Write bottlenecks

Hi,

My write pipelines have a bottleneck at the bottom (LMDB writing). Work piles on top up-front, waiting for LMDB writes to complete.

Therefore the earlier work cannot really parallelize as much as it could if I could somehow make the writes go faster.

Are there any tips/tricks around this? I wonder if it could pay off to parallelize the writes into a few LMDB environments - and merge them in the end. This could have the advantage of more parallelization on the up-front work, but disadvantage of the final merging work.

Regards, Sam