The problem was a race the forceSync() assert.soon, whereby the node would not have a chance to begin syncing before being instructed to change sync source again. This failure was only observed on linux32 and linux32-debug since they have slower builders. The fix involves checking for replication progress in an assert.soon inside the original assert.soon before retrying.

Here are the failures. It seems to be some sort of startup failure. It is likely 32-bit specific given this set of failures.

http://mci.10gen.com/ui/task/mongodb_mongo_master_linux_32_43d2ae25b1872273cb227ada251315cbaf817534_14_06_06_15_55_07_replicasets_linux_32
http://buildlogs.mongodb.org/build/5391f913d2a60f4011000429/test/5391ff19d2a60f4566000838/

http://mci.10gen.com/ui/task/mongodb_mongo_master_linux_32_debug_adcf6601f49b8afbe5b0c2b23d0f83ceeeca1fa1_14_06_05_20_14_08_replicasets_linux_32_debug
http://buildlogs.mongodb.org/build/5390dcb1d2a60f3104000245/test/5390e483d2a60f3481000af7/