That presumably fixed things from a deployment point of view, but if there was a concurrency bug involving a hashmap, the service may have been emitting incorrect results.

For example: calculate hash code of string, determine index, find apparent match, hashmap is modified by another thread, return value at that index which no longer matches.

I don't think that particular issue can happen with Java's HashMap, but there's probably some sort of similar goofiness.