LeaderInitiatedRecoveryThread looks at a random replica to figure out if it should re-publish LIR state to "down". It does however publish the LIR state for the correct replica.
The bug has always been there. The thread used ZkStateReader.getReplicaProps method with the coreName to find the correct replica. However, the coreName parameter in getReplicaProps was un-used and I removed it in
SOLR-6240 but I didn't find and fix this bug then.
The possible side-effects of this bug would be that we may be republish LIR state multiple times and/or in rare cases, cause double 'requestrecovery' to be executed on a replica.