Crowd JVM exceeding XMX when performing Crowd db to HipChat db sync

Summary

When embedded Crowd sync's user data from the Crowd db to HipChat db, the process spawns thousands of threads and exceeds the allotted JVM XMX setting of 1024MB. This is only triggered when an excessive amount of users are brought into Crowd per the directory integration specifically sync'ing users from an external Crowd/JIRA server instance.

Environment

HipChat Server v2.2.0
Directory integration linked to BTF Crowd/JIRA
Hardware: 4CPU's 8GB RAM

Steps to Reproduce

HipChat Server v2.2.0
JIRA BTF instance with 23k users

1. Setup the directory integration to the Crowd/JIRA instance with 23k users
2. Run the initial sync
3. Watch the 'top' output

Expected Results

The sync runs and users are brought over (or denied due to license constraints)

Actual Results

The sync overcommits memory to the Crowd JVM, eventually eating up all system memory and causing a crash.

Notes

This is only triggered when there are several thousand users over the 5k limit getting sync'ed over in a machine that has limited RAM capactity (tested with ~8GB, but it's possible that systems with more may experience the same issue).

Workaround

The only workarounds that I know of bypass Crowd/JIRA as a user management system:

A.) Use SAML (only available in v2.2.0+)
B.) Use LDAP (or AD) with LDAP filtering - Configuring User Directories
C.) Use email invite link to invite only users you want in HCS (they will not be directory linked, but will have access to HC)