@Test
public void testPersistMergeCaseInsensitive() throws Exception {
final long timestamp = System.currentTimeMillis();
IncrementalIndex toPersist1 = IncrementalIndexTest.createCaseInsensitiveIndex(timestamp);
IncrementalIndex toPersist2 =
new IncrementalIndex(0L, QueryGranularity.NONE, new AggregatorFactory[] {});
toPersist2.add(
new MapBasedInputRow(
timestamp,
Arrays.asList("DIm1", "DIM2"),
ImmutableMap.<String, Object>of(
"dim1", "1", "dim2", "2", "DIm1", "10000", "DIM2", "100000000")));
toPersist2.add(
new MapBasedInputRow(
timestamp,
Arrays.asList("dIM1", "dIm2"),
ImmutableMap.<String, Object>of("DIm1", "1", "DIM2", "2", "dim1", "5", "dim2", "6")));
final File tempDir1 = Files.createTempDir();
final File tempDir2 = Files.createTempDir();
final File mergedDir = Files.createTempDir();
try {
QueryableIndex index1 = IndexIO.loadIndex(IndexMerger.persist(toPersist1, tempDir1));
Assert.assertEquals(2, index1.getTimeColumn().getLength());
Assert.assertEquals(
Arrays.asList("dim1", "dim2"), Lists.newArrayList(index1.getAvailableDimensions()));
Assert.assertEquals(2, index1.getColumnNames().size());
QueryableIndex index2 = IndexIO.loadIndex(IndexMerger.persist(toPersist2, tempDir2));
Assert.assertEquals(2, index2.getTimeColumn().getLength());
Assert.assertEquals(
Arrays.asList("dim1", "dim2"), Lists.newArrayList(index2.getAvailableDimensions()));
Assert.assertEquals(2, index2.getColumnNames().size());
QueryableIndex merged =
IndexIO.loadIndex(
IndexMerger.mergeQueryableIndex(
Arrays.asList(index1, index2), new AggregatorFactory[] {}, mergedDir));
Assert.assertEquals(3, merged.getTimeColumn().getLength());
Assert.assertEquals(
Arrays.asList("dim1", "dim2"), Lists.newArrayList(merged.getAvailableDimensions()));
Assert.assertEquals(2, merged.getColumnNames().size());
} finally {
FileUtils.deleteQuietly(tempDir1);
FileUtils.deleteQuietly(tempDir2);
FileUtils.deleteQuietly(mergedDir);
}
}
private static IncrementalIndex buildIncrementalIndexWithRows(
IncrementalIndexSchema schema, int maxRows, Iterable<InputRow> rows) {
Preconditions.checkNotNull(schema, "schema");
final IncrementalIndex incrementalIndex = new OnheapIncrementalIndex(schema, true, maxRows);
for (InputRow row : rows) {
try {
incrementalIndex.add(row);
} catch (IndexSizeExceededException e) {
throw Throwables.propagate(e);
}
}
return incrementalIndex;
}
@Test
public void testPersistEmptyColumn() throws Exception {
final IncrementalIndex toPersist1 =
new IncrementalIndex(0L, QueryGranularity.NONE, new AggregatorFactory[] {});
final IncrementalIndex toPersist2 =
new IncrementalIndex(0L, QueryGranularity.NONE, new AggregatorFactory[] {});
final File tmpDir1 = Files.createTempDir();
final File tmpDir2 = Files.createTempDir();
final File tmpDir3 = Files.createTempDir();
try {
toPersist1.add(
new MapBasedInputRow(
1L,
ImmutableList.of("dim1", "dim2"),
ImmutableMap.<String, Object>of("dim1", ImmutableList.of(), "dim2", "foo")));
toPersist2.add(
new MapBasedInputRow(
1L,
ImmutableList.of("dim1", "dim2"),
ImmutableMap.<String, Object>of("dim1", ImmutableList.of(), "dim2", "bar")));
final QueryableIndex index1 = IndexIO.loadIndex(IndexMerger.persist(toPersist1, tmpDir1));
final QueryableIndex index2 = IndexIO.loadIndex(IndexMerger.persist(toPersist1, tmpDir2));
final QueryableIndex merged =
IndexIO.loadIndex(
IndexMerger.mergeQueryableIndex(
Arrays.asList(index1, index2), new AggregatorFactory[] {}, tmpDir3));
Assert.assertEquals(1, index1.getTimeColumn().getLength());
Assert.assertEquals(
ImmutableList.of("dim2"), ImmutableList.copyOf(index1.getAvailableDimensions()));
Assert.assertEquals(1, index2.getTimeColumn().getLength());
Assert.assertEquals(
ImmutableList.of("dim2"), ImmutableList.copyOf(index2.getAvailableDimensions()));
Assert.assertEquals(1, merged.getTimeColumn().getLength());
Assert.assertEquals(
ImmutableList.of("dim2"), ImmutableList.copyOf(merged.getAvailableDimensions()));
} finally {
FileUtils.deleteQuietly(tmpDir1);
FileUtils.deleteQuietly(tmpDir2);
FileUtils.deleteQuietly(tmpDir3);
}
}
private void createTestIndex(File segmentDir) throws Exception {
List<String> rows =
Lists.newArrayList(
"2014102200,host1,10",
"2014102200,host2,20",
"2014102200,host3,30",
"2014102201,host1,10",
"2014102201,host2,20",
"2014102201,host3,30",
"2014102202,host1,10",
"2014102202,host2,20",
"2014102202,host3,30");
StringInputRowParser parser =
new StringInputRowParser(
new CSVParseSpec(
new TimestampSpec("timestamp", "yyyyMMddHH", null),
new DimensionsSpec(ImmutableList.of("host"), null, null),
null,
ImmutableList.of("timestamp", "host", "visited")),
Charsets.UTF_8.toString());
AggregatorFactory[] aggregators =
new AggregatorFactory[] {new LongSumAggregatorFactory("visited_sum", "visited")};
IncrementalIndex index = null;
try {
index = new OnheapIncrementalIndex(0, QueryGranularity.NONE, aggregators, true, 5000);
for (String line : rows) {
index.add(parser.parse(line));
}
IndexMerger.persist(index, segmentDir, null, new IndexSpec());
} finally {
if (index != null) {
index.close();
}
}
}
@BeforeClass
public static void setupClass() throws Exception {
incrementalIndex =
new OnheapIncrementalIndex(
0,
QueryGranularity.NONE,
new AggregatorFactory[] {new CountAggregatorFactory("count")},
true,
5000);
StringInputRowParser parser =
new StringInputRowParser(
new CSVParseSpec(
new TimestampSpec("timestamp", "iso", null),
new DimensionsSpec(ImmutableList.of("product", "tags"), null, null),
"\t",
ImmutableList.of("timestamp", "product", "tags")),
"UTF-8");
String[] rows =
new String[] {
"2011-01-12T00:00:00.000Z,product_1,t1\tt2\tt3",
"2011-01-13T00:00:00.000Z,product_2,t3\tt4\tt5",
"2011-01-14T00:00:00.000Z,product_3,t5\tt6\tt7",
};
for (String row : rows) {
incrementalIndex.add(parser.parse(row));
}
persistedSegmentDir = Files.createTempDir();
TestHelper.getTestIndexMerger()
.persist(
incrementalIndex,
persistedSegmentDir,
ImmutableMap.<String, Object>of(),
new IndexSpec());
queryableIndex = TestHelper.getTestIndexIO().loadIndex(persistedSegmentDir);
}
@Setup
public void setup() throws IOException {
log.info("SETUP CALLED AT " + System.currentTimeMillis());
if (ComplexMetrics.getSerdeForType("hyperUnique") == null) {
ComplexMetrics.registerSerde("hyperUnique", new HyperUniquesSerde(Hashing.murmur3_128()));
}
executorService = Execs.multiThreaded(numSegments, "TimeseriesThreadPool");
setupQueries();
String[] schemaQuery = schemaAndQuery.split("\\.");
String schemaName = schemaQuery[0];
String queryName = schemaQuery[1];
schemaInfo = BenchmarkSchemas.SCHEMA_MAP.get(schemaName);
query = SCHEMA_QUERY_MAP.get(schemaName).get(queryName);
incIndexes = new ArrayList<>();
for (int i = 0; i < numSegments; i++) {
log.info("Generating rows for segment " + i);
BenchmarkDataGenerator gen =
new BenchmarkDataGenerator(
schemaInfo.getColumnSchemas(),
RNG_SEED + i,
schemaInfo.getDataInterval(),
rowsPerSegment);
IncrementalIndex incIndex = makeIncIndex();
for (int j = 0; j < rowsPerSegment; j++) {
InputRow row = gen.nextRow();
if (j % 10000 == 0) {
log.info(j + " rows generated.");
}
incIndex.add(row);
}
log.info(rowsPerSegment + " rows generated");
incIndexes.add(incIndex);
}
File tmpFile = Files.createTempDir();
log.info("Using temp dir: " + tmpFile.getAbsolutePath());
tmpFile.deleteOnExit();
qIndexes = new ArrayList<>();
for (int i = 0; i < numSegments; i++) {
File indexFile = INDEX_MERGER_V9.persist(incIndexes.get(i), tmpFile, new IndexSpec());
QueryableIndex qIndex = INDEX_IO.loadIndex(indexFile);
qIndexes.add(qIndex);
}
factory =
new TimeseriesQueryRunnerFactory(
new TimeseriesQueryQueryToolChest(
QueryBenchmarkUtil.NoopIntervalChunkingQueryRunnerDecorator()),
new TimeseriesQueryEngine(),
QueryBenchmarkUtil.NOOP_QUERYWATCHER);
}
@AfterClass
public static void cleanup() throws Exception {
queryableIndex.close();
incrementalIndex.close();
FileUtils.deleteDirectory(persistedSegmentDir);
}
@Override
public Sequence<Row> processSubqueryResult(
GroupByQuery subquery, GroupByQuery query, Sequence<Row> subqueryResult) {
final Set<AggregatorFactory> aggs = Sets.newHashSet();
// Nested group-bys work by first running the inner query and then materializing the results in
// an incremental
// index which the outer query is then run against. To build the incremental index, we use the
// fieldNames from
// the aggregators for the outer query to define the column names so that the index will match
// the query. If
// there are multiple types of aggregators in the outer query referencing the same fieldName, we
// will try to build
// multiple columns of the same name using different aggregator types and will fail. Here, we
// permit multiple
// aggregators of the same type referencing the same fieldName (and skip creating identical
// columns for the
// subsequent ones) and return an error if the aggregator types are different.
for (AggregatorFactory aggregatorFactory : query.getAggregatorSpecs()) {
for (final AggregatorFactory transferAgg : aggregatorFactory.getRequiredColumns()) {
if (Iterables.any(
aggs,
new Predicate<AggregatorFactory>() {
@Override
public boolean apply(AggregatorFactory agg) {
return agg.getName().equals(transferAgg.getName()) && !agg.equals(transferAgg);
}
})) {
throw new IAE(
"Inner aggregator can currently only be referenced by a single type of outer aggregator"
+ " for '%s'",
transferAgg.getName());
}
aggs.add(transferAgg);
}
}
// We need the inner incremental index to have all the columns required by the outer query
final GroupByQuery innerQuery =
new GroupByQuery.Builder(subquery)
.setAggregatorSpecs(Lists.newArrayList(aggs))
.setInterval(subquery.getIntervals())
.setPostAggregatorSpecs(Lists.<PostAggregator>newArrayList())
.build();
final GroupByQuery outerQuery =
new GroupByQuery.Builder(query)
.setLimitSpec(query.getLimitSpec().merge(subquery.getLimitSpec()))
.build();
final IncrementalIndex innerQueryResultIndex =
makeIncrementalIndex(
innerQuery.withOverriddenContext(
ImmutableMap.<String, Object>of(GroupByQueryHelper.CTX_KEY_SORT_RESULTS, true)),
subqueryResult);
// Outer query might have multiple intervals, but they are expected to be non-overlapping and
// sorted which
// is ensured by QuerySegmentSpec.
// GroupByQueryEngine can only process one interval at a time, so we need to call it once per
// interval
// and concatenate the results.
final IncrementalIndex outerQueryResultIndex =
makeIncrementalIndex(
outerQuery,
Sequences.concat(
Sequences.map(
Sequences.simple(outerQuery.getIntervals()),
new Function<Interval, Sequence<Row>>() {
@Override
public Sequence<Row> apply(Interval interval) {
return process(
outerQuery.withQuerySegmentSpec(
new MultipleIntervalSegmentSpec(ImmutableList.of(interval))),
new IncrementalIndexStorageAdapter(innerQueryResultIndex));
}
})));
innerQueryResultIndex.close();
return new ResourceClosingSequence<>(
outerQuery.applyLimit(GroupByQueryHelper.postAggregate(query, outerQueryResultIndex)),
outerQueryResultIndex);
}