There is a crash when:
* App is configured. So, oldConfig now has {loc, access, portb }
* Cloudron is restored. The restore code path accesses the oldConfig.manifest.addons.
oldConfig is basically a messaging passing thing. It's not really a
db field. With that spirit, we simply pass an empty message in setup_infra.sh
We explicitly mark /tmp, /run and /var/log as writable volumes.
Docker creates such volumes in it's own volumes directory. Note
that these volumes are separate from host binds (/app/data).
When removing the container the docker created volumes are
removed (but not host binds).
Fixes#196
And abort status checking after the first one fails. Otherwise, this
bombards the appstore unnecessarily. And checks for status of other
things unnecessarily.
Add LDAP_BIND_DN and LDAP_BIND_PASSWORD that allow
apps to bind before a search. There appear to be two kinds of
ldap flows:
1. App simply binds using cn=<username>,$LDAP_USERS_BASE_DN. This
works swimmingly today.
2. App searches the username under a "bind_dn" using some admin
credentials. It takes the result and uses the first dn in the
result as the user dn. It then binds as step 1.
This commit tries to help out the case 2) apps. These apps really
insist on having some credentials for searching.
The backup cron job ensures backups every 4 hours which simply does
a 'box' backup listing. If we do only a 'box' backup during update,
this means that this cron job skips doing a backup and thus the apps
are not backed up.
This results in the janitor on the CaaS side complaining that the
app backups are too old.
Since we don't stop apps anymore during updates, it makes sense
to simply backup everything for updates as well. This is probably
what the user wants anyway.
