file mounting is fraught with problems wrt change notifications.
first, we must be carefule that the inode does not change.
second, changes outside container do not result in fs events inside the container.
haraka cache settings files and relies on fs events. So, even
though the file gets updated inside the container, haraka doesn't
see it.
https://github.com/docker/docker/issues/15793
Because of the docker upgrade, dnsbl queries are failing again
since we are not using the unbound server from the containers.
For some reason, docker cannot query 127.0.0.1 (https://github.com/docker/docker/issues/14627).
Make unbound listed on the cloudron network and let docker proxy
DNS calls to unbound (docker always use the embedded DNS server
when using UDN).
See also #130
For reasons unknown, the images build by the buildbot (which currently
uses btrfs), does not work with devicemapper.
Existing cloudrons with aufs will not be affected because docker will
just ignore it.
devmapper: Base device already exists and has filesystem xfs on it. User specified filesystem will be ignored.
Existing AUFS users can move to devicemapper either by restoring to
a new cloudron (recommended) OR
* systemctl stop box
* systemctl stop docker
* rm -rf /var/lib/docker
* Edit /home/yellowtent/data/INFRA_VERSION. Change the "version" field to "1"
* systemctl start docker
* systemctl start box # this will download images all over
Fixes#182
The default swappiness is 60 on ubuntu. This sets the tendency
to swap out memory pages to be more frequent, which in our case
means swapping out to networked disks and increase the cpu load
a lot, which is especially bad on EC2
This is done because:
1. The box code can install certs for addons (like mail addon) when
required.
2. The box code initialize/teardown addons on demand. This is not planned
currently.
