.IGNORE_NEWER_MUTATIONS) != null;
    byte[] indexUUID = scan.getAttribute(PhoenixIndexCodec.INDEX_UUID);
    byte[] txState = scan.getAttribute(BaseScannerRegionObserver.TX_STATE);
    List<Expression> selectExpressions = null;
    byte[] upsertSelectTable = scan.getAttribute(BaseScannerRegionObserver.UPSERT_SELECT_TABLE);
    boolean isUpsert = false;
    boolean isDelete = false;
    byte[] deleteCQ = null;
    byte[] deleteCF = null;
    byte[] emptyCF = null;
    HTable targetHTable = null;
    boolean areMutationInSameRegion = true;
    ImmutableBytesWritable ptr = new ImmutableBytesWritable();
    if (upsertSelectTable != null) {
        isUpsert = true;
        projectedTable = deserializeTable(upsertSelectTable);
        targetHTable = new HTable(env.getConfiguration(), projectedTable.getPhysicalName().getBytes());
        selectExpressions = deserializeExpressions(scan.getAttribute(BaseScannerRegionObserver.UPSERT_SELECT_EXPRS));
        values = new byte[projectedTable.getPKColumns().size()][];
        areMutationInSameRegion = Bytes.compareTo(targetHTable.getTableName(), region.getTableDesc().getTableName().getName()) == 0 && !ExpressionUtil.isPkPositionChanging(new TableRef(projectedTable), selectExpressions);
    } else {
        byte[] isDeleteAgg = scan.getAttribute(BaseScannerRegionObserver.DELETE_AGG);
        isDelete = isDeleteAgg != null && Bytes.compareTo(PDataType.TRUE_BYTES, isDeleteAgg) == 0;
        if (!isDelete) {
            deleteCF = scan.getAttribute(BaseScannerRegionObserver.DELETE_CF);
            deleteCQ = scan.getAttribute(BaseScannerRegionObserver.DELETE_CQ);
        }
        emptyCF = scan.getAttribute(BaseScannerRegionObserver.EMPTY_CF);
    }
    TupleProjector tupleProjector = null;
    byte[][] viewConstants = null;
    ColumnReference[] dataColumns = IndexUtil.deserializeDataTableColumnsToJoin(scan);
    final TupleProjector p = TupleProjector.deserializeProjectorFromScan(scan);
    final HashJoinInfo j = HashJoinInfo.deserializeHashJoinFromScan(scan);
    boolean useQualifierAsIndex = EncodedColumnsUtil.useQualifierAsIndex(EncodedColumnsUtil.getMinMaxQualifiersFromScan(scan));
    if ((localIndexScan && !isDelete && !isDescRowKeyOrderUpgrade) || (j == null && p != null)) {
        if (dataColumns != null) {
            tupleProjector = IndexUtil.getTupleProjector(scan, dataColumns);
            viewConstants = IndexUtil.deserializeViewConstantsFromScan(scan);
        }
        ImmutableBytesWritable tempPtr = new ImmutableBytesWritable();
        theScanner = getWrappedScanner(c, theScanner, offset, scan, dataColumns, tupleProjector, region, indexMaintainers == null ? null : indexMaintainers.get(0), viewConstants, p, tempPtr, useQualifierAsIndex);
    }
    if (j != null) {
        theScanner = new HashJoinRegionScanner(theScanner, p, j, ScanUtil.getTenantId(scan), env, useQualifierAsIndex, useNewValueColumnQualifier);
    }
    int maxBatchSize = 0;
    long maxBatchSizeBytes = 0L;
    MutationList mutations = new MutationList();
    boolean needToWrite = false;
    Configuration conf = c.getEnvironment().getConfiguration();
    long flushSize = region.getTableDesc().getMemStoreFlushSize();
    if (flushSize <= 0) {
        flushSize = conf.getLong(HConstants.HREGION_MEMSTORE_FLUSH_SIZE, HTableDescriptor.DEFAULT_MEMSTORE_FLUSH_SIZE);
    }
    /**
         * Slow down the writes if the memstore size more than
         * (hbase.hregion.memstore.block.multiplier - 1) times hbase.hregion.memstore.flush.size
         * bytes. This avoids flush storm to hdfs for cases like index building where reads and
         * write happen to all the table regions in the server.
         */
    final long blockingMemStoreSize = flushSize * (conf.getLong(HConstants.HREGION_MEMSTORE_BLOCK_MULTIPLIER, HConstants.DEFAULT_HREGION_MEMSTORE_BLOCK_MULTIPLIER) - 1);
    boolean buildLocalIndex = indexMaintainers != null && dataColumns == null && !localIndexScan;
    if (isDescRowKeyOrderUpgrade || isDelete || isUpsert || (deleteCQ != null && deleteCF != null) || emptyCF != null || buildLocalIndex) {
        needToWrite = true;
        maxBatchSize = env.getConfiguration().getInt(MUTATE_BATCH_SIZE_ATTRIB, QueryServicesOptions.DEFAULT_MUTATE_BATCH_SIZE);
        mutations = new MutationList(Ints.saturatedCast(maxBatchSize + maxBatchSize / 10));
        maxBatchSizeBytes = env.getConfiguration().getLong(MUTATE_BATCH_SIZE_BYTES_ATTRIB, QueryServicesOptions.DEFAULT_MUTATE_BATCH_SIZE_BYTES);
    }
    Aggregators aggregators = ServerAggregators.deserialize(scan.getAttribute(BaseScannerRegionObserver.AGGREGATORS), env.getConfiguration());
    Aggregator[] rowAggregators = aggregators.getAggregators();
    boolean hasMore;
    boolean hasAny = false;
    Pair<Integer, Integer> minMaxQualifiers = EncodedColumnsUtil.getMinMaxQualifiersFromScan(scan);
    Tuple result = useQualifierAsIndex ? new PositionBasedMultiKeyValueTuple() : new MultiKeyValueTuple();
    if (logger.isDebugEnabled()) {
        logger.debug(LogUtil.addCustomAnnotations("Starting ungrouped coprocessor scan " + scan + " " + region.getRegionInfo(), ScanUtil.getCustomAnnotations(scan)));
    }
    int rowCount = 0;
    final RegionScanner innerScanner = theScanner;
    boolean useIndexProto = true;
    byte[] indexMaintainersPtr = scan.getAttribute(PhoenixIndexCodec.INDEX_PROTO_MD);
    // for backward compatiblity fall back to look by the old attribute
    if (indexMaintainersPtr == null) {
        indexMaintainersPtr = scan.getAttribute(PhoenixIndexCodec.INDEX_MD);
        useIndexProto = false;
    }
    boolean acquiredLock = false;
    try {
        if (needToWrite) {
            synchronized (lock) {
                scansReferenceCount++;
            }
        }
        region.startRegionOperation();
        acquiredLock = true;
        synchronized (innerScanner) {
            do {
                List<Cell> results = useQualifierAsIndex ? new EncodedColumnQualiferCellsList(minMaxQualifiers.getFirst(), minMaxQualifiers.getSecond(), encodingScheme) : new ArrayList<Cell>();
                // Results are potentially returned even when the return value of s.next is false
                // since this is an indication of whether or not there are more values after the
                // ones returned
                hasMore = innerScanner.nextRaw(results);
                if (!results.isEmpty()) {
                    rowCount++;
                    result.setKeyValues(results);
                    if (isDescRowKeyOrderUpgrade) {
                        Arrays.fill(values, null);
                        Cell firstKV = results.get(0);
                        RowKeySchema schema = projectedTable.getRowKeySchema();
                        int maxOffset = schema.iterator(firstKV.getRowArray(), firstKV.getRowOffset() + offset, firstKV.getRowLength(), ptr);
                        for (int i = 0; i < schema.getFieldCount(); i++) {
                            Boolean hasValue = schema.next(ptr, i, maxOffset);
                            if (hasValue == null) {
                                break;
                            }
                            Field field = schema.getField(i);
                            if (field.getSortOrder() == SortOrder.DESC) {
                                // Special case for re-writing DESC ARRAY, as the actual byte value needs to change in this case
                                if (field.getDataType().isArrayType()) {
                                    field.getDataType().coerceBytes(ptr, null, field.getDataType(), field.getMaxLength(), field.getScale(), field.getSortOrder(), field.getMaxLength(), field.getScale(), field.getSortOrder(), // force to use correct separator byte
                                    true);
                                } else // Special case for re-writing DESC CHAR or DESC BINARY, to force the re-writing of trailing space characters
                                if (field.getDataType() == PChar.INSTANCE || field.getDataType() == PBinary.INSTANCE) {
                                    int len = ptr.getLength();
                                    while (len > 0 && ptr.get()[ptr.getOffset() + len - 1] == StringUtil.SPACE_UTF8) {
                                        len--;
                                    }
                                    ptr.set(ptr.get(), ptr.getOffset(), len);
                                // Special case for re-writing DESC FLOAT and DOUBLE, as they're not inverted like they should be (PHOENIX-2171)
                                } else if (field.getDataType() == PFloat.INSTANCE || field.getDataType() == PDouble.INSTANCE) {
                                    byte[] invertedBytes = SortOrder.invert(ptr.get(), ptr.getOffset(), ptr.getLength());
                                    ptr.set(invertedBytes);
                                }
                            } else if (field.getDataType() == PBinary.INSTANCE) {
                                // Remove trailing space characters so that the setValues call below will replace them
                                // with the correct zero byte character. Note this is somewhat dangerous as these
                                // could be legit, but I don't know what the alternative is.
                                int len = ptr.getLength();
                                while (len > 0 && ptr.get()[ptr.getOffset() + len - 1] == StringUtil.SPACE_UTF8) {
                                    len--;
                                }
                                ptr.set(ptr.get(), ptr.getOffset(), len);
                            }
                            values[i] = ptr.copyBytes();
                        }
                        writeToTable.newKey(ptr, values);
                        if (Bytes.compareTo(firstKV.getRowArray(), firstKV.getRowOffset() + offset, firstKV.getRowLength(), ptr.get(), ptr.getOffset() + offset, ptr.getLength()) == 0) {
                            continue;
                        }
                        byte[] newRow = ByteUtil.copyKeyBytesIfNecessary(ptr);
                        if (offset > 0) {
                            // for local indexes (prepend region start key)
                            byte[] newRowWithOffset = new byte[offset + newRow.length];
                            System.arraycopy(firstKV.getRowArray(), firstKV.getRowOffset(), newRowWithOffset, 0, offset);
                            ;
                            System.arraycopy(newRow, 0, newRowWithOffset, offset, newRow.length);
                            newRow = newRowWithOffset;
                        }
                        byte[] oldRow = Bytes.copy(firstKV.getRowArray(), firstKV.getRowOffset(), firstKV.getRowLength());
                        for (Cell cell : results) {
                            // Copy existing cell but with new row key
                            Cell newCell = new KeyValue(newRow, 0, newRow.length, cell.getFamilyArray(), cell.getFamilyOffset(), cell.getFamilyLength(), cell.getQualifierArray(), cell.getQualifierOffset(), cell.getQualifierLength(), cell.getTimestamp(), KeyValue.Type.codeToType(cell.getTypeByte()), cell.getValueArray(), cell.getValueOffset(), cell.getValueLength());
                            switch(KeyValue.Type.codeToType(cell.getTypeByte())) {
                                case Put:
                                    // If Put, point delete old Put
                                    Delete del = new Delete(oldRow);
                                    del.addDeleteMarker(new KeyValue(cell.getRowArray(), cell.getRowOffset(), cell.getRowLength(), cell.getFamilyArray(), cell.getFamilyOffset(), cell.getFamilyLength(), cell.getQualifierArray(), cell.getQualifierOffset(), cell.getQualifierLength(), cell.getTimestamp(), KeyValue.Type.Delete, ByteUtil.EMPTY_BYTE_ARRAY, 0, 0));
                                    mutations.add(del);
                                    Put put = new Put(newRow);
                                    put.add(newCell);
                                    mutations.add(put);
                                    break;
                                case Delete:
                                case DeleteColumn:
                                case DeleteFamily:
                                case DeleteFamilyVersion:
                                    Delete delete = new Delete(newRow);
                                    delete.addDeleteMarker(newCell);
                                    mutations.add(delete);
                                    break;
                            }
                        }
                    } else if (buildLocalIndex) {
                        for (IndexMaintainer maintainer : indexMaintainers) {
                            if (!results.isEmpty()) {
                                result.getKey(ptr);
                                ValueGetter valueGetter = maintainer.createGetterFromKeyValues(ImmutableBytesPtr.copyBytesIfNecessary(ptr), results);
                                Put put = maintainer.buildUpdateMutation(kvBuilder, valueGetter, ptr, results.get(0).getTimestamp(), env.getRegion().getRegionInfo().getStartKey(), env.getRegion().getRegionInfo().getEndKey());
                                indexMutations.add(put);
                            }
                        }
                        result.setKeyValues(results);
                    } else if (isDelete) {
                        // FIXME: the version of the Delete constructor without the lock
                        // args was introduced in 0.94.4, thus if we try to use it here
                        // we can no longer use the 0.94.2 version of the client.
                        Cell firstKV = results.get(0);
                        Delete delete = new Delete(firstKV.getRowArray(), firstKV.getRowOffset(), firstKV.getRowLength(), ts);
                        if (replayMutations) {
                            delete.setAttribute(IGNORE_NEWER_MUTATIONS, PDataType.TRUE_BYTES);
                        }
                        mutations.add(delete);
                        // force tephra to ignore this deletes
                        delete.setAttribute(TxConstants.TX_ROLLBACK_ATTRIBUTE_KEY, new byte[0]);
                    } else if (isUpsert) {
                        Arrays.fill(values, null);
                        int bucketNumOffset = 0;
                        if (projectedTable.getBucketNum() != null) {
                            values[0] = new byte[] { 0 };
                            bucketNumOffset = 1;
                        }
                        int i = bucketNumOffset;
                        List<PColumn> projectedColumns = projectedTable.getColumns();
                        for (; i < projectedTable.getPKColumns().size(); i++) {
                            Expression expression = selectExpressions.get(i - bucketNumOffset);
                            if (expression.evaluate(result, ptr)) {
                                values[i] = ptr.copyBytes();
                                // column being projected into then invert the bits.
                                if (expression.getSortOrder() != projectedColumns.get(i).getSortOrder()) {
                                    SortOrder.invert(values[i], 0, values[i], 0, values[i].length);
                                }
                            } else {
                                values[i] = ByteUtil.EMPTY_BYTE_ARRAY;
                            }
                        }
                        projectedTable.newKey(ptr, values);
                        PRow row = projectedTable.newRow(kvBuilder, ts, ptr, false);
                        for (; i < projectedColumns.size(); i++) {
                            Expression expression = selectExpressions.get(i - bucketNumOffset);
                            if (expression.evaluate(result, ptr)) {
                                PColumn column = projectedColumns.get(i);
                                if (!column.getDataType().isSizeCompatible(ptr, null, expression.getDataType(), expression.getSortOrder(), expression.getMaxLength(), expression.getScale(), column.getMaxLength(), column.getScale())) {
                                    throw new DataExceedsCapacityException(column.getDataType(), column.getMaxLength(), column.getScale(), column.getName().getString(), ptr);
                                }
                                column.getDataType().coerceBytes(ptr, null, expression.getDataType(), expression.getMaxLength(), expression.getScale(), expression.getSortOrder(), column.getMaxLength(), column.getScale(), column.getSortOrder(), projectedTable.rowKeyOrderOptimizable());
                                byte[] bytes = ByteUtil.copyKeyBytesIfNecessary(ptr);
                                row.setValue(column, bytes);
                            }
                        }
                        for (Mutation mutation : row.toRowMutations()) {
                            if (replayMutations) {
                                mutation.setAttribute(IGNORE_NEWER_MUTATIONS, PDataType.TRUE_BYTES);
                            }
                            mutations.add(mutation);
                        }
                        for (i = 0; i < selectExpressions.size(); i++) {
                            selectExpressions.get(i).reset();
                        }
                    } else if (deleteCF != null && deleteCQ != null) {
                        // if no empty key value is being set
                        if (emptyCF == null || result.getValue(deleteCF, deleteCQ) != null) {
                            Delete delete = new Delete(results.get(0).getRowArray(), results.get(0).getRowOffset(), results.get(0).getRowLength());
                            delete.deleteColumns(deleteCF, deleteCQ, ts);
                            // force tephra to ignore this deletes
                            delete.setAttribute(TxConstants.TX_ROLLBACK_ATTRIBUTE_KEY, new byte[0]);
                            mutations.add(delete);
                        }
                    }
                    if (emptyCF != null) {
                        /*
                             * If we've specified an emptyCF, then we need to insert an empty
                             * key value "retroactively" for any key value that is visible at
                             * the timestamp that the DDL was issued. Key values that are not
                             * visible at this timestamp will not ever be projected up to
                             * scans past this timestamp, so don't need to be considered.
                             * We insert one empty key value per row per timestamp.
                             */
                        Set<Long> timeStamps = Sets.newHashSetWithExpectedSize(results.size());
                        for (Cell kv : results) {
                            long kvts = kv.getTimestamp();
                            if (!timeStamps.contains(kvts)) {
                                Put put = new Put(kv.getRowArray(), kv.getRowOffset(), kv.getRowLength());
                                put.add(emptyCF, QueryConstants.EMPTY_COLUMN_BYTES, kvts, ByteUtil.EMPTY_BYTE_ARRAY);
                                mutations.add(put);
                            }
                        }
                    }
                    if (readyToCommit(rowCount, mutations.byteSize(), maxBatchSize, maxBatchSizeBytes)) {
                        commit(region, mutations, indexUUID, blockingMemStoreSize, indexMaintainersPtr, txState, areMutationInSameRegion, targetHTable, useIndexProto);
                        mutations.clear();
                    }
                    if (readyToCommit(rowCount, indexMutations.byteSize(), maxBatchSize, maxBatchSizeBytes)) {
                        commitBatch(region, indexMutations, null, blockingMemStoreSize, null, txState, useIndexProto);
                        indexMutations.clear();
                    }
                    aggregators.aggregate(rowAggregators, result);
                    hasAny = true;
                }
            } while (hasMore);
            if (!mutations.isEmpty()) {
                commit(region, mutations, indexUUID, blockingMemStoreSize, indexMaintainersPtr, txState, areMutationInSameRegion, targetHTable, useIndexProto);
                mutations.clear();
            }
            if (!indexMutations.isEmpty()) {
                commitBatch(region, indexMutations, null, blockingMemStoreSize, indexMaintainersPtr, txState, useIndexProto);
                indexMutations.clear();
            }
        }
    } finally {
        if (needToWrite) {
            synchronized (lock) {
                scansReferenceCount--;
            }
        }
        if (targetHTable != null) {
            targetHTable.close();
        }
        try {
            innerScanner.close();
        } finally {
            if (acquiredLock)
                region.closeRegionOperation();
        }
    }
    if (logger.isDebugEnabled()) {
        logger.debug(LogUtil.addCustomAnnotations("Finished scanning " + rowCount + " rows for ungrouped coprocessor scan " + scan, ScanUtil.getCustomAnnotations(scan)));
    }
    final boolean hadAny = hasAny;
    KeyValue keyValue = null;
    if (hadAny) {
        byte[] value = aggregators.toBytes(rowAggregators);
        keyValue = KeyValueUtil.newKeyValue(UNGROUPED_AGG_ROW_KEY, SINGLE_COLUMN_FAMILY, SINGLE_COLUMN, AGG_TIMESTAMP, value, 0, value.length);
    }
    final KeyValue aggKeyValue = keyValue;
    RegionScanner scanner = new BaseRegionScanner(innerScanner) {

        private boolean done = !hadAny;

        @Override
        public boolean isFilterDone() {
            return done;
        }

        @Override
        public boolean next(List<Cell> results) throws IOException {
            if (done)
                return false;
            done = true;
            results.add(aggKeyValue);
            return false;
        }

        @Override
        public long getMaxResultSize() {
            return scan.getMaxResultSize();
        }
    };
    return scanner;
}
Also used : Delete(org.apache.hadoop.hbase.client.Delete) Configuration(org.apache.hadoop.conf.Configuration) TupleProjector(org.apache.phoenix.execute.TupleProjector) PTable(org.apache.phoenix.schema.PTable) ValueGetter(org.apache.phoenix.hbase.index.ValueGetter) EncodedColumnQualiferCellsList(org.apache.phoenix.schema.tuple.EncodedColumnQualiferCellsList) ArrayList(java.util.ArrayList) List(java.util.List) Cell(org.apache.hadoop.hbase.Cell) ImmutableBytesWritable(org.apache.hadoop.hbase.io.ImmutableBytesWritable) DataExceedsCapacityException(org.apache.phoenix.exception.DataExceedsCapacityException) PositionBasedMultiKeyValueTuple(org.apache.phoenix.schema.tuple.PositionBasedMultiKeyValueTuple) Aggregators(org.apache.phoenix.expression.aggregator.Aggregators) ServerAggregators(org.apache.phoenix.expression.aggregator.ServerAggregators) PLong(org.apache.phoenix.schema.types.PLong) Region(org.apache.hadoop.hbase.regionserver.Region) Mutation(org.apache.hadoop.hbase.client.Mutation) ColumnReference(org.apache.phoenix.hbase.index.covered.update.ColumnReference) KeyValue(org.apache.hadoop.hbase.KeyValue) StatisticsCollector(org.apache.phoenix.schema.stats.StatisticsCollector) SQLException(java.sql.SQLException) HTable(org.apache.hadoop.hbase.client.HTable) PRow(org.apache.phoenix.schema.PRow) PColumn(org.apache.phoenix.schema.PColumn) Field(org.apache.phoenix.schema.ValueSchema.Field) RegionCoprocessorEnvironment(org.apache.hadoop.hbase.coprocessor.RegionCoprocessorEnvironment) IndexMaintainer(org.apache.phoenix.index.IndexMaintainer) EncodedColumnQualiferCellsList(org.apache.phoenix.schema.tuple.EncodedColumnQualiferCellsList) Aggregator(org.apache.phoenix.expression.aggregator.Aggregator) RowKeySchema(org.apache.phoenix.schema.RowKeySchema) Put(org.apache.hadoop.hbase.client.Put) RegionScanner(org.apache.hadoop.hbase.regionserver.RegionScanner) Expression(org.apache.phoenix.expression.Expression) HashJoinInfo(org.apache.phoenix.join.HashJoinInfo) MultiKeyValueTuple(org.apache.phoenix.schema.tuple.MultiKeyValueTuple) PositionBasedMultiKeyValueTuple(org.apache.phoenix.schema.tuple.PositionBasedMultiKeyValueTuple) TableRef(org.apache.phoenix.schema.TableRef) MultiKeyValueTuple(org.apache.phoenix.schema.tuple.MultiKeyValueTuple) Tuple(org.apache.phoenix.schema.tuple.Tuple) PositionBasedMultiKeyValueTuple(org.apache.phoenix.schema.tuple.PositionBasedMultiKeyValueTuple)

Example 4 with PRow

use of org.apache.phoenix.schema.PRow in project phoenix by apache.

the class MutationState method generateMutations.

private void generateMutations(final TableRef tableRef, long timestamp, final Map<ImmutableBytesPtr, RowMutationState> values, final List<Mutation> mutationList, final List<Mutation> mutationsPertainingToIndex) {
    final PTable table = tableRef.getTable();
    boolean tableWithRowTimestampCol = table.getRowTimestampColPos() != -1;
    Iterator<Map.Entry<ImmutableBytesPtr, RowMutationState>> iterator = values.entrySet().iterator();
    long timestampToUse = timestamp;
    Map<ImmutableBytesPtr, RowMutationState> modifiedValues = Maps.newHashMap();
    while (iterator.hasNext()) {
        Map.Entry<ImmutableBytesPtr, RowMutationState> rowEntry = iterator.next();
        byte[] onDupKeyBytes = rowEntry.getValue().getOnDupKeyBytes();
        boolean hasOnDupKey = onDupKeyBytes != null;
        ImmutableBytesPtr key = rowEntry.getKey();
        RowMutationState state = rowEntry.getValue();
        if (tableWithRowTimestampCol) {
            RowTimestampColInfo rowTsColInfo = state.getRowTimestampColInfo();
            if (rowTsColInfo.useServerTimestamp()) {
                // since we are about to modify the byte[] stored in key (which changes its hashcode)
                // we need to remove the entry from the values map and add a new entry with the modified byte[]
                modifiedValues.put(key, state);
                iterator.remove();
                // regenerate the key with this timestamp.
                key = getNewRowKeyWithRowTimestamp(key, timestampToUse, table);
            } else {
                if (rowTsColInfo.getTimestamp() != null) {
                    timestampToUse = rowTsColInfo.getTimestamp();
                }
            }
        }
        PRow row = tableRef.getTable().newRow(connection.getKeyValueBuilder(), timestampToUse, key, hasOnDupKey);
        List<Mutation> rowMutations, rowMutationsPertainingToIndex;
        if (rowEntry.getValue().getColumnValues() == PRow.DELETE_MARKER) {
            // means delete
            row.delete();
            rowMutations = row.toRowMutations();
            // Row deletes for index tables are processed by running a re-written query
            // against the index table (as this allows for flexibility in being able to
            // delete rows).
            rowMutationsPertainingToIndex = Collections.emptyList();
        } else {
            for (Map.Entry<PColumn, byte[]> valueEntry : rowEntry.getValue().getColumnValues().entrySet()) {
                row.setValue(valueEntry.getKey(), valueEntry.getValue());
            }
            rowMutations = row.toRowMutations();
            // TODO: use our ServerCache 
            for (Mutation mutation : rowMutations) {
                if (onDupKeyBytes != null) {
                    mutation.setAttribute(PhoenixIndexBuilder.ATOMIC_OP_ATTRIB, onDupKeyBytes);
                }
            }
            rowMutationsPertainingToIndex = rowMutations;
        }
        mutationList.addAll(rowMutations);
        if (connection.isReplayMutations()) {
            // correct index rows on replay.
            for (Mutation mutation : rowMutations) {
                mutation.setAttribute(BaseScannerRegionObserver.IGNORE_NEWER_MUTATIONS, PDataType.TRUE_BYTES);
            }
        }
        if (mutationsPertainingToIndex != null)
            mutationsPertainingToIndex.addAll(rowMutationsPertainingToIndex);
    }
    values.putAll(modifiedValues);
}
Also used : ImmutableBytesPtr(org.apache.phoenix.hbase.index.util.ImmutableBytesPtr) PTable(org.apache.phoenix.schema.PTable) PRow(org.apache.phoenix.schema.PRow) PColumn(org.apache.phoenix.schema.PColumn) Entry(java.util.Map.Entry) Mutation(org.apache.hadoop.hbase.client.Mutation) Map(java.util.Map)

Aggregations

PColumn (org.apache.phoenix.schema.PColumn)4 PRow (org.apache.phoenix.schema.PRow)4 List (java.util.List)3 Cell (org.apache.hadoop.hbase.Cell)3 Delete (org.apache.hadoop.hbase.client.Delete)3 Mutation (org.apache.hadoop.hbase.client.Mutation)3 PTable (org.apache.phoenix.schema.PTable)3 Map (java.util.Map)2 Put (org.apache.hadoop.hbase.client.Put)2 ImmutableBytesWritable (org.apache.hadoop.hbase.io.ImmutableBytesWritable)2 DataExceedsCapacityException (org.apache.phoenix.exception.DataExceedsCapacityException)2 Expression (org.apache.phoenix.expression.Expression)2 RowKeySchema (org.apache.phoenix.schema.RowKeySchema)2 MultiKeyValueTuple (org.apache.phoenix.schema.tuple.MultiKeyValueTuple)2 ByteArrayInputStream (java.io.ByteArrayInputStream)1 DataInputStream (java.io.DataInputStream)1 EOFException (java.io.EOFException)1 SQLException (java.sql.SQLException)1 ArrayList (java.util.ArrayList)1 BitSet (java.util.BitSet)1