We have a scaled-out web application architecture with multiple servers in a load-balance. These nodes share a common broker database. Since upgrading to 2011 (currently SP1), When a file-system asset is unpublished, this succeeds on the first node, but not on the others. Our current thinking is that this first undeployment alters the state of the broker database in a way that prevents undeployments of the same asset from other nodes from taking place.
Obvious solutions would be to give each node its own database, or to arrange for a single deployment destination with further replications placed out of the responsibility of Tridion. Neither of these is attractive, as the system is already deployed to production, and large infrastructure changes would be difficult.
Is this a known behaviour? Does our current thinking make sense? Can you suggest a solution or workaround that would match these needs?