Idle replication change log consumes CPU and disk IO

On an idle server the change number indexer thread wakes up regularly to check the changelog head.

This can be observed with strace -p <change number indexer thread>

Attaching a debugger it seems that it's being woken up when heartbeat messages are received. In a large topology this could be noticeable number of times per second and leads to a small constant usage of resources.

This is not a significant concern, since I believe as load increases, wake-ups due to heartbeat messages should become irrelevant; the indexer will be active anyway due to real changes coming in.