harvest job shows gather errors from previous run

[sabinem]

CKAN Version 2.8.10,
ckanext-harvest version v1.4.0

I have the following problem with my harvesters:

• when I run it on a harvest source url that does not exist, I get a gather error which is correct
• after that I fix the url and rerun the harvester: it still shows the same errors in the job log panel

The previous errors are created at this call:

ckanext-harvest/ckanext/harvest/logic/action/update.py

Line 863 in 2560315

publisher.send({'harvest_job_id': job['id']})

Before the call the harvester job is there, but it does not have any errors.
After the call the errors have been stored for the harvester, despite the fact that the gather method of the harvester has not even started yet.

This happened both for the run_test paster command but also when I start the harvester via the WebUI.

I suspect that it has something to do with redis. Is there any hint on what might cause the problem and how to prevent it?