[BUG] left join style merge reorders resultant dataframe

[beckernick]

Description

As a user, I expect the resultant dataframe from a left join to preserve the row ordering from my left dataframe. This reordering occurs when the left and right dataframes have all keys in common and when the right dataframe only has a subset of the left's keys.

Example

import cudf
import numpy as np
import pandas as pd
import dask_cudf
import dask.dataframe as dd

## Subset of keys
df_a = cudf.DataFrame()
df_a['key'] = [0, 1, 2, 3, 4]
df_a['vals_a'] = [float(i + 10) for i in range(5)]

df_b = cudf.DataFrame()
df_b['key'] = [1, 2, 4]
df_b['vals_b'] = [float(i+100) for i in range(3)]

ddf_a = dask_cudf.from_cudf(df_a, npartitions=2)
ddf_b = dask_cudf.from_cudf(df_b, npartitions=2)

merged = ddf_a.merge(ddf_b, on=['key'], how='left').compute()
print(merged)
   key  vals_a  vals_b
0    1    11.0   100.0
1    2    12.0   101.0
2    4    14.0   102.0
3    0    10.0        
0    3    13.0   


## All keys in common
df_a = cudf.DataFrame()
df_a['key'] = [0, 1, 2, 3, 4]
df_a['vals_a'] = [float(i + 10) for i in range(5)]

df_b = cudf.DataFrame()
df_b['key'] = [0, 1, 2, 3, 4]
df_b['vals_b'] = [float(i+100) for i in range(5)]

ddf_a = dask_cudf.from_cudf(df_a, npartitions=2)
ddf_b = dask_cudf.from_cudf(df_b, npartitions=2)

merged = ddf_a.merge(ddf_b, on=['key'], how='left').compute()
print(merged)
   key  vals_a  vals_b
0    0    10.0   100.0
1    1    11.0   101.0
2    2    12.0   102.0
3    4    14.0   104.0
0    3    13.0   103.0

[kkraus14]

Duplicate of #1781