Open
Conversation
myrrc
added
action/benchmark
github-actions
bot
removed
the
action/benchmark
Contributor
Benchmarks: TPC-H SF=10 on NVMESummary
Detailed Results Table
|
Contributor
Benchmarks: FineWeb S3Summary
Detailed Results Table
|
Contributor
Benchmarks: Statistical and Population GeneticsSummary
Detailed Results Table
|
Contributor
Benchmarks: TPC-H SF=1 on S3Summary
Detailed Results Table
|
Contributor
Benchmarks: TPC-H SF=10 on S3Summary
Detailed Results Table
|
Contributor
Benchmarks: Clickbench on NVMESummary
Detailed Results Table
|
Contributor
Benchmarks: PolarSignals ProfilingSummary
Detailed Results Table
|
Contributor
Benchmarks: TPC-DS SF=1 on NVMESummary
Detailed Results Table
|
Contributor
Benchmarks: FineWeb NVMeSummary
Detailed Results Table
|
Contributor
Benchmarks: TPC-H SF=1 on NVMESummary
Detailed Results Table
|
github-actions
bot
removed
the
action/benchmark
github-actions
bot
removed
the
action/benchmark
Contributor
Polar Signals Profiling ResultsLatest Run
Previous Runs (16)
Powered by Polar Signals Cloud |
github-actions
bot
removed
the
action/benchmark
myrrc
force-pushed
the
myrrc/duckdbfs-perf-regression
branch
4 times, most recently
from
1bbec73 to
11ada49
Compare
github-actions
bot
removed
the
action/benchmark
Contributor
Benchmarks: Random AccessSummary
Detailed Results Table
|
Contributor
Benchmarks: CompressionSummary
Detailed Results Table
|
Contributor
Author
|
Testing this change is tricky as 1.5 is not released, thus there are no artifacts. I'll open a patch to spin up our own building jobs and cache artifacts in own S3, so we could test pre-releases. |
Contributor
Author
^ Stacktrace for the segfault |
Contributor
Author
|
Looks like we're copying some wrong-sized memory region |
Contributor
Author
|
I believe the issue is in somewhere else. If I comment the specific copying code, it errors elsewhere: |
Contributor
Author
|
Full stacktrace: The segfault itself is trying to memcpy from an invalid string_t: both length This error reproduces from time to time. Our reproduction: LD_LIBRARY_PATH override as otherwise duckdb-bench uses copied libduckdb.so Asan build: Crashes occur only on vortex and vortex-compact Asan/ubsan build Tsan build Asan crash |
Contributor
Author
|
Sometimes I also get the following assertion in duckdb: |
Contributor
Author
|
When building with leak sanitizer, I also get the following stack trace: |
Contributor
Author
|
The segfault can be reproduced without vortex-bench, just running query 5 from tpc-ds given the pre-generated database by |
Contributor
Author
|
The simplest reproduction is |
Contributor
|
Turns out we need to initialize columns for chunks that are not part of the projection to fix tpc-ds: |
Contributor
|
What, why are they even reading unreferenced columns? |
Contributor
|
Yah this seems weird? We never had to do it in the past. Is this a new 1.5 thing? |
Contributor
|
I would be curious to know the query and the number of non projected columns. Maybe we do need to fill out a dummy column in case of 0 column projections? |
Contributor
Yes, it's a 1.5 thing. |
Contributor
tpcds query 5. gotta check back on the exact counts. Note that the zero-projection scan in the doc is just an example, and not the tpcds 5 case. We should check what the Parquet extension does in DuckDB cc: @myrrc in terms of initializing (or not) unprojected columns. In my understanding the Parquet ext only writes to projected columns, so the bug is likely on the Vortex side. We're probably missing on writing some of the columns here, which should not just be NULL initialized. It's not clear whether the issue popped up bc of bumping DDB to 1.5 or bumping Vortex to the commit we're using here. |
Signed-off-by: Mikhail Kot <mikhail@spiraldb.com>
Signed-off-by: Mikhail Kot <mikhail@spiraldb.com>
Signed-off-by: Mikhail Kot <mikhail@spiraldb.com>
Signed-off-by: Mikhail Kot <mikhail@spiraldb.com>
Signed-off-by: Mikhail Kot <mikhail@spiraldb.com>
Signed-off-by: Alexander Droste <alexander.droste@protonmail.com> Signed-off-by: Mikhail Kot <mikhail@spiraldb.com>
Signed-off-by: Mikhail Kot <mikhail@spiraldb.com>
Signed-off-by: Mikhail Kot <mikhail@spiraldb.com>
0ax1
requested changes
Mar 4, 2026
| } else { | ||
| // The default DuckDB version to use when DUCKDB_VERSION env var is not set. | ||
| DuckDBVersion::Release("1.4.2".to_owned()) | ||
| DuckDBVersion::Commit("7f992e5a96662d66134577cf9d8a5bad9053847e".to_owned()) |
Contributor
There was a problem hiding this comment.
We have to wait until 1.5.0 is released to merge this and wrap this as DuckDBVersion::Release, in order to not build from source by default.
Contributor
Author
|
On the invalid chunk passed, the bug is also reproduced without a GROUP BY: The origin is unclear, but for the first correctly handled query duckdb tells us in init_global and init_local On the other query, duckdb tells us there is 1 column but zero projections. Vortex then forms an empty array. However, duckdb returns an output chunk with 1 column, so the semantics of TableFunctionInitInput's column_ids vs. projection_ids is unclear. Our current match is that chunks columns === projections_ids.size(), but this doesn't hold for second query. My guess is that optimizer in 1.5 correctly assumes we don't need an output column (because first over a constant is a constant) but then something strange happens in output chunk allocation. |
Contributor
Author
|
Even simpler reproduction: My second guess is a bug on our side because duckdb 1.5 does pass a chunk with 1 column (or 2, or columns present in the input file) for a .parquet, so we don't handle these correctly |
Contributor
Author
|
A non-related but ugly error is on request to write to non-existent file: The error message should be more clear |
Contributor
Author
|
Uninitialized read fixed in #6831 |
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
1.5 API changes:
#5767
Continuation of #5901