Closed gmweaver closed 2 years ago
Hey @gmweaver, realistically I think 3.1.x versions only. Spark tends to break (spark-sql) bin compat between minor versions.
@gmweaver Yeh, unfortunately, if you want Spark 2.x, you should use RasterFrames < 0.10. Fortunately, there have been no major API changes between RasterFrames 0.9.x and 0.10.x, so your code should work against both.
Got it, thanks for the feedback!
For future releases, is it possible have requirements be flexible within minor versions (i.e. a release that works with both 3.0.0
and 3.0.1
)? Also, is there a release for 3.0.x? These questions are more related to constraints within my own dev environment and the need to use a certain Spark version until we approve/test an upgrade, so I understand that this may not be high priority.
Hey @gmweaver it is theoretically possible but fairly complicates the release process. It is not only a tiny constraint on the Python side, it is also the spark jar dependency. In frameless we resolved it by releasing separate artifacts - so not only cross scala version releases but also cross spark.
Is there a specific dependency on the currently pinned pyspark version or will this work with older versions (I assume latest release depends on at least 3.x.x)?
https://github.com/locationtech/rasterframes/blob/develop/pyrasterframes/src/main/python/setup.py#L143
If it does work with prior versions, would it be possible to update requirements to be >= as opposed to a specific version. This would be very helpful for using with automated build tools that depend on correct dependency resolution.