use of org.apache.cassandra.db.compaction.SizeTieredCompactionStrategy in project eiger by wlloyd.
the class CollationController method collectTimeOrderedData.
/**
* Collects data in order of recency, using the sstable maxtimestamp data.
* Once we have data for all requests columns that is newer than the newest remaining maxtimestamp,
* we stop.
*/
private ColumnFamily collectTimeOrderedData() {
logger.debug("collectTimeOrderedData");
ISortedColumns.Factory factory = mutableColumns ? AtomicSortedColumns.factory() : TreeMapBackedSortedColumns.factory();
ColumnFamily container = ColumnFamily.create(cfs.metadata, factory, filter.filter.isReversed());
List<IColumnIterator> iterators = new ArrayList<IColumnIterator>();
ColumnFamilyStore.ViewFragment view = cfs.markReferenced(filter.key);
try {
for (Memtable memtable : view.memtables) {
IColumnIterator iter = filter.getMemtableColumnIterator(memtable);
if (iter != null) {
iterators.add(iter);
container.delete(iter.getColumnFamily());
while (iter.hasNext()) container.addColumn(iter.next());
}
}
// avoid changing the filter columns of the original filter
// (reduceNameFilter removes columns that are known to be irrelevant)
TreeSet<ByteBuffer> filterColumns = new TreeSet<ByteBuffer>(((NamesQueryFilter) filter.filter).columns);
QueryFilter reducedFilter = new QueryFilter(filter.key, filter.path, new NamesQueryFilter(filterColumns));
/* add the SSTables on disk */
Collections.sort(view.sstables, SSTable.maxTimestampComparator);
// read sorted sstables
for (SSTableReader sstable : view.sstables) {
long currentMaxTs = sstable.getMaxTimestamp();
reduceNameFilter(reducedFilter, container, currentMaxTs);
if (((NamesQueryFilter) reducedFilter.filter).columns.isEmpty())
break;
IColumnIterator iter = reducedFilter.getSSTableColumnIterator(sstable);
iterators.add(iter);
if (iter.getColumnFamily() != null) {
container.delete(iter.getColumnFamily());
sstablesIterated++;
while (iter.hasNext()) container.addColumn(iter.next());
}
}
// and "there used to be data, but it's gone now" (we should cache the empty CF so we don't need to rebuild that slower)
if (iterators.isEmpty())
return null;
// do a final collate. toCollate is boilerplate required to provide a CloseableIterator
final ColumnFamily c2 = container;
CloseableIterator<IColumn> toCollate = new SimpleAbstractColumnIterator() {
final Iterator<IColumn> iter = c2.iterator();
protected IColumn computeNext() {
return iter.hasNext() ? iter.next() : endOfData();
}
public ColumnFamily getColumnFamily() {
return c2;
}
public DecoratedKey getKey() {
return filter.key;
}
};
ColumnFamily returnCF = container.cloneMeShallow();
filter.collateColumns(returnCF, Collections.singletonList(toCollate), gcBefore);
// "hoist up" the requested data into a more recent sstable
if (sstablesIterated > cfs.getMinimumCompactionThreshold() && !cfs.isCompactionDisabled() && cfs.getCompactionStrategy() instanceof SizeTieredCompactionStrategy) {
RowMutation rm = new RowMutation(cfs.table.name, new Row(filter.key, returnCF.cloneMe()));
try {
// skipping commitlog and index updates is fine since we're just de-fragmenting existing data
Table.open(rm.getTable()).apply(rm, false, false);
} catch (IOException e) {
// log and allow the result to be returned
logger.error("Error re-writing read results", e);
}
}
// Caller is responsible for final removeDeletedCF. This is important for cacheRow to work correctly:
return returnCF;
} finally {
for (IColumnIterator iter : iterators) FileUtils.closeQuietly(iter);
SSTableReader.releaseReferences(view.sstables);
}
}
Aggregations