private static byte[] nextKey(byte[] key, RowKeySchema schema, ImmutableBytesWritable ptr) {
int pos = 0;
int maxOffset = schema.iterator(key, ptr);
while (schema.next(ptr, pos, maxOffset) != null) {
pos++;
}
Field field = schema.getField(pos - 1);
if (!field.getDataType().isFixedWidth()) {
byte[] newLowerRange = new byte[key.length + 1];
System.arraycopy(key, 0, newLowerRange, 0, key.length);
newLowerRange[key.length] =
SchemaUtil.getSeparatorByte(schema.rowKeyOrderOptimizable(), key.length == 0, field);
key = newLowerRange;
} else {
key = Arrays.copyOf(key, key.length);
}
ByteUtil.nextKey(key, key.length);
return key;
}
public static int setKey(
RowKeySchema schema,
List<List<KeyRange>> slots,
int[] slotSpan,
int[] position,
Bound bound,
byte[] key,
int byteOffset,
int slotStartIndex,
int slotEndIndex,
int schemaStartIndex) {
int offset = byteOffset;
boolean lastInclusiveUpperSingleKey = false;
boolean anyInclusiveUpperRangeKey = false;
// The index used for slots should be incremented by 1,
// but the index for the field it represents in the schema
// should be incremented by 1 + value in the current slotSpan index
// slotSpan stores the number of columns beyond one that the range spans
Field field = null;
int i = slotStartIndex, fieldIndex = ScanUtil.getRowKeyPosition(slotSpan, slotStartIndex);
for (i = slotStartIndex; i < slotEndIndex; i++) {
// Build up the key by appending the bound of each key range
// from the current position of each slot.
KeyRange range = slots.get(i).get(position[i]);
// Use last slot in a multi-span column to determine if fixed width
field = schema.getField(fieldIndex + slotSpan[i]);
boolean isFixedWidth = field.getDataType().isFixedWidth();
fieldIndex += slotSpan[i] + 1;
/*
* If the current slot is unbound then stop if:
* 1) setting the upper bound. There's no value in
* continuing because nothing will be filtered.
* 2) setting the lower bound when the type is fixed length
* for the same reason. However, if the type is variable width
* continue building the key because null values will be filtered
* since our separator byte will be appended and incremented.
*/
if (range.isUnbound(bound) && (bound == Bound.UPPER || isFixedWidth)) {
break;
}
byte[] bytes = range.getRange(bound);
System.arraycopy(bytes, 0, key, offset, bytes.length);
offset += bytes.length;
/*
* We must add a terminator to a variable length key even for the last PK column if
* the lower key is non inclusive or the upper key is inclusive. Otherwise, we'd be
* incrementing the key value itself, and thus bumping it up too much.
*/
boolean inclusiveUpper = range.isInclusive(bound) && bound == Bound.UPPER;
boolean exclusiveLower = !range.isInclusive(bound) && bound == Bound.LOWER;
// If we are setting the upper bound of using inclusive single key, we remember
// to increment the key if we exit the loop after this iteration.
//
// We remember to increment the last slot if we are setting the upper bound with an
// inclusive range key.
//
// We cannot combine the two flags together in case for single-inclusive key followed
// by the range-exclusive key. In that case, we do not need to increment the end at the
// end. But if we combine the two flag, the single inclusive key in the middle of the
// key slots would cause the flag to become true.
lastInclusiveUpperSingleKey = range.isSingleKey() && inclusiveUpper;
anyInclusiveUpperRangeKey |= !range.isSingleKey() && inclusiveUpper;
// A null or empty byte array is always represented as a zero byte
byte sepByte =
SchemaUtil.getSeparatorByte(schema.rowKeyOrderOptimizable(), bytes.length == 0, field);
if (!isFixedWidth
&& (fieldIndex < schema.getMaxFields()
|| inclusiveUpper
|| exclusiveLower
|| sepByte == QueryConstants.DESC_SEPARATOR_BYTE)) {
key[offset++] = sepByte;
// Set lastInclusiveUpperSingleKey back to false if this is the last pk column
// as we don't want to increment the null byte in this case
lastInclusiveUpperSingleKey &= i < schema.getMaxFields() - 1;
}
// If we are setting the lower bound with an exclusive range key, we need to bump the
// slot up for each key part. For an upper bound, we bump up an inclusive key, but
// only after the last key part.
if (exclusiveLower) {
if (!ByteUtil.nextKey(key, offset)) {
// Special case for not being able to increment.
// In this case we return a negative byteOffset to
// remove this part from the key being formed. Since the
// key has overflowed, this means that we should not
// have an end key specified.
return -byteOffset;
}
// We're filtering on values being non null here, but we still need the 0xFF
// terminator, since DESC keys ignore the last byte as it's expected to be
// the terminator. Without this, we'd ignore the separator byte that was
// just added and incremented.
if (!isFixedWidth
&& bytes.length == 0
&& SchemaUtil.getSeparatorByte(schema.rowKeyOrderOptimizable(), false, field)
== QueryConstants.DESC_SEPARATOR_BYTE) {
key[offset++] = QueryConstants.DESC_SEPARATOR_BYTE;
}
}
}
if (lastInclusiveUpperSingleKey || anyInclusiveUpperRangeKey) {
if (!ByteUtil.nextKey(key, offset)) {
// Special case for not being able to increment.
// In this case we return a negative byteOffset to
// remove this part from the key being formed. Since the
// key has overflowed, this means that we should not
// have an end key specified.
return -byteOffset;
}
}
// Remove trailing separator bytes, since the columns may have been added
// after the table has data, in which case there won't be a separator
// byte.
if (bound == Bound.LOWER) {
while (--i >= schemaStartIndex
&& offset > byteOffset
&& !(field = schema.getField(--fieldIndex)).getDataType().isFixedWidth()
&& field.getSortOrder() == SortOrder.ASC
&& key[offset - 1] == QueryConstants.SEPARATOR_BYTE) {
offset--;
fieldIndex -= slotSpan[i];
}
}
return offset - byteOffset;
}
public static ScanRanges create(
RowKeySchema schema,
List<List<KeyRange>> ranges,
int[] slotSpan,
KeyRange minMaxRange,
Integer nBuckets,
boolean useSkipScan,
int rowTimestampColIndex) {
int offset = nBuckets == null ? 0 : SaltingUtil.NUM_SALTING_BYTES;
int nSlots = ranges.size();
if (nSlots == offset && minMaxRange == KeyRange.EVERYTHING_RANGE) {
return EVERYTHING;
} else if (minMaxRange == KeyRange.EMPTY_RANGE
|| (nSlots == 1 + offset
&& ranges.get(offset).size() == 1
&& ranges.get(offset).get(0) == KeyRange.EMPTY_RANGE)) {
return NOTHING;
}
TimeRange rowTimestampRange = getRowTimestampColumnRange(ranges, schema, rowTimestampColIndex);
boolean isPointLookup = isPointLookup(schema, ranges, slotSpan, useSkipScan);
if (isPointLookup) {
// TODO: consider keeping original to use for serialization as it would be smaller?
List<byte[]> keys = ScanRanges.getPointKeys(ranges, slotSpan, schema, nBuckets);
List<KeyRange> keyRanges = Lists.newArrayListWithExpectedSize(keys.size());
KeyRange unsaltedMinMaxRange = minMaxRange;
if (nBuckets != null && minMaxRange != KeyRange.EVERYTHING_RANGE) {
unsaltedMinMaxRange =
KeyRange.getKeyRange(
stripPrefix(minMaxRange.getLowerRange(), offset),
minMaxRange.lowerUnbound(),
stripPrefix(minMaxRange.getUpperRange(), offset),
minMaxRange.upperUnbound());
}
// We have full keys here, so use field from our varbinary schema
BytesComparator comparator = ScanUtil.getComparator(SchemaUtil.VAR_BINARY_SCHEMA.getField(0));
for (byte[] key : keys) {
// Filter now based on unsalted minMaxRange and ignore the point key salt byte
if (unsaltedMinMaxRange.compareLowerToUpperBound(
key, offset, key.length - offset, true, comparator)
<= 0
&& unsaltedMinMaxRange.compareUpperToLowerBound(
key, offset, key.length - offset, true, comparator)
>= 0) {
keyRanges.add(KeyRange.getKeyRange(key));
}
}
ranges = Collections.singletonList(keyRanges);
useSkipScan = keyRanges.size() > 1;
// Treat as binary if descending because we've got a separator byte at the end
// which is not part of the value.
if (keys.size() > 1
|| SchemaUtil.getSeparatorByte(schema.rowKeyOrderOptimizable(), false, schema.getField(0))
== QueryConstants.DESC_SEPARATOR_BYTE) {
schema = SchemaUtil.VAR_BINARY_SCHEMA;
slotSpan = ScanUtil.SINGLE_COLUMN_SLOT_SPAN;
} else {
// Keep original schema and don't use skip scan as it's not necessary
// when there's a single key.
slotSpan = new int[] {schema.getMaxFields() - 1};
}
}
List<List<KeyRange>> sortedRanges = Lists.newArrayListWithExpectedSize(ranges.size());
for (int i = 0; i < ranges.size(); i++) {
List<KeyRange> sorted = Lists.newArrayList(ranges.get(i));
Collections.sort(sorted, KeyRange.COMPARATOR);
sortedRanges.add(ImmutableList.copyOf(sorted));
}
// Don't set minMaxRange for point lookup because it causes issues during intersect
// by going across region boundaries
KeyRange scanRange = KeyRange.EVERYTHING_RANGE;
// if (!isPointLookup && (nBuckets == null || !useSkipScanFilter)) {
// if (! ( isPointLookup || (nBuckets != null && useSkipScanFilter) ) ) {
// if (nBuckets == null || (nBuckets != null && (!isPointLookup || !useSkipScanFilter))) {
if (nBuckets == null || !isPointLookup || !useSkipScan) {
byte[] minKey = ScanUtil.getMinKey(schema, sortedRanges, slotSpan);
byte[] maxKey = ScanUtil.getMaxKey(schema, sortedRanges, slotSpan);
// If the maxKey has crossed the salt byte boundary, then we do not
// have anything to filter at the upper end of the range
if (ScanUtil.crossesPrefixBoundary(maxKey, ScanUtil.getPrefix(minKey, offset), offset)) {
maxKey = KeyRange.UNBOUND;
}
// We won't filter anything at the low end of the range if we just have the salt byte
if (minKey.length <= offset) {
minKey = KeyRange.UNBOUND;
}
scanRange = KeyRange.getKeyRange(minKey, maxKey);
}
if (minMaxRange != KeyRange.EVERYTHING_RANGE) {
minMaxRange =
ScanUtil.convertToInclusiveExclusiveRange(
minMaxRange, schema, new ImmutableBytesWritable());
scanRange = scanRange.intersect(minMaxRange);
}
if (scanRange == KeyRange.EMPTY_RANGE) {
return NOTHING;
}
return new ScanRanges(
schema,
slotSpan,
sortedRanges,
scanRange,
minMaxRange,
useSkipScan,
isPointLookup,
nBuckets,
rowTimestampRange);
}
