Merge Replication slow reinitialization

  • I have a Merge Replication where I had to reinitialize one of the subscribers. The network connection to this subscriber is slow but the reinitialization has taken less than an hour in the past. This time the Reinit has been processing for 24 hours and it is still not complete.

    One of the tables is very large - the bcp file is over 2G. sp_who2 on the subscriber shows the spid as SUSPENDED on the BULK INSERT for this very large table. The DiskIO and CPU time is very slowly going up. sys.dm_os_waiting_tasks shows the wait type ASYNC_NETWORK_IO. The wait time goes up to large numbers (up to at least 26000 sometimes) then will go down again.

    I assume that this means it is processing the file but extremely slowly. Is this a correct assumption?

    The NIC card on the subscriber says it is half-duplex, the publisher says auto detect. If I change these settings to full-duplex mid-stream will it restart the reinitialization?

    I have read that ASYNC_NETWORK_IO means that the client is not processing the records fast enough. Who is the client in this case? I assume it is perhaps another process on the subscriber SQL Server but is the client the publisher?

    My customer has been tolerant up til now. I've got him running on a backup copy of the database and will switch him over but I need to get this done as soon as possible. Any help would be greatly appreciated.

  • Jennifer, any luck, I am having a similar problem. When I initialize to another Database in the same SQL Server instance, it's very fast. But I have an initialization to a remote server that has been running for more than 24 hours. The agent is still in progress, I finally see that data at the remore server, but not the rowguid that's used for synchronization, and cannot do any inserts or updates on the table.

  • Yes. I also posted the question to another forum.

    You can see the whole conversation at

    http://social.technet.microsoft.com/Forums/en-US/sqlreplication/thread/e0bc86cc-2305-4921-933f-af49257d2b81?prof=required

    But basically the answer is to get the snapshot files sent (we used snail mail in my case) over to the subscriber server and add

    -AltSnapshotFolder //Server/pathname

    to the merge job, then stop and restart the synchronization. The snapshot bcp files should be in the same subdirectory structure as the snapshot folder set on your publisher.

Viewing 3 posts - 1 through 2 (of 2 total)

You must be logged in to reply to this topic. Login to reply