/** * To handle multiple INSERT... statements in a single transaction, we want to make sure * to generate unique {@code rowId} for all inserted rows of the transaction. * @return largest rowId created by previous statements (maybe 0) * @ */ private long findRowIdOffsetForInsert() { /* * 1. need to know bucket we are writing to * 2. need to know which delta dir it's in * Then, * 1. find the same bucket file in previous delta dir for this txn * 2. read the footer and get AcidStats which has insert count * 2.1 if AcidStats.inserts>0 done * else go to previous delta file * For example, consider insert/update/insert case...*/ if (options.getStatementId() <= 0) { return(0);//there is only 1 statement in this transaction (so far) } for (int pastStmt = options.getStatementId() - 1; pastStmt >= 0; pastStmt--) { Path matchingBucket = AcidUtils.createFilename(options.getFinalDestination(), options.clone().statementId(pastStmt)); if (!fs.exists(matchingBucket)) { continue; } Reader reader = OrcFile.createReader(matchingBucket, OrcFile.readerOptions(options.getConfiguration())); //no close() on Reader?! AcidStats acidStats = parseAcidStats(reader); if (acidStats.inserts > 0) { return(acidStats.inserts); } } //if we got here, we looked at all delta files in this txn, prior to current statement and didn't //find any inserts... return(0); }
/** * Create a reader that merge sorts the ACID events together. * @param conf the configuration * @param collapseEvents should the events on the same row be collapsed * @param isOriginal is the base file a pre-acid file * @param bucket the bucket we are reading * @param options the options to read with * @param deltaDirectory the list of delta directories to include * @ */ OrcRawRecordMerger(Configuration conf, bool collapseEvents, Reader reader, bool isOriginal, int bucket, ValidTxnList validTxnList, Reader.Options options, Path[] deltaDirectory) { this.conf = conf; this.collapse = collapseEvents; this.offset = options.getOffset(); this.length = options.getLength(); this.validTxnList = validTxnList; TypeDescription typeDescr = OrcUtils.getDesiredRowTypeDescr(conf); if (typeDescr == null) { throw new IOException(ErrorMsg.SCHEMA_REQUIRED_TO_READ_ACID_TABLES.getErrorCodedMsg()); } objectInspector = OrcRecordUpdater.createEventSchema (OrcStruct.createObjectInspector(0, OrcUtils.getOrcTypes(typeDescr))); // modify the options to reflect the event instead of the base row Reader.Options eventOptions = createEventOptions(options); if (reader == null) { baseReader = null; } else { // find the min/max based on the offset and length if (isOriginal) { discoverOriginalKeyBounds(reader, bucket, options); } else { discoverKeyBounds(reader, options); } LOG.info("min key = " + minKey + ", max key = " + maxKey); // use the min/max instead of the byte range ReaderPair pair; ReaderKey key = new ReaderKey(); if (isOriginal) { options = options.clone(); options.range(options.getOffset(), Long.MAX_VALUE); pair = new OriginalReaderPair(key, reader, bucket, minKey, maxKey, options); } else { pair = new ReaderPair(key, reader, bucket, minKey, maxKey, eventOptions, 0); } // if there is at least one record, put it in the map if (pair.nextRecord != null) { readers.put(key, pair); } baseReader = pair.recordReader; } // we always want to read all of the deltas eventOptions.range(0, Long.MAX_VALUE); if (deltaDirectory != null) { foreach (Path delta in deltaDirectory) { ReaderKey key = new ReaderKey(); Path deltaFile = AcidUtils.createBucketFile(delta, bucket); AcidUtils.ParsedDelta deltaDir = AcidUtils.parsedDelta(delta); FileSystem fs = deltaFile.getFileSystem(conf); long length = getLastFlushLength(fs, deltaFile); if (length != -1 && fs.exists(deltaFile)) { Reader deltaReader = OrcFile.createReader(deltaFile, OrcFile.readerOptions(conf).maxLength(length)); Reader.Options deltaEventOptions = null; if (eventOptions.getSearchArgument() != null) { // Turn off the sarg before pushing it to delta. We never want to push a sarg to a delta as // it can produce wrong results (if the latest valid version of the record is filtered out by // the sarg) or ArrayOutOfBounds errors (when the sarg is applied to a delete record) // unless the delta only has insert events OrcRecordUpdater.AcidStats acidStats = OrcRecordUpdater.parseAcidStats(deltaReader); if (acidStats.deletes > 0 || acidStats.updates > 0) { deltaEventOptions = eventOptions.clone().searchArgument(null, null); } } ReaderPair deltaPair; deltaPair = new ReaderPair(key, deltaReader, bucket, minKey, maxKey, deltaEventOptions != null ? deltaEventOptions : eventOptions, deltaDir.getStatementId()); if (deltaPair.nextRecord != null) { readers.put(key, deltaPair); } } } } // get the first record Map.Entry <ReaderKey, ReaderPair> entry = readers.pollFirstEntry(); if (entry == null) { columns = 0; primary = null; } else { primary = entry.getValue(); if (readers.isEmpty()) { secondaryKey = null; } else { secondaryKey = readers.firstKey(); } // get the number of columns in the user's rows columns = primary.getColumns(); } }
public static void printJsonMetaData(List <string> files, Configuration conf, List <int> rowIndexCols, bool prettyPrint, bool printTimeZone) { JsonWriter writer = new JsonWriter(); bool multiFile = files.Count > 1; if (multiFile) { writer.array(); } else { writer.newObject(); } foreach (string filename in files) { if (multiFile) { writer.newObject(); } writer.key("fileName").value(Path.GetFileName(filename)); Reader reader = OrcFile.createReader(filename, OrcFile.readerOptions(conf)); writer.key("fileVersion").value(OrcFile.VersionHelper.getName(reader.getFileVersion())); writer.key("writerVersion").value(reader.getWriterVersion().ToString()); using (RecordReaderImpl rows = (RecordReaderImpl)reader.rows()) { writer.key("numberOfRows").value(reader.getNumberOfRows()); writer.key("compression").value(reader.getCompression().ToString()); if (reader.getCompression() != CompressionKind.NONE) { writer.key("compressionBufferSize").value(reader.getCompressionSize()); } writer.key("schemaString").value(reader.getObjectInspector().getTypeName()); writer.key("schema").array(); writeSchema(writer, reader.getTypes()); writer.endArray(); writer.key("stripeStatistics").array(); List <StripeStatistics> stripeStatistics = reader.getStripeStatistics(); for (int n = 0; n < stripeStatistics.Count; n++) { writer.newObject(); writer.key("stripeNumber").value(n + 1); StripeStatistics ss = stripeStatistics[n]; writer.key("columnStatistics").array(); for (int i = 0; i < ss.getColumnStatistics().Length; i++) { writer.newObject(); writer.key("columnId").value(i); writeColumnStatistics(writer, ss.getColumnStatistics()[i]); writer.endObject(); } writer.endArray(); writer.endObject(); } writer.endArray(); ColumnStatistics[] stats = reader.getStatistics(); int colCount = stats.Length; writer.key("fileStatistics").array(); for (int i = 0; i < stats.Length; ++i) { writer.newObject(); writer.key("columnId").value(i); writeColumnStatistics(writer, stats[i]); writer.endObject(); } writer.endArray(); writer.key("stripes").array(); int stripeIx = -1; foreach (StripeInformation stripe in reader.getStripes()) { ++stripeIx; long stripeStart = stripe.getOffset(); OrcProto.StripeFooter footer = rows.readStripeFooter(stripe); writer.newObject(); // start of stripe information writer.key("stripeNumber").value(stripeIx + 1); writer.key("stripeInformation"); writeStripeInformation(writer, stripe); if (printTimeZone) { writer.key("writerTimezone").value( footer.HasWriterTimezone ? footer.WriterTimezone : FileDump.UNKNOWN); } long sectionStart = stripeStart; writer.key("streams").array(); foreach (OrcProto.Stream section in footer.StreamsList) { writer.newObject(); string kind = section.HasKind ? section.Kind.ToString() : FileDump.UNKNOWN; writer.key("columnId").value(section.Column); writer.key("section").value(kind); writer.key("startOffset").value(sectionStart); writer.key("length").value(section.Length); sectionStart += (long)section.Length; writer.endObject(); } writer.endArray(); writer.key("encodings").array(); for (int i = 0; i < footer.ColumnsCount; ++i) { writer.newObject(); OrcProto.ColumnEncoding encoding = footer.ColumnsList[i]; writer.key("columnId").value(i); writer.key("kind").value(encoding.Kind.ToString()); if (encoding.Kind == OrcProto.ColumnEncoding.Types.Kind.DICTIONARY || encoding.Kind == OrcProto.ColumnEncoding.Types.Kind.DICTIONARY_V2) { writer.key("dictionarySize").value(encoding.DictionarySize); } writer.endObject(); } writer.endArray(); if (rowIndexCols != null && rowIndexCols.Count != 0) { // include the columns that are specified, only if the columns are included, bloom filter // will be read bool[] sargColumns = new bool[colCount]; foreach (int colIdx in rowIndexCols) { sargColumns[colIdx] = true; } RecordReaderImpl.Index indices = rows.readRowIndex(stripeIx, null, sargColumns); writer.key("indexes").array(); foreach (int col in rowIndexCols) { writer.newObject(); writer.key("columnId").value(col); writeRowGroupIndexes(writer, col, indices.getRowGroupIndex()); writeBloomFilterIndexes(writer, col, indices.getBloomFilterIndex()); writer.endObject(); } writer.endArray(); } writer.endObject(); // end of stripe information } writer.endArray(); long fileLen = new FileInfo(filename).Length; long paddedBytes = FileDump.getTotalPaddingSize(reader); // empty ORC file is ~45 bytes. Assumption here is file length always >0 double percentPadding = ((double)paddedBytes / (double)fileLen) * 100; writer.key("fileLength").value(fileLen); writer.key("paddingLength").value(paddedBytes); writer.key("paddingRatio").value(percentPadding); rows.close(); } writer.endObject(); } if (multiFile) { writer.endArray(); } if (prettyPrint) { #if false string prettyJson; if (multiFile) { JSONArray jsonArray = new JSONArray(writer.toString()); prettyJson = jsonArray.toString(2); } else { JSONObject jsonObject = new JSONObject(writer.toString()); prettyJson = jsonObject.toString(2); } #else string prettyJson = writer.ToString(); #endif System.Console.WriteLine(prettyJson); } else { System.Console.WriteLine(writer.ToString()); } }