Skip to content

Reduce parquet writer peak memory #64424

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
vdimir merged 7 commits into from
Jun 3, 2024
Merged

Reduce parquet writer peak memory #64424

Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
7bfd97a
reduce parquet writer peak memory
liuneng1994 May 27, 2024
c3882e6
some fix
liuneng1994 May 27, 2024
ac7f2ea
fix build tidy failed
liuneng1994 May 28, 2024
ab84878
Merge remote-tracking branch 'refs/remotes/origin/master' into reduce…
liuneng1994 May 28, 2024
cc4e1b5
Merge branch 'master' into reduce-parquet-writer-peak-memory
vdimir May 28, 2024
d4a7304
Merge remote-tracking branch 'refs/remotes/origin/master' into reduce…
liuneng1994 May 29, 2024
acccc20
Merge remote-tracking branch 'mine/reduce-parquet-writer-peak-memory'…
liuneng1994 May 29, 2024
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
17 changes: 10 additions & 7 deletions src/Processors/Formats/Impl/ParquetBlockOutputFormat.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -145,11 +145,10 @@ void ParquetBlockOutputFormat::consume(Chunk chunk)
/// Because the real SquashingTransform is only used for INSERT, not for SELECT ... INTO OUTFILE.
/// The latter doesn't even have a pipeline where a transform could be inserted, so it's more
/// convenient to do the squashing here. It's also parallelized here.

if (chunk.getNumRows() != 0)
{
staging_rows += chunk.getNumRows();
staging_bytes += chunk.bytes();
staging_bytes += chunk.allocatedBytes();
staging_chunks.push_back(std::move(chunk));
}

Expand Down Expand Up @@ -282,11 +281,15 @@ void ParquetBlockOutputFormat::writeRowGroup(std::vector<Chunk> chunks)
writeUsingArrow(std::move(chunks));
else
{
Chunk concatenated = std::move(chunks[0]);
for (size_t i = 1; i < chunks.size(); ++i)
concatenated.append(chunks[i]);
chunks.clear();

Chunk concatenated;
Copy link
Copy Markdown
Contributor

@arthurpassos arthurpassos Nov 14, 2025

Choose a reason for hiding this comment

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

As I understand, this affects the ordering of the data. I am not sure this is ok... @vdimir @liuneng1994

while (!chunks.empty())
{
if (concatenated.empty())
concatenated.swap(chunks.back());
else
concatenated.append(chunks.back());
chunks.pop_back();
}
writeRowGroupInOneThread(std::move(concatenated));
}
}
Expand Down