jo-me opened a new issue, #40125:
URL: https://github.com/apache/arrow/issues/40125

   ### Describe the usage question you have. Please include as many useful 
details as  possible.
   
   
   Hi,
   
   beein new to Apache Arrow I'm a little confused about the different options 
to interact with Parquet files. The documentation in the Go library is in many 
places very sparse and existing examples from various sources don't seem to 
match my use case.
   
   **The question is:** 
   Given that you have a parquet file containing serveral thousand rows each 
with an _ID_ column and a _Data_ column, where the data column holds some 
larger blob, how do you seek certain rows based on their ID column and extract 
the data of the _Data_ column in an efficient and memory-friendly way.  
   
   By 'memory-friendly' I mean that only the relevant values should be read 
from the parquet files and loaded into memory, not the whole column, rowgroup, 
batch or chunk. Reading the ID column completely into memory would be fine, but 
not the blob data.
   
   I tried the variant with creating a pqarrow.RecordReader based on a 
pqarrow.FileReader based on parquet.Reader, but it seems that the Record 
batches always load the the whole batch (incl. all column data) into memory, 
not just when loading the value of of a column entry by index. While this 
approach works as desired, it has a very high memory usage due to the large 
blobs.
   
   I also tried to extract the relevant row-indexes in a first sweep to then 
somehow only retrieve these rows from the Data column in a second sweep, but I 
could not find a way that improved the first approach. 
   
   There is probably a simple way (without using pqarrow?) by just iterating 
over the parquet file rowgroups, but the usage of the available datastructures 
are not reallly documented well (FieldReaders, chunks, etc.)
   
   Btw, doing the same thing with DuckDb works very well and is noticably 
lighter on memory than the RecordReader approach, but including that library 
for the simple seek&extract use case is somewhat overkill and I would prefer to 
avoid it.
   
   Thanks for any hints
   
   
   Jochen Mehlhorn <[email protected]>, Mercedes-Benz Tech 
Innovation GmbH
   
   [Provider 
Information](https://github.com/Daimler/daimler-foss/blob/master/PROVIDER_INFORMATION.md)
   
   
   
   
   
   ### Component(s)
   
   Go


-- 
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]

Reply via email to