This is an interesting direction. I had looked at Arrow as a default Coder
replacement for schema PCollections, and that didn't seem fruitful as we
would need to create Arrow batches of size 1. However using Arrow to encode
batches over the FnAPI might indeed be an interesting approach. Streaming
It really needs someone to take a deep dive and look into whether Arrow is
a good fit now considering all the use cases that Apache Beam has. I did a
look about a year ago when designing the Fn Data API and concluded at that
point in time it wasn't great for several reasons but mainly due to the
Kenn, it can be done but requires explicit flow control communication
between the Runner -> SDK and SDK -> Runner to be developed to support
sub-bundle groupings.
Transports and in memory layouts are related but improving our coders to
use in memory layouts would give us most of the benefit. For
For the latter, can we have the Fn API data plane transmit sub-bundle
groupings to benefit from the memory layout? On input the runner controls,
on output the SDK controls (spilling)? Just random thoughts.
Kenn
On Thu, May 31, 2018 at 8:21 AM Lukasz Cwik wrote:
> Tyler and I had reached out to
Tyler and I had reached out to Arrow folks[1] asking about how could we
support the KV> when the iterable of values is beyond
memory size limits. There is an open JIRA about adding support for large
byte[] and strings and list types in ARROW-750[2]. Robert had pointed out
that we could do the same
I've looked at arrow, and there's some trickiness. Beam has a record model
and arrow works best with large batches of records. We could do per record
encoding, but that might be inefficient in arrow.
On Thu, May 31, 2018, 5:50 PM Ismaël Mejía wrote:
> If I understand correctly Arrow allows a
If I understand correctly Arrow allows a common multi language
in-memory data representation, so basically it is a columnar data
format that you can use to transfer data betweeen libraries in python
(pandas, numpy, etc), Java and other languages. This avoids the
round-trip to disk to do so. So we
On Wed, May 30, 2018 at 4:43 PM Lukasz Cwik wrote:
> For Python Parquet support, hopefully we can have cross language pipelines
> solve this so we only need to implement it once. If it is really popular,
> having it implemented more then once may be worthwhile.
>
I'd say Parquet format is
For Python Parquet support, hopefully we can have cross language pipelines
solve this so we only need to implement it once. If it is really popular,
having it implemented more then once may be worthwhile.
Would the point of Arrow be to treat it as an IO connector similar to
ParquetIO or JdbcIO (I
I can see great use cases with s3/Parquet - so that's a great addition
(which JB is addressing, for Java)!
It would be even more ideal for the use cases I find myself around for
there to be python parquet support, so for perhaps this next release:
Would it make sense to be exploring:
10 matches
Mail list logo