If a work item is being executed and the workflow runtime system tries to access the work item then it will not be successful due to the lock or enqueue currently on the work item.
An asynchronous task is being executed by a user. While the work item is being executed the terminating event is raised in the system and tries to set the work item status to COMPLETE but cannot due to the lock/enqueue.In this case the event is buffered in the event queue. As soon as the work item is released, it will deliver the buffered event immediately.
You use parallel processing where you have a fork with 2 branches (1 Branch necessary for completion). In one branch is a dialog activity step and in the other is a 'Wait for Event' step. While the dialog step is being executed by a user the Wait for Event step receives its event and continues along the branch and completes the fork (Remember only 1 branch needed for completion). Once the end of fork is reached the dialog work item should be set to status Logically Deleted but this does not occur due to the lock/enqueue on the work item while the user is executing it. Since a callback is essential for a workflow to continue running, this callback is suspended (stored in the SWP_SUSPEN table). These callbacks are started again via the RSWWERRE report. If you have not scheduled report RSWWERRE then the work items will remain in table SWP_SUSPEN and the work items will not receive their callback and will therefor will not continue.
If there are examples of work items hanging then check table SWP_SUSPEN to see if the callback work item ID is there. If it is then make sure you have the RSWWERRE job running in order to redeliver it. If RSWWERRE is running and the entry is not being delivered then please check for notes using the search term "RSWWERRE" and "SWP_SUSPEN". If there is no entry in SWP_SUSPEN, then check the workflow definition to see if the work item is asynchronous i.e. needs a terminating event as in Example 1 above. Check the event queue via transaction SWEQADM to see if the terminating event is being buffered there. If it is then it should automatically be redelivered so do a notes search in relation to the event queue.