use of org.apache.cassandra.spark.stats.Stats in project spark-cassandra-bulkreader by jberragan.
the class IndexDbTests method testReadToken.
@Test
public void testReadToken() {
qt().withExamples(500).forAll(TestUtils.partitioners(), integers().all()).checkAssert((partitioner, value) -> {
final IPartitioner iPartitioner = FourZero.getPartitioner(partitioner);
final BigInteger expectedToken = token(iPartitioner, value);
try (final DataInputStream in = mockDataInputStream(value, 0)) {
IndexDbUtils.readNextToken(iPartitioner, in, new Stats() {
public void readPartitionIndexDb(ByteBuffer key, BigInteger token) {
assertEquals(value.intValue(), key.getInt());
assertEquals(expectedToken, token);
}
});
} catch (IOException e) {
throw new RuntimeException(e);
}
});
}
use of org.apache.cassandra.spark.stats.Stats in project spark-cassandra-bulkreader by jberragan.
the class SSTableReaderTests method testSkipNoPartitions.
@Test
public void testSkipNoPartitions() {
runTest((partitioner, dir, bridge) -> {
// write an SSTable
final TestSchema schema = TestSchema.basic(bridge);
TestUtils.writeSSTable(bridge, dir, partitioner, schema, (writer) -> {
for (int i = 0; i < NUM_ROWS; i++) {
for (int j = 0; j < NUM_COLS; j++) {
writer.write(i, j, i + j);
}
}
});
assertEquals(1, countSSTables(dir));
final Path dataFile = getFirstFileType(dir, DataLayer.FileType.DATA);
final Path summaryFile = getFirstFileType(dir, DataLayer.FileType.SUMMARY);
final TableMetadata metaData = schema.schemaBuilder(partitioner).tableMetaData();
final TestDataLayer dataLayer = new TestDataLayer(bridge, Collections.singletonList(dataFile));
SummaryDbUtils.Summary summary;
try (final InputStream in = new BufferedInputStream(Files.newInputStream(summaryFile))) {
summary = SummaryDbUtils.readSummary(in, metaData.partitioner, metaData.params.minIndexInterval, metaData.params.maxIndexInterval);
}
// set Spark token range equal to SSTable token range
final Range<BigInteger> sparkTokenRange = Range.closed(FourZeroUtils.tokenToBigInteger(summary.first().getToken()), FourZeroUtils.tokenToBigInteger(summary.last().getToken()));
final SparkRangeFilter rangeFilter = SparkRangeFilter.create(sparkTokenRange);
final AtomicBoolean skipped = new AtomicBoolean(false);
final Stats stats = new Stats() {
@Override
public void skippedPartition(ByteBuffer key, BigInteger token) {
LOGGER.error("Skipped partition when should not: " + token);
skipped.set(true);
}
};
final FourZeroSSTableReader reader = openReader(metaData, dataLayer.listSSTables().findFirst().orElseThrow(() -> new RuntimeException("Could not find SSTable")), Collections.singletonList(rangeFilter), true, stats);
// shouldn't skip any partitions here
assertEquals(NUM_ROWS * NUM_COLS, countAndValidateRows(reader));
assertFalse(skipped.get());
});
}
use of org.apache.cassandra.spark.stats.Stats in project spark-cassandra-bulkreader by jberragan.
the class SSTableReaderTests method testPartialFilterMatch.
@Test
public void testPartialFilterMatch() {
runTest((partitioner, dir, bridge) -> {
// write an SSTable
final TestSchema schema = TestSchema.basic(bridge);
TestUtils.writeSSTable(bridge, dir, partitioner, schema, (writer) -> {
for (int i = 0; i < NUM_ROWS; i++) {
for (int j = 0; j < NUM_COLS; j++) {
writer.write(i, j, i + j);
}
}
});
assertEquals(1, countSSTables(dir));
final Path dataFile = getFirstFileType(dir, DataLayer.FileType.DATA);
final TableMetadata metaData = new FourZeroSchemaBuilder(schema.createStmt, schema.keyspace, new ReplicationFactor(ReplicationFactor.ReplicationStrategy.SimpleStrategy, ImmutableMap.of("replication_factor", 1)), partitioner).tableMetaData();
final TestDataLayer dataLayer = new TestDataLayer(bridge, Collections.singletonList(dataFile));
final ByteBuffer key1 = Int32Type.instance.fromString("0");
final BigInteger token1 = bridge.hash(partitioner, key1);
final PartitionKeyFilter keyInSSTable = PartitionKeyFilter.create(key1, token1);
final SparkRangeFilter rangeFilter = SparkRangeFilter.create(Range.closed(token1, token1));
final ByteBuffer key2 = Int32Type.instance.fromString("55");
final BigInteger token2 = bridge.hash(partitioner, key2);
final PartitionKeyFilter keyNotInSSTable = PartitionKeyFilter.create(key2, token2);
final List<CustomFilter> filters = Arrays.asList(rangeFilter, keyInSSTable, keyNotInSSTable);
final AtomicBoolean pass = new AtomicBoolean(true);
final AtomicInteger skipCount = new AtomicInteger(0);
final Stats stats = new Stats() {
@Override
public void skippedPartition(ByteBuffer key, BigInteger token) {
LOGGER.info("Skipping partition: " + token);
skipCount.incrementAndGet();
if (filters.stream().anyMatch(filter -> !filter.skipPartition(key, token))) {
LOGGER.info("Should not skip partition: " + token);
pass.set(false);
}
}
};
final FourZeroSSTableReader reader = openReader(metaData, dataLayer.listSSTables().findFirst().orElseThrow(() -> new RuntimeException("Could not find SSTable")), filters, false, stats);
final int rows = countAndValidateRows(reader);
assertTrue(skipCount.get() > 0);
assertEquals(NUM_COLS, rows);
// should skip partitions not matching filters
assertEquals((NUM_ROWS - skipCount.get()) * NUM_COLS, rows);
assertTrue(pass.get());
});
}
use of org.apache.cassandra.spark.stats.Stats in project spark-cassandra-bulkreader by jberragan.
the class SSTableReaderTests method testSkipPartitionsCompactionScanner.
@Test
public void testSkipPartitionsCompactionScanner() {
runTest((partitioner, dir, bridge) -> {
// write an SSTable
final TestSchema schema = TestSchema.basic(bridge);
TestUtils.writeSSTable(bridge, dir, partitioner, schema, (writer) -> {
for (int i = 0; i < NUM_ROWS; i++) {
for (int j = 0; j < NUM_COLS; j++) {
writer.write(i, j, i + j);
}
}
});
assertEquals(1, countSSTables(dir));
final Path dataFile = getFirstFileType(dir, DataLayer.FileType.DATA);
final TableMetadata metaData = schema.schemaBuilder(partitioner).tableMetaData();
final Set<SparkSSTableReader> readers = new HashSet<>(1);
final TestDataLayer dataLayer = new TestDataLayer(bridge, Collections.singletonList(dataFile), schema.buildSchema()) {
public SSTablesSupplier sstables(final List<CustomFilter> filters) {
return new SSTablesSupplier() {
public <T extends SparkSSTableReader> Set<T> openAll(ReaderOpener<T> readerOpener) {
return (Set<T>) readers;
}
};
}
};
final Range<BigInteger> sparkTokenRange;
switch(partitioner) {
case Murmur3Partitioner:
sparkTokenRange = Range.closed(BigInteger.valueOf(-9223372036854775808L), BigInteger.valueOf(3074457345618258602L));
break;
case RandomPartitioner:
sparkTokenRange = Range.closed(BigInteger.ZERO, new BigInteger("916176208424801638531839357843455255"));
break;
default:
throw new RuntimeException("Unexpected partitioner: " + partitioner);
}
final SparkRangeFilter rangeFilter = SparkRangeFilter.create(sparkTokenRange);
final AtomicBoolean pass = new AtomicBoolean(true);
final AtomicInteger skipCount = new AtomicInteger(0);
final Stats stats = new Stats() {
@Override
public void skippedPartition(ByteBuffer key, BigInteger token) {
LOGGER.info("Skipping partition: " + token);
skipCount.incrementAndGet();
if (sparkTokenRange.contains(token)) {
LOGGER.info("Should not skip partition: " + token);
pass.set(false);
}
}
};
final FourZeroSSTableReader reader = openReader(metaData, dataLayer.listSSTables().findFirst().orElseThrow(() -> new RuntimeException("Could not find SSTable")), Collections.singletonList(rangeFilter), false, stats);
readers.add(reader);
// read the SSTable end-to-end using SparkRowIterator and verify it skips the required partitions
// and all the partitions returned are within the Spark token range.
final SparkRowIterator it = new SparkRowIterator(dataLayer);
int count = 0;
while (it.next()) {
final InternalRow row = it.get();
assertEquals(row.getInt(2), row.getInt(0) + row.getInt(1));
final DecoratedKey key = FourZero.getPartitioner(partitioner).decorateKey((ByteBuffer) ByteBuffer.allocate(4).putInt(row.getInt(0)).flip());
final BigInteger token = FourZeroUtils.tokenToBigInteger(key.getToken());
assertTrue(sparkTokenRange.contains(token));
count++;
}
assertTrue(skipCount.get() > 0);
// should skip out of range partitions here
assertEquals((NUM_ROWS - skipCount.get()) * NUM_COLS, count);
assertTrue(pass.get());
});
}
use of org.apache.cassandra.spark.stats.Stats in project spark-cassandra-bulkreader by jberragan.
the class SSTableReaderTests method testIncrementalRepair.
// incremental repair
@Test
public void testIncrementalRepair() {
runTest((partitioner, dir, bridge) -> {
final TestSchema schema = TestSchema.basic(bridge);
final int numSSTables = 4;
final int numRepaired = 2;
final int numUnRepaired = numSSTables - numRepaired;
// write some SSTables
for (int a = 0; a < numSSTables; a++) {
final int pos = a * NUM_ROWS;
TestUtils.writeSSTable(bridge, dir, partitioner, schema, (writer) -> {
for (int i = pos; i < pos + NUM_ROWS; i++) {
for (int j = 0; j < NUM_COLS; j++) {
writer.write(i, j, i + j);
}
}
});
}
assertEquals(numSSTables, countSSTables(dir));
final TableMetadata metaData = new FourZeroSchemaBuilder(schema.createStmt, schema.keyspace, new ReplicationFactor(ReplicationFactor.ReplicationStrategy.SimpleStrategy, ImmutableMap.of("replication_factor", 1)), partitioner).tableMetaData();
final TestDataLayer dataLayer = new TestDataLayer(bridge, getFileType(dir, DataLayer.FileType.DATA).collect(Collectors.toList()));
final AtomicInteger skipCount = new AtomicInteger(0);
final Stats stats = new Stats() {
@Override
public void skippedRepairedSSTable(DataLayer.SSTable ssTable, long repairedAt) {
skipCount.incrementAndGet();
}
};
// mark some SSTables as repaired
final Map<DataLayer.SSTable, Boolean> isRepaired = dataLayer.listSSTables().collect(Collectors.toMap(Function.identity(), a -> false));
int count = 0;
for (final DataLayer.SSTable ssTable : isRepaired.keySet()) {
if (count < numRepaired) {
isRepaired.put(ssTable, true);
count++;
}
}
final List<FourZeroSSTableReader> primaryReaders = dataLayer.listSSTables().map(ssTable -> openIncrementalReader(metaData, ssTable, stats, true, isRepaired.get(ssTable))).filter(reader -> !reader.ignore()).collect(Collectors.toList());
final List<FourZeroSSTableReader> nonPrimaryReaders = dataLayer.listSSTables().map(ssTable -> openIncrementalReader(metaData, ssTable, stats, false, isRepaired.get(ssTable))).filter(reader -> !reader.ignore()).collect(Collectors.toList());
// primary repair replica should read all sstables
assertEquals(numSSTables, primaryReaders.size());
// non-primary repair replica should only read unrepaired sstables
assertEquals(numUnRepaired, nonPrimaryReaders.size());
for (final FourZeroSSTableReader reader : nonPrimaryReaders) {
assertFalse(isRepaired.get(reader.sstable()));
}
assertEquals(numUnRepaired, skipCount.get());
final Set<FourZeroSSTableReader> toCompact = Stream.concat(primaryReaders.stream().filter(r -> isRepaired.get(r.sstable())), nonPrimaryReaders.stream()).collect(Collectors.toSet());
assertEquals(numSSTables, toCompact.size());
int rowCount = 0;
boolean[] found = new boolean[numSSTables * NUM_ROWS];
try (final CompactionStreamScanner scanner = new CompactionStreamScanner(metaData, partitioner, toCompact)) {
// iterate through CompactionScanner and verify we have all the partition keys we are looking for
final Rid rid = scanner.getRid();
while (scanner.hasNext()) {
scanner.next();
final int a = rid.getPartitionKey().asIntBuffer().get();
found[a] = true;
// extract clustering key value and column name
final ByteBuffer colBuf = rid.getColumnName();
final ByteBuffer clusteringKey = ByteBufUtils.readBytesWithShortLength(colBuf);
colBuf.get();
final String colName = ByteBufUtils.string(ByteBufUtils.readBytesWithShortLength(colBuf));
colBuf.get();
if (StringUtils.isEmpty(colName)) {
continue;
}
assertEquals("c", colName);
final int b = clusteringKey.asIntBuffer().get();
// extract value column
final int c = rid.getValue().asIntBuffer().get();
assertEquals(c, a + b);
rowCount++;
}
}
assertEquals(numSSTables * NUM_ROWS * NUM_COLS, rowCount);
for (final boolean b : found) {
assertTrue(b);
}
});
}
Aggregations