use of org.apache.hudi.common.table.timeline.HoodieTimeline in project hudi by apache.
the class TestCleaner method insertFirstBigBatchForClientCleanerTest.
/**
* Helper method to do first batch of insert for clean by versions/commits tests.
*
* @param cfg Hoodie Write Config
* @param client Hoodie Client
* @param recordGenFunction Function to generate records for insertion
* @param insertFn Insertion API for testing
* @throws Exception in case of error
*/
private Pair<String, JavaRDD<WriteStatus>> insertFirstBigBatchForClientCleanerTest(HoodieWriteConfig cfg, SparkRDDWriteClient client, Function2<List<HoodieRecord>, String, Integer> recordGenFunction, Function3<JavaRDD<WriteStatus>, SparkRDDWriteClient, JavaRDD<HoodieRecord>, String> insertFn, HoodieCleaningPolicy cleaningPolicy) throws Exception {
/*
* do a big insert (this is basically same as insert part of upsert, just adding it here so we can catch breakages
* in insert(), if the implementation diverges.)
*/
String newCommitTime = client.startCommit();
List<HoodieRecord> records = recordGenFunction.apply(newCommitTime, BIG_BATCH_INSERT_SIZE);
JavaRDD<HoodieRecord> writeRecords = jsc.parallelize(records, 5);
JavaRDD<WriteStatus> statuses = insertFn.apply(client, writeRecords, newCommitTime);
// Verify there are no errors
assertNoWriteErrors(statuses.collect());
// verify that there is a commit
metaClient = HoodieTableMetaClient.reload(metaClient);
HoodieTimeline timeline = new HoodieActiveTimeline(metaClient).getCommitTimeline();
assertEquals(1, timeline.findInstantsAfter("000", Integer.MAX_VALUE).countInstants(), "Expecting a single commit.");
// Should have 100 records in table (check using Index), all in locations marked at commit
HoodieTable table = HoodieSparkTable.create(client.getConfig(), context, metaClient);
if (client.getConfig().shouldAutoCommit()) {
assertFalse(table.getCompletedCommitsTimeline().empty());
}
// We no longer write empty cleaner plans when there is nothing to be cleaned.
assertTrue(table.getCompletedCleanTimeline().empty());
if (client.getConfig().shouldAutoCommit()) {
HoodieIndex index = SparkHoodieIndexFactory.createIndex(cfg);
List<HoodieRecord> taggedRecords = tagLocation(index, jsc.parallelize(records, 1), table).collect();
checkTaggedRecords(taggedRecords, newCommitTime);
}
return Pair.of(newCommitTime, statuses);
}
use of org.apache.hudi.common.table.timeline.HoodieTimeline in project hudi by apache.
the class TestUpsertPartitioner method testAverageBytesPerRecordForNonEmptyCommitTimeLine.
@Test
public void testAverageBytesPerRecordForNonEmptyCommitTimeLine() throws Exception {
HoodieTimeline commitTimeLine = mock(HoodieTimeline.class);
HoodieWriteConfig config = makeHoodieClientConfigBuilder().withCompactionConfig(HoodieCompactionConfig.newBuilder().compactionSmallFileSize(1000).build()).build();
when(commitTimeLine.empty()).thenReturn(false);
when(commitTimeLine.getReverseOrderedInstants()).thenReturn(setupHoodieInstants().stream());
LinkedList<Option<byte[]>> commits = generateCommitMetadataList();
when(commitTimeLine.getInstantDetails(any(HoodieInstant.class))).thenAnswer(invocationOnMock -> commits.pop());
long expectAvgSize = (long) Math.ceil((1.0 * 7500) / 1500);
long actualAvgSize = averageBytesPerRecord(commitTimeLine, config);
assertEquals(expectAvgSize, actualAvgSize);
}
use of org.apache.hudi.common.table.timeline.HoodieTimeline in project hudi by apache.
the class TestHoodieMergeHandle method testUpsertsForMultipleRecordsInSameFile.
@ParameterizedTest
@MethodSource("testArguments")
public void testUpsertsForMultipleRecordsInSameFile(ExternalSpillableMap.DiskMapType diskMapType, boolean isCompressionEnabled) throws Exception {
// Create records in a single partition
String partitionPath = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS[0];
dataGen = new HoodieTestDataGenerator(new String[] { partitionPath });
// Build a common config with diff configs
Properties properties = new Properties();
properties.setProperty(HoodieCommonConfig.SPILLABLE_DISK_MAP_TYPE.key(), diskMapType.name());
properties.setProperty(HoodieCommonConfig.DISK_MAP_BITCASK_COMPRESSION_ENABLED.key(), String.valueOf(isCompressionEnabled));
// Build a write config with bulkinsertparallelism set
HoodieWriteConfig cfg = getConfigBuilder().withProperties(properties).build();
try (SparkRDDWriteClient client = getHoodieWriteClient(cfg)) {
FileSystem fs = FSUtils.getFs(basePath, hadoopConf);
/**
* Write 1 (only inserts) This will do a bulk insert of 44 records of which there are 2 records repeated 21 times
* each. id1 (21 records), id2 (21 records), id3, id4
*/
String newCommitTime = "001";
client.startCommitWithTime(newCommitTime);
List<HoodieRecord> records = dataGen.generateInserts(newCommitTime, 4);
HoodieRecord record1 = records.get(0);
HoodieRecord record2 = records.get(1);
for (int i = 0; i < 20; i++) {
HoodieRecord dup = dataGen.generateUpdateRecord(record1.getKey(), newCommitTime);
records.add(dup);
}
for (int i = 0; i < 20; i++) {
HoodieRecord dup = dataGen.generateUpdateRecord(record2.getKey(), newCommitTime);
records.add(dup);
}
JavaRDD<HoodieRecord> writeRecords = jsc.parallelize(records, 1);
List<WriteStatus> statuses = client.bulkInsert(writeRecords, newCommitTime).collect();
assertNoWriteErrors(statuses);
// verify that there is a commit
metaClient = HoodieTableMetaClient.reload(metaClient);
HoodieTimeline timeline = new HoodieActiveTimeline(metaClient).getCommitTimeline();
assertEquals(1, timeline.findInstantsAfter("000", Integer.MAX_VALUE).countInstants(), "Expecting a single commit.");
assertEquals(newCommitTime, timeline.lastInstant().get().getTimestamp(), "Latest commit should be 001");
assertEquals(records.size(), HoodieClientTestUtils.readCommit(basePath, sqlContext, timeline, newCommitTime).count(), "Must contain 44 records");
/**
* Write 2 (insert) This will do a bulk insert of 1 record with the same row_key as record1 in the previous insert
* - id1. At this point, we will have 2 files with the row_keys as shown here - File 1 - id1 (21 records), id2 (21
* records), id3, id4 File 2 - id1
*/
newCommitTime = "002";
client.startCommitWithTime(newCommitTime);
// Do 1 more bulk insert with the same dup record1
List<HoodieRecord> newRecords = new ArrayList<>();
HoodieRecord sameAsRecord1 = dataGen.generateUpdateRecord(record1.getKey(), newCommitTime);
newRecords.add(sameAsRecord1);
writeRecords = jsc.parallelize(newRecords, 1);
statuses = client.bulkInsert(writeRecords, newCommitTime).collect();
assertNoWriteErrors(statuses);
// verify that there are 2 commits
metaClient = HoodieTableMetaClient.reload(metaClient);
timeline = new HoodieActiveTimeline(metaClient).getCommitTimeline();
assertEquals(2, timeline.findInstantsAfter("000", Integer.MAX_VALUE).countInstants(), "Expecting two commits.");
assertEquals(newCommitTime, timeline.lastInstant().get().getTimestamp(), "Latest commit should be 002");
Dataset<Row> dataSet = getRecords();
assertEquals(45, dataSet.count(), "Must contain 45 records");
/**
* Write 3 (insert) This will bulk insert 2 new completely new records. At this point, we will have 2 files with
* the row_keys as shown here - File 1 - id1 (21 records), id2 (21 records), id3, id4 File 2 - id1 File 3 - id5,
* id6
*/
newCommitTime = "003";
client.startCommitWithTime(newCommitTime);
newRecords = dataGen.generateInserts(newCommitTime, 2);
writeRecords = jsc.parallelize(newRecords, 1);
statuses = client.bulkInsert(writeRecords, newCommitTime).collect();
assertNoWriteErrors(statuses);
// verify that there are now 3 commits
metaClient = HoodieTableMetaClient.reload(metaClient);
timeline = new HoodieActiveTimeline(metaClient).getCommitTimeline();
assertEquals(3, timeline.findInstantsAfter("000", Integer.MAX_VALUE).countInstants(), "Expecting three commits.");
assertEquals(newCommitTime, timeline.lastInstant().get().getTimestamp(), "Latest commit should be 003");
dataSet = getRecords();
assertEquals(47, dataSet.count(), "Must contain 47 records");
/**
* Write 4 (updates) This will generate 2 upsert records with id1 and id2. The rider and driver names in the
* update records will be rider-004 and driver-004. After the upsert is complete, all the records with id1 in File
* 1 and File 2 must be updated, all the records with id2 in File 2 must also be updated. Also, none of the other
* records in File 1, File 2 and File 3 must be updated.
*/
newCommitTime = "004";
client.startCommitWithTime(newCommitTime);
List<HoodieRecord> updateRecords = new ArrayList<>();
// This exists in 001 and 002 and should be updated in both
sameAsRecord1 = dataGen.generateUpdateRecord(record1.getKey(), newCommitTime);
updateRecords.add(sameAsRecord1);
// This exists in 001 and should be updated
HoodieRecord sameAsRecord2 = dataGen.generateUpdateRecord(record2.getKey(), newCommitTime);
updateRecords.add(sameAsRecord2);
JavaRDD<HoodieRecord> updateRecordsRDD = jsc.parallelize(updateRecords, 1);
statuses = client.upsert(updateRecordsRDD, newCommitTime).collect();
// Verify there are no errors
assertNoWriteErrors(statuses);
// verify there are now 4 commits
timeline = new HoodieActiveTimeline(metaClient).getCommitTimeline();
assertEquals(4, timeline.findInstantsAfter("000", Integer.MAX_VALUE).countInstants(), "Expecting four commits.");
assertEquals(timeline.lastInstant().get().getTimestamp(), newCommitTime, "Latest commit should be 004");
// Check the entire dataset has 47 records still
dataSet = getRecords();
assertEquals(47, dataSet.count(), "Must contain 47 records");
Row[] rows = (Row[]) dataSet.collect();
int record1Count = 0;
int record2Count = 0;
for (Row row : rows) {
if (row.getAs("_hoodie_record_key").equals(record1.getKey().getRecordKey())) {
record1Count++;
// assert each duplicate record is updated
assertEquals(row.getAs("rider"), "rider-004");
assertEquals(row.getAs("driver"), "driver-004");
} else if (row.getAs("_hoodie_record_key").equals(record2.getKey().getRecordKey())) {
record2Count++;
// assert each duplicate record is updated
assertEquals(row.getAs("rider"), "rider-004");
assertEquals(row.getAs("driver"), "driver-004");
} else {
assertNotEquals(row.getAs("rider"), "rider-004");
assertNotEquals(row.getAs("driver"), "rider-004");
}
}
// Assert that id1 record count which has been updated to rider-004 and driver-004 is 22, which is the total
// number of records with row_key id1
assertEquals(22, record1Count);
// Assert that id2 record count which has been updated to rider-004 and driver-004 is 21, which is the total
// number of records with row_key id2
assertEquals(21, record2Count);
}
}
use of org.apache.hudi.common.table.timeline.HoodieTimeline in project hudi by apache.
the class TestHoodieTimelineArchiver method testArchiveCommitSavepointNoHole.
@ParameterizedTest
@ValueSource(booleans = { true, false })
public void testArchiveCommitSavepointNoHole(boolean enableMetadataTable) throws Exception {
init();
HoodieWriteConfig cfg = HoodieWriteConfig.newBuilder().withPath(basePath).withSchema(HoodieTestDataGenerator.TRIP_EXAMPLE_SCHEMA).withParallelism(2, 2).forTable("test-trip-table").withCompactionConfig(HoodieCompactionConfig.newBuilder().retainCommits(1).archiveCommitsWith(2, 5).build()).withFileSystemViewConfig(FileSystemViewStorageConfig.newBuilder().withRemoteServerPort(timelineServicePort).build()).withMetadataConfig(HoodieMetadataConfig.newBuilder().enable(enableMetadataTable).build()).build();
HoodieTestDataGenerator.createCommitFile(basePath, "100", wrapperFs.getConf());
HoodieTestDataGenerator.createCommitFile(basePath, "101", wrapperFs.getConf());
HoodieTestDataGenerator.createSavepointFile(basePath, "101", wrapperFs.getConf());
HoodieTestDataGenerator.createCommitFile(basePath, "102", wrapperFs.getConf());
HoodieTestDataGenerator.createCommitFile(basePath, "103", wrapperFs.getConf());
HoodieTestDataGenerator.createCommitFile(basePath, "104", wrapperFs.getConf());
HoodieTestDataGenerator.createCommitFile(basePath, "105", wrapperFs.getConf());
HoodieTable table = HoodieSparkTable.create(cfg, context);
HoodieTimelineArchiver archiver = new HoodieTimelineArchiver(cfg, table);
if (enableMetadataTable) {
// Simulate a compaction commit in metadata table timeline
// so the archival in data table can happen
createCompactionCommitInMetadataTable(hadoopConf, wrapperFs, basePath, "105");
}
HoodieTimeline timeline = metaClient.getActiveTimeline().getCommitsTimeline().filterCompletedInstants();
assertEquals(6, timeline.countInstants(), "Loaded 6 commits and the count should match");
assertTrue(archiver.archiveIfRequired(context));
timeline = metaClient.getActiveTimeline().reload().getCommitsTimeline().filterCompletedInstants();
assertEquals(5, timeline.countInstants(), "Since we have a savepoint at 101, we should never archive any commit after 101 (we only archive 100)");
assertTrue(timeline.containsInstant(new HoodieInstant(false, HoodieTimeline.COMMIT_ACTION, "101")), "Archived commits should always be safe");
assertTrue(timeline.containsInstant(new HoodieInstant(false, HoodieTimeline.COMMIT_ACTION, "102")), "Archived commits should always be safe");
assertTrue(timeline.containsInstant(new HoodieInstant(false, HoodieTimeline.COMMIT_ACTION, "103")), "Archived commits should always be safe");
}
use of org.apache.hudi.common.table.timeline.HoodieTimeline in project hudi by apache.
the class TestHoodieTimelineArchiver method testArchiveCommitsWithCompactionCommitInMetadataTableTimeline.
@Test
public void testArchiveCommitsWithCompactionCommitInMetadataTableTimeline() throws Exception {
HoodieWriteConfig writeConfig = initTestTableAndGetWriteConfig(true, 2, 4, 20);
int startInstantTime = 100;
int numCommits = 15;
// "100" till "105" should be archived in this case
int numExpectedArchived = 6;
for (int i = startInstantTime; i < startInstantTime + numCommits; i++) {
HoodieTestDataGenerator.createCommitFile(basePath, Integer.toString(i), wrapperFs.getConf());
}
// Simulate a compaction commit in metadata table timeline
// so the archival in data table can happen
createCompactionCommitInMetadataTable(hadoopConf, wrapperFs, basePath, "105");
HoodieTable table = HoodieSparkTable.create(writeConfig, context);
HoodieTimelineArchiver archiveLog = new HoodieTimelineArchiver(writeConfig, table);
HoodieTimeline timeline = metaClient.getActiveTimeline().getCommitsTimeline().filterCompletedInstants();
assertEquals(numCommits, timeline.countInstants(), String.format("Loaded %d commits and the count should match", numCommits));
assertTrue(archiveLog.archiveIfRequired(context));
timeline = metaClient.getActiveTimeline().reload().getCommitsTimeline().filterCompletedInstants();
assertEquals(numCommits - numExpectedArchived, timeline.countInstants(), "Since we have a compaction commit of 105 in metadata table timeline, we should never archive any commit after that");
for (int i = startInstantTime + numExpectedArchived; i < startInstantTime + numCommits; i++) {
assertTrue(timeline.containsInstant(new HoodieInstant(false, HoodieTimeline.COMMIT_ACTION, Integer.toString(i))), String.format("Commit %d should not be archived", i));
}
}
Aggregations