diff --git ql/src/java/org/apache/hadoop/hive/ql/exec/FileSinkOperator.java ql/src/java/org/apache/hadoop/hive/ql/exec/FileSinkOperator.java index 3e0943251e..d4f93c6c46 100644 --- ql/src/java/org/apache/hadoop/hive/ql/exec/FileSinkOperator.java +++ ql/src/java/org/apache/hadoop/hive/ql/exec/FileSinkOperator.java @@ -38,6 +38,7 @@ import org.apache.hadoop.hive.ql.io.RecordUpdater; import org.apache.hadoop.hive.ql.io.StatsProvidingRecordWriter; import org.apache.hadoop.hive.ql.io.StreamingOutputFormat; +import org.apache.hadoop.hive.ql.io.orc.OrcRecordUpdater; import org.apache.hadoop.hive.ql.metadata.HiveException; import org.apache.hadoop.hive.ql.metadata.HiveFatalException; import org.apache.hadoop.hive.ql.plan.DynamicPartitionCtx; @@ -768,8 +769,10 @@ public void process(Object row, int tag) throws HiveException { // Find the bucket id, and switch buckets if need to ObjectInspector rowInspector = bDynParts ? subSetOI : outputObjInspector; Object recId = ((StructObjectInspector)rowInspector).getStructFieldData(row, recIdField); - int bucketNum = + int bucketValue = bucketInspector.get(recIdInspector.getStructFieldData(recId, bucketField)); + int bucketNum = + OrcRecordUpdater.BucketFormat.determineVersion(bucketValue).decodeWriterId(bucketValue); if (fpaths.acidLastBucket != bucketNum) { fpaths.acidLastBucket = bucketNum; // Switch files diff --git ql/src/java/org/apache/hadoop/hive/ql/exec/ReduceSinkOperator.java ql/src/java/org/apache/hadoop/hive/ql/exec/ReduceSinkOperator.java index 92741ee62c..cb5ace92f9 100644 --- ql/src/java/org/apache/hadoop/hive/ql/exec/ReduceSinkOperator.java +++ ql/src/java/org/apache/hadoop/hive/ql/exec/ReduceSinkOperator.java @@ -127,14 +127,14 @@ * This two dimensional array holds key data and a corresponding Union object * which contains the tag identifying the aggregate expression for distinct columns. * - * If there is no distict expression, cachedKeys is simply like this. + * If there is no distinct expression, cachedKeys is simply like this. * cachedKeys[0] = [col0][col1] * - * with two distict expression, union(tag:key) is attatched for each distinct expression + * with two distinct expression, union(tag:key) is attached for each distinct expression * cachedKeys[0] = [col0][col1][0:dist1] * cachedKeys[1] = [col0][col1][1:dist2] * - * in this case, child GBY evaluates distict values with expression like KEY.col2:0.dist1 + * in this case, child GBY evaluates distinct values with expression like KEY.col2:0.dist1 * see {@link ExprNodeColumnEvaluator} */ // TODO: we only ever use one row of these at a time. Why do we need to cache multiple? diff --git ql/src/java/org/apache/hadoop/hive/ql/io/orc/OrcRawRecordMerger.java ql/src/java/org/apache/hadoop/hive/ql/io/orc/OrcRawRecordMerger.java index 95b8806e70..be0c6362ac 100644 --- ql/src/java/org/apache/hadoop/hive/ql/io/orc/OrcRawRecordMerger.java +++ ql/src/java/org/apache/hadoop/hive/ql/io/orc/OrcRawRecordMerger.java @@ -602,7 +602,9 @@ public boolean next(RecordIdentifier recordIdentifier, * event. If we did want to pass it along, we'd have to include statementId in the row * returned so that compaction could write it out or make minor minor compaction understand * how to write out delta files in delta_xxx_yyy_stid format. There doesn't seem to be any - * value in this.*/ + * value in this. + * TODO: if stmtID is used for multi-insert, then a row that differs only by stmtId is not the same row so this should pay attention to + * event type and collapse only if update (delete?)*/ boolean isSameRow = prevKey.isSameRow((ReaderKey)recordIdentifier); // if we are collapsing, figure out if this is a new row if (collapse || isSameRow) { diff --git ql/src/java/org/apache/hadoop/hive/ql/io/orc/OrcRecordUpdater.java ql/src/java/org/apache/hadoop/hive/ql/io/orc/OrcRecordUpdater.java index 65f4a24750..34fe9f6bb5 100644 --- ql/src/java/org/apache/hadoop/hive/ql/io/orc/OrcRecordUpdater.java +++ ql/src/java/org/apache/hadoop/hive/ql/io/orc/OrcRecordUpdater.java @@ -52,7 +52,11 @@ /** * A RecordUpdater where the files are stored as ORC. - */ + * A note on various record structures: the {@code row} coming in (as in {@link #insert(long, Object)} + * for example), is a struct like but what is written to the file + * * is > (see {@link #createEventSchema(ObjectInspector)}) + * So there are OIs here to make the translation. + * */ public class OrcRecordUpdater implements RecordUpdater { private static final Logger LOG = LoggerFactory.getLogger(OrcRecordUpdater.class); @@ -111,6 +115,7 @@ private LongObjectInspector rowIdInspector; // OI for the long row id inside the recordIdentifier private LongObjectInspector origTxnInspector; // OI for the original txn inside the record // identifer + private IntObjectInspector bucketInspector; static int getOperation(OrcStruct struct) { return ((IntWritable) struct.getFieldValue(OPERATION)).get(); @@ -185,6 +190,105 @@ static StructObjectInspector createEventSchema(ObjectInspector rowInspector) { return new OrcStruct.OrcStructInspector(fields); } + /** + * This class makes sense of {@link RecordIdentifier#bucketId}. Up until ASF Hive 3.0 this field + * was simply the bucket ID. Since 3.0 it does bit packing to store several things: + * top 3 bits - version describing the format (we can only have 8). + * The rest is version specific. + * + * TODO: 1) add decent toString() here and use it all over + * 2) did this break Mutate API? + * 3) fix {@link OrcRawRecordMerger} where it skips based on statementID + * 4) new tests (backwards compat) + * 5) run all tests + * 6) in {@link OrcRecordUpdater} fix setting/resetting the bucket id + */ + public enum BucketFormat { + /** + * This is the "legacy" version. The whole {@code bucket} value just has the bucket ID in it. + * The numeric code for this version is 0. (Assumes bucket ID takes less than 29 bits... which + * implies top 3 bits are 000 so data written before Hive 3.0 is readable with this scheme). + */ + V0(0) { + @Override + public int decodeWriterId(int value) { + return value; + } + @Override + public int decodeStatementId(int value) { + return 0; + } + @Override + public int encode(AcidOutputFormat.Options options) { + return options.getBucket(); + } + }, + /** + * Represents format of "bucket" value in Hive 3.0. + * top 3 bits - version code. + * next 14 bits - the bucket ID + * remaining 15 bits - the statement ID - 0-based numbering of all statements within a + * transaction. Each leg of a multi-insert statement gets a separate statement ID. + * + * Constructs like Merge and Multi-Insert may have multiple tasks writing data that belongs to + * the same physical bucket file. For example, a Merge stmt with update and insert clauses, + * (and split update enabled - should be the default in 3.0). A task on behalf of insert may + * be writing a row into bucket 0 and another task in the update branch may be writing an insert + * event into bucket 0. Each of these task are writing to different delta directory - distinguished + * by statement ID. By including both bucket ID and statement ID in {@link RecordIdentifier} + * we ensure that {@link RecordIdentifier} is unique. + * + * The intent is that sorting rows by {@link RecordIdentifier} groups rows in the same physical + * bucket next to each other. + * For any row created by a given version of Hive, top 3 bits are constant. The next + * most significant bits are the bucket ID, then the statement ID. This ensures that + * {@link org.apache.hadoop.hive.ql.optimizer.SortedDynPartitionOptimizer} works which is + * designed so that each task only needs to keep 1 writer opened at a time. It could be + * configured such that a single writer sees data for multiple buckets so it must "group" data + * by bucket ID (and then sort within each bucket as required) which is achieved via sorting + * by {@link RecordIdentifier} which includes the {@link RecordIdentifier#bucketId} which has + * the actual bucket ID in the high order bits. This scheme also ensures that + * {@link org.apache.hadoop.hive.ql.exec.FileSinkOperator#process(Object, int)} works in case + * there numBuckets > numReducers. (The later could be fixed by changing how writers are + * initialized in "if (fpaths.acidLastBucket != bucketNum) {") + */ + V1(1) { + @Override + public int decodeWriterId(int bucket) { + return (bucket & 0b0001_1111_1111_1111_1000_0000_0000_0000) >>> 15; + } + @Override + public int decodeStatementId(int bucket) { + return (bucket & 0b0000_0000_0000_0000_0111_1111_1111_1111); + } + @Override + public int encode(AcidOutputFormat.Options options) { + return this.version << 29 | options.getBucket() << 15 | + (options.getStatementId() >= 0 ? options.getStatementId() : 0); + } + }; + private static int TOP3BITS_MASK = 0b1110_0000_0000_0000_0000_0000_0000_0000; + public static BucketFormat determineVersion(int bucket) { + assert 7 << 29 == BucketFormat.TOP3BITS_MASK; + //look at top 3 bits and return appropriate enum + int version = (BucketFormat.TOP3BITS_MASK & bucket) >>> 29; + switch (version) { + case 0: + return BucketFormat.V0; + case 1: + return BucketFormat.V1; + default: + throw new IllegalArgumentException("Illegal 'bucket' format. Version=" + version + " from int " + bucket); + } + } + final int version; + BucketFormat(int version) { + this.version = version; + } + public abstract int decodeWriterId(int bucket); + public abstract int decodeStatementId(int bucket); + public abstract int encode(AcidOutputFormat.Options options); + } OrcRecordUpdater(Path path, AcidOutputFormat.Options options) throws IOException { this.options = options; @@ -200,7 +304,11 @@ static StructObjectInspector createEventSchema(ObjectInspector rowInspector) { this.acidOperationalProperties = AcidUtils.getAcidOperationalProperties(options.getConfiguration()); } - this.bucket.set(options.getBucket()); + + this.bucket.set(BucketFormat.V1.encode(options)); + if(options.getBucket() > 100) { + System.out.print("WTF"); + } this.path = AcidUtils.createFilename(path, options); this.deleteEventWriter = null; this.deleteEventPath = null; @@ -338,6 +446,7 @@ private ObjectInspector findRecId(ObjectInspector inspector, int rowIdColNum) { originalTxnField = fields.get(0); origTxnInspector = (LongObjectInspector)originalTxnField.getFieldObjectInspector(); bucketField = fields.get(1); + bucketInspector = (IntObjectInspector) bucketField.getFieldObjectInspector(); rowIdField = fields.get(2); rowIdInspector = (LongObjectInspector)rowIdField.getFieldObjectInspector(); @@ -351,6 +460,7 @@ private void addSimpleEvent(int operation, long currentTransaction, long rowId, throws IOException { this.operation.set(operation); this.currentTransaction.set(currentTransaction); + int currentBucket = bucket.get(); // If this is an insert, originalTransaction should be set to this transaction. If not, // it will be reset by the following if anyway. long originalTransaction = currentTransaction; @@ -359,6 +469,8 @@ private void addSimpleEvent(int operation, long currentTransaction, long rowId, originalTransaction = origTxnInspector.get( recIdInspector.getStructFieldData(rowIdValue, originalTxnField)); rowId = rowIdInspector.get(recIdInspector.getStructFieldData(rowIdValue, rowIdField)); + int bucketValue = bucketInspector.get(recIdInspector.getStructFieldData(rowIdValue, bucketField)); + bucket.set(bucketValue); } else if(operation == INSERT_OPERATION) { rowId += rowIdOffset; @@ -372,6 +484,7 @@ else if(operation == INSERT_OPERATION) { writer = OrcFile.createWriter(path, writerOptions); } writer.addRow(item); + bucket.set(currentBucket); } private void addSplitUpdateEvent(int operation, long currentTransaction, long rowId, Object row) @@ -388,8 +501,11 @@ private void addSplitUpdateEvent(int operation, long currentTransaction, long ro recIdInspector.getStructFieldData(rowValue, originalTxnField)); rowId = rowIdInspector.get( recIdInspector.getStructFieldData(rowValue, rowIdField)); + int currentBucket = bucket.get(); if (operation == DELETE_OPERATION || operation == UPDATE_OPERATION) { + int bucketValue = bucketInspector.get(recIdInspector.getStructFieldData(rowValue, bucketField)); + bucket.set(bucketValue); // Initialize a deleteEventWriter if not yet done. (Lazy initialization) if (deleteEventWriter == null) { // Initialize an indexBuilder for deleteEvents. @@ -414,6 +530,7 @@ private void addSplitUpdateEvent(int operation, long currentTransaction, long ro item.setFieldValue(OrcRecordUpdater.ROW, null); // ROW is null for delete events. deleteEventIndexBuilder.addKey(DELETE_OPERATION, originalTransaction, bucket.get(), rowId); deleteEventWriter.addRow(item); + bucket.set(currentBucket); } if (operation == UPDATE_OPERATION) { diff --git ql/src/java/org/apache/hadoop/hive/ql/io/orc/VectorizedOrcAcidRowBatchReader.java ql/src/java/org/apache/hadoop/hive/ql/io/orc/VectorizedOrcAcidRowBatchReader.java index 75c7680e26..c4b797996b 100644 --- ql/src/java/org/apache/hadoop/hive/ql/io/orc/VectorizedOrcAcidRowBatchReader.java +++ ql/src/java/org/apache/hadoop/hive/ql/io/orc/VectorizedOrcAcidRowBatchReader.java @@ -551,6 +551,9 @@ public boolean next(DeleteRecordKey deleteRecordKey) throws IOException { if (indexPtrInBatch >= batch.size) { // We have exhausted our current batch, read the next batch. if (recordReader.nextBatch(batch)) { + /* + * todo: this check only happens for 2nd batch + * */ // Whenever we are reading a batch, we must ensure that all the records in the batch // have the same bucket id as the bucket id of the split. If not, throw exception. // NOTE: this assertion might not hold, once virtual bucketing is in place. However, @@ -621,6 +624,17 @@ public int compareTo(CompressedOtid other) { } } + /** + * this is a bit problematic - in order to load ColumnizedDeleteEventRegistry we still open + * all delete deltas at once - possibly causing OOM same as for {@link SortMergedDeleteEventRegistry} + * which uses {@link OrcRawRecordMerger}. Why not load all delete_delta sequentially. Each + * dd is sorted by {@link RecordIdentifier} so we could create a BTree like structure where the + * 1st level is an array of originalTransactionId where each entry points at an array + * of bucketIds where each entry points at an array of rowIds. We could probably use ArrayList to + * manage insertion as the structure is built (LinkedList?). This should reduce memory foot print + * (as far as OrcReader to a single reader) + * + */ private TreeMap sortMerger; private long rowIds[]; private CompressedOtid compressedOtids[]; diff --git ql/src/java/org/apache/hadoop/hive/ql/parse/UpdateDeleteSemanticAnalyzer.java ql/src/java/org/apache/hadoop/hive/ql/parse/UpdateDeleteSemanticAnalyzer.java index 0541a4035a..78c511b016 100644 --- ql/src/java/org/apache/hadoop/hive/ql/parse/UpdateDeleteSemanticAnalyzer.java +++ ql/src/java/org/apache/hadoop/hive/ql/parse/UpdateDeleteSemanticAnalyzer.java @@ -696,15 +696,7 @@ WHEN NOT MATCHED THEN INSERT VALUES(source.a2, source.b2) if(numWhenMatchedUpdateClauses > 1) { throw new SemanticException(ErrorMsg.MERGE_TOO_MANY_UPDATE, ctx.getCmd()); } - assert numInsertClauses < 2; - if(numInsertClauses == 1 && numWhenMatchedUpdateClauses == 1) { - if(AcidUtils.getAcidOperationalProperties(targetTable).isSplitUpdate()) { - throw new IllegalStateException("Tables with " + - hive_metastoreConstants.TABLE_TRANSACTIONAL_PROPERTIES + "=" + - TransactionalValidationListener.DEFAULT_TRANSACTIONAL_PROPERTY + " currently do not " + - "support MERGE with both Insert and Update clauses."); - } - } + assert numInsertClauses < 2: "too many Insert clauses"; } if(numWhenMatchedDeleteClauses + numWhenMatchedUpdateClauses == 2 && extraPredicate == null) { throw new SemanticException(ErrorMsg.MERGE_PREDIACTE_REQUIRED, ctx.getCmd()); diff --git ql/src/java/org/apache/hadoop/hive/ql/udf/UDFToInteger.java ql/src/java/org/apache/hadoop/hive/ql/udf/UDFToInteger.java index 461ef86b83..c57ad04f70 100755 --- ql/src/java/org/apache/hadoop/hive/ql/udf/UDFToInteger.java +++ ql/src/java/org/apache/hadoop/hive/ql/udf/UDFToInteger.java @@ -25,6 +25,7 @@ import org.apache.hadoop.hive.ql.exec.vector.expressions.CastStringToLong; import org.apache.hadoop.hive.ql.exec.vector.expressions.CastTimestampToLong; import org.apache.hadoop.hive.ql.io.RecordIdentifier; +import org.apache.hadoop.hive.ql.io.orc.OrcRecordUpdater; import org.apache.hadoop.hive.serde2.io.ByteWritable; import org.apache.hadoop.hive.serde2.io.DoubleWritable; import org.apache.hadoop.hive.serde2.io.HiveDecimalWritable; @@ -221,6 +222,9 @@ public IntWritable evaluate(RecordIdentifier i) { return null; } else { intWritable.set(i.getBucketId()); + OrcRecordUpdater.BucketFormat decoder = + OrcRecordUpdater.BucketFormat.determineVersion(i.getBucketId()); + intWritable.set(decoder.decodeWriterId(i.getBucketId())); return intWritable; } } diff --git ql/src/test/org/apache/hadoop/hive/ql/TestTxnCommands.java ql/src/test/org/apache/hadoop/hive/ql/TestTxnCommands.java index 7c66955e14..530c7def88 100644 --- ql/src/test/org/apache/hadoop/hive/ql/TestTxnCommands.java +++ ql/src/test/org/apache/hadoop/hive/ql/TestTxnCommands.java @@ -492,53 +492,10 @@ private static void pause(int timeMillis) { * sorts rows in dictionary order */ private List stringifyValues(int[][] rowsIn) { - assert rowsIn.length > 0; - int[][] rows = rowsIn.clone(); - Arrays.sort(rows, new RowComp()); - List rs = new ArrayList(); - for(int[] row : rows) { - assert row.length > 0; - StringBuilder sb = new StringBuilder(); - for(int value : row) { - sb.append(value).append("\t"); - } - sb.setLength(sb.length() - 1); - rs.add(sb.toString()); - } - return rs; - } - private static final class RowComp implements Comparator { - @Override - public int compare(int[] row1, int[] row2) { - assert row1 != null && row2 != null && row1.length == row2.length; - for(int i = 0; i < row1.length; i++) { - int comp = Integer.compare(row1[i], row2[i]); - if(comp != 0) { - return comp; - } - } - return 0; - } + return TestTxnCommands2.stringifyValues(rowsIn); } private String makeValuesClause(int[][] rows) { - assert rows.length > 0; - StringBuilder sb = new StringBuilder("values"); - for(int[] row : rows) { - assert row.length > 0; - if(row.length > 1) { - sb.append("("); - } - for(int value : row) { - sb.append(value).append(","); - } - sb.setLength(sb.length() - 1);//remove trailing comma - if(row.length > 1) { - sb.append(")"); - } - sb.append(","); - } - sb.setLength(sb.length() - 1);//remove trailing comma - return sb.toString(); + return TestTxnCommands2.makeValuesClause(rows); } private List runStatementOnDriver(String stmt) throws Exception { @@ -558,7 +515,6 @@ private CommandProcessorResponse runStatementOnDriverNegative(String stmt) throw throw new RuntimeException("Didn't get expected failure!"); } -// @Ignore @Test public void exchangePartition() throws Exception { runStatementOnDriver("create database ex1"); @@ -756,9 +712,9 @@ public void testMergeUpdateDelete() throws Exception { runStatementOnDriver("insert into " + Table.ACIDTBL + " " + makeValuesClause(vals)); String query = "merge into " + Table.ACIDTBL + " as t using " + Table.NONACIDORCTBL + " s ON t.a = s.a " + - "WHEN MATCHED AND s.a < 3 THEN update set b = 0 " + - "WHEN MATCHED and t.a > 3 and t.a < 5 THEN DELETE " + - "WHEN NOT MATCHED THEN INSERT VALUES(s.a, s.b) "; + "WHEN MATCHED AND s.a < 3 THEN update set b = 0 " + //updates (2,1) -> (2,0) + "WHEN MATCHED and t.a > 3 and t.a < 5 THEN DELETE " +//deletes (4,3) + "WHEN NOT MATCHED THEN INSERT VALUES(s.a, s.b) ";//inserts (11,11) runStatementOnDriver(query); List r = runStatementOnDriver("select a,b from " + Table.ACIDTBL + " order by a,b"); diff --git ql/src/test/org/apache/hadoop/hive/ql/TestTxnCommands2.java ql/src/test/org/apache/hadoop/hive/ql/TestTxnCommands2.java index 5786c4f659..254a998e8e 100644 --- ql/src/test/org/apache/hadoop/hive/ql/TestTxnCommands2.java +++ ql/src/test/org/apache/hadoop/hive/ql/TestTxnCommands2.java @@ -87,19 +87,27 @@ protected Driver d; protected static enum Table { ACIDTBL("acidTbl"), - ACIDTBLPART("acidTblPart"), + ACIDTBLPART("acidTblPart", "p"), NONACIDORCTBL("nonAcidOrcTbl"), - NONACIDPART("nonAcidPart"), - NONACIDPART2("nonAcidPart2"), - ACIDNESTEDPART("acidNestedPart"); + NONACIDPART("nonAcidPart", "p"), + NONACIDPART2("nonAcidPart2", "p2"), + ACIDNESTEDPART("acidNestedPart", "p,q"); private final String name; + private final String partitionColumns; @Override public String toString() { return name; } + String getPartitionColumns() { + return partitionColumns; + } Table(String name) { + this(name, null); + } + Table(String name, String partitionColumns) { this.name = name; + this.partitionColumns = partitionColumns; } } @@ -599,7 +607,7 @@ public void testNonAcidToAcidConversion3() throws Exception { List rs = runStatementOnDriver("select a,b from " + Table.NONACIDORCTBL); int [][] resultData = new int[][] {{1, 2}}; Assert.assertEquals(stringifyValues(resultData), rs); - rs = runStatementOnDriver("select count(*) from " + Table.NONACIDORCTBL); + rs = runStatementOnDriver("select count(*) from " + Table.NONACIDORCTBL);//todo: what is the point of this if we just did select *? int resultCount = 1; Assert.assertEquals(resultCount, Integer.parseInt(rs.get(0))); @@ -615,7 +623,7 @@ public void testNonAcidToAcidConversion3() throws Exception { rs = runStatementOnDriver("select a,b from " + Table.NONACIDORCTBL); resultData = new int[][] {{1, 2}}; Assert.assertEquals(stringifyValues(resultData), rs); - rs = runStatementOnDriver("select count(*) from " + Table.NONACIDORCTBL); + rs = runStatementOnDriver("select count(*) from " + Table.NONACIDORCTBL);//todo: what is the point of this if we just did select *? resultCount = 1; Assert.assertEquals(resultCount, Integer.parseInt(rs.get(0))); @@ -686,7 +694,7 @@ public void testNonAcidToAcidConversion3() throws Exception { rs = runStatementOnDriver("select a,b from " + Table.NONACIDORCTBL); resultData = new int[][] {{1, 3}, {3, 4}}; Assert.assertEquals(stringifyValues(resultData), rs); - rs = runStatementOnDriver("select count(*) from " + Table.NONACIDORCTBL); + rs = runStatementOnDriver("select count(*) from " + Table.NONACIDORCTBL);//todo: what is the point of this if we just did select *? resultCount = 2; Assert.assertEquals(resultCount, Integer.parseInt(rs.get(0))); @@ -722,7 +730,7 @@ public void testNonAcidToAcidConversion3() throws Exception { rs = runStatementOnDriver("select a,b from " + Table.NONACIDORCTBL); resultData = new int[][] {{1, 3}, {3, 4}}; Assert.assertEquals(stringifyValues(resultData), rs); - rs = runStatementOnDriver("select count(*) from " + Table.NONACIDORCTBL); + rs = runStatementOnDriver("select count(*) from " + Table.NONACIDORCTBL);//todo: what is the point of this if we just did select *? resultCount = 2; Assert.assertEquals(resultCount, Integer.parseInt(rs.get(0))); @@ -1438,12 +1446,13 @@ public void testMergeWithPredicate() throws Exception { String query = "merge into " + Table.ACIDTBL + " t using " + Table.NONACIDPART2 + " s ON t.a = s.a2 " + "WHEN MATCHED AND t.b between 1 and 3 THEN UPDATE set b = s.b2 " + - "WHEN NOT MATCHED and s.b2 >= 11 THEN INSERT VALUES(s.a2, s.b2)"; + "WHEN NOT MATCHED and s.b2 >= 8 THEN INSERT VALUES(s.a2, s.b2)"; runStatementOnDriver(query); r = runStatementOnDriver("select a,b from " + Table.ACIDTBL + " order by a,b"); - int[][] rExpected = {{2,2},{4,3},{5,6},{7,8},{11,11}}; + int[][] rExpected = {{2,2},{4,3},{5,6},{7,8},{8,8},{11,11}}; Assert.assertEquals(stringifyValues(rExpected), r); + assertUniqueID(Table.ACIDTBL); } /** @@ -1471,6 +1480,7 @@ public void testMerge2() throws Exception { r = runStatementOnDriver("select a,b from " + Table.ACIDTBL + " order by a,b"); int[][] rExpected = {{2,2},{4,44},{5,5},{7,8},{11,11}}; Assert.assertEquals(stringifyValues(rExpected), r); + assertUniqueID(Table.ACIDTBL); } /** @@ -1497,27 +1507,21 @@ public void testMerge3() throws Exception { int[][] rExpected = {{7,8},{11,11}}; Assert.assertEquals(stringifyValues(rExpected), r); } - /** - * https://hortonworks.jira.com/browse/BUG-66580 - * @throws Exception - */ - @Ignore @Test public void testMultiInsert() throws Exception { - runStatementOnDriver("create table if not exists srcpart (a int, b int, c int) " + - "partitioned by (z int) clustered by (a) into 2 buckets " + - "stored as orc tblproperties('transactional'='true')"); runStatementOnDriver("create temporary table if not exists data1 (x int)"); -// runStatementOnDriver("create temporary table if not exists data2 (x int)"); - runStatementOnDriver("insert into data1 values (1),(2),(3)"); -// runStatementOnDriver("insert into data2 values (4),(5),(6)"); d.destroy(); hiveConf.setVar(HiveConf.ConfVars.DYNAMICPARTITIONINGMODE, "nonstrict"); d = new Driver(hiveConf); - List r = runStatementOnDriver(" from data1 " + - "insert into srcpart partition(z) select 0,0,1,x " + - "insert into srcpart partition(z=1) select 0,0,1"); + + runStatementOnDriver(" from data1 " + + "insert into " + Table.ACIDTBLPART + " partition(p) select 0,x, 'p' || x " + + + "insert into " + Table.ACIDTBLPART + " partition(p='p1') select 0,x"); + List r = runStatementOnDriver("select p,a,b from " + Table.ACIDTBLPART + " order by p, a, b"); + Assert.assertEquals("[p1\t0\t1, p1\t0\t1, p1\t0\t2, p1\t0\t3, p2\t0\t2, p3\t0\t3]", r.toString()); + assertUniqueID(Table.ACIDTBLPART); } /** * Investigating DP and WriteEntity, etc @@ -1583,6 +1587,7 @@ public void testDynamicPartitionsMerge() throws Exception { r1 = runStatementOnDriver("select p,a,b from " + Table.ACIDTBLPART + " order by p, a, b"); String result= r1.toString(); Assert.assertEquals("[new part\t5\t5, new part\t11\t11, p1\t1\t1, p1\t2\t15, p1\t3\t3, p2\t4\t44]", result); + assertUniqueID(Table.ACIDTBLPART); } /** * Using nested partitions and thus DummyPartition @@ -1605,6 +1610,7 @@ public void testDynamicPartitionsMerge2() throws Exception { "when not matched then insert values(s.a, s.b, 3,4)"); r1 = runStatementOnDriver("select p,q,a,b from " + Table.ACIDNESTEDPART + " order by p,q, a, b"); Assert.assertEquals(stringifyValues(new int[][] {{1,1,1,1},{1,1,3,3},{1,2,2,15},{1,2,4,44},{3,4,5,5},{3,4,11,11}}), r1); + assertUniqueID(Table.ACIDNESTEDPART); } @Ignore("Covered elsewhere") @Test @@ -1661,7 +1667,7 @@ public void testValuesSource() throws Exception { } return rs; } - private static final class RowComp implements Comparator { + static class RowComp implements Comparator { @Override public int compare(int[] row1, int[] row2) { assert row1 != null && row2 != null && row1.length == row2.length; @@ -1674,7 +1680,7 @@ public int compare(int[] row1, int[] row2) { return 0; } } - String makeValuesClause(int[][] rows) { + static String makeValuesClause(int[][] rows) { assert rows.length > 0; StringBuilder sb = new StringBuilder("values"); for(int[] row : rows) { @@ -1705,4 +1711,19 @@ String makeValuesClause(int[][] rows) { d.getResults(rs); return rs; } + final void assertUniqueID(Table table) throws Exception { + String partCols = table.getPartitionColumns(); + //check to make sure there are no duplicate ROW__IDs - HIVE-16832 + StringBuilder sb = new StringBuilder("select "); + if(partCols != null && partCols.length() > 0) { + sb.append(partCols).append(","); + } + sb.append(" ROW__ID, count(*) from ").append(table).append(" group by "); + if(partCols != null && partCols.length() > 0) { + sb.append(partCols).append(","); + } + sb.append("ROW__ID having count(*) > 1"); + List r = runStatementOnDriver(sb.toString()); + Assert.assertTrue(r.toString(),r.size() == 0); + } } diff --git ql/src/test/org/apache/hadoop/hive/ql/TestTxnCommands2WithSplitUpdate.java ql/src/test/org/apache/hadoop/hive/ql/TestTxnCommands2WithSplitUpdate.java index ea5ecbc842..520e958af3 100644 --- ql/src/test/org/apache/hadoop/hive/ql/TestTxnCommands2WithSplitUpdate.java +++ ql/src/test/org/apache/hadoop/hive/ql/TestTxnCommands2WithSplitUpdate.java @@ -542,45 +542,4 @@ public void testNonAcidToAcidConversion3() throws Exception { resultCount = 2; Assert.assertEquals(resultCount, Integer.parseInt(rs.get(0))); } - @Ignore("HIVE-14947") - @Test - @Override - public void testDynamicPartitionsMerge() throws Exception {} - @Ignore("HIVE-14947") - @Test - @Override - public void testDynamicPartitionsMerge2() throws Exception {} - @Ignore("HIVE-14947") - @Test - @Override - public void testMerge() throws Exception {} - - /** - * todo: remove this test once HIVE-14947 is done (parent class has a better version) - */ - @Test - @Override - public void testMerge2() throws Exception { - int[][] baseValsOdd = {{5,5},{11,11}}; - int[][] baseValsEven = {{2,2},{4,44}}; - runStatementOnDriver("insert into " + Table.NONACIDPART2 + " PARTITION(p2='odd') " + makeValuesClause(baseValsOdd)); - runStatementOnDriver("insert into " + Table.NONACIDPART2 + " PARTITION(p2='even') " + makeValuesClause(baseValsEven)); - int[][] vals = {{2,1},{4,3},{5,6},{7,8}}; - runStatementOnDriver("insert into " + Table.ACIDTBL + " " + makeValuesClause(vals)); - List r = runStatementOnDriver("select a,b from " + Table.ACIDTBL + " order by a,b"); - Assert.assertEquals(stringifyValues(vals), r); - String query = "merge into " + Table.ACIDTBL + - " using " + Table.NONACIDPART2 + " source ON " + Table.ACIDTBL + ".a = source.a2 " + - "WHEN MATCHED THEN UPDATE set b = source.b2 "; - r = runStatementOnDriver(query); - - r = runStatementOnDriver("select a,b from " + Table.ACIDTBL + " order by a,b"); - int[][] rExpected = {{2,2},{4,44},{5,5},{7,8}}; - Assert.assertEquals(stringifyValues(rExpected), r); - - } - @Ignore("HIVE-14947") - @Test - @Override - public void testMergeWithPredicate() throws Exception {} } diff --git ql/src/test/org/apache/hadoop/hive/ql/TestTxnCommands2WithSplitUpdateAndVectorization.java ql/src/test/org/apache/hadoop/hive/ql/TestTxnCommands2WithSplitUpdateAndVectorization.java index 44a94127ec..34df95d29c 100644 --- ql/src/test/org/apache/hadoop/hive/ql/TestTxnCommands2WithSplitUpdateAndVectorization.java +++ ql/src/test/org/apache/hadoop/hive/ql/TestTxnCommands2WithSplitUpdateAndVectorization.java @@ -19,9 +19,12 @@ package org.apache.hadoop.hive.ql; import org.apache.hadoop.hive.conf.HiveConf; +import org.junit.Assert; import org.junit.Before; import org.junit.Test; +import java.util.List; + /** * Same as TestTxnCommands2WithSplitUpdate but tests ACID tables with vectorization turned on by * default, and having 'transactional_properties' set to 'default'. This specifically tests the @@ -48,5 +51,28 @@ public void testFailureOnAlteringTransactionalProperties() throws Exception { // To not override this test, that temporary table needs to be renamed. However, as // mentioned this does not serve any purpose, as this test does not relate to vectorization. } + @Test + public void testMultiInsertVectorized() throws Exception { + runStatementOnDriver("create temporary table if not exists data1 (x int)"); + runStatementOnDriver("insert into data1 values (1)"); + d.destroy(); + hiveConf.setVar(HiveConf.ConfVars.DYNAMICPARTITIONINGMODE, "nonstrict"); + d = new Driver(hiveConf); + + runStatementOnDriver(" from data1 " + + "insert into " + Table.ACIDTBLPART + " partition(p) select 0, 0, 'p' || x " + + + "insert into " + Table.ACIDTBLPART + " partition(p='p1') select 0, 1"); + /*w/o the fix we'd have 2 rows with the same ID in the same bucket of the same partition + * acidtblpart/p=p1/delta_0000015_0000015_0000/bucket_00000 [length: 578] {"operation":0,"originalTransaction":15,"bucket":0,"rowId":0,"currentTransaction":15,"row":{"_col0":0,"_col1":1}} + acidtblpart/p=p1/delta_0000015_0000015_0001/bucket_00000 [length: 568] {"operation":0,"originalTransaction":15,"bucket":0,"rowId":0,"currentTransaction":15,"row":{"_col0":0,"_col1":0}}*/ + List r = runStatementOnDriver("select p,a,b from " + Table.ACIDTBLPART + " order by p, a, b"); + Assert.assertEquals("[p1\t0\t0, p1\t0\t1]", r.toString()); + assertUniqueID(Table.ACIDTBLPART); + runStatementOnDriver("delete from " + Table.ACIDTBLPART); + r = runStatementOnDriver("select p,a,b from " + Table.ACIDTBLPART + " order by p, a, b"); + Assert.assertEquals("[]", r.toString()); + + } }