The following is a loose collection of potential features that fit the future scope of the operator. Their exact implementation and viability are subject to further discussion.
The Operator creates an empty ConfigMap of the name <prometheus-name>-rules
if it
doesn't exist yet. It is left to the user to populate it with the desired rules.
It is still up for discussion whether it should be possible to include rule files living
in arbitrary ConfigMaps by their labels.
Intuitively, it seems fitting to define in each ServiceMonitor
which rule files (based
label selections over ConfigMaps) should be deployed with it.
However, rules act upon all metrics in a Prometheus server. Hence, defining the
relationship in each ServiceMonitor
may cause undesired interference.
In the future, the Prometheus Operator should register new Prometheus setups it brought up as data sources in potential Grafana deployments.
Prometheus has basic capabilities to run horizontally sharded setups. This is only necessary in the largest of clusters. The Operator is an ideal candidate to manage the sharding process and make it appear seamless to the user.
Prometheus supports federation patterns, which have to be setup manually. Direct support in the operator is a desirable feature that could potentially tightly integrate with Kubernetes cluster federation to minimize user-defined configuration.
The Operator expects a ConfigMap
of the name <alertmanager-name>
which
contains the configuration for the Alertmanager instances to run. It is left to
the user to populate it with the desired configuration. Note, that the
Alertmanager pods will stay in a Pending
state as long as the ConfigMap
does not exist.
The Alertmanager, in high availability mode, is a distributed system. A desired deployment ensures no data loss and zero downtime while performing a deployment. Zero downtime is simply done as the Alertmanager is running high availability mode. No data loss is achieved by using PVCs and attaching the same volumes a previous Alertmanager instance had to a new instance. The hard part, however, is knowing whether a new instance is healthy or not.
A healthy node would be one that has joined the existing mesh network and has been communicated the state that it missed while that particular instance was down for the upgrade.
Currently there is no way to tell whether an Alertmanager instance is healthy under the above conditions. There are discussions of using vector clocks to resolve merges in the above mentioned situation, and ensure on a best effort basis that joining the network was successful.
Note that single instance Alertmanager setups will therefore not have zero downtime on deployments.
The current implementation of rolling deployments simply decides based on the Pod state whether an instance is considered healthy. This mechanism may be part of an implementation with the characteristics that are mentioned above.