It turns out that the "repeated too quickly" simply says that the redis server died when it gets started through systemd. Many test sequences look correct because they miss the crucial part that SystemD is adding: the switch to user:group "redis:redis" before running the command. That makes the redis process to be unable to read /etc/redis/default.conf and it is unable to write to /var/log/redis/default.log because they are still owned by "root".
The docker test scenario works this way: https://github.com/gdraheim/do…ensuse15-redis.dockerfile
ensured: chown redis:redis /etc/redis/default.conf
ensured: chown redis:redis /var/log/redis/default.log