I finished migrating my wiki site from CentOS to CoreOS yesterday. After shutdown a CentOS instance, it is discovered some problems. It seems that a docker-gen unit and a BitTorrent Sync unit have not been in proper working condition.
May be a central etcd registry problems
I make a etcd be a central registry intentionally, but sometimes a docker host ps states and fleet unit states are in correct.
BitTorrent synched files permission problems
I make a root shared directory be owned by root user, and set GUID bit. Whenever a systemd timer kicks a sync job between data volume containers, the shared subsidiary directories ownership are reset to root user.
docker-gen and systemd dependency chain problems
In my fleet unit files relationship, a docker-gen unit shoud be start the second. A docker-gen container suddenly dies the reason why is not known. Then succeeding units all fail.
Workaround
As a workaround, I stopped a BitTorrent Sync data volume container and systemd timer on IDCF Cloud. After A MoinMoin unit started successfully and a docker-gen generated a nginx.conf, I stopped a docker-gen container on Vultr with fleet. In addition I removed a die-on-term frog from uWSGI start command. We have a long way to go.