This happened in Documentum 5.3 SP3 environment but potentially could happen in others as well.
Start a workflow with a document attached to it. Delete this document before the last activity in the workflow has kicked off. When the workflow agent tries to package this document for an activity transition, it throws an exception. Then it retries repeatedly, in the process opening new sessions. This is evident from the content server log. Depending on the rate at which the open sessions are reclaimed, clients may encounter connection problems.
Since the document is gone, the workflow instance can usually be aborted. The API command for aborting a workflow is:
If you also want to clean up the objects related to this workflow, you can run the
-clean_aborted_wf option. See the section on Using dmclean in Content Server Administrator’s Guide.