ajantha-bhat commented on a change in pull request #3196: [CARBONDATA-3366]
Support SDK reader to read blocklet level split
URL: https://github.com/apache/carbondata/pull/3196#discussion_r285516314
##########
File path:
store/sdk/src/main/java/org/apache/carbondata/sdk/file/CarbonReaderBuilder.java
##########
@@ -184,48 +175,116 @@ public CarbonReaderBuilder withRowRecordReader() {
}
format.setColumnProjection(job.getConfiguration(), projectionColumns);
}
+ if ((disableLoadBlockDataMap) && (filterExpression == null)) {
+ job.getConfiguration().set("filter_blocks", "false");
+ }
+ return format;
+ }
+ private <T> RecordReader getRecordReader(Job job, CarbonFileInputFormat
format,
+ List<RecordReader<Void, T>> readers, InputSplit split)
+ throws IOException, InterruptedException {
+ TaskAttemptContextImpl attempt =
+ new TaskAttemptContextImpl(job.getConfiguration(), new
TaskAttemptID());
+ RecordReader reader;
+ QueryModel queryModel = format.createQueryModel(split, attempt);
+ boolean hasComplex = false;
+ for (ProjectionDimension projectionDimension :
queryModel.getProjectionDimensions()) {
+ if (projectionDimension.getDimension().isComplex()) {
+ hasComplex = true;
+ break;
+ }
+ }
+ if (useVectorReader && !hasComplex) {
+ queryModel.setDirectVectorFill(filterExpression == null);
+ reader = new CarbonVectorizedRecordReader(queryModel);
+ } else {
+ reader = format.createRecordReader(split, attempt);
+ }
try {
+ reader.initialize(split, attempt);
+ } catch (Exception e) {
+ CarbonUtil.closeStreams(readers.toArray(new RecordReader[0]));
+ throw e;
+ }
+ return reader;
+ }
- if (filterExpression == null) {
- job.getConfiguration().set("filter_blocks", "false");
- }
+ /**
+ * Build CarbonReader
+ *
+ * @param <T>
+ * @return CarbonReader
+ * @throws IOException
+ * @throws InterruptedException
+ */
+ public <T> CarbonReader<T> build()
+ throws IOException, InterruptedException {
+ if (hadoopConf == null) {
+ hadoopConf = FileFactory.getConfiguration();
+ }
+ final Job job = new Job(new JobConf(hadoopConf));
+ CarbonFileInputFormat format = prepareFileInputFormat(job, false, true);
+ try {
List<InputSplit> splits =
format.getSplits(new JobContextImpl(job.getConfiguration(), new
JobID()));
List<RecordReader<Void, T>> readers = new ArrayList<>(splits.size());
for (InputSplit split : splits) {
- TaskAttemptContextImpl attempt =
- new TaskAttemptContextImpl(job.getConfiguration(), new
TaskAttemptID());
- RecordReader reader;
- QueryModel queryModel = format.createQueryModel(split, attempt);
- boolean hasComplex = false;
- for (ProjectionDimension projectionDimension :
queryModel.getProjectionDimensions()) {
- if (projectionDimension.getDimension().isComplex()) {
- hasComplex = true;
- break;
- }
- }
- if (useVectorReader && !hasComplex) {
- queryModel.setDirectVectorFill(filterExpression == null);
- reader = new CarbonVectorizedRecordReader(queryModel);
- } else {
- reader = format.createRecordReader(split, attempt);
- }
- try {
- reader.initialize(split, attempt);
- readers.add(reader);
- } catch (Exception e) {
- CarbonUtil.closeStreams(readers.toArray(new RecordReader[0]));
- throw e;
- }
+ RecordReader reader = getRecordReader(job, format, readers, split);
+ readers.add(reader);
}
return new CarbonReader<>(readers);
} catch (Exception ex) {
// Clear the datamap cache as it can get added in getSplits() method
DataMapStoreManager.getInstance()
- .clearDataMaps(table.getAbsoluteTableIdentifier());
+ .clearDataMaps(format.getAbsoluteTableIdentifier(hadoopConf));
throw ex;
}
}
+ /**
+ * Build blocklet level CarbonReaders
+ *
+ * @param <T>
+ * @throws IOException
+ * @throws InterruptedException
+ */
+ public <T> CarbonReader<T>[] buildBlockletReaders() throws IOException,
InterruptedException {
Review comment:
Have some integration issues in concurrent multithread scenario when SDK is
used from python. For that purpose getSplits() give an easy integration.
In future, I will check about masking the split inside readers and handling
the same.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
