Iterate over Blocks

You can manually apply a function to each block by iterating over the blocks in the array.

import dask.array as da

arr = da.random.random(size=(1_000, 1_000), chunks=(250, 500))

result = 0
for block in arr.blocks:
    result += block.sum()
    
result.visualize()

For dataframes this looks very similar. Here we iterate over partitions in a list comprehension.

import dask

ddf = dask.datasets.timeseries()

result = sum(partition.groupby("name").sum() for partition in ddf.partitions)
result.visualize()

result.visualize(optimize_graph=True)

Potentially better graph

Another approach to this problem would be to add neighbors together repeatedly until you only have one left.

outputs = [partition.groupby("name").sum() for partition in ddf.partitions]
    
while len(outputs) > 1:
    start = len(outputs) % 2  # 1 if odd, 0 if even
    outputs = [
        *outputs[:start],
        *[outputs[i] + outputs[i + 1] for i in range(start, len(outputs), 2)]
    ]

output = outputs[0]
output.visualize()

That graph optimizes to the same as our original.

output.visualize(optimize_graph=True)