If container is restarted on the node (docker restart <id>) fluentd.conf is not built from scratch, configuration is duplicated (appended once more), with multiple restarts it may lead to significant log multiplication.
I think it is this line fault (and any other using cat << EOF >>).
I think we could start with empty file on each fluentd boot, any objections? or better ideas how to handle this? If not I can prepare PR.