Testing results:
It looks, that koji (or 'requests' lib) has its own mechanism of waiting/reconnecting.
When I interrupted a network connection and reconnected it after, it survived and the process of watching continued after some time.
But in the past, it wasn't enough every time. Our method adds extra attempts. Question is, whether these extra re-executed (watch)tasks will significantly help in these cases.

We are able to catch the specific exception and re-execute (watch)task few times. That's it.

better remove.

I prefer "attempts" instead of repetitions. It looks more natural to me.

requests.exceptions.ConnectionError

This is too general. Maybe something like "Couldn't finish the 'watch task' after N attempts. Reason: " ?

Link to the JIRA into the commit message, please.

Reconsider adding some delay (a few seconds) between attempts. Not to execute another attempt right after failing the previous one.

Koji ClientSession already performs retries on any API call. For details see
https://pagure.io/koji/blob/master/f/koji/init.py#_2786 (_callMethod is relevant in case the line number changes in the future).
I'm not sure adding another layer of retries here is a good idea.

It looks, that koji (or 'requests' lib) has its own mechanism of waiting/reconnecting.

that corresponds to my observations. Thanks for notifying us of the specific part in the koji's code.

OK, now I see, that returning a different code in case of koji fail is sufficient for the original request. My bad.

OK, sorry I have not noticed this piece of code in Koji. I will adjust this commit accordingly.

Looks, good.