When receiving calls because the import died, they do not care about a stack trace in a style you like. They care: did any rows write, which ones, and can we press the button again.
I log start, source hash, row counts, the last primary key committed, and a clear rerun: safe or not. I print that in the UI, not only in a file on my laptop.
Correlate with the physical world
A log line with a tracking number beats one with a UUID. Humans search Slack for the tracking number.
Keep logs where the operators are
A folder on a share, a table in Postgres, even an emailed summary. If logs require VPN and a tail command, you will be the pager forever.
