The problem is that sometimes rule branches may execute on another node than the one the rule was originally triggered from. Jira's index replication in datacenter may be lagging behind though, so JQL conditions may not be able to find an issue yet on another node in DC.
An example rule that produces this problem is:
Trigger: Issue created
Related branch with 'Current issue'
The related branch may execute on any node in the DC cluster, but those other nodes indexes may not include the created issue yet, causing the JQL condition to always fail.