[BUG] databricks nightly window_function_test failures

[tgravescs]

Describe the bug

[2022-03-11T03:51:14.568Z] FAILED ../../src/main/python/window_function_test.py::test_window_running_rank[Decimal(20,2)][IGNORE_ORDER({'local': True})]
[2022-03-11T03:51:14.568Z] FAILED ../../src/main/python/window_function_test.py::test_window_running[Decimal(20,2)-Long-1000][IGNORE_ORDER({'local': True})]
[2022-03-11T03:51:14.568Z] FAILED ../../src/main/python/window_function_test.py::test_window_running[Decimal(20,2)-Long-1g][IGNORE_ORDER({'local': True})]

Perhaps related to #4924?

[2022-03-11T03:51:14.566Z] java.lang.IllegalArgumentException: Part of the plan is not columnar class com.databricks.sql.execution.window.RunningWindowFunctionExec
[2022-03-11T03:51:14.566Z] RunningWindowFunction [b#215344, a#215343L, row_number() windowspecdefinition(b#215344, a#215343L ASC NULLS FIRST, specifiedwindowframe(RowFrame, unboundedpreceding$(), currentrow$())) AS row_num#215352, rank(a#215343L) windowspecdefinition(b#215344, a#215343L ASC NULLS FIRST, specifiedwindowframe(RowFrame, unboundedpreceding$(), currentrow$())) AS rank_val#215353, dense_rank(a#215343L) windowspecdefinition(b#215344, a#215343L ASC NULLS FIRST, specifiedwindowframe(RowFrame, unboundedpreceding$(), currentrow$())) AS dense_rank_val#215354, count(c#215345L) windowspecdefinition(b#215344, a#215343L ASC NULLS FIRST, specifiedwindowframe(RowFrame, unboundedpreceding$(), currentrow$())) AS count_col#215355L, min(c#215345L) windowspecdefinition(b#215344, a#215343L ASC NULLS FIRST, specifiedwindowframe(RowFrame, unboundedpreceding$(), currentrow$())) AS min_col#215356L, max(c#215345L) windowspecdefinition(b#215344, a#215343L ASC NULLS FIRST, specifiedwindowframe(RowFrame, unboundedpreceding$(), currentrow$())) AS max_col#215357L, sum(c#215345L) windowspecdefinition(b#215344, a#215343L ASC NULLS FIRST, specifiedwindowframe(RowFrame, unboundedpreceding$(), currentrow$())) AS sum_col#215358L], [b#215344], [a#215343L ASC NULLS FIRST]
[2022-03-11T03:51:14.567Z] +- GpuColumnarToRow false
[2022-03-11T03:51:14.567Z]    +- GpuSort [b#215344 ASC NULLS FIRST, a#215343L ASC NULLS FIRST], false, com.nvidia.spark.rapids.OutOfCoreSort$@269e4d10
[2022-03-11T03:51:14.567Z]       +- GpuShuffleCoalesce 1073741824
[2022-03-11T03:51:14.567Z]          +- GpuColumnarExchange gpuhashpartitioning(b#215344, 12), true, [id=#264892]
[2022-03-11T03:51:14.567Z]             +- GpuProject [b#215344, a#215343L, c#215345L]
[2022-03-11T03:51:14.567Z]                +- GpuRowToColumnar targetsize(1073741824)
[2022-03-11T03:51:14.567Z]                   +- Scan ExistingRDD[a#215343L,b#215344,c#215345L]

[sameerz]

Since PR #4942 is a workaround, can we lower the priority of this issue?

Is there a long term fix we should track here?

[NVnavkumar]

I have a better fix than this workaround almost done which will track the actual xfail properly.

[tgravescs]

we should not lower the priority as in the very least we should xfail and ideally we fixed Databricks so it supports it properly or if we don't want to somehow indicate in our docs its not supported there.