Add the Serval API and controller for the non-cluster host gateway - #5067
Draft
hjiawei wants to merge 1 commit into
Draft
Add the Serval API and controller for the non-cluster host gateway#5067hjiawei wants to merge 1 commit into
hjiawei wants to merge 1 commit into
Conversation
hjiawei
force-pushed
the
serval-non-cluster-host-gateway
branch
5 times, most recently
from
July 26, 2026 06:07
bce1311 to
c042677
Compare
Render serval, the single HTTPS entrypoint for non-cluster hosts, from the existing NonClusterHost CR. spec.typhaEndpoint is deprecated: unset selects the tunnel and the nonclusterhost controller renders serval; set keeps today's direct-to-typha path and renders calico-typha-noncluster-host. Exactly one exists at a time, both owned by the same CR, so flipping the field is an ordinary reconcile. There is no Serval CR and no separate controller. Serval runs Typha in-process, sourced from the in-cluster calico-typha, so it needs a Typha client keypair (serval-typha-client-certs, common name typha-client) and no datastore RBAC. Its egress rule to calico-typha matches the Service rather than an endpoint selector, because a plain selector does not match ClusterIP-DNAT'd traffic; a service selector implies the Service's ports, so Ports must not be set alongside it. Serval carries the per-host fan-out, so the HostEndpoint-count autoscaler drives its replica count, as it did for calico-typha-noncluster-host. The in-cluster calico-typha keeps scaling on nodes alone: each serval replica is a single client of it however many hosts it serves. The HostEndpoint count now includes only the endpoints the node init labels as non-cluster hosts, so auto and in-cluster host endpoints no longer inflate it. The non-cluster-host Typha policy is deleted alongside the other v3 policies rather than with that deployment and service. A v3 resource is only reconcilable while the API server is healthy, so deleting it from the Typha component made every reconcile fail on a cluster that has no Calico API server -- including the FV environment, where it stalled unrelated components behind a degraded installation. Also refresh the imported FelixConfiguration CRDs, which had drifted from calico master (nftables flowtable offload fields). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
hjiawei
force-pushed
the
serval-non-cluster-host-gateway
branch
from
July 27, 2026 19:31
c042677 to
07751de
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
New feature. Adds the
ServalAPI (operator.tigera.io/v1) and controller. Serval is a layer-7 gateway that becomes the single entrypoint for non-cluster hosts: one HTTPS endpoint serving the impersonating kube-apiserver proxy, the log ingestion routes, and a WebSocket tunnel that carries the felix-typha sync connection. It supersedes the deprecatedNonClusterHostresource;NonClusterHostkeeps working unchanged as a legacy mode until a later removal release.The controller renders the gateway Deployment (combined calico image,
controlPlaneReplicaswith pod anti-affinity,enableServiceLinks: false), a ClusterIP Service, an operator-issued TLS keypair including the external endpoint SAN, RBAC (impersonation, TokenReview/SubjectAccessReview create,geton serviceaccounts for the Tigera-JWT authenticator), and allow-tigera network policies. The fluent-bit input policy admits serval as a source, and the host ClusterRole gains read onservals.Testing: render unit tests, plus end-to-end validation of the rendered object shapes on a kubeadm Enterprise cluster with a Rocky 9 non-cluster host (felix in sync through the tunnel, apiserver proxy serving the cert-init and token-minting flows, logs flowing to linseed).
Companion PR (calico-private): https://github.com/tigera/calico-private/pull/12765
Design: tigera/designs
2026/serval-non-cluster-host-gatewayRelease Note
🤖 Generated with Claude Code