Where Merge Replication actually stores data?

Question

I've read some non-official articles that say merge replication actually stores data as transactions and replicate them as transactions. But no official info about that on MSDN or any kind of official sources I could reach.

Please, can someone make it clear for me? If it stores transactions - is there any link in MSMerge_contents to the corresponding transaction? How come that it filters inserts/updates/deletes as it said here?

If it does not replicate transactions, where it stores actual data for replication?

In my own opinion Merge Replication does not use transaction log to store data.

Table triggers (MSmerge triggers) transforms each transaction in incremental generation information and stores that information in system metadata tables MSmerge_contents, MSmerge_tombstone and MSmerge_genhistory where you can join each table by tablenick and use rowguid column to find row (for MSmerge_contents and MSmerge_tombstone).

Replication Agent compares contents of MSmerge_contents for publisher and subscriber, it copy new rows and change existing depending on rowguid column and generation number of each row. It uses actual rows in actual tables, joining by rowguid. Same thing for MSmerge_tombstone table.

It does not use transaction log. It also does not even have Log Reader active.

Answer 1

Merge Replication or for that matter any type of the replication always starts with the initial step of Snapshot initialisation, A snapshot is created on the publisher, of all the data and the objects and sent to all the subscribers. (In this step the actual data is moving from Publisher to Subscriber(s)).

Transactional Replication

Once the initial Snapshot has been delivered to subscribers, for transactional replication SQL Server read Transaction logs from the Publisher and push them to Distributor and Distributor send (or the subscriber pulls, depending on what type of subscriber it is) logs to all subscribers. These logs are replayed on the subscribers via triggers and some metadata tables.

Merge Replication

Once the initial Snapshot has been delivered to the subscribers in a Merge Replications, The SQL Server starts reading Transaction logs from the Publisher and from all the Subscribers and send them to the Merge Agent (this agent is only specific to Merge Replication, it does not exist in any other type of replication), the Merge Agent uses a specific algorithm to sort out the order in which Inserts/updates/deletes are applied (Usually Deletes are first and then the updates and finally inserts are applied) to all participating Subscribers and Publisher, once the merge agent has sorted out the order, the logs get queued on the distributor like in Transactional replication and gets pushed/pulled to the subscribers.

The data movement only happens in the very first step (Snapshot initialisation), after that its all about Logs and agents moving around and synchronising data. I hope this helps.