for (Memtable memtable : view.memtables)
{
ColumnFamily cf = memtable.getColumnFamily(filter.key);
if (cf != null)
{
OnDiskAtomIterator iter = filter.getColumnFamilyIterator(cf);
if (copyOnHeap)
{
ColumnFamily newCf = cf.cloneMeShallow(ArrayBackedSortedColumns.factory, false);
for (Cell cell : cf)
{
newCf.addColumn(cell.localCopy(cfs.metadata, HeapAllocator.instance));
}
cf = newCf;
iter = filter.getColumnFamilyIterator(cf);
}
filter.delete(returnDeletionInfo, cf);
iterators.add(iter);
}
}
/*
* We can't eliminate full sstables based on the timestamp of what we've already read like
* in collectTimeOrderedData, but we still want to eliminate sstable whose maxTimestamp < mostRecentTombstone
* we've read. We still rely on the sstable ordering by maxTimestamp since if
* maxTimestamp_s1 > maxTimestamp_s0,
* we're guaranteed that s1 cannot have a row tombstone such that
* timestamp(tombstone) > maxTimestamp_s0
* since we necessarily have
* timestamp(tombstone) <= maxTimestamp_s1
* In othere words, iterating in maxTimestamp order allow to do our mostRecentTombstone elimination
* in one pass, and minimize the number of sstables for which we read a rowTombstone.
*/
Collections.sort(view.sstables, SSTableReader.maxTimestampComparator);
List<SSTableReader> skippedSSTables = null;
long mostRecentRowTombstone = Long.MIN_VALUE;
long minTimestamp = Long.MAX_VALUE;
int nonIntersectingSSTables = 0;
for (SSTableReader sstable : view.sstables)
{
minTimestamp = Math.min(minTimestamp, sstable.getMinTimestamp());
// if we've already seen a row tombstone with a timestamp greater
// than the most recent update to this sstable, we can skip it
if (sstable.getMaxTimestamp() < mostRecentRowTombstone)
break;
if (!filter.shouldInclude(sstable))
{
nonIntersectingSSTables++;
// sstable contains no tombstone if maxLocalDeletionTime == Integer.MAX_VALUE, so we can safely skip those entirely
if (sstable.getSSTableMetadata().maxLocalDeletionTime != Integer.MAX_VALUE)
{
if (skippedSSTables == null)
skippedSSTables = new ArrayList<>();
skippedSSTables.add(sstable);
}
continue;
}
sstable.incrementReadCount();
OnDiskAtomIterator iter = filter.getSSTableColumnIterator(sstable);
iterators.add(iter);
if (iter.getColumnFamily() != null)
{
ColumnFamily cf = iter.getColumnFamily();
if (cf.isMarkedForDelete())
mostRecentRowTombstone = cf.deletionInfo().getTopLevelDeletion().markedForDeleteAt;
returnCF.delete(cf);
sstablesIterated++;
}
}
int includedDueToTombstones = 0;
// Check for row tombstone in the skipped sstables
if (skippedSSTables != null)
{
for (SSTableReader sstable : skippedSSTables)
{
if (sstable.getMaxTimestamp() <= minTimestamp)
continue;
sstable.incrementReadCount();
OnDiskAtomIterator iter = filter.getSSTableColumnIterator(sstable);
ColumnFamily cf = iter.getColumnFamily();
// we are only interested in row-level tombstones here, and only if markedForDeleteAt is larger than minTimestamp
if (cf != null && cf.deletionInfo().getTopLevelDeletion().markedForDeleteAt > minTimestamp)
{
includedDueToTombstones++;
iterators.add(iter);