When we reprotect after an SRM failover it seems that SRA steps on itself a bit in the process. As noted in the snapmirror_audit snippet below, SRA creates the SnapMirror relationship and attempts to resync (actually it issues an Initialize request first) which fails then immediately retries with a successful resync operation.
The main problem here is that vSphere hangs onto the FAILURE so a rediscover of devices in SRA is required in order to clear the error and make everyone (i.e. our management) happy. I suspect this is a bug?