Snap refresh

Charms that comprise Charmed Kubernetes install required software such as etcd, kubelet, and kube-apiserver using snap packages. The snapd daemon periodically scans installed snaps for updates and will automatically refresh upgradeable packages to ensure the software is current.

All charms built from layer-snap include a snapd_refresh configuration option that can be used to adjust the snapd refresh interval. By default, Charmed Kubernetes charms set this option to the maximum amount of time that snapd will allow between scans.

Refresh interval configuration

Display the currently configured snapd_refresh option for a given charm:

juju config <charm> snapd_refresh

Change the snapd_refresh option with the following:

juju config <charm> snapd_refresh=<value>

The value returned or set above should be an explicit timer, an empty string, or the special keyword max.

• Explicit timer

  An explicit timer may be a simple mon (scan every Monday) or a more complex mon3,23:00 (scan on the third Monday of the month at 23:00). See possible values for explicit timers in the the refresh.timer section of the system options documentation.

• Empty string

  An empty string instructs snapd to refresh snap packages according to the default system policy. This is currently 4 times per day.

• max

  When set to max, refresh scans will be delayed for the maximum amount of time allowed by snapd. This is currently once per month based on the date this option was set.

Determine the actual max value

Use snap get on a deployed system to determine the value that snapd uses when a charm is configured with snapd_refresh=max. An example with etcd shows that the max option used when etcd/0 was deployed has mapped to the last Sunday of every month (sun5):

juju config etcd snapd_refresh
max
juju run --unit etcd/0 'snap get system refresh.timer'
sun5

Refresh interval among peers

Charmed Kubernetes applications that support peering will use Juju leadership to configure a consistent refresh interval among peers. The lead unit for etcd, kubernetes-master, and kubernetes-worker applications will set an initial refresh value. Subsequent units that join as followers will use the leader value as their snap refresh interval. This ensures all units in a peer group will refresh at approximately the same time.

Force snapd to refresh installed snaps

If an immediate snap refresh is desired, invoke snap refresh on the applicable cluster components. For example, refresh all snaps on all etcd units with:

juju run --application etcd 'snap refresh'

As another example, only refresh the cdk-addons snap on the kubernetes-master/0 unit with:

juju run --unit kubernetes-master/0 'snap refresh cdk-addons'