Skip to content

docs: improve listener documentation #318

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 7 commits into
base: main
Choose a base branch
from
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,6 @@
apiVersion: listeners.stackable.tech/v1alpha1
kind: ListenerClass
metadata:
name: public
name: external-stable
spec:
serviceType: LoadBalancer
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,6 @@
apiVersion: listeners.stackable.tech/v1alpha1
kind: ListenerClass
metadata:
name: public
name: external-stable
spec:
serviceType: NodePort
96 changes: 82 additions & 14 deletions docs/modules/listener-operator/pages/listener.adoc
Original file line number Diff line number Diff line change
@@ -1,31 +1,99 @@
= Listener
:description: The Listener exposes Pods based on ListenerClass rules, provides address info via Ingress_addresses, supports PVC mounting, and enables sticky scheduling.

A Listener object exposes a set of Pods according to the rules of a xref:listenerclass.adoc[], but it also adds a couple of other features that are useful for the Stackable data platform at large.
:fn-kubernetes-service: footnote:[It is actually implemented using them, but don't rely on that.]

== ListenerClass
A Listener object represents a single exposed (possibly) load-balanced service that clients can connect to.
It can be thought of as the Stackable Data Platform equivalent of a Kubernetes https://kubernetes.io/docs/concepts/services-networking/service/[Service,window=_blank].{fn-kubernetes-service}

The exact rules of pod exposure are dictated by the specified xref:listenerclass.adoc[], which allow a single Listener definition to be reused in different clusters, regardless of the Kubernetes distribution or cloud provider.
The mechanism for the service is controlled by the xref:listenerclass.adoc[].
This way, a single Listener definition can be reused in different clusters, expressing the same _intent_ regardless of the Kubernetes distribution or cloud provider's limitations.

Listeners only direct traffic to Pods that also mount them as a xref:volume.adoc[volume].
The volume allows the operator to xref:#pinning[pin] the Pod to a specific Node, and provides an API for workloads to retrieve their external address.

[#address-api]
== Address API

A Listener writes back all addresses that it can be reached on to `Listener.status.ingress_addresses`, which can then be used to generate discovery information.
Contrary to Kubernetes' Service, this is done regardless of the type of service, and transparently also contains information about remapped ports.
NOTE: The CRD-based API is intended for external clients that need to retrieve the address.
The workload can retrieve _its own_ address(es) by using the xref:volume.adoc#downwards-api[downwards API].

== Address volume projection
A Listener writes back all addresses that it can be reached on to `Listener.status.ingressAddresses`, which can then be used to connect to the service (generate discovery information).
Compared to Kubernetes' Services, this list is provided _regardless_ of the type of the backing Service.

Listener objects can be mounted into a Pod as a PersistentVolumeClaim (PVC), which contains information about how the Pod should request that external clients refer to it.
NOTE: All addresses may not be able to reach all backing Pods. Clients should connect to a _random_ address in the list, not just the first one.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is not clear to me: does the address list act like a quorum, or not? Does that mean that I can pick a random address, that might not be able to reach all backing pods?


For example, if the volume is mounted to `/stackable/listener`, the primary address can be read from `/stackable/listener/default-address/address`, and the public `http` port number can be read from `/stackable/listener/default-address/ports/http`.
Ports may be remapped from the Service definition.
Never assume that the exposed port on an address will match your declared port.
Instead, read the port numbers from `.ports.\{portname\}`.
Otherwise, it will break when using NodePort services.

== Per-replica listeners

A Listener PVC can also specify a xref:listenerclass.adoc[] rather than a Listener, in which case a Listener object is created automatically.
These PVCs can automatically be created for each replica using either StatefulSet's `volumeClaimTemplates` (for long-lived listeners that will be kept across replica restarts and upgrades) or Pod's `volumes[].ephemeral` (for temporary listeners that are deleted when their corresponding Pod is deleted).
A Listener volume can also specify a xref:listenerclass.adoc[] rather than a Listener, in which case a Listener object is created automatically for each volume.

These volumes, in turn, can automatically be created for each replica using either:

- StatefulSet's `volumeClaimTemplates` (for long-lived listeners that will be kept across replica restarts and upgrades), or
- Pod's `volumes[].ephemeral` (for temporary listeners that are deleted when their corresponding Pod is deleted)

[#pinning]
== Pinning

When mounting a Listener volume, it will be "pinned" to that node if the xref:listenerclass.adoc[] uses a strategy that depends on the node that the workload is running on.

Keep in mind that this will only work correctly when using long-lived volumes (such as via StatefulSet's `volumeClaimTemplates`).
Ephemeral volumes will be "reset" for every pod that is created, even if they refer to a long-lived Listener object.

== Sticky scheduling
[#reference]
== Reference

When mounting a Listener PVC, it will be made "sticky" to that node if the xref:listenerclass.adoc[] uses a strategy that depends on the node that the workload is running on.
[source,yaml]
----
apiVersion: listeners.stackable.tech/v1alpha1
kind: Listener
metadata:
name: my-listener
spec:
className: external-unstable
ports:
- name: http
port: 9864
protocol: TCP
extraPodSelectorLabels:
foo: bar
publishNotReadyAddresses: true
status:
ingressAddresses:
- address: 172.18.0.3
addressType: IP
ports:
http: 32222
nodePorts:
http: 32222
serviceName: my-listener
----

Keep in mind that this will only work correctly when using long-lived PVCs (such as via StatefulSet's `volumeClaimTemplates`).
Ephemeral PVCs will be "reset" for every pod that is created, even if they refer to a long-lived Listener object.
`spec.className`:: The name of the xref:listenerclass.adoc[] to use.
`spec.ports`:: The ports exposed from the backing Pods.
`spec.ports.name`:: The name of the port.
`spec.ports.port`:: The number of the port.
This must match the port number exposed by the container.
`spec.ports.protocol`:: The IP protocol (TCP/UDP/SCTP).
Defaults to TCP.
`spec.extraPodSelectorLabels`:: Traffic will only be forwarded to Pods that apply these labels.
This field exists for exceptional cases, where Pods sometimes want to stop receiving traffic based on some dynamic condition.
Normal target selection should use xref:volume.adoc[Listener volumes] instead.
(Volumes are still required when using `extraPodSelectorLabels`.)
`spec.publishNotReadyAddresses`:: If false, traffic will only be directed to Pods that are Ready.
If true, traffic will be directed to any running Pod. Defaults to true.
`status.ingressAddresses`:: A list of all addresses that the Listener can be reached on.
See xref:#address-api[].
`status.ingressAddresses.address`:: The hostname or IP address of this Listener.
`status.ingressAddresses.addressType`:: `IP` if `address` is an IP address, `Hostname` if it is a hostname.
`status.ingressAddresses.ports.\{portName\}`:: The _exposed_ port number for a given port name (as defined in `.spec.ports`).
Note that this may be different than the port specified in `.spec.ports.port``.
`status.nodePorts.\{portName\}`:: For internal use only.
You probably want to use `.status.ingressAddresses` instead.
_If_ the ListenerClass is configured to use xref:listenerclass.adoc#servicetype-nodeport[NodePort] then this is the port number that each port is accessible on on its respective Node.
`status.serviceName`:: The name of the Kubernetes Service object backing this Listener.
2 changes: 2 additions & 0 deletions docs/modules/listener-operator/pages/usage.adoc
Original file line number Diff line number Diff line change
Expand Up @@ -25,3 +25,5 @@ Or like this for on-premise environments:
----
include::example$listenerclass-public-onprem.yaml[]
----

These are normally installed by the appropriate xref:listenerclass.adoc#presets[preset].
3 changes: 2 additions & 1 deletion docs/modules/listener-operator/pages/volume.adoc
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,8 @@ When this address must be configured statically in clients (such as for HDFS Nam
Mounting listeners into Pods as PersistentVolume allows the Listener Operator to pin these workloads to one node.
Note that this only happens for xref:listenerclass.adoc[]es that actually benefit from pinning.

== Pod metadata injection
[#downwards-api]
== Downwards API

Some services (such as Kafka) need to know their external address, so that they can advertize it to their own replica discovery mechanism.
xref:listener.adoc[] volumes contain a file tree that exposes this information:
Expand Down