/** * Licensed to the Apache Software Foundation (ASF) under one * or more contributor license agreements. See the NOTICE file * distributed with this work for additional information * regarding copyright ownership. The ASF licenses this file * to you under the Apache License, Version 2.0 (the * "License"); you may not use this file except in compliance * with the License. You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, * software distributed under the License is distributed on an * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY * KIND, either express or implied. See the License for the * specific language governing permissions and limitations * under the License. */ package org.apache.cassandra.io.sstable.bitidx; import java.io.File; import java.io.IOException; import java.util.*; import org.apache.avro.file.DataFileReader; import org.apache.avro.file.DataFileWriter; import org.apache.avro.specific.SpecificDatumReader; import org.apache.avro.specific.SpecificDatumWriter; import org.apache.cassandra.io.sstable.bitidx.avro.*; import org.apache.cassandra.io.sstable.Component; import org.apache.cassandra.io.sstable.Descriptor; import org.apache.cassandra.io.util.FileUtils; import org.apache.cassandra.utils.Pair; /** * Maintains a stack of scratch files which are periodically merged. */ class ScratchStack { private final Descriptor desc; private final Component.IdGenerator gen; // number of stack components of the same size to merge private final int mergeFactor; // stack of scratch components with their merge counts private final ArrayDeque<Pair<Integer,Component>> stack = new ArrayDeque<Pair<Integer,Component>>(); public ScratchStack(Descriptor desc, Component.IdGenerator gen, int mergeFactor) { this.desc = desc; this.gen = gen; this.mergeFactor = mergeFactor; } /** * @param scratch A scratch component to add to the stack. */ public void add(Component scratch) throws IOException { stack.push(new Pair<Integer,Component>(0, scratch)); // merge while merges are available int depth; while ((depth = mergeNeeded()) > 0) { // merge the top mergeFactor scratch files to a new file Component dest = Component.get(Component.Type.SCRATCH, gen.getNextId()); merge(mergeFactor, new Pair<Integer,Component>(depth, dest)); } } /** * Finalizes this stack by merging all components to a single output component. * @param component The output component to merge stack contents to. */ public void complete(Component output) throws IOException { if (stack.size() == 1) // already contains a single scratch component: just rename FileUtils.renameWithConfirm(new File(desc.filenameFor(stack.pop().right)), new File(desc.filenameFor(output))); else { // force all files to be merged to the destination merge(stack.size(), new Pair<Integer,Component>(-1, output)); assert stack.size() == 1; } } /** @return A target depth to merge the top 'mergeFactor' components to, or 0 if not needed. */ private int mergeNeeded() { int depth = 0; int sequential = 0; Iterator<Pair<Integer,Component>> iter = stack.iterator(); if (iter.hasNext()) depth = iter.next().left; while (iter.hasNext()) { Pair<Integer,Component> component = iter.next(); if (component.left != depth) // not enough components of this depth break; if (++sequential >= mergeFactor) // at least 'mergeFactor' components at this depth: return next depth return depth + 1; } return 0; } /** * Merges/replaces the top n components from the stack to/with the given output. */ private void merge(int n, Pair<Integer,Component> output) throws IOException { /** * Open readers for the top n components and merge them. Files are stacked in order by * rowid/creation time, and all of them contain the same bins, so we empty the segments * for each bin from the readers in order, outputting sorted bins containing sorted rowids. */ List<File> files = popTopNFiles(n); List<DataFileReader<BinSegment>> readers = new ArrayList<DataFileReader<BinSegment>>(); for (File file : files) readers.add(new DataFileReader<BinSegment>(file, new SpecificDatumReader<BinSegment>())); // get metadata from the first reader to apply to the output byte[] indexMetadata = readers.get(0).getMeta(BitmapIndex.META_KEY); // open the output DataFileWriter writer = new DataFileWriter(new SpecificDatumWriter<BinSegment>()); writer.setMeta(BitmapIndex.META_KEY, indexMetadata); writer.create(BinSegment.SCHEMA$, new File(desc.filenameFor(output.right))); ReusableBinSegment segment = ReusableBinSegment.poolGet(); try { // loop over bins: all readers will reach eof after the same bin boolean eof = false; int headersSeen = 0; long cardinality = 0; while (!eof) { // for each reader in order, write out all data segments for this bin for (DataFileReader<BinSegment> reader : readers) { while (reader.hasNext()) { segment.readFrom(reader); if (segment.data != null) // data segment: write to output writer.append(segment); else { // header segment: reached end of bin for this reader cardinality += segment.header.cardinality; if (++headersSeen == n) { // write once all headers have been seen for a bin segment.header.cardinality = cardinality; writer.append(segment); writer.sync(); } break; } } if (!reader.hasNext()) // reached the last bin: finish emptying readers eof = true; } // reset for the next bin headersSeen = 0; cardinality = 0; } } finally { ReusableBinSegment.poolReturn(segment); writer.close(); for (DataFileReader<BinSegment> reader : readers) reader.close(); } // add the newly written component to the stack stack.push(output); // cleanup the retired readers for (File file : files) FileUtils.deleteWithConfirm(file); } /** * @return The top n files from the stack, in ascending rowid order. */ private List<File> popTopNFiles(int n) { List<File> files = new ArrayList<File>(n); for (; n > 0; n--) files.add(new File(desc.filenameFor(stack.pop().right))); Collections.reverse(files); return files; } }