use of org.apache.hudi.avro.model.HoodieRequestedReplaceMetadata in project hudi by apache.
the class TestHoodieClientOnCopyOnWriteStorage method testPendingClusteringRollback.
@Test
public void testPendingClusteringRollback() throws Exception {
boolean populateMetaFields = true;
// setup clustering config.
HoodieClusteringConfig clusteringConfig = HoodieClusteringConfig.newBuilder().withClusteringMaxNumGroups(10).withClusteringTargetPartitions(0).withInlineClusteringNumCommits(1).withInlineClustering(true).build();
// start clustering, but don't commit
List<HoodieRecord> allRecords = testInsertAndClustering(clusteringConfig, populateMetaFields, false);
HoodieTableMetaClient metaClient = HoodieTableMetaClient.builder().setConf(hadoopConf).setBasePath(basePath).build();
List<Pair<HoodieInstant, HoodieClusteringPlan>> pendingClusteringPlans = ClusteringUtils.getAllPendingClusteringPlans(metaClient).collect(Collectors.toList());
assertEquals(1, pendingClusteringPlans.size());
HoodieInstant pendingClusteringInstant = pendingClusteringPlans.get(0).getLeft();
// complete another commit after pending clustering
HoodieWriteConfig.Builder cfgBuilder = getConfigBuilder(EAGER);
addConfigsForPopulateMetaFields(cfgBuilder, populateMetaFields);
HoodieWriteConfig config = cfgBuilder.build();
SparkRDDWriteClient client = getHoodieWriteClient(config);
dataGen = new HoodieTestDataGenerator();
String commitTime = HoodieActiveTimeline.createNewInstantTime();
allRecords.addAll(dataGen.generateInserts(commitTime, 200));
assertThrows(HoodieUpsertException.class, () -> writeAndVerifyBatch(client, allRecords, commitTime, populateMetaFields));
// verify pending clustering can be rolled back (even though there is a completed commit greater than pending clustering)
client.rollback(pendingClusteringInstant.getTimestamp());
metaClient.reloadActiveTimeline();
// verify there are no pending clustering instants
assertEquals(0, ClusteringUtils.getAllPendingClusteringPlans(metaClient).count());
// delete rollback.completed instant to mimic failed rollback of clustering. and then trigger rollback of clustering again. same rollback instant should be used.
HoodieInstant rollbackInstant = metaClient.getActiveTimeline().getRollbackTimeline().lastInstant().get();
FileCreateUtils.deleteRollbackCommit(metaClient.getBasePath(), rollbackInstant.getTimestamp());
metaClient.reloadActiveTimeline();
// create replace commit requested meta file so that rollback will not throw FileNotFoundException
// create file slice with instantTime 001 and build clustering plan including this created 001 file slice.
HoodieClusteringPlan clusteringPlan = ClusteringTestUtils.createClusteringPlan(metaClient, pendingClusteringInstant.getTimestamp(), "1");
// create requested replace commit
HoodieRequestedReplaceMetadata requestedReplaceMetadata = HoodieRequestedReplaceMetadata.newBuilder().setClusteringPlan(clusteringPlan).setOperationType(WriteOperationType.CLUSTER.name()).build();
FileCreateUtils.createRequestedReplaceCommit(metaClient.getBasePath(), pendingClusteringInstant.getTimestamp(), Option.of(requestedReplaceMetadata));
// trigger clustering again. no new rollback instants should be generated.
try {
client.cluster(pendingClusteringInstant.getTimestamp(), false);
// new replace commit metadata generated is fake one. so, clustering will fail. but the intention of test is ot check for duplicate rollback instants.
} catch (Exception e) {
// ignore.
}
metaClient.reloadActiveTimeline();
// verify that there is no new rollback instant generated
HoodieInstant newRollbackInstant = metaClient.getActiveTimeline().getRollbackTimeline().lastInstant().get();
assertEquals(rollbackInstant.getTimestamp(), newRollbackInstant.getTimestamp());
}
use of org.apache.hudi.avro.model.HoodieRequestedReplaceMetadata in project hudi by apache.
the class TestCleaner method testCleanWithReplaceCommits.
@Test
public void testCleanWithReplaceCommits() throws Exception {
HoodieWriteConfig config = HoodieWriteConfig.newBuilder().withPath(basePath).withMetadataConfig(HoodieMetadataConfig.newBuilder().withMaxNumDeltaCommitsBeforeCompaction(1).withAssumeDatePartitioning(true).build()).withCompactionConfig(HoodieCompactionConfig.newBuilder().withCleanerPolicy(HoodieCleaningPolicy.KEEP_LATEST_COMMITS).retainCommits(2).build()).build();
HoodieTableMetadataWriter metadataWriter = SparkHoodieBackedTableMetadataWriter.create(hadoopConf, config, context);
HoodieTestTable testTable = HoodieMetadataTestTable.of(metaClient, metadataWriter);
String p0 = "2020/01/01";
String p1 = "2020/01/02";
// make 1 commit, with 1 file per partition
String file1P0C0 = UUID.randomUUID().toString();
String file1P1C0 = UUID.randomUUID().toString();
testTable.addInflightCommit("00000000000001").withBaseFilesInPartition(p0, file1P0C0).withBaseFilesInPartition(p1, file1P1C0);
HoodieCommitMetadata commitMetadata = generateCommitMetadata("00000000000001", Collections.unmodifiableMap(new HashMap<String, List<String>>() {
{
put(p0, CollectionUtils.createImmutableList(file1P0C0));
put(p1, CollectionUtils.createImmutableList(file1P1C0));
}
}));
metadataWriter.update(commitMetadata, "00000000000001", false);
metaClient.getActiveTimeline().saveAsComplete(new HoodieInstant(State.INFLIGHT, HoodieTimeline.COMMIT_ACTION, "00000000000001"), Option.of(commitMetadata.toJsonString().getBytes(StandardCharsets.UTF_8)));
metaClient = HoodieTableMetaClient.reload(metaClient);
List<HoodieCleanStat> hoodieCleanStatsOne = runCleaner(config);
assertEquals(0, hoodieCleanStatsOne.size(), "Must not scan any partitions and clean any files");
assertTrue(testTable.baseFileExists(p0, "00000000000001", file1P0C0));
assertTrue(testTable.baseFileExists(p1, "00000000000001", file1P1C0));
// make next replacecommit, with 1 clustering operation. logically delete p0. No change to p1
// notice that clustering generates empty inflight commit files
Map<String, String> partitionAndFileId002 = testTable.forReplaceCommit("00000000000002").getFileIdsWithBaseFilesInPartitions(p0);
String file2P0C1 = partitionAndFileId002.get(p0);
Pair<HoodieRequestedReplaceMetadata, HoodieReplaceCommitMetadata> replaceMetadata = generateReplaceCommitMetadata("00000000000002", p0, file1P0C0, file2P0C1);
testTable.addReplaceCommit("00000000000002", Option.of(replaceMetadata.getKey()), Option.empty(), replaceMetadata.getValue());
// run cleaner
List<HoodieCleanStat> hoodieCleanStatsTwo = runCleaner(config);
assertEquals(0, hoodieCleanStatsTwo.size(), "Must not scan any partitions and clean any files");
assertTrue(testTable.baseFileExists(p0, "00000000000002", file2P0C1));
assertTrue(testTable.baseFileExists(p0, "00000000000001", file1P0C0));
assertTrue(testTable.baseFileExists(p1, "00000000000001", file1P1C0));
// make next replacecommit, with 1 clustering operation. Replace data in p1. No change to p0
// notice that clustering generates empty inflight commit files
Map<String, String> partitionAndFileId003 = testTable.forReplaceCommit("00000000000003").getFileIdsWithBaseFilesInPartitions(p1);
String file3P1C2 = partitionAndFileId003.get(p1);
replaceMetadata = generateReplaceCommitMetadata("00000000000003", p1, file1P1C0, file3P1C2);
testTable.addReplaceCommit("00000000000003", Option.of(replaceMetadata.getKey()), Option.empty(), replaceMetadata.getValue());
// run cleaner
List<HoodieCleanStat> hoodieCleanStatsThree = runCleaner(config);
assertEquals(0, hoodieCleanStatsThree.size(), "Must not scan any partitions and clean any files");
assertTrue(testTable.baseFileExists(p0, "00000000000002", file2P0C1));
assertTrue(testTable.baseFileExists(p0, "00000000000001", file1P0C0));
assertTrue(testTable.baseFileExists(p1, "00000000000003", file3P1C2));
assertTrue(testTable.baseFileExists(p1, "00000000000001", file1P1C0));
// make next replacecommit, with 1 clustering operation. Replace data in p0 again
// notice that clustering generates empty inflight commit files
Map<String, String> partitionAndFileId004 = testTable.forReplaceCommit("00000000000004").getFileIdsWithBaseFilesInPartitions(p0);
String file4P0C3 = partitionAndFileId004.get(p0);
replaceMetadata = generateReplaceCommitMetadata("00000000000004", p0, file2P0C1, file4P0C3);
testTable.addReplaceCommit("00000000000004", Option.of(replaceMetadata.getKey()), Option.empty(), replaceMetadata.getValue());
// run cleaner
List<HoodieCleanStat> hoodieCleanStatsFour = runCleaner(config, 5);
assertTrue(testTable.baseFileExists(p0, "00000000000004", file4P0C3));
assertTrue(testTable.baseFileExists(p0, "00000000000002", file2P0C1));
assertTrue(testTable.baseFileExists(p1, "00000000000003", file3P1C2));
assertFalse(testTable.baseFileExists(p0, "00000000000001", file1P0C0));
// file1P1C0 still stays because its not replaced until 3 and its the only version available
assertTrue(testTable.baseFileExists(p1, "00000000000001", file1P1C0));
// make next replacecommit, with 1 clustering operation. Replace all data in p1. no new files created
// notice that clustering generates empty inflight commit files
Map<String, String> partitionAndFileId005 = testTable.forReplaceCommit("00000000000006").getFileIdsWithBaseFilesInPartitions(p1);
String file4P1C4 = partitionAndFileId005.get(p1);
replaceMetadata = generateReplaceCommitMetadata("00000000000006", p0, file3P1C2, file4P1C4);
testTable.addReplaceCommit("00000000000006", Option.of(replaceMetadata.getKey()), Option.empty(), replaceMetadata.getValue());
List<HoodieCleanStat> hoodieCleanStatsFive = runCleaner(config, 7);
assertTrue(testTable.baseFileExists(p0, "00000000000004", file4P0C3));
assertTrue(testTable.baseFileExists(p0, "00000000000002", file2P0C1));
assertTrue(testTable.baseFileExists(p1, "00000000000003", file3P1C2));
assertFalse(testTable.baseFileExists(p0, "00000000000001", file1P0C0));
assertFalse(testTable.baseFileExists(p1, "00000000000001", file1P1C0));
}
Aggregations