We're seeing 2 seconds lag before each step is executed. Is there a way to reduce that?

using the

forkserver

start_method

for the multiprocess executor is one way to reduce this overhead https://docs.dagster.io/concepts/ops-jobs-graphs/job-execution#default-job-executor