public ColumnFamilyStore testSingleSSTableCompaction(String strategyClassName) throws Exception {
Keyspace keyspace = Keyspace.open(KEYSPACE1);
ColumnFamilyStore store = keyspace.getColumnFamilyStore(CF_STANDARD1);
store.clearUnsafe();
store.metadata.gcGraceSeconds(1);
store.setCompactionStrategyClass(strategyClassName);
// disable compaction while flushing
store.disableAutoCompaction();
long timestamp = populate(KEYSPACE1, CF_STANDARD1, 0, 9, 3); // ttl=3s
store.forceBlockingFlush();
assertEquals(1, store.getSSTables().size());
long originalSize = store.getSSTables().iterator().next().uncompressedLength();
// wait enough to force single compaction
TimeUnit.SECONDS.sleep(5);
// enable compaction, submit background and wait for it to complete
store.enableAutoCompaction();
FBUtilities.waitOnFutures(CompactionManager.instance.submitBackground(store));
while (CompactionManager.instance.getPendingTasks() > 0
|| CompactionManager.instance.getActiveCompactions() > 0) TimeUnit.SECONDS.sleep(1);
// and sstable with ttl should be compacted
assertEquals(1, store.getSSTables().size());
long size = store.getSSTables().iterator().next().uncompressedLength();
assertTrue("should be less than " + originalSize + ", but was " + size, size < originalSize);
// make sure max timestamp of compacted sstables is recorded properly after compaction.
assertMaxTimestamp(store, timestamp);
return store;
}
@Test
public void testCheckForExpiredSSTableBlockers() throws InterruptedException {
String KEYSPACE1 = "Keyspace1";
ColumnFamilyStore cfs = Keyspace.open("Keyspace1").getColumnFamilyStore("Standard1");
cfs.truncateBlocking();
cfs.disableAutoCompaction();
cfs.metadata.gcGraceSeconds(0);
RowMutation rm = new RowMutation(KEYSPACE1, Util.dk("test").key);
rm.add(
"Standard1",
ByteBufferUtil.bytes("col1"),
ByteBufferUtil.EMPTY_BYTE_BUFFER,
System.currentTimeMillis());
rm.applyUnsafe();
cfs.forceBlockingFlush();
SSTableReader blockingSSTable = cfs.getSSTables().iterator().next();
for (int i = 0; i < 10; i++) {
rm = new RowMutation(KEYSPACE1, Util.dk("test").key);
rm.delete("Standard1", System.currentTimeMillis());
rm.applyUnsafe();
cfs.forceBlockingFlush();
}
Multimap<SSTableReader, SSTableReader> blockers =
SSTableExpiredBlockers.checkForExpiredSSTableBlockers(
cfs.getSSTables(), (int) (System.currentTimeMillis() / 1000) + 100);
assertEquals(1, blockers.keySet().size());
assertTrue(blockers.keySet().contains(blockingSSTable));
assertEquals(10, blockers.get(blockingSSTable).size());
}
@Test
public void testEchoedRow() throws IOException, ExecutionException, InterruptedException {
// This test check that EchoedRow doesn't skipp rows: see CASSANDRA-2653
Keyspace keyspace = Keyspace.open(KEYSPACE1);
ColumnFamilyStore cfs = keyspace.getColumnFamilyStore("Standard2");
// disable compaction while flushing
cfs.disableAutoCompaction();
// Insert 4 keys in two sstables. We need the sstables to have 2 rows
// at least to trigger what was causing CASSANDRA-2653
for (int i = 1; i < 5; i++) {
DecoratedKey key = Util.dk(String.valueOf(i));
RowMutation rm = new RowMutation(KEYSPACE1, key.key);
rm.add(
"Standard2",
ByteBufferUtil.bytes(String.valueOf(i)),
ByteBufferUtil.EMPTY_BYTE_BUFFER,
i);
rm.apply();
if (i % 2 == 0) cfs.forceBlockingFlush();
}
Collection<SSTableReader> toCompact = cfs.getSSTables();
assert toCompact.size() == 2;
// Reinserting the same keys. We will compact only the previous sstable, but we need those new
// ones
// to make sure we use EchoedRow, otherwise it won't be used because purge can be done.
for (int i = 1; i < 5; i++) {
DecoratedKey key = Util.dk(String.valueOf(i));
RowMutation rm = new RowMutation(KEYSPACE1, key.key);
rm.add(
"Standard2",
ByteBufferUtil.bytes(String.valueOf(i)),
ByteBufferUtil.EMPTY_BYTE_BUFFER,
i);
rm.apply();
}
cfs.forceBlockingFlush();
SSTableReader tmpSSTable = null;
for (SSTableReader sstable : cfs.getSSTables())
if (!toCompact.contains(sstable)) tmpSSTable = sstable;
assert tmpSSTable != null;
// Force compaction on first sstables. Since each row is in only one sstable, we will be using
// EchoedRow.
Util.compact(cfs, toCompact);
assertEquals(2, cfs.getSSTables().size());
// Now, we remove the sstable that was just created to force the use of EchoedRow (so that it
// doesn't hide the problem)
cfs.markObsolete(Collections.singleton(tmpSSTable), OperationType.UNKNOWN);
assertEquals(1, cfs.getSSTables().size());
// Now assert we do have the 4 keys
assertEquals(4, Util.getRangeSlice(cfs).size());
}
private void testDontPurgeAccidentaly(String k, String cfname)
throws IOException, ExecutionException, InterruptedException {
// This test catches the regression of CASSANDRA-2786
Keyspace keyspace = Keyspace.open(KEYSPACE1);
ColumnFamilyStore cfs = keyspace.getColumnFamilyStore(cfname);
// disable compaction while flushing
cfs.clearUnsafe();
cfs.disableAutoCompaction();
// Add test row
DecoratedKey key = Util.dk(k);
RowMutation rm = new RowMutation(KEYSPACE1, key.key);
rm.add(
cfname,
CompositeType.build(ByteBufferUtil.bytes("sc"), ByteBufferUtil.bytes("c")),
ByteBufferUtil.EMPTY_BYTE_BUFFER,
0);
rm.apply();
cfs.forceBlockingFlush();
Collection<SSTableReader> sstablesBefore = cfs.getSSTables();
QueryFilter filter = QueryFilter.getIdentityFilter(key, cfname, System.currentTimeMillis());
assert !(cfs.getColumnFamily(filter).getColumnCount() == 0);
// Remove key
rm = new RowMutation(KEYSPACE1, key.key);
rm.delete(cfname, 2);
rm.apply();
ColumnFamily cf = cfs.getColumnFamily(filter);
assert cf == null || cf.getColumnCount() == 0 : "should be empty: " + cf;
// Sleep one second so that the removal is indeed purgeable even with gcgrace == 0
Thread.sleep(1000);
cfs.forceBlockingFlush();
Collection<SSTableReader> sstablesAfter = cfs.getSSTables();
Collection<SSTableReader> toCompact = new ArrayList<SSTableReader>();
for (SSTableReader sstable : sstablesAfter)
if (!sstablesBefore.contains(sstable)) toCompact.add(sstable);
Util.compact(cfs, toCompact);
cf = cfs.getColumnFamily(filter);
assert cf == null || cf.getColumnCount() == 0 : "should be empty: " + cf;
}
public ColumnFamilyStore testSingleSSTableCompaction(String strategyClassName) throws Exception {
Keyspace keyspace = Keyspace.open(KEYSPACE1);
ColumnFamilyStore store = keyspace.getColumnFamilyStore("Standard1");
store.clearUnsafe();
store.metadata.gcGraceSeconds(1);
store.setCompactionStrategyClass(strategyClassName);
// disable compaction while flushing
store.disableAutoCompaction();
long timestamp = System.currentTimeMillis();
for (int i = 0; i < 10; i++) {
DecoratedKey key = Util.dk(Integer.toString(i));
RowMutation rm = new RowMutation(KEYSPACE1, key.key);
for (int j = 0; j < 10; j++)
rm.add(
"Standard1",
ByteBufferUtil.bytes(Integer.toString(j)),
ByteBufferUtil.EMPTY_BYTE_BUFFER,
timestamp,
j > 0
? 3
: 0); // let first column never expire, since deleting all columns does not produce
// sstable
rm.apply();
}
store.forceBlockingFlush();
assertEquals(1, store.getSSTables().size());
long originalSize = store.getSSTables().iterator().next().uncompressedLength();
// wait enough to force single compaction
TimeUnit.SECONDS.sleep(5);
// enable compaction, submit background and wait for it to complete
store.enableAutoCompaction();
FBUtilities.waitOnFutures(CompactionManager.instance.submitBackground(store));
while (CompactionManager.instance.getPendingTasks() > 0
|| CompactionManager.instance.getActiveCompactions() > 0) TimeUnit.SECONDS.sleep(1);
// and sstable with ttl should be compacted
assertEquals(1, store.getSSTables().size());
long size = store.getSSTables().iterator().next().uncompressedLength();
assertTrue("should be less than " + originalSize + ", but was " + size, size < originalSize);
// make sure max timestamp of compacted sstables is recorded properly after compaction.
assertMaxTimestamp(store, timestamp);
return store;
}
@Test
public void testCompactionLog() throws Exception {
SystemKeyspace.discardCompactionsInProgress();
String cf = "Standard4";
ColumnFamilyStore cfs = Keyspace.open(KEYSPACE1).getColumnFamilyStore(cf);
SchemaLoader.insertData(KEYSPACE1, cf, 0, 1);
cfs.forceBlockingFlush();
Collection<SSTableReader> sstables = cfs.getSSTables();
assertFalse(sstables.isEmpty());
Set<Integer> generations =
Sets.newHashSet(
Iterables.transform(
sstables,
new Function<SSTableReader, Integer>() {
public Integer apply(SSTableReader sstable) {
return sstable.descriptor.generation;
}
}));
UUID taskId = SystemKeyspace.startCompaction(cfs, sstables);
Map<Pair<String, String>, Map<Integer, UUID>> compactionLogs =
SystemKeyspace.getUnfinishedCompactions();
Set<Integer> unfinishedCompactions = compactionLogs.get(Pair.create(KEYSPACE1, cf)).keySet();
assertTrue(unfinishedCompactions.containsAll(generations));
SystemKeyspace.finishCompaction(taskId);
compactionLogs = SystemKeyspace.getUnfinishedCompactions();
assertFalse(compactionLogs.containsKey(Pair.create(KEYSPACE1, cf)));
}
@Test
public void testSuperColumnTombstones()
throws IOException, ExecutionException, InterruptedException {
Keyspace keyspace = Keyspace.open(KEYSPACE1);
ColumnFamilyStore cfs = keyspace.getColumnFamilyStore("Super1");
cfs.disableAutoCompaction();
DecoratedKey key = Util.dk("tskey");
ByteBuffer scName = ByteBufferUtil.bytes("TestSuperColumn");
// a subcolumn
RowMutation rm = new RowMutation(KEYSPACE1, key.key);
rm.add(
"Super1",
CompositeType.build(scName, ByteBufferUtil.bytes(0)),
ByteBufferUtil.EMPTY_BYTE_BUFFER,
FBUtilities.timestampMicros());
rm.apply();
cfs.forceBlockingFlush();
// shadow the subcolumn with a supercolumn tombstone
rm = new RowMutation(KEYSPACE1, key.key);
rm.deleteRange(
"Super1",
SuperColumns.startOf(scName),
SuperColumns.endOf(scName),
FBUtilities.timestampMicros());
rm.apply();
cfs.forceBlockingFlush();
CompactionManager.instance.performMaximal(cfs);
assertEquals(1, cfs.getSSTables().size());
// check that the shadowed column is gone
SSTableReader sstable = cfs.getSSTables().iterator().next();
Range keyRange =
new Range<RowPosition>(key, sstable.partitioner.getMinimumToken().maxKeyBound());
SSTableScanner scanner = sstable.getScanner(DataRange.forKeyRange(keyRange));
OnDiskAtomIterator iter = scanner.next();
assertEquals(key, iter.getKey());
assert iter.next() instanceof RangeTombstone;
assert !iter.hasNext();
}
@Test
public void testNoExpire() throws ExecutionException, InterruptedException {
ColumnFamilyStore cfs = Keyspace.open("Keyspace1").getColumnFamilyStore("Standard1");
cfs.disableAutoCompaction();
cfs.metadata.gcGraceSeconds(0);
long timestamp = System.currentTimeMillis();
RowMutation rm = new RowMutation("Keyspace1", Util.dk("ttl").key);
rm.add(
"Standard1", ByteBufferUtil.bytes("col"), ByteBufferUtil.EMPTY_BYTE_BUFFER, timestamp, 1);
rm.add(
"Standard1", ByteBufferUtil.bytes("col7"), ByteBufferUtil.EMPTY_BYTE_BUFFER, timestamp, 1);
rm.apply();
cfs.forceBlockingFlush();
rm = new RowMutation("Keyspace1", Util.dk("ttl").key);
rm.add(
"Standard1", ByteBufferUtil.bytes("col2"), ByteBufferUtil.EMPTY_BYTE_BUFFER, timestamp, 1);
rm.apply();
cfs.forceBlockingFlush();
rm = new RowMutation("Keyspace1", Util.dk("ttl").key);
rm.add(
"Standard1", ByteBufferUtil.bytes("col3"), ByteBufferUtil.EMPTY_BYTE_BUFFER, timestamp, 1);
rm.apply();
cfs.forceBlockingFlush();
DecoratedKey noTTLKey = Util.dk("nottl");
rm = new RowMutation("Keyspace1", noTTLKey.key);
rm.add(
"Standard1", ByteBufferUtil.bytes("col311"), ByteBufferUtil.EMPTY_BYTE_BUFFER, timestamp);
rm.apply();
cfs.forceBlockingFlush();
Thread.sleep(2000); // wait for ttl to expire
assertEquals(4, cfs.getSSTables().size());
cfs.enableAutoCompaction(true);
assertEquals(1, cfs.getSSTables().size());
SSTableReader sstable = cfs.getSSTables().iterator().next();
SSTableScanner scanner = sstable.getScanner(DataRange.allData(sstable.partitioner));
assertTrue(scanner.hasNext());
while (scanner.hasNext()) {
OnDiskAtomIterator iter = scanner.next();
assertEquals(noTTLKey, iter.getKey());
}
}
@Test
public void testUserDefinedCompaction() throws Exception {
Keyspace keyspace = Keyspace.open(KEYSPACE1);
final String cfname = "Standard3"; // use clean(no sstable) CF
ColumnFamilyStore cfs = keyspace.getColumnFamilyStore(cfname);
// disable compaction while flushing
cfs.disableAutoCompaction();
final int ROWS_PER_SSTABLE = 10;
for (int i = 0; i < ROWS_PER_SSTABLE; i++) {
DecoratedKey key = Util.dk(String.valueOf(i));
Mutation rm = new Mutation(KEYSPACE1, key.getKey());
rm.add(
cfname,
Util.cellname("col"),
ByteBufferUtil.EMPTY_BYTE_BUFFER,
System.currentTimeMillis());
rm.applyUnsafe();
}
cfs.forceBlockingFlush();
Collection<SSTableReader> sstables = cfs.getSSTables();
assertEquals(1, sstables.size());
SSTableReader sstable = sstables.iterator().next();
int prevGeneration = sstable.descriptor.generation;
String file = new File(sstable.descriptor.filenameFor(Component.DATA)).getAbsolutePath();
// submit user defined compaction on flushed sstable
CompactionManager.instance.forceUserDefinedCompaction(file);
// wait until user defined compaction finishes
do {
Thread.sleep(100);
} while (CompactionManager.instance.getPendingTasks() > 0
|| CompactionManager.instance.getActiveCompactions() > 0);
// CF should have only one sstable with generation number advanced
sstables = cfs.getSSTables();
assertEquals(1, sstables.size());
assertEquals(prevGeneration + 1, sstables.iterator().next().descriptor.generation);
}
@Test
public void testSimpleExpire() throws ExecutionException, InterruptedException {
ColumnFamilyStore cfs = Keyspace.open("Keyspace1").getColumnFamilyStore("Standard1");
cfs.disableAutoCompaction();
cfs.metadata.gcGraceSeconds(0);
long timestamp = System.currentTimeMillis();
RowMutation rm = new RowMutation("Keyspace1", Util.dk("ttl").key);
rm.add(
"Standard1", ByteBufferUtil.bytes("col"), ByteBufferUtil.EMPTY_BYTE_BUFFER, timestamp, 1);
rm.add(
"Standard1", ByteBufferUtil.bytes("col7"), ByteBufferUtil.EMPTY_BYTE_BUFFER, timestamp, 1);
rm.apply();
cfs.forceBlockingFlush();
rm = new RowMutation("Keyspace1", Util.dk("ttl").key);
rm.add(
"Standard1", ByteBufferUtil.bytes("col2"), ByteBufferUtil.EMPTY_BYTE_BUFFER, timestamp, 1);
rm.apply();
cfs.forceBlockingFlush();
rm = new RowMutation("Keyspace1", Util.dk("ttl").key);
rm.add(
"Standard1", ByteBufferUtil.bytes("col3"), ByteBufferUtil.EMPTY_BYTE_BUFFER, timestamp, 1);
rm.apply();
cfs.forceBlockingFlush();
rm = new RowMutation("Keyspace1", Util.dk("ttl").key);
rm.add(
"Standard1",
ByteBufferUtil.bytes("col311"),
ByteBufferUtil.EMPTY_BYTE_BUFFER,
timestamp,
1);
rm.apply();
cfs.forceBlockingFlush();
Thread.sleep(2000); // wait for ttl to expire
assertEquals(4, cfs.getSSTables().size());
cfs.enableAutoCompaction(true);
assertEquals(0, cfs.getSSTables().size());
}
@Test
public void testAggressiveFullyExpired() {
String KEYSPACE1 = "Keyspace1";
ColumnFamilyStore cfs = Keyspace.open("Keyspace1").getColumnFamilyStore("Standard1");
cfs.disableAutoCompaction();
cfs.metadata.gcGraceSeconds(0);
DecoratedKey ttlKey = Util.dk("ttl");
RowMutation rm = new RowMutation("Keyspace1", ttlKey.key);
rm.add("Standard1", ByteBufferUtil.bytes("col1"), ByteBufferUtil.EMPTY_BYTE_BUFFER, 1, 1);
rm.add("Standard1", ByteBufferUtil.bytes("col2"), ByteBufferUtil.EMPTY_BYTE_BUFFER, 3, 1);
rm.applyUnsafe();
cfs.forceBlockingFlush();
rm = new RowMutation(KEYSPACE1, ttlKey.key);
rm.add("Standard1", ByteBufferUtil.bytes("col1"), ByteBufferUtil.EMPTY_BYTE_BUFFER, 2, 1);
rm.add("Standard1", ByteBufferUtil.bytes("col2"), ByteBufferUtil.EMPTY_BYTE_BUFFER, 5, 1);
rm.applyUnsafe();
cfs.forceBlockingFlush();
rm = new RowMutation(KEYSPACE1, ttlKey.key);
rm.add("Standard1", ByteBufferUtil.bytes("col1"), ByteBufferUtil.EMPTY_BYTE_BUFFER, 4, 1);
rm.add("Standard1", ByteBufferUtil.bytes("shadow"), ByteBufferUtil.EMPTY_BYTE_BUFFER, 7, 1);
rm.applyUnsafe();
cfs.forceBlockingFlush();
rm = new RowMutation(KEYSPACE1, ttlKey.key);
rm.add("Standard1", ByteBufferUtil.bytes("shadow"), ByteBufferUtil.EMPTY_BYTE_BUFFER, 6, 3);
rm.add("Standard1", ByteBufferUtil.bytes("col2"), ByteBufferUtil.EMPTY_BYTE_BUFFER, 8, 1);
rm.applyUnsafe();
cfs.forceBlockingFlush();
Set<SSTableReader> sstables = Sets.newHashSet(cfs.getSSTables());
int now = (int) (System.currentTimeMillis() / 1000);
int gcBefore = now + 2;
Set<SSTableReader> expired =
CompactionController.getFullyExpiredSSTables(
cfs, sstables, Collections.EMPTY_SET, gcBefore);
assertEquals(2, expired.size());
cfs.clearUnsafe();
}
/**
* Starts sending/receiving our list of differences to/from the remote endpoint: creates a
* callback that will be called out of band once the streams complete.
*/
void performStreamingRepair() throws IOException {
logger.info(
"Performing streaming repair of "
+ differences.size()
+ " ranges with "
+ remote
+ " for "
+ range);
ColumnFamilyStore cfstore = Table.open(tablename).getColumnFamilyStore(cfname);
try {
Collection<SSTableReader> sstables = cfstore.getSSTables();
Callback callback = new Callback();
// send ranges to the remote node
StreamOutSession outsession = StreamOutSession.create(tablename, remote, callback);
StreamOut.transferSSTables(outsession, sstables, differences, OperationType.AES);
// request ranges from the remote node
StreamIn.requestRanges(remote, tablename, differences, callback, OperationType.AES);
} catch (Exception e) {
throw new IOException("Streaming repair failed.", e);
}
}
public CommitLogReplayer() {
this.tablesRecovered = new NonBlockingHashSet<Table>();
this.futures = new ArrayList<Future<?>>();
this.buffer = new byte[4096];
this.invalidMutations = new HashMap<Integer, AtomicInteger>();
// count the number of replayed mutation. We don't really care about atomicity, but we need it
// to be a reference.
this.replayedCount = new AtomicInteger();
// compute per-CF and global replay positions
this.cfPositions = new HashMap<Integer, ReplayPosition>();
for (ColumnFamilyStore cfs : ColumnFamilyStore.all()) {
// it's important to call RP.gRP per-cf, before aggregating all the positions w/ the
// Ordering.min call
// below: gRP will return NONE if there are no flushed sstables, which is important to have in
// the
// list (otherwise we'll just start replay from the first flush position that we do have,
// which is not correct).
ReplayPosition rp = ReplayPosition.getReplayPosition(cfs.getSSTables());
cfPositions.put(cfs.metadata.cfId, rp);
}
this.globalPosition = Ordering.from(ReplayPosition.comparator).min(cfPositions.values());
this.checksum = new PureJavaCrc32();
}
/**
* Retrieves a local subBlock
*
* @param blockId row key
* @param sblockId SubBlock column name
* @param offset inside the sblock
* @return a local sublock
* @throws TException
*/
private LocalBlock getLocalSubBlock(
String subBlockCFName, ByteBuffer blockId, ByteBuffer sblockId, int offset)
throws TException {
DecoratedKey<Token<?>> decoratedKey =
new DecoratedKey<Token<?>>(StorageService.getPartitioner().getToken(blockId), blockId);
Table table = Table.open(cfsKeyspace);
ColumnFamilyStore sblockStore = table.getColumnFamilyStore(subBlockCFName);
Collection<SSTableReader> sstables = sblockStore.getSSTables();
for (SSTableReader sstable : sstables) {
long position = sstable.getPosition(decoratedKey, Operator.EQ);
if (position == -1) continue;
String filename = sstable.descriptor.filenameFor(Component.DATA);
RandomAccessFile raf = null;
int mappedLength = -1;
MappedByteBuffer mappedData = null;
MappedFileDataInput file = null;
try {
raf = new RandomAccessFile(filename, "r");
assert position < raf.length();
mappedLength =
(raf.length() - position) < Integer.MAX_VALUE
? (int) (raf.length() - position)
: Integer.MAX_VALUE;
mappedData = raf.getChannel().map(FileChannel.MapMode.READ_ONLY, position, mappedLength);
file = new MappedFileDataInput(mappedData, filename, 0);
if (file == null) continue;
// Verify key was found in data file
DecoratedKey keyInDisk =
SSTableReader.decodeKey(
sstable.partitioner, sstable.descriptor, ByteBufferUtil.readWithShortLength(file));
assert keyInDisk.equals(decoratedKey)
: String.format("%s != %s in %s", keyInDisk, decoratedKey, file.getPath());
long rowSize = SSTableReader.readRowSize(file, sstable.descriptor);
assert rowSize > 0;
assert rowSize < mappedLength;
Filter bf = IndexHelper.defreezeBloomFilter(file, sstable.descriptor.usesOldBloomFilter);
// verify this column in in this version of the row.
if (!bf.isPresent(sblockId)) continue;
List<IndexHelper.IndexInfo> indexList = IndexHelper.deserializeIndex(file);
// we can stop early if bloom filter says none of the
// columns actually exist -- but,
// we can't stop before initializing the cf above, in
// case there's a relevant tombstone
ColumnFamilySerializer serializer = ColumnFamily.serializer();
try {
ColumnFamily cf =
serializer.deserializeFromSSTableNoColumns(
ColumnFamily.create(sstable.metadata), file);
if (cf.isMarkedForDelete()) continue;
} catch (Exception e) {
e.printStackTrace();
throw new IOException(
serializer
+ " failed to deserialize "
+ sstable.getColumnFamilyName()
+ " with "
+ sstable.metadata
+ " from "
+ file,
e);
}
Integer sblockLength = null;
if (indexList == null) sblockLength = seekToSubColumn(sstable.metadata, file, sblockId);
else sblockLength = seekToSubColumn(sstable.metadata, file, sblockId, indexList);
if (sblockLength == null || sblockLength < 0) continue;
int bytesReadFromStart = mappedLength - (int) file.bytesRemaining();
if (logger.isDebugEnabled())
logger.debug("BlockLength = " + sblockLength + " Availible " + file.bytesRemaining());
assert offset <= sblockLength : String.format("%d > %d", offset, sblockLength);
long dataOffset = position + bytesReadFromStart;
if (file.bytesRemaining() == 0 || sblockLength == 0) continue;
return new LocalBlock(file.getPath(), dataOffset + offset, sblockLength - offset);
} catch (IOException e) {
throw new TException(e);
} finally {
FileUtils.closeQuietly(raf);
}
}
return null;
}
public static void assertMaxTimestamp(ColumnFamilyStore cfs, long maxTimestampExpected) {
long maxTimestampObserved = Long.MIN_VALUE;
for (SSTableReader sstable : cfs.getSSTables())
maxTimestampObserved = Math.max(sstable.getMaxTimestamp(), maxTimestampObserved);
assertEquals(maxTimestampExpected, maxTimestampObserved);
}
@Test
public void testParallelLeveledCompaction() throws Exception {
String ksname = "Keyspace1";
String cfname = "StandardLeveled";
Keyspace keyspace = Keyspace.open(ksname);
ColumnFamilyStore store = keyspace.getColumnFamilyStore(cfname);
store.disableAutoCompaction();
LeveledCompactionStrategy lcs = (LeveledCompactionStrategy) store.getCompactionStrategy();
ByteBuffer value =
ByteBuffer.wrap(new byte[100 * 1024]); // 100 KB value, make it easy to have multiple files
// Enough data to have a level 1 and 2
int rows = 128;
int columns = 10;
// Adds enough data to trigger multiple sstable per level
for (int r = 0; r < rows; r++) {
DecoratedKey key = Util.dk(String.valueOf(r));
RowMutation rm = new RowMutation(ksname, key.key);
for (int c = 0; c < columns; c++) {
rm.add(cfname, ByteBufferUtil.bytes("column" + c), value, 0);
}
rm.apply();
store.forceBlockingFlush();
}
// Execute LCS in parallel
ExecutorService executor =
new ThreadPoolExecutor(
4, 4, Long.MAX_VALUE, TimeUnit.SECONDS, new LinkedBlockingDeque<Runnable>());
List<Runnable> tasks = new ArrayList<Runnable>();
while (true) {
while (true) {
final AbstractCompactionTask t = lcs.getMaximalTask(Integer.MIN_VALUE);
if (t == null) break;
tasks.add(
new Runnable() {
public void run() {
t.execute(null);
}
});
}
if (tasks.isEmpty()) break;
List<Future<?>> futures = new ArrayList<Future<?>>(tasks.size());
for (Runnable r : tasks) futures.add(executor.submit(r));
FBUtilities.waitOnFutures(futures);
tasks.clear();
}
// Assert all SSTables are lined up correctly.
LeveledManifest manifest = lcs.manifest;
int levels = manifest.getLevelCount();
for (int level = 0; level < levels; level++) {
List<SSTableReader> sstables = manifest.getLevel(level);
// score check
assert (double) SSTable.getTotalBytes(sstables) / manifest.maxBytesForLevel(level) < 1.00;
// overlap check for levels greater than 0
if (level > 0) {
for (SSTableReader sstable : sstables) {
Set<SSTableReader> overlaps = LeveledManifest.overlapping(sstable, sstables);
assert overlaps.size() == 1 && overlaps.contains(sstable);
}
}
}
for (SSTableReader sstable : store.getSSTables()) {
assert sstable.getSSTableLevel() == sstable.getSSTableLevel();
}
}
public static Future<?> compactAll(ColumnFamilyStore cfs, int gcBefore) {
List<Descriptor> descriptors = new ArrayList<>();
for (SSTableReader sstable : cfs.getSSTables()) descriptors.add(sstable.descriptor);
return CompactionManager.instance.submitUserDefined(cfs, descriptors, gcBefore);
}
@Test
public void testUncheckedTombstoneSizeTieredCompaction() throws Exception {
Keyspace keyspace = Keyspace.open(KEYSPACE1);
ColumnFamilyStore store = keyspace.getColumnFamilyStore(CF_STANDARD1);
store.clearUnsafe();
store.metadata.gcGraceSeconds(1);
store.metadata.compactionStrategyOptions.put("tombstone_compaction_interval", "1");
store.metadata.compactionStrategyOptions.put("unchecked_tombstone_compaction", "false");
store.reload();
store.setCompactionStrategyClass(SizeTieredCompactionStrategy.class.getName());
// disable compaction while flushing
store.disableAutoCompaction();
// Populate sstable1 with with keys [0..9]
populate(KEYSPACE1, CF_STANDARD1, 0, 9, 3); // ttl=3s
store.forceBlockingFlush();
// Populate sstable2 with with keys [10..19] (keys do not overlap with SSTable1)
long timestamp2 = populate(KEYSPACE1, CF_STANDARD1, 10, 19, 3); // ttl=3s
store.forceBlockingFlush();
assertEquals(2, store.getSSTables().size());
Iterator<SSTableReader> it = store.getSSTables().iterator();
long originalSize1 = it.next().uncompressedLength();
long originalSize2 = it.next().uncompressedLength();
// wait enough to force single compaction
TimeUnit.SECONDS.sleep(5);
// enable compaction, submit background and wait for it to complete
store.enableAutoCompaction();
FBUtilities.waitOnFutures(CompactionManager.instance.submitBackground(store));
while (CompactionManager.instance.getPendingTasks() > 0
|| CompactionManager.instance.getActiveCompactions() > 0) TimeUnit.SECONDS.sleep(1);
// even though both sstables were candidate for tombstone compaction
// it was not executed because they have an overlapping token range
assertEquals(2, store.getSSTables().size());
it = store.getSSTables().iterator();
long newSize1 = it.next().uncompressedLength();
long newSize2 = it.next().uncompressedLength();
assertEquals(
"candidate sstable should not be tombstone-compacted because its key range overlap with other sstable",
originalSize1,
newSize1);
assertEquals(
"candidate sstable should not be tombstone-compacted because its key range overlap with other sstable",
originalSize2,
newSize2);
// now let's enable the magic property
store.metadata.compactionStrategyOptions.put("unchecked_tombstone_compaction", "true");
store.reload();
// submit background task again and wait for it to complete
FBUtilities.waitOnFutures(CompactionManager.instance.submitBackground(store));
while (CompactionManager.instance.getPendingTasks() > 0
|| CompactionManager.instance.getActiveCompactions() > 0) TimeUnit.SECONDS.sleep(1);
// we still have 2 sstables, since they were not compacted against each other
assertEquals(2, store.getSSTables().size());
it = store.getSSTables().iterator();
newSize1 = it.next().uncompressedLength();
newSize2 = it.next().uncompressedLength();
assertTrue(
"should be less than " + originalSize1 + ", but was " + newSize1, newSize1 < originalSize1);
assertTrue(
"should be less than " + originalSize2 + ", but was " + newSize2, newSize2 < originalSize2);
// make sure max timestamp of compacted sstables is recorded properly after compaction.
assertMaxTimestamp(store, timestamp2);
}
@Test
@Ignore("making ranges based on the keys, not on the tokens")
public void testNeedsCleanup() {
Keyspace keyspace = Keyspace.open(KEYSPACE1);
ColumnFamilyStore store = keyspace.getColumnFamilyStore("CF_STANDARD1");
store.clearUnsafe();
// disable compaction while flushing
store.disableAutoCompaction();
// write three groups of 9 keys: 001, 002, ... 008, 009
// 101, 102, ... 108, 109
// 201, 202, ... 208, 209
for (int i = 1; i < 10; i++) {
insertRowWithKey(i);
insertRowWithKey(i + 100);
insertRowWithKey(i + 200);
}
store.forceBlockingFlush();
assertEquals(1, store.getSSTables().size());
SSTableReader sstable = store.getSSTables().iterator().next();
// contiguous range spans all data
assertFalse(CompactionManager.needsCleanup(sstable, makeRanges(0, 209)));
assertFalse(CompactionManager.needsCleanup(sstable, makeRanges(0, 210)));
// separate ranges span all data
assertFalse(
CompactionManager.needsCleanup(
sstable,
makeRanges(
0, 9,
100, 109,
200, 209)));
assertFalse(
CompactionManager.needsCleanup(
sstable,
makeRanges(
0, 109,
200, 210)));
assertFalse(
CompactionManager.needsCleanup(
sstable,
makeRanges(
0, 9,
100, 210)));
// one range is missing completely
assertTrue(
CompactionManager.needsCleanup(
sstable,
makeRanges(
100, 109,
200, 209)));
assertTrue(
CompactionManager.needsCleanup(
sstable,
makeRanges(
0, 9,
200, 209)));
assertTrue(
CompactionManager.needsCleanup(
sstable,
makeRanges(
0, 9,
100, 109)));
// the beginning of one range is missing
assertTrue(
CompactionManager.needsCleanup(
sstable,
makeRanges(
1, 9,
100, 109,
200, 209)));
assertTrue(
CompactionManager.needsCleanup(
sstable,
makeRanges(
0, 9,
101, 109,
200, 209)));
assertTrue(
CompactionManager.needsCleanup(
sstable,
makeRanges(
0, 9,
100, 109,
201, 209)));
// the end of one range is missing
assertTrue(
CompactionManager.needsCleanup(
sstable,
makeRanges(
0, 8,
100, 109,
200, 209)));
assertTrue(
CompactionManager.needsCleanup(
sstable,
makeRanges(
0, 9,
100, 108,
200, 209)));
assertTrue(
CompactionManager.needsCleanup(
sstable,
makeRanges(
0, 9,
100, 109,
200, 208)));
// some ranges don't contain any data
assertFalse(
CompactionManager.needsCleanup(
sstable,
makeRanges(
0, 0,
0, 9,
50, 51,
100, 109,
150, 199,
200, 209,
300, 301)));
// same case, but with a middle range not covering some of the existing data
assertFalse(
CompactionManager.needsCleanup(
sstable,
makeRanges(
0, 0,
0, 9,
50, 51,
100, 103,
150, 199,
200, 209,
300, 301)));
}
@Test
public void testRangeTombstones() throws IOException, ExecutionException, InterruptedException {
Keyspace keyspace = Keyspace.open(KEYSPACE1);
ColumnFamilyStore cfs = keyspace.getColumnFamilyStore("Standard2");
cfs.clearUnsafe();
// disable compaction while flushing
cfs.disableAutoCompaction();
final CFMetaData cfmeta = cfs.metadata;
Directories dir = cfs.directories;
ArrayList<DecoratedKey> keys = new ArrayList<DecoratedKey>();
for (int i = 0; i < 4; i++) {
keys.add(Util.dk("" + i));
}
ArrayBackedSortedColumns cf = ArrayBackedSortedColumns.factory.create(cfmeta);
cf.addColumn(Util.column("01", "a", 1)); // this must not resurrect
cf.addColumn(Util.column("a", "a", 3));
cf.deletionInfo()
.add(
new RangeTombstone(
Util.cellname("0"),
Util.cellname("b"),
2,
(int) (System.currentTimeMillis() / 1000)),
cfmeta.comparator);
SSTableWriter writer =
new SSTableWriter(
cfs.getTempSSTablePath(dir.getDirectoryForNewSSTables()),
0,
0,
cfs.metadata,
StorageService.getPartitioner(),
new MetadataCollector(cfs.metadata.comparator));
writer.append(Util.dk("0"), cf);
writer.append(Util.dk("1"), cf);
writer.append(Util.dk("3"), cf);
cfs.addSSTable(writer.closeAndOpenReader());
writer =
new SSTableWriter(
cfs.getTempSSTablePath(dir.getDirectoryForNewSSTables()),
0,
0,
cfs.metadata,
StorageService.getPartitioner(),
new MetadataCollector(cfs.metadata.comparator));
writer.append(Util.dk("0"), cf);
writer.append(Util.dk("1"), cf);
writer.append(Util.dk("2"), cf);
writer.append(Util.dk("3"), cf);
cfs.addSSTable(writer.closeAndOpenReader());
Collection<SSTableReader> toCompact = cfs.getSSTables();
assert toCompact.size() == 2;
// Force compaction on first sstables. Since each row is in only one sstable, we will be using
// EchoedRow.
Util.compact(cfs, toCompact);
assertEquals(1, cfs.getSSTables().size());
// Now assert we do have the 4 keys
assertEquals(4, Util.getRangeSlice(cfs).size());
ArrayList<DecoratedKey> k = new ArrayList<DecoratedKey>();
for (Row r : Util.getRangeSlice(cfs)) {
k.add(r.key);
assertEquals(ByteBufferUtil.bytes("a"), r.cf.getColumn(Util.cellname("a")).value());
assertNull(r.cf.getColumn(Util.cellname("01")));
assertEquals(3, r.cf.getColumn(Util.cellname("a")).timestamp());
}
for (SSTableReader sstable : cfs.getSSTables()) {
StatsMetadata stats = sstable.getSSTableMetadata();
assertEquals(ByteBufferUtil.bytes("0"), stats.minColumnNames.get(0));
assertEquals(ByteBufferUtil.bytes("b"), stats.maxColumnNames.get(0));
}
assertEquals(keys, k);
}