The LISTALIASES command might return a stale response due to the default eventual consistency of reads of ZooKeeper. I think if someone calls this command (which generally won't be rapid-fire), they deserve an up to date response. This is easily done with a one-liner; patch forthcoming.
Returning stale alias info is the only plausible explanation I have for why a recent CI failure for AliasesIntegrationTest.tearDown() failed to detect aliases to be deleted. It calls listAliases to know which aliases exist so it can then delete them 1st.
tearDown then calls MiniSolrCloudCluster.deleteAllCollections() which interestingly grabs a ZkStateReader.createClusterStateWatchersAndUpdate().... perhaps this ought to delete all aliases as well since, after all, if there were any aliases then well deleting all collections is bound to fail. Should I file a separate issue or just handle this together?