lidavidm commented on pull request #10664:
URL: https://github.com/apache/arrow/pull/10664#issuecomment-875091620
Now that I'm running it again it just seems that count_rows with a filter
got very slow.
```
Before:
Min: 24.285114823000185 s
Median: 25.309564675000274 s
Mean: 25.45627587750014 s
Max: 26.564001424999333 s
After:
Min: 113.42360810700029 s
Median: 117.11463808250028 s
Mean: 118.39458996600005 s
Max: 123.9104565839998 s
```
<details>
<summary>Test script</summary>
```python
import statistics
import time
import pyarrow
import pyarrow.dataset
import pyarrow.fs
fs = pyarrow.fs.S3FileSystem(region="us-east-2")
print("PyArrow:", pyarrow.__version__)
times = []
for _ in range(10):
start = time.monotonic()
ds = pyarrow.dataset.dataset([
"ursa-labs-taxi-data/2009/01/data.parquet",
"ursa-labs-taxi-data/2009/02/data.parquet",
"ursa-labs-taxi-data/2009/03/data.parquet",
"ursa-labs-taxi-data/2009/04/data.parquet",
],
format="parquet",
partitioning=["year", "month"],
filesystem=fs,
)
print("Rows:", ds.scanner(use_async=True, use_threads=True).count_rows())
times.append(time.monotonic() - start)
print("No filter")
print("Min:", min(times), "s")
print("Median:", statistics.median(times), "s")
print("Mean:", statistics.mean(times), "s")
print("Max:", max(times), "s")
print("With filter")
times = []
expr = pyarrow.dataset.field("passenger_count") > 1
for _ in range(10):
start = time.monotonic()
ds = pyarrow.dataset.dataset([
"ursa-labs-taxi-data/2009/01/data.parquet",
"ursa-labs-taxi-data/2009/02/data.parquet",
"ursa-labs-taxi-data/2009/03/data.parquet",
"ursa-labs-taxi-data/2009/04/data.parquet",
],
format="parquet",
partitioning=["year", "month"],
filesystem=fs,
)
print("Rows:", ds.scanner(use_async=True, use_threads=True,
filter=expr).count_rows())
times.append(time.monotonic() - start)
print("Min:", min(times), "s")
print("Median:", statistics.median(times), "s")
print("Mean:", statistics.mean(times), "s")
print("Max:", max(times), "s")
```
</details>
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
