Skip to content

Read all row group metadata only when position column is projected #1716

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
rdblue merged 2 commits into from
Nov 7, 2020
Merged

Read all row group metadata only when position column is projected #1716

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f0faa61
Read all row group metadata only when position column is projected
chenjunjiedada Nov 4, 2020
55bb184
address comments
chenjunjiedada Nov 6, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 1 addition & 3 deletions parquet/src/main/java/org/apache/iceberg/parquet/ParquetReader.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -135,11 +135,9 @@ private void advance() {
throw new RuntimeIOException(e);
}

long rowPosition = rowGroupsStartRowPos[nextRowGroup];
model.setPageSource(pages, rowGroupsStartRowPos == null ? 0 : rowGroupsStartRowPos[nextRowGroup]);
Copy link
Collaborator Author

@chenjunjiedada chenjunjiedada Nov 4, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The rowPosition will be ignored if the position column is not projected.

Copy link
Contributor

@rdblue rdblue Nov 6, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't think this file should change. If the determination not to use correct start positions in made in ReadConf, then it should just set the position to 0 there. This can use whatever rowGroupStartRowPos contains.

nextRowGroupStart += pages.getRowCount();
nextRowGroup += 1;

model.setPageSource(pages, rowPosition);
}

@Override
Expand Down
15 changes: 11 additions & 4 deletions parquet/src/main/java/org/apache/iceberg/parquet/ReadConf.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,6 +27,7 @@
import java.util.Set;
import java.util.function.Function;
import java.util.stream.Collectors;
import org.apache.iceberg.MetadataColumns;
import org.apache.iceberg.Schema;
import org.apache.iceberg.exceptions.RuntimeIOException;
import org.apache.iceberg.expressions.Expression;
Expand Down Expand Up @@ -58,7 +59,8 @@ class ReadConf<T> {
private final long totalValues;
private final boolean reuseContainers;
private final Integer batchSize;
private final long[] startRowPositions;

private long[] startRowPositions;

// List of column chunk metadata for each row group
private final List<Map<ColumnPath, ColumnChunkMetaData>> columnChunkMetaDataForRowGroups;
Expand Down Expand Up @@ -89,8 +91,11 @@ class ReadConf<T> {
this.shouldSkip = new boolean[rowGroups.size()];

// Fetch all row groups starting positions to compute the row offsets of the filtered row groups
Map<Long, Long> offsetToStartPos = generateOffsetToStartPos();
this.startRowPositions = new long[rowGroups.size()];
Map<Long, Long> offsetToStartPos = null;
if (expectedSchema.findField(MetadataColumns.ROW_POSITION.fieldId()) != null) {
offsetToStartPos = generateOffsetToStartPos();
this.startRowPositions = new long[rowGroups.size()];
Copy link
Contributor

@rdblue rdblue Nov 6, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think this should be defined for all cases so that ParquetReader doesn't need to check whether it is defined. In that case, it should still be a final variable.

The only change for this PR should be whether offsetToStartPos is an empty map or the result of generateOffsetToStartPos(). We may also want to do the ROW_POSITION check in that method instead of here, depending on what is cleaner.

Copy link
Collaborator Author

@chenjunjiedada chenjunjiedada Nov 7, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Make sense to me. Updated.

}

ParquetMetricsRowGroupFilter statsFilter = null;
ParquetDictionaryRowGroupFilter dictFilter = null;
Expand All @@ -102,7 +107,9 @@ class ReadConf<T> {
long computedTotalValues = 0L;
for (int i = 0; i < shouldSkip.length; i += 1) {
BlockMetaData rowGroup = rowGroups.get(i);
startRowPositions[i] = offsetToStartPos.get(rowGroup.getStartingPos());
if (offsetToStartPos != null) {
startRowPositions[i] = offsetToStartPos.get(rowGroup.getStartingPos());
Copy link
Contributor

@rdblue rdblue Nov 6, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think this should use offsetToStartPos.getOrDefault(0L).

}
boolean shouldRead = filter == null || (
statsFilter.shouldRead(typeWithIds, rowGroup) &&
dictFilter.shouldRead(typeWithIds, rowGroup, reader.getDictionaryReader(rowGroup)));
Expand Down