Add experimental filesystem="arrow" support in dask_cudf.read_parquet
#16684
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
rjzamora
added
feature request
rapids-bot bot
pushed a commit
to rapidsai/dask-cuda
that referenced
this pull request
Aug 30, 2024
Adds new benchmark for parquet read performance using a `LocalCUDACluster`. The user can pass in `--key` and `--secret` options to specify S3 credentials. E.g. ``` $ python ./local_read_parquet.py --devs 0,1,2,3,4,5,6,7 --filesystem fsspec --type gpu --file-count 48 --aggregate-files Parquet read benchmark -------------------------------------------------------------------------------- Path | s3://dask-cudf-parquet-testing/dedup_parquet Columns | None Backend | cudf Filesystem | fsspec Blocksize | 244.14 MiB Aggregate files | True Row count | 372066 Size on disk | 1.03 GiB Number of workers | 8 ================================================================================ Wall clock | Throughput -------------------------------------------------------------------------------- 36.75 s | 28.78 MiB/s 21.29 s | 49.67 MiB/s 17.91 s | 59.05 MiB/s ================================================================================ Throughput | 41.77 MiB/s +/- 7.81 MiB/s Bandwidth | 0 B/s +/- 0 B/s Wall clock | 25.32 s +/- 8.20 s ================================================================================ ... ``` **Notes**: - S3 Performance generally scales with the number of workers (multiplied the number of threads per worker) - The example shown above was not executed from an EC2 instance - The example shown above *should* perform better after rapidsai/cudf#16657 - Using `--filesystem arrow` together with `--type gpu` performs well, but depends on rapidsai/cudf#16684 Authors: - Richard (Rick) Zamora (https://github.com/rjzamora) Approvers: - Mads R. B. Kristensen (https://github.com/madsbk) - Peter Andreas Entschev (https://github.com/pentschev) URL: #1371
3 tasks
Closes rapidsai#14537. Authors: - Matthew Murray (https://github.com/Matt711) - GALI PREM SAGAR (https://github.com/galipremsagar) Approvers: - GALI PREM SAGAR (https://github.com/galipremsagar) - Matthew Roeschke (https://github.com/mroeschke) - Vyas Ramasubramani (https://github.com/vyasr) URL: rapidsai#16601
…#16574) Improves performance of wide strings (avg > 64 bytes) when using `cudf::strings::slice_strings`. Addresses some concerns from issue rapidsai#15924 Authors: - David Wendt (https://github.com/davidwendt) Approvers: - Bradley Dice (https://github.com/bdice) - Muhammad Haseeb (https://github.com/mhaseeb123) URL: rapidsai#16574
rjzamora
added
3 - Ready for Review
Thanks for the nudge @GregoryKimball - I spent some time today isolating the "experimental" logic used for this feature. Therefore, the PR should now be relatively "low risk". I'd like to grab an ec2 instance later to day to test that the performance is still good. The usual dask-cudf and python/IO reviewers include @pentschev @charlesbluca @galipremsagar @wence- @madsbk @quasiben @vyasr - I'd I welcome a review from anyone available :) |
bdice
reviewed
Sep 25, 2024
galipremsagar
approved these changes
Sep 25, 2024
galipremsagar
added
5 - Ready to Merge
wence-
approved these changes
Sep 25, 2024
Comment on lines
752
to
756
| warnings.warn( | ||
| f"Support for `filesystem={filesystem}` is experimental. " | ||
| "Using PyArrow to perform IO on multiple CPU threads. " | ||
| "Behavior may change in the future (without deprecation). " | ||
| ) |
There was a problem hiding this comment.
question: Do we really want a user-visible (?) warning here? If curator uses this, it would presumably mean end users would see this warning but not know what to do about it (or have any facility to).
There was a problem hiding this comment.
@ayushdg @VibhuJawa - What are your feelings on this. I agree that this is probably a bit annoying :)
There was a problem hiding this comment.
Okay - I added a note to the best-practices docs about this feature, and removed the explicit warning.
This PR adds `cudf-polars` to the top level build script. Authors: - https://github.com/brandon-b-miller - GALI PREM SAGAR (https://github.com/galipremsagar) Approvers: - GALI PREM SAGAR (https://github.com/galipremsagar) - Jake Awe (https://github.com/AyodeAwe) URL: rapidsai#16898
Before when `columns=` was a `cudf.Series/Index` we would call `return array.unique.to_pandas()`, but `.unique` is a method not a property so this would have raised an error. Also took the time to refactor the helper methods here and push down the `errors=` keyword to `Frame._drop_column` Authors: - Matthew Roeschke (https://github.com/mroeschke) Approvers: - Bradley Dice (https://github.com/bdice) URL: rapidsai#16712
This PR is a first pass at rapidsai#15937. We will close rapidsai#15937 after rapidsai#15162 is closed Authors: - Matthew Murray (https://github.com/Matt711) Approvers: - GALI PREM SAGAR (https://github.com/galipremsagar) URL: rapidsai#16810
Fixes rapidsai#16625 This PR fixes a slow implementation of the centroid merging step during the tdigest merge aggregation. Previously it was doing a linear march over the individual tdigests per group and merging them one by one. This led to terrible performance for large numbers of groups. In principle though, all this really was doing was a segmented sort of centroid values. So that's what this PR changes it to. Speedup for 1,000,000 input tidests with 1,000,000 individual groups is ~1000x, ``` Old --------------------------------------------------------------------------------------------------------------- Benchmark Time CPU Iterations --------------------------------------------------------------------------------------------------------------- TDigest/many_tiny_groups/1000000/1/1/10000/iterations:8/manual_time 7473 ms 7472 ms 8 TDigest/many_tiny_groups2/1000000/1/1/1000/iterations:8/manual_time 7433 ms 7431 ms 8 ``` ``` New --------------------------------------------------------------------------------------------------------------- Benchmark Time CPU Iterations --------------------------------------------------------------------------------------------------------------- TDigest/many_tiny_groups/1000000/1/1/10000/iterations:8/manual_time 6.72 ms 6.79 ms 8 TDigest/many_tiny_groups2/1000000/1/1/1000/iterations:8/manual_time 1.24 ms 1.32 ms 8 ``` Authors: - https://github.com/nvdbaranec - Muhammad Haseeb (https://github.com/mhaseeb123) - GALI PREM SAGAR (https://github.com/galipremsagar) Approvers: - Muhammad Haseeb (https://github.com/mhaseeb123) - Nghia Truong (https://github.com/ttnghia) - Mike Wilson (https://github.com/hyperbolic2346) URL: rapidsai#16780
This PR displays delta's for CPU and GPU usage metrics that are extracted from `cudf.pandas` pytests. Authors: - GALI PREM SAGAR (https://github.com/galipremsagar) Approvers: - Jake Awe (https://github.com/AyodeAwe) URL: rapidsai#16864
|
/merge |
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.
Description
This PR piggybacks on the existing CPU/Arrow Parquet infrastructure in dask-expr. With this PR,
will produce a
cudf-backed collection using PyArrow for IO (i.e. disk->pa.Table->cudf.DataFrame). Before this PR, passingfilesystem="arrow"will simply result in an error.Although this code path is not ideal for fast/local storage, it can be very efficient for remote storage (e.g. S3).
Checklist