[GitHub] [arrow] tustvold commented on issue #34510: Reading FixedSizeList from parquet is slower than reading values into more rows

["tustvold (via GitHub)" <gi...@apache.org>]

tustvold commented on issue #34510:
URL: https://github.com/apache/arrow/issues/34510#issuecomment-1464215953

   > @alamb @tustvold I saw your blog post about this for arrow-rs. Do you handle this differently in Rust?
   
   We don't support FixedSizeList in arrow-rs AFAIK. Parquet to my knowledge does not have an equivalent [logical](https://github.com/apache/parquet-format/blob/master/LogicalTypes.md) construct, and so it isn't particularly clear to me what support would mean other than implicitly casting between a regular list and a fixed size list.
   
   > Calculating the def and rep levels for 100k rows with all non-null values takes 2x time as reading 8M doubles
   
   Assuming the doubles are PLAIN encoded this is not surprising, you are comparing the performance of what is effectively a `memcpy` that will run at the memory bandwidth, to a fairly complex [bit-packing](https://github.com/apache/parquet-format/blob/master/Encodings.md#run-length-encoding--bit-packing-hybrid-rle--3) scheme used for the definition and repetition levels.
   
   In the Rust implementation we have a couple of tricks that help here, but it is still relatively expensive (at least compared to primitive decoding):
   
   * We decode definition levels directly to the null buffer if there are only nulls at the leaf level (i.e. no lists or nested nulls), allowing us to preserve the bit-packing
   * We have vectorised unpack implementations specialised for each bit width (I believe arrow C++ does this also)
   
   > Definition level data is all 1 and supposed to be RLE encoded
   
   It will actually all be 2, unless the doubles are themselves not nullable
   
   > Repetition level data is a vector of 0 followed by 79x1 repeated 100k times for our case. I'm not sure if RLE will help here, sounds like an unnecessary complex structure for fixed size lists
   
   These repetition levels will be [RLE encoded](https://github.com/apache/parquet-format/blob/master/Encodings.md#run-length-encoding--bit-packing-hybrid-rle--3). Theoretically a reader could preserve this, but the record shredding logic is extremely fiddly and so might run the risk of adding complexity to an already very complex piece of code. At least in arrow-rs we decode to an array of `i16`


-- 
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: github-unsubscribe@arrow.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org