K8SPG-866 use discovery.k8s.io/v1 EndpointSlice

[gkech]

DESCRIPTION

Problem:
v1 Endpoints is deprecated in v1.33+; use discovery.k8s.io/v1 EndpointSlice

Solution:
Before this PR, the operator created only the -primary Endpoints resource. All other Endpoints resources are created by patroni. Patroni currently does not support creating EndpointSlices in place of the deprecated Endpoints.

This PR adds a feature gate EndpointSlices which replaces the -primary Endpoints resource with an EndpointSlice once version 2.9.0 is reached. As part of the migration, the deprecated Endpoints resource and the mirrored EndpointSlice created by endpointslicemirroring-controller.k8s.io from the deprecated Endpoints resource are deleted.

This PR also addresses the fact that EndpointSlices are not supported on GKE out of the box. The default DNS provider on GKE is kube-dns, which does not support EndpointSlices. To use EndpointSlices, users must manually deploy CoreDNS or use Cloud DNS. To fix tests, every test deploys CoreDNS in the kube-system namespace if it is not already present.

Please note that Endpoints resources are mirrored to EndpointSlices by endpointslicemirroring-controller.k8s.io, which is why the upgrade-consistency test includes assertions for EndpointSlices.

CHECKLIST

Jira

• Is the Jira ticket created and referenced properly?
• Does the Jira ticket have the proper statuses for documentation (Needs Doc) and QA (Needs QA)?
• Does the Jira ticket link to the proper milestone (Fix Version field)?

Tests

• Is an E2E test/test case added for the new feature/change?
• Are unit tests added where appropriate?

Config/Logging/Testability

• Are all needed new/changed options added to default YAML files?
• Are all needed new/changed options added to the Helm Chart?
• Did we add proper logging messages for operator actions?
• Did we ensure compatibility with the previous version or cluster upgrade process?
• Does the change support oldest and newest supported PG version?
• Does the change support oldest and newest supported Kubernetes version?

[egegunes]

This PR also addresses the fact that EndpointSlices are not supported on GKE out of the box. The default DNS provider on GKE is kube-dns, which does not support EndpointSlices. To use EndpointSlices, users must manually deploy CoreDNS or use Cloud DNS. To fix tests, every test deploys CoreDNS in the kube-system namespace if it is not already present.

this worries me. this means we are breaking GKE installations out of the box. and we don't know when will Endpoints support be dropped. maybe we need make using EndpointSlices optional for now by introducing a cr field. wdyt @hors ?

[mayankshah1607]

maybe we need make using EndpointSlices optional for now by introducing a cr field

It looks more like a system-wide preference. Do we imagine a use-case where some clusters use Endpoints and some use EndpointSlices? Maybe it should be a CLI / feature flag?

[egegunes]

maybe we need make using EndpointSlices optional for now by introducing a cr field

It looks more like a system-wide preference. Do we imagine a use-case where some clusters use Endpoints and some use EndpointSlices? Maybe it should be a CLI / feature flag?

yes, you have a point. maybe we need a feature flag

[hors]

maybe we need make using EndpointSlices optional for now by introducing a cr field

It looks more like a system-wide preference. Do we imagine a use-case where some clusters use Endpoints and some use EndpointSlices? Maybe it should be a CLI / feature flag?

yes, you have a point. maybe we need a feature flag

Let's use a feature flag and disable it by default.

[JNKPercona]

Test Name	Result	Time
backup-enable-disable	passed	00:09:36
builtin-extensions	passed	00:05:50
cert-manager-tls	passed	00:05:33
custom-envs	passed	00:19:15
custom-extensions	passed	00:15:41
custom-tls	passed	00:05:53
database-init-sql	passed	00:02:20
demand-backup	passed	00:33:33
demand-backup-offline-snapshot	failure	00:31:08
dynamic-configuration	passed	00:03:30
finalizers	passed	00:03:59
init-deploy	passed	00:02:54
huge-pages	passed	00:03:18
ldap	passed	00:07:16
ldap-tls	passed	00:06:33
monitoring	passed	00:08:17
monitoring-pmm3	passed	00:10:48
one-pod	passed	00:06:24
operator-self-healing	failure	01:30:22
pg-tde	passed	00:10:32
pitr	passed	00:13:00
scaling	passed	00:06:52
scheduled-backup	passed	00:29:58
self-healing	passed	00:08:57
sidecars	passed	00:02:46
standby-pgbackrest	passed	00:13:35
standby-streaming	passed	00:09:42
start-from-backup	passed	00:16:29
tablespaces	passed	00:08:06
telemetry-transfer	passed	00:04:01
upgrade-consistency	passed	00:05:15
upgrade-minor	passed	00:05:51
users	passed	00:04:50

Summary	Value
Tests Run	33/33
Job Duration	02:54:18
Total Test Time	06:52:18

commit: bd37c95
image: perconalab/percona-postgresql-operator:PR-1307-bd37c959e

[gkech]

going to close this pr for now and we can reopen if needed