/*
- * Copyright 2017-present Open Networking Foundation
+ * Copyright 2017-2022 Open Networking Foundation and others. All rights reserved.
+ * Copyright (c) 2024 PANTHEON.tech, s.r.o.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
*/
package io.atomix.storage.journal;
+import static io.atomix.storage.journal.SegmentEntry.HEADER_BYTES;
+
import com.esotericsoftware.kryo.KryoException;
import io.atomix.storage.journal.index.JournalIndex;
-
import java.io.IOException;
import java.nio.BufferOverflowException;
-import java.nio.BufferUnderflowException;
import java.nio.ByteBuffer;
import java.nio.MappedByteBuffer;
+import java.nio.channels.FileChannel;
import java.util.zip.CRC32;
+import org.eclipse.jdt.annotation.NonNull;
/**
* Segment writer.
*
* @author <a href="http://github.com/kuujo">Jordan Halterman</a>
*/
-class MappedJournalSegmentWriter<E> implements JournalWriter<E> {
- private final MappedByteBuffer mappedBuffer;
+final class MappedJournalSegmentWriter<E> extends JournalSegmentWriter<E> {
+ private final @NonNull MappedByteBuffer mappedBuffer;
private final ByteBuffer buffer;
- private final JournalSegment<E> segment;
- private final int maxEntrySize;
- private final JournalIndex index;
- private final JournalSerdes namespace;
- private final long firstIndex;
+
private Indexed<E> lastEntry;
+ private int currentPosition;
MappedJournalSegmentWriter(
- MappedByteBuffer buffer,
- JournalSegment<E> segment,
- int maxEntrySize,
- JournalIndex index,
- JournalSerdes namespace) {
- this.mappedBuffer = buffer;
- this.buffer = buffer.slice();
- this.segment = segment;
- this.maxEntrySize = maxEntrySize;
- this.index = index;
- this.namespace = namespace;
- this.firstIndex = segment.index();
+ final FileChannel channel,
+ final JournalSegment<E> segment,
+ final int maxEntrySize,
+ final JournalIndex index,
+ final JournalSerdes namespace) {
+ super(channel, segment, maxEntrySize, index, namespace);
+ mappedBuffer = mapBuffer(channel, maxSegmentSize);
+ buffer = mappedBuffer.slice();
reset(0);
}
- /**
- * Returns the mapped buffer underlying the segment writer.
- *
- * @return the mapped buffer underlying the segment writer
- */
- MappedByteBuffer buffer() {
+ MappedJournalSegmentWriter(final JournalSegmentWriter<E> previous, final int position) {
+ super(previous);
+ mappedBuffer = mapBuffer(channel, maxSegmentSize);
+ buffer = mappedBuffer.slice();
+ currentPosition = position;
+ lastEntry = previous.getLastEntry();
+ }
+
+ private static @NonNull MappedByteBuffer mapBuffer(final FileChannel channel, final int maxSegmentSize) {
+ try {
+ return channel.map(FileChannel.MapMode.READ_WRITE, 0, maxSegmentSize);
+ } catch (IOException e) {
+ throw new StorageException(e);
+ }
+ }
+
+ @Override
+ @NonNull MappedByteBuffer buffer() {
return mappedBuffer;
}
@Override
- public void reset(long index) {
+ MappedJournalSegmentWriter<E> toMapped() {
+ return this;
+ }
+
+ @Override
+ DiskJournalSegmentWriter<E> toFileChannel() {
+ close();
+ return new DiskJournalSegmentWriter<>(this, currentPosition);
+ }
+
+ @Override
+ void reset(final long index) {
long nextIndex = firstIndex;
// Clear the buffer indexes.
- buffer.position(JournalSegmentDescriptor.BYTES);
+ currentPosition = JournalSegmentDescriptor.BYTES;
- // Record the current buffer position.
- int position = buffer.position();
+ int length = buffer.getInt(currentPosition);
- // Read the entry length.
- buffer.mark();
+ // If the length is non-zero, read the entry.
+ while (0 < length && length <= maxEntrySize && (index == 0 || nextIndex <= index)) {
- try {
- int length = buffer.getInt();
-
- // If the length is non-zero, read the entry.
- while (0 < length && length <= maxEntrySize && (index == 0 || nextIndex <= index)) {
-
- // Read the checksum of the entry.
- final long checksum = buffer.getInt() & 0xFFFFFFFFL;
-
- // Compute the checksum for the entry bytes.
- final CRC32 crc32 = new CRC32();
- ByteBuffer slice = buffer.slice();
- slice.limit(length);
- crc32.update(slice);
-
- // If the stored checksum equals the computed checksum, return the entry.
- if (checksum == crc32.getValue()) {
- slice.rewind();
- final E entry = namespace.deserialize(slice);
- lastEntry = new Indexed<>(nextIndex, entry, length);
- this.index.index(nextIndex, position);
- nextIndex++;
- } else {
- break;
- }
+ // Read the checksum of the entry.
+ final long checksum = buffer.getInt(currentPosition + Integer.BYTES);
- // Update the current position for indexing.
- position = buffer.position() + length;
- buffer.position(position);
+ // Slice off the entry's bytes
+ final var entryBytes = buffer.slice(currentPosition + SegmentEntry.HEADER_BYTES, length);
- buffer.mark();
- length = buffer.getInt();
- }
+ // Compute the checksum for the entry bytes.
+ final var crc32 = new CRC32();
+ crc32.update(entryBytes);
- // Reset the buffer to the previous mark.
- buffer.reset();
- } catch (BufferUnderflowException e) {
- buffer.reset();
- }
- }
+ // If the stored checksum does not equal the computed checksum, do not proceed further
+ if (checksum != (int) crc32.getValue()) {
+ break;
+ }
- @Override
- public long getLastIndex() {
- return lastEntry != null ? lastEntry.index() : segment.index() - 1;
- }
+ entryBytes.rewind();
+ final E entry = namespace.deserialize(entryBytes);
+ lastEntry = new Indexed<>(nextIndex, entry, length);
+ this.index.index(nextIndex, currentPosition);
+ nextIndex++;
- @Override
- public Indexed<E> getLastEntry() {
- return lastEntry;
- }
+ // Update the current position for indexing.
+ currentPosition = currentPosition + SegmentEntry.HEADER_BYTES + length;
- @Override
- public long getNextIndex() {
- if (lastEntry != null) {
- return lastEntry.index() + 1;
- } else {
- return firstIndex;
+ if (currentPosition + SegmentEntry.HEADER_BYTES >= maxSegmentSize) {
+ break;
+ }
+ length = buffer.getInt(currentPosition);
}
}
@Override
- public void append(Indexed<E> entry) {
- final long nextIndex = getNextIndex();
-
- // If the entry's index is greater than the next index in the segment, skip some entries.
- if (entry.index() > nextIndex) {
- throw new IndexOutOfBoundsException("Entry index is not sequential");
- }
-
- // If the entry's index is less than the next index, truncate the segment.
- if (entry.index() < nextIndex) {
- truncate(entry.index() - 1);
- }
- append(entry.entry());
+ Indexed<E> getLastEntry() {
+ return lastEntry;
}
@Override
@SuppressWarnings("unchecked")
- public <T extends E> Indexed<T> append(T entry) {
+ <T extends E> Indexed<T> append(final T entry) {
// Store the entry index.
final long index = getNextIndex();
// Serialize the entry.
- int position = buffer.position();
- if (position + Integer.BYTES + Integer.BYTES > buffer.limit()) {
+ final int bodyPosition = currentPosition + HEADER_BYTES;
+ final int avail = maxSegmentSize - bodyPosition;
+ if (avail < 0) {
throw new BufferOverflowException();
}
- buffer.position(position + Integer.BYTES + Integer.BYTES);
-
+ final var entryBytes = buffer.slice(bodyPosition, Math.min(avail, maxEntrySize));
try {
- namespace.serialize(entry, buffer);
+ namespace.serialize(entry, entryBytes);
} catch (KryoException e) {
- throw new BufferOverflowException();
- }
-
- final int length = buffer.position() - (position + Integer.BYTES + Integer.BYTES);
+ if (entryBytes.capacity() != maxEntrySize) {
+ // We have not provided enough capacity, signal to roll to next segment
+ throw new BufferOverflowException();
+ }
- // If the entry length exceeds the maximum entry size then throw an exception.
- if (length > maxEntrySize) {
// Just reset the buffer. There's no need to zero the bytes since we haven't written the length or checksum.
- buffer.position(position);
- throw new StorageException.TooLarge("Entry size " + length + " exceeds maximum allowed bytes (" + maxEntrySize + ")");
+ throw new StorageException.TooLarge("Entry size exceeds maximum allowed bytes (" + maxEntrySize + ")");
}
+ final int length = entryBytes.position();
+
// Compute the checksum for the entry.
- final CRC32 crc32 = new CRC32();
- buffer.position(position + Integer.BYTES + Integer.BYTES);
- ByteBuffer slice = buffer.slice();
- slice.limit(length);
- crc32.update(slice);
- final long checksum = crc32.getValue();
+ final var crc32 = new CRC32();
+ crc32.update(entryBytes.flip());
// Create a single byte[] in memory for the entire entry and write it as a batch to the underlying buffer.
- buffer.position(position);
- buffer.putInt(length);
- buffer.putInt((int) checksum);
- buffer.position(position + Integer.BYTES + Integer.BYTES + length);
+ buffer.putInt(currentPosition, length).putInt(currentPosition + Integer.BYTES, (int) crc32.getValue());
// Update the last entry with the correct index/term/length.
Indexed<E> indexedEntry = new Indexed<>(index, entry, length);
- this.lastEntry = indexedEntry;
- this.index.index(index, position);
- return (Indexed<T>) indexedEntry;
- }
-
- @Override
- public void commit(long index) {
+ lastEntry = indexedEntry;
+ this.index.index(index, currentPosition);
+ currentPosition = currentPosition + HEADER_BYTES + length;
+ return (Indexed<T>) indexedEntry;
}
@Override
- public void truncate(long index) {
+ void truncate(final long index) {
// If the index is greater than or equal to the last index, skip the truncate.
if (index >= getLastIndex()) {
return;
// Truncate the index.
this.index.truncate(index);
- if (index < segment.index()) {
+ if (index < firstIndex) {
// Reset the writer to the first entry.
- buffer.position(JournalSegmentDescriptor.BYTES);
+ currentPosition = JournalSegmentDescriptor.BYTES;
} else {
// Reset the writer to the given index.
reset(index);
}
// Zero the entry header at current buffer position.
- int position = buffer.position();
// Note: we issue a single putLong() instead of two putInt()s.
- buffer.putLong(0);
- buffer.position(position);
+ buffer.putLong(currentPosition, 0L);
}
@Override
- public void flush() {
+ void flush() {
mappedBuffer.force();
}
@Override
- public void close() {
+ void close() {
flush();
try {
BufferCleaner.freeBuffer(mappedBuffer);