when the cert of a stopped app gets auto-cleaned up, nginx does not
start anymore since the config references the cert.
there are two possible fixes:
* do not cleanup cert of stopped apps
* remove the nginx config of stopped apps
this implements the second approach
the primary subdomain was previously called 'location'. but the alias/secondary/redirect
subdomain is called 'subdomain'. this makes it all consistent.
location terminology is now used for { subdomain, domain } pair
this adds and waits for AAAA records based on setting. we have to wait
for both A and AAAA because we don't know if the user is accessing via
IPv4 or IPv6. For Let's Encrypt, IPv6 is preferred (but not sure if it
retries if IPv6 is unreachable).
part of #264
the notification wasn't working because this was in apptask and the apptask died
before it could send out the email. we now move the notification to box process
and also remove the email notification.
Currently, the export route only creates the snapshot (the other side
of in-place import). In the future, the export route can export to a
custom backup config (like import).
We removed httpPort with the assumption that docker allocated IPs
and kept them as long as the container is around. This turned out
to be not true because the IP changes on even container restart.
So we now allocate IPs statically. The iprange makes sure we don't
overlap with addons and other CI app or JupyterHub apps.
https://github.com/moby/moby/issues/6743https://github.com/moby/moby/pull/19001
we can just use container IP instead of all this httpPort exporting magic.
this is also required for exposing httpPaths feature (we have to otherwise
have multiple httpPorts).
BindsTo will kill all the tasks when systemctl stop box is executed.
But when restarted, it keeps the tasks running. Because of this behavior,
we kill the tasks on startup and stop of the box code.
