JIRA imports are slow, and a major contributor to this is that multiple (3 or more) re-indexes occur during the import. One re-index is triggered by the standard JIRA import process that the OnDemand import process wraps, then, as described in JRA-31473, several re-indexes are triggered by upgrade tasks.

All of these re-indexes are ultimately pointless, because a re-index is triggered explicitly at the end of the OnDemand import anyway.

We should look into the feasibility of disabling all re-indexing at the beginning of the OnDemand import process (so that re-index request become no-ops), and re-enabling it only at the very end of the process for the final request.