王昊
05/22/2023, 8:47 AMs3_pickle_io_manager
, yielding DynamicOutput
is slow, it takes about one second to yield one. Even if I push the DynamicOutput
to a list and return it all at once, it is still like this. My s3 service is a local s3 service built with MinIO, so there is no network problem. Is there a way to optimize it?owen
05/22/2023, 7:57 PMs3_pickle_io_manager
is not particularly complex (basically just calling this function for each output). So my initial guess is that this is likely bumping into some bottleneck either in the pickling step or the s3 upload step (or both), rather than something in the Dagster machinery. How large are your output objects? It might also be worthwhile benchmarking your local s3 service in isolation to get a better idea of what performance you could expect王昊
05/23/2023, 3:08 AMowen
05/23/2023, 3:59 PM