If you are publishing through custom workflow step make sure to take care of cluster awarness.
Below steps are must & to be honest need a call to access the situation. Please file official support request.
#1 Install the hotfixes NPR-3728 and NPR-3729 (to upgrade sling.event and workflow bundles) from ticket and http://helpx.adobe.com/experience-manager/kb/cq561-available-hotfixes.html
#2 Disable job distribution and increase topology heart beat interval
#3 Reorder the sling ids in the topology so the master id is ordered first
#4 Clean up all workflows and jobs (as the data previous to the hot fixes might be corrupt)
- Go to http://host:port/crx/explorer
- Login as admin
- Browse to /var/eventing
- Move all subnodes of /var/eventing under /tmp for later deletion. (this includes jobs, distribution, timed-jobs, etc.)
- Save All
- Restart the CQ cluster (stop slaves, restart master, then start slaves)
7.Go to http://host:port/workflow to go to the workflow console
- Terminate all running workflows
- Validate whether or not everything is working correctly.
- Assuming everything is now working, go back to http://host:port/crx/explorer and login as admin
- Browse to /tmp and and right click the jobs node then select "Delete Recursively"
- Disable "preliminary scan" and run (this will delete the bad jobs)
- Repeat step 11 and 12 for each of the nodes moved under /tmp in step 4.