[
https://issues.apache.org/jira/browse/DRILL-7062?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16812142#comment-16812142
]
ASF GitHub Bot commented on DRILL-7062:
---------------------------------------
amansinha100 commented on pull request #1738: DRILL-7062: Initial
implementation of run-time row-group pruning
URL: https://github.com/apache/drill/pull/1738#discussion_r272898010
##########
File path:
exec/java-exec/src/main/java/org/apache/drill/exec/store/parquet/AbstractParquetScanBatchCreator.java
##########
@@ -68,76 +83,131 @@ protected ScanBatch getBatch(ExecutorFragmentContext
context, AbstractParquetRow
List<RecordReader> readers = new LinkedList<>();
List<Map<String, String>> implicitColumns = new ArrayList<>();
Map<String, String> mapWithMaxColumns = new LinkedHashMap<>();
- for (RowGroupReadEntry rowGroup : rowGroupScan.getRowGroupReadEntries()) {
- /*
- Here we could store a map from file names to footers, to prevent
re-reading the footer for each row group in a file
- TODO - to prevent reading the footer again in the parquet record reader
(it is read earlier in the ParquetStorageEngine)
- we should add more information to the RowGroupInfo that will be
populated upon the first read to
- provide the reader with all of th file meta-data it needs
- These fields will be added to the constructor below
- */
- try {
- Stopwatch timer = logger.isTraceEnabled() ?
Stopwatch.createUnstarted() : null;
- DrillFileSystem fs = fsManager.get(rowGroupScan.getFsConf(rowGroup),
rowGroup.getPath());
- ParquetReaderConfig readerConfig = rowGroupScan.getReaderConfig();
- if (!footers.containsKey(rowGroup.getPath())) {
- if (timer != null) {
- timer.start();
+ ParquetReaderConfig readerConfig = rowGroupScan.getReaderConfig();
+ RowGroupReadEntry firstRowGroup = null; // to be scanned in case ALL row
groups are pruned out
+ ParquetMetadata firstFooter = null;
+ long rowgroupsPruned = 0; // for stats
+
+ try {
+
+ LogicalExpression filterExpr = rowGroupScan.getFilter();
+ Path selectionRoot = rowGroupScan.getSelectionRoot();
+ // Runtime pruning: Avoid recomputing metadata objects for each
row-group in case they use the same file
+ // by keeping the following objects computed earlier (relies on same
file being in consecutive rowgroups)
+ Path prevRowGroupPath = null;
+ Metadata_V3.ParquetTableMetadata_v3 tableMetadataV3 = null;
+ Metadata_V3.ParquetFileMetadata_v3 fileMetadataV3 = null;
+ FileSelection fileSelection = null;
+ ParquetTableMetadataProviderImpl metadataProvider = null;
+
+ for (RowGroupReadEntry rowGroup : rowGroupScan.getRowGroupReadEntries())
{
+ /*
+ Here we could store a map from file names to footers, to prevent
re-reading the footer for each row group in a file
+ TODO - to prevent reading the footer again in the parquet record
reader (it is read earlier in the ParquetStorageEngine)
+ we should add more information to the RowGroupInfo that will be
populated upon the first read to
+ provide the reader with all of th file meta-data it needs
+ These fields will be added to the constructor below
+ */
+
+ Stopwatch timer = logger.isTraceEnabled() ?
Stopwatch.createUnstarted() : null;
+ DrillFileSystem fs = fsManager.get(rowGroupScan.getFsConf(rowGroup),
rowGroup.getPath());
+ if (!footers.containsKey(rowGroup.getPath())) {
+ if (timer != null) {
+ timer.start();
+ }
+
+ ParquetMetadata footer = readFooter(fs.getConf(),
rowGroup.getPath(), readerConfig);
+ if (timer != null) {
+ long timeToRead = timer.elapsed(TimeUnit.MICROSECONDS);
+ logger.trace("ParquetTrace,Read Footer,{},{},{},{},{},{},{}",
"", rowGroup.getPath(), "", 0, 0, 0, timeToRead);
+ }
+ footers.put(rowGroup.getPath(), footer);
}
+ ParquetMetadata footer = footers.get(rowGroup.getPath());
+
+ //
+ // If a filter is given (and it is not just "TRUE") - then use it
to perform run-time pruning
+ //
+ if ( filterExpr != null && ! (filterExpr instanceof
ValueExpressions.BooleanExpression) ) { // skip when no filter or filter is
TRUE
+
+ int rowGroupIndex = rowGroup.getRowGroupIndex();
+ long footerRowCount =
footer.getBlocks().get(rowGroupIndex).getRowCount();
+
+ if ( timer != null ) { // restart the timer, if tracing
+ timer.reset();
+ timer.start();
+ }
+
+ // When starting a new file, or at the first time - Initialize
path specific metadata etc
+ if ( ! rowGroup.getPath().equals(prevRowGroupPath) ) {
+ // Get the table metadata (V3)
+ tableMetadataV3 = Metadata.getParquetTableMetadata(footer, fs,
rowGroup.getPath().toString(), readerConfig);
+
+ // The file status for this file
+ FileStatus fileStatus = fs.getFileStatus(rowGroup.getPath());
+ List<FileStatus> listFileStatus = new
ArrayList<>(Arrays.asList(fileStatus));
+ List<Path> listRowGroupPath = new
ArrayList<>(Arrays.asList(rowGroup.getPath()));
+ List<ReadEntryWithPath> entries = new
ArrayList<>(Arrays.asList(new ReadEntryWithPath(rowGroup.getPath())));
+ fileSelection = new FileSelection(listFileStatus,
listRowGroupPath, selectionRoot);
+
+ metadataProvider = new ParquetTableMetadataProviderImpl(entries,
selectionRoot, fileSelection.cacheFileRoot, readerConfig, fs,false);
+ // The file metadata (for all columns)
+ fileMetadataV3 =
Metadata.getParquetFileMetadata_v3(tableMetadataV3, footer, fileStatus, fs,
true, null, readerConfig);
+
+ prevRowGroupPath = rowGroup.getPath(); // for next time
+ }
+
+ MetadataBase.RowGroupMetadata rowGroupMetadata =
fileMetadataV3.getRowGroups().get(rowGroup.getRowGroupIndex());
+
+ Map<SchemaPath, ColumnStatistics> columnsStatistics =
ParquetTableMetadataUtils.getRowGroupColumnStatistics(tableMetadataV3,
rowGroupMetadata);
+
+ List<SchemaPath> columns =
columnsStatistics.keySet().stream().collect(Collectors.toList());
+
+ ParquetGroupScan parquetGroupScan = new ParquetGroupScan(
context.getQueryUserName(), metadataProvider, fileSelection, columns,
readerConfig, filterExpr);
Review comment:
I am actually less concerned about the elapsed time and more about the
number of objects created on the Java heap. The `ParquetGroupScan` is at least
100 bytes each and since we are targeting use cases with about million row
groups per minor fragment, it means 100 MB per minor fragment (assuming the
garbage collector is not doing the cleanup while the Foo loop is executing).
If there are 24 minor frags, this would add up to 2.4GB heap per node which
would cause quite a bit of GC activity subsequently.
I have made some changes on top of your branch which avoids this by directly
calling a static method to create a FilterPredicate. The pruning unit tests
are passing with these changes. I will send my changes to you offline and we
can discuss.
----------------------------------------------------------------
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:
us...@infra.apache.org
> Run-time row group pruning
> --------------------------
>
> Key: DRILL-7062
> URL: https://issues.apache.org/jira/browse/DRILL-7062
> Project: Apache Drill
> Issue Type: Sub-task
> Components: Metadata
> Reporter: Venkata Jyothsna Donapati
> Assignee: Boaz Ben-Zvi
> Priority: Major
> Fix For: 1.16.0
>
> Original Estimate: 504h
> Remaining Estimate: 504h
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
