For a large set of UMWLs the unused-umwl function of workloader will fail with an EOF while doing a GET API call
This can happen anywhere from 5 hours to 10 hours into a set of processing, usually between 2100 and 2600 UMWLs processed.
Normal sequence for the API calls is PUT / GET / GET / GET with download
The failure happens after the first GET which completes successfully. The second GET does not reach the PCE.
This happens with either direct calls to the PCE or when there is Zscaler in the middle.
One possibility is the connection pool either exhausting or attempting to reuse a now closed connection.