When a VM is either added to SG or removed from SG, address set changes come down to the host. In this use case, the same SG comes down as a set of IPs for source and destination for the FW rule and as vNICs for where the rule is applied to. Generally, they come as separate messages. But if there are large number of SGs that get changed at the same time (due to nesting in this case), they come down aggregated in the same message. Earlier, the host processed one at a time. With the new aggregation optimization, it picks up multiple messages at a time. If the manager is slow or host is slow, the messages are not aggregated as much and hence chances of this happening is less.