/**
* 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);
}
OrcRecordUpdater(Path path,
AcidOutputFormat.Options options)
{
this.options = options;
this.bucket.set(options.getBucket());
this.path = AcidUtils.createFilename(path, options);
FileSystem fs = options.getFilesystem();
if (fs == null)
{
fs = path.getFileSystem(options.getConfiguration());
}
this.fs = fs;
try
{
FSDataOutputStream strm = fs.create(new Path(path, ACID_FORMAT), false);
strm.writeInt(ORC_ACID_VERSION);
strm.close();
}
catch (IOException ioe)
{
if (LOG.isDebugEnabled())
{
LOG.debug("Failed to create " + path + "/" + ACID_FORMAT + " with " +
ioe);
}
}
if (options.getMinimumTransactionId() != options.getMaximumTransactionId() &&
!options.isWritingBase())
{
flushLengths = fs.create(getSideFile(this.path), true, 8,
options.getReporter());
}
else
{
flushLengths = null;
}
OrcFile.WriterOptions writerOptions = null;
if (options is OrcOptions)
{
writerOptions = ((OrcOptions)options).getOrcOptions();
}
if (writerOptions == null)
{
writerOptions = OrcFile.writerOptions( /* options.getTableProperties(), */
options.getConfiguration());
}
writerOptions.fileSystem(fs).callback(indexBuilder);
if (!options.isWritingBase())
{
writerOptions.blockPadding(false);
writerOptions.bufferSize(DELTA_BUFFER_SIZE);
writerOptions.stripeSize(DELTA_STRIPE_SIZE);
}
rowInspector = (StructObjectInspector)options.getInspector();
writerOptions.inspector(createEventSchema(findRecId(options.getInspector(),
options.getRecordIdColumn())));
this.writer = OrcFile.createWriter(this.path, writerOptions);
item = new OrcStruct(FIELDS);
item.setFieldValue(OPERATION, operation);
item.setFieldValue(CURRENT_TRANSACTION, currentTransaction);
item.setFieldValue(ORIGINAL_TRANSACTION, originalTransaction);
item.setFieldValue(BUCKET, bucket);
item.setFieldValue(ROW_ID, rowId);
}
/**
* 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();
}
}
