feat: Optimize PrestoBatchVectorSerializer [1/7]: Serialize FlatVectors #12060
+743
−56
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
kevinwilfong
requested review from
assignUser and
majetideepak
as code owners
facebook-github-bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D68037258 |
✅ Deploy Preview for meta-velox canceled.
|
kevinwilfong
force-pushed
the
export-D68037258
branch
from
c06148e to
3164264
Compare
kevinwilfong
pushed a commit
to kevinwilfong/velox
that referenced
this pull request
Jan 10, 2025
…rs (facebookincubator#12060) Summary: Context: This is a series of diffs in which I reimplement PrestoBatchVectorSerializer to write directly to the output stream, rather than the indirect route it currently uses via VectorStreams. Reusing VectorStreams and much of the code for PrestoIterativeVectorSerializer prevented us from capturing all of the performance benefits of writing data in batches rather than row by row. These changes combined will speed up PrestoBatchVectorSerializer 2-3x (as measured in Presto queries and other use cases). In the final diff I will integrate the new serialization functions into PrestoBatchVectorSerializer's serialize function which will switch it to the new optimized writing path, therefore I will land these changes as a stack. In this diff: I provide the implementations for serializing FlatVectors. Differential Revision: D68037258
|
This pull request was exported from Phabricator. Differential Revision: D68037258 |
kevinwilfong
pushed a commit
to kevinwilfong/velox
that referenced
this pull request
Jan 10, 2025
…rs (facebookincubator#12060) Summary: Context: This is a series of diffs in which I reimplement PrestoBatchVectorSerializer to write directly to the output stream, rather than the indirect route it currently uses via VectorStreams. Reusing VectorStreams and much of the code for PrestoIterativeVectorSerializer prevented us from capturing all of the performance benefits of writing data in batches rather than row by row. These changes combined will speed up PrestoBatchVectorSerializer 2-3x (as measured in Presto queries and other use cases). In the final diff I will integrate the new serialization functions into PrestoBatchVectorSerializer's serialize function which will switch it to the new optimized writing path, therefore I will land these changes as a stack. In this diff: I provide the implementations for serializing FlatVectors. Differential Revision: D68037258
kevinwilfong
force-pushed
the
export-D68037258
branch
from
3164264 to
02b90c2
Compare
|
This pull request was exported from Phabricator. Differential Revision: D68037258 |
kevinwilfong
pushed a commit
to kevinwilfong/velox
that referenced
this pull request
Jan 11, 2025
…rs (facebookincubator#12060) Summary: Pull Request resolved: facebookincubator#12060 Context: This is a series of diffs in which I reimplement PrestoBatchVectorSerializer to write directly to the output stream, rather than the indirect route it currently uses via VectorStreams. Reusing VectorStreams and much of the code for PrestoIterativeVectorSerializer prevented us from capturing all of the performance benefits of writing data in batches rather than row by row. These changes combined will speed up PrestoBatchVectorSerializer 2-3x (as measured in Presto queries and other use cases). In the final diff I will integrate the new serialization functions into PrestoBatchVectorSerializer's serialize function which will switch it to the new optimized writing path, therefore I will land these changes as a stack. In this diff: I provide the implementations for serializing FlatVectors. Differential Revision: D68037258
kevinwilfong
force-pushed
the
export-D68037258
branch
from
02b90c2 to
e75e0f8
Compare
|
This pull request was exported from Phabricator. Differential Revision: D68037258 |
kevinwilfong
force-pushed
the
export-D68037258
branch
from
e75e0f8 to
3c6c7e7
Compare
kevinwilfong
pushed a commit
to kevinwilfong/velox
that referenced
this pull request
Jan 11, 2025
…rs (facebookincubator#12060) Summary: Pull Request resolved: facebookincubator#12060 Context: This is a series of diffs in which I reimplement PrestoBatchVectorSerializer to write directly to the output stream, rather than the indirect route it currently uses via VectorStreams. Reusing VectorStreams and much of the code for PrestoIterativeVectorSerializer prevented us from capturing all of the performance benefits of writing data in batches rather than row by row. These changes combined will speed up PrestoBatchVectorSerializer 2-3x (as measured in Presto queries and other use cases). In the final diff I will integrate the new serialization functions into PrestoBatchVectorSerializer's serialize function which will switch it to the new optimized writing path, therefore I will land these changes as a stack. In this diff: I provide the implementations for serializing FlatVectors. Differential Revision: D68037258
|
This pull request was exported from Phabricator. Differential Revision: D68037258 |
1 similar comment
|
This pull request was exported from Phabricator. Differential Revision: D68037258 |
kevinwilfong
pushed a commit
to kevinwilfong/velox
that referenced
this pull request
Jan 13, 2025
…rs (facebookincubator#12060) Summary: Pull Request resolved: facebookincubator#12060 Context: This is a series of diffs in which I reimplement PrestoBatchVectorSerializer to write directly to the output stream, rather than the indirect route it currently uses via VectorStreams. Reusing VectorStreams and much of the code for PrestoIterativeVectorSerializer prevented us from capturing all of the performance benefits of writing data in batches rather than row by row. These changes combined will speed up PrestoBatchVectorSerializer 2-3x (as measured in Presto queries and other use cases). In the final diff I will integrate the new serialization functions into PrestoBatchVectorSerializer's serialize function which will switch it to the new optimized writing path, therefore I will land these changes as a stack. In this diff: I provide the implementations for serializing FlatVectors. Differential Revision: D68037258
kevinwilfong
force-pushed
the
export-D68037258
branch
from
3c6c7e7 to
edf01b5
Compare
kevinwilfong
pushed a commit
to kevinwilfong/velox
that referenced
this pull request
Jan 13, 2025
…rs (facebookincubator#12060) Summary: Pull Request resolved: facebookincubator#12060 Context: This is a series of diffs in which I reimplement PrestoBatchVectorSerializer to write directly to the output stream, rather than the indirect route it currently uses via VectorStreams. Reusing VectorStreams and much of the code for PrestoIterativeVectorSerializer prevented us from capturing all of the performance benefits of writing data in batches rather than row by row. These changes combined will speed up PrestoBatchVectorSerializer 2-3x (as measured in Presto queries and other use cases). In the final diff I will integrate the new serialization functions into PrestoBatchVectorSerializer's serialize function which will switch it to the new optimized writing path, therefore I will land these changes as a stack. In this diff: I provide the implementations for serializing FlatVectors. Differential Revision: D68037258
kevinwilfong
force-pushed
the
export-D68037258
branch
from
edf01b5 to
29bb319
Compare
|
This pull request was exported from Phabricator. Differential Revision: D68037258 |
added 3 commits
January 14, 2025 14:06
Summary: Context: I plan to update the PrestoBatchVectorSerializer to write directly to the OutputStream rather than going through VectorStreams as I've seen this can greatly improve the speed. However, this ends up with many small writes to the OutputStream, which is typically a ByteOutputStream, so this leads to a lot of small allocations which is counter productive to the optimization I'm trying to make. To address this I add a BufferedOutputStream which wraps around another OutputStream, coalesces writes in a buffer, and flushes those as large writes to the wrapped OutputStream as the buffer fills up, or as needed. In my experiments I've seen the cost of the additional copy is far less then the cost of the tiny allocations. Differential Revision: D67997655
…incubator#12050) Summary: In PrestoSerializer we write the null bits to a ByteOutputStream. Presto's serialization format inverts the null flags from what Velox uses. This change adds an option to ByteOutputStream to do that negation on flush, rather than requiring the user to negate them when calling append. This makes it easier to write ranges of null bits without an additional copy, and reduces the level of mental gymnastics needed to reason about the null flags in the serializer. (It's analogous to the option ByteOutputStream already has to reverse the order of bits within a byte, again because of Presto's serialization format for null flags). Reviewed By: xiaoxmeng Differential Revision: D67994045
…rs (facebookincubator#12060) Summary: Context: This is a series of diffs in which I reimplement PrestoBatchVectorSerializer to write directly to the output stream, rather than the indirect route it currently uses via VectorStreams. Reusing VectorStreams and much of the code for PrestoIterativeVectorSerializer prevented us from capturing all of the performance benefits of writing data in batches rather than row by row. These changes combined will speed up PrestoBatchVectorSerializer 2-3x (as measured in Presto queries and other use cases). In the final diff I will integrate the new serialization functions into PrestoBatchVectorSerializer's serialize function which will switch it to the new optimized writing path, therefore I will land these changes as a stack. In this diff: I provide the implementations for serializing FlatVectors. Differential Revision: D68037258
kevinwilfong
force-pushed
the
export-D68037258
branch
from
29bb319 to
bac05f9
Compare
|
This pull request was exported from Phabricator. Differential Revision: D68037258 |
kevinwilfong
pushed a commit
to kevinwilfong/velox
that referenced
this pull request
Jan 14, 2025
…rs (facebookincubator#12060) Summary: Context: This is a series of diffs in which I reimplement PrestoBatchVectorSerializer to write directly to the output stream, rather than the indirect route it currently uses via VectorStreams. Reusing VectorStreams and much of the code for PrestoIterativeVectorSerializer prevented us from capturing all of the performance benefits of writing data in batches rather than row by row. These changes combined will speed up PrestoBatchVectorSerializer 2-3x (as measured in Presto queries and other use cases). In the final diff I will integrate the new serialization functions into PrestoBatchVectorSerializer's serialize function which will switch it to the new optimized writing path, therefore I will land these changes as a stack. In this diff: I provide the implementations for serializing FlatVectors. Differential Revision: D68037258
kevinwilfong
pushed a commit
to kevinwilfong/velox
that referenced
this pull request
Jan 14, 2025
…rs (facebookincubator#12060) Summary: Context: This is a series of diffs in which I reimplement PrestoBatchVectorSerializer to write directly to the output stream, rather than the indirect route it currently uses via VectorStreams. Reusing VectorStreams and much of the code for PrestoIterativeVectorSerializer prevented us from capturing all of the performance benefits of writing data in batches rather than row by row. These changes combined will speed up PrestoBatchVectorSerializer 2-3x (as measured in Presto queries and other use cases). In the final diff I will integrate the new serialization functions into PrestoBatchVectorSerializer's serialize function which will switch it to the new optimized writing path, therefore I will land these changes as a stack. In this diff: I provide the implementations for serializing FlatVectors. Differential Revision: D68037258
kevinwilfong
pushed a commit
to kevinwilfong/velox
that referenced
this pull request
Jan 14, 2025
…rs (facebookincubator#12060) Summary: Context: This is a series of diffs in which I reimplement PrestoBatchVectorSerializer to write directly to the output stream, rather than the indirect route it currently uses via VectorStreams. Reusing VectorStreams and much of the code for PrestoIterativeVectorSerializer prevented us from capturing all of the performance benefits of writing data in batches rather than row by row. These changes combined will speed up PrestoBatchVectorSerializer 2-3x (as measured in Presto queries and other use cases). In the final diff I will integrate the new serialization functions into PrestoBatchVectorSerializer's serialize function which will switch it to the new optimized writing path, therefore I will land these changes as a stack. In this diff: I provide the implementations for serializing FlatVectors. Differential Revision: D68037258
kevinwilfong
pushed a commit
to kevinwilfong/velox
that referenced
this pull request
Jan 14, 2025
…rs (facebookincubator#12060) Summary: Context: This is a series of diffs in which I reimplement PrestoBatchVectorSerializer to write directly to the output stream, rather than the indirect route it currently uses via VectorStreams. Reusing VectorStreams and much of the code for PrestoIterativeVectorSerializer prevented us from capturing all of the performance benefits of writing data in batches rather than row by row. These changes combined will speed up PrestoBatchVectorSerializer 2-3x (as measured in Presto queries and other use cases). In the final diff I will integrate the new serialization functions into PrestoBatchVectorSerializer's serialize function which will switch it to the new optimized writing path, therefore I will land these changes as a stack. In this diff: I provide the implementations for serializing FlatVectors. Differential Revision: D68037258
kevinwilfong
pushed a commit
to kevinwilfong/velox
that referenced
this pull request
Jan 14, 2025
…rs (facebookincubator#12060) Summary: Context: This is a series of diffs in which I reimplement PrestoBatchVectorSerializer to write directly to the output stream, rather than the indirect route it currently uses via VectorStreams. Reusing VectorStreams and much of the code for PrestoIterativeVectorSerializer prevented us from capturing all of the performance benefits of writing data in batches rather than row by row. These changes combined will speed up PrestoBatchVectorSerializer 2-3x (as measured in Presto queries and other use cases). In the final diff I will integrate the new serialization functions into PrestoBatchVectorSerializer's serialize function which will switch it to the new optimized writing path, therefore I will land these changes as a stack. In this diff: I provide the implementations for serializing FlatVectors. Differential Revision: D68037258
kevinwilfong
pushed a commit
to kevinwilfong/velox
that referenced
this pull request
Jan 14, 2025
…rs (facebookincubator#12060) Summary: Context: This is a series of diffs in which I reimplement PrestoBatchVectorSerializer to write directly to the output stream, rather than the indirect route it currently uses via VectorStreams. Reusing VectorStreams and much of the code for PrestoIterativeVectorSerializer prevented us from capturing all of the performance benefits of writing data in batches rather than row by row. These changes combined will speed up PrestoBatchVectorSerializer 2-3x (as measured in Presto queries and other use cases). In the final diff I will integrate the new serialization functions into PrestoBatchVectorSerializer's serialize function which will switch it to the new optimized writing path, therefore I will land these changes as a stack. In this diff: I provide the implementations for serializing FlatVectors. Differential Revision: D68037258
kevinwilfong
pushed a commit
to kevinwilfong/velox
that referenced
this pull request
Jan 14, 2025
…rs (facebookincubator#12060) Summary: Context: This is a series of diffs in which I reimplement PrestoBatchVectorSerializer to write directly to the output stream, rather than the indirect route it currently uses via VectorStreams. Reusing VectorStreams and much of the code for PrestoIterativeVectorSerializer prevented us from capturing all of the performance benefits of writing data in batches rather than row by row. These changes combined will speed up PrestoBatchVectorSerializer 2-3x (as measured in Presto queries and other use cases). In the final diff I will integrate the new serialization functions into PrestoBatchVectorSerializer's serialize function which will switch it to the new optimized writing path, therefore I will land these changes as a stack. In this diff: I provide the implementations for serializing FlatVectors. Differential Revision: D68037258
kevinwilfong
pushed a commit
to kevinwilfong/velox
that referenced
this pull request
Jan 14, 2025
…rs (facebookincubator#12060) Summary: Context: This is a series of diffs in which I reimplement PrestoBatchVectorSerializer to write directly to the output stream, rather than the indirect route it currently uses via VectorStreams. Reusing VectorStreams and much of the code for PrestoIterativeVectorSerializer prevented us from capturing all of the performance benefits of writing data in batches rather than row by row. These changes combined will speed up PrestoBatchVectorSerializer 2-3x (as measured in Presto queries and other use cases). In the final diff I will integrate the new serialization functions into PrestoBatchVectorSerializer's serialize function which will switch it to the new optimized writing path, therefore I will land these changes as a stack. In this diff: I provide the implementations for serializing FlatVectors. Differential Revision: D68037258
kevinwilfong
pushed a commit
to kevinwilfong/velox
that referenced
this pull request
Jan 14, 2025
…rs (facebookincubator#12060) Summary: Context: This is a series of diffs in which I reimplement PrestoBatchVectorSerializer to write directly to the output stream, rather than the indirect route it currently uses via VectorStreams. Reusing VectorStreams and much of the code for PrestoIterativeVectorSerializer prevented us from capturing all of the performance benefits of writing data in batches rather than row by row. These changes combined will speed up PrestoBatchVectorSerializer 2-3x (as measured in Presto queries and other use cases). In the final diff I will integrate the new serialization functions into PrestoBatchVectorSerializer's serialize function which will switch it to the new optimized writing path, therefore I will land these changes as a stack. In this diff: I provide the implementations for serializing FlatVectors. Differential Revision: D68037258
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary:
Context:
This is a series of diffs in which I reimplement PrestoBatchVectorSerializer to write directly to the output stream,
rather than the indirect route it currently uses via VectorStreams. Reusing VectorStreams and much of the code
for PrestoIterativeVectorSerializer prevented us from capturing all of the performance benefits of writing data in
batches rather than row by row. These changes combined will speed up PrestoBatchVectorSerializer 2-3x (as
measured in Presto queries and other use cases).
In the final diff I will integrate the new serialization functions into PrestoBatchVectorSerializer's serialize
function which will switch it to the new optimized writing path, therefore I will land these changes as a stack.
In this diff:
I provide the implementations for serializing FlatVectors.
Differential Revision: D68037258