Processing Unit remained in SCHEDULED state although there was a container that could host the missing instance. It was observed that after the XAP Manager had relinquished leadership the pending Processing Unit instance was removed from the pending list (as it should) and when the XAP Manager was re-offered leadership it ignored the request due to a mismatch in allocated instance count.
Fix in 12.3.1 makes sure that removing a pending request also aligns the allocated instance ID count.
This issue started appearing after fix which closes the leader-selector and speeds up the re-election process only to be relinquished because of lack of majority. Before this fix, the ephemeral node lease would elapse and only then would a leader be selected. This exposed a bug in the GSM as described above.