Assuming there's a large number of unique messageIDs this is one of those tricky problems to do for any distributed data store.
You'll likely need to maintain an entity-centric index keyed on the message ID rather than attempting this analysis on a purely log-centric index.
Here's a link to why entity centric indexes are sometimes required. It includes some example scripts to build an entity-centric index but we also now have the dataframes feature in 7.2 which can also fuse related data around an ID.