Resolution: Cannot Reproduce
Affects Version/s: OpenIDM 22.214.171.124, 6.5.0, OpenIDM 126.96.36.199, OpenIDM 188.8.131.52, OpenIDM 184.108.40.206, OpenIDM 220.127.116.11, 18.104.22.168, 7.0.0, 22.214.171.124, 126.96.36.199
Fix Version/s: 7.0.0
Intermittently, about once a month for a test which runs daily, clustered recon will exhibit strange behavior. The test runs as follows:
- 100,000 users are reconciled in from DS into a bare IDM (sync with ldap bidirectional)
- the phone number of the 100,000 users is updated in DS
- another reconciliation from ldap->managed is triggered
The error manifests itself with the following summary:
The initial, 'create' recon:
Reconciliation completed. SOURCE_IGNORED: 0 FOUND_ALREADY_LINKED: 0 UNQUALIFIED: 0 ABSENT: 100000 TARGET_IGNORED: 0 MISSING: 0 ALL_GONE: 0 UNASSIGNED: 0 AMBIGUOUS: 0 CONFIRMED: 0 LINK_ONLY: 0 SOURCE_MISSING: 0 FOUND: 0
The subsequent, 'update' recon:
INFO: Reconciliation completed. SOURCE_IGNORED: 0 FOUND_ALREADY_LINKED: 0 UNQUALIFIED: 0 ABSENT: 1 TARGET_IGNORED: 0 MISSING: 0 ALL_GONE: 0 UNASSIGNED: 1 AMBIGUOUS: 0 CONFIRMED: 99999 LINK_ONLY: 0 SOURCE_MISSING: 0 FOUND: 0
The logs are clear of error when this occurs.
There is no correlation query in the sync-with-ldap-bidirectional sample. Thus, the presence/absence of link state determines the distinction between the ABSENT/CONFIRMED(UNASSIGNED) state. In other words, in the initial recon, if no link can be found for the ldap source object, the situation is ABSENT, and the object is created in managed. In the subsequent recon, if the link cannot be found, the situation will be ABSENT for the source phase, which correlates to UNASSIGNED for the target phase. In other words, the fact that 99999 records were CONFIRMED, and 1 was ABSENT and 1 was UNASSIGNED does not indicate that 100001 records were reconciled, but rather a missing link record will result in the ABSENT situation in the source phase, and an UNASSIGNED situation in the target phase.
See https://ea.forgerock.com/docs/idm/integrators-guide/index.html#source-reconciliation for a helpful table of situations.
An open mystery is why there are no error logs, as, in the 'update' recon, the ABSENT situation should result in a create attempt, which should be rejected due to a policy violation (duplicate userName).
Initial investigations revealed several memory visibility issues, prompting the filing of this JIRA. Note that direct causality between these memory visibility issues, and the state characterizing this intermittent failure, has not been established.