Operations
Operate the sidecar as part of the application rollout. Track process health, cache behavior, and recoverable state separately.
Pin the operator chart and workload images to the release you intend to run. From the new Swytch checkout root, update the CRD before upgrading Helm:
kubectl apply -f operator/charts/swytch-operator/crds/
helm upgrade swytch ./operator/charts/swytch-operator \
-n swytch-system --reuse-values --wait
Review reused Helm values, especially image pins: --reuse-values retains previous overrides. Set the intended operator
image and swytch.image explicitly when upgrading them.
Server profile changes reconcile a Deployment. Sidecar profile and Secret changes only affect new Pods. Plan an application rollout after confirming the data recovery path.
DNS profiles create swytch-PROFILE-peers, a headless Service that publishes Pod addresses before readiness. Peers
advertise Pod IPs on UDP 7379. Check DNS, network policy, matching credentials, and peer reachability when a node cannot
join.
The injection webhook fails closed for annotated Pod creates. If those Pods cannot be created, inspect the operator and webhook certificate before retrying. Unannotated Pods do not depend on injection.
kubectl -n YOUR_NAMESPACE get swytch
kubectl -n YOUR_NAMESPACE describe swytch YOUR_PROFILE
kubectl -n YOUR_NAMESPACE logs YOUR_POD -c swytch
status.observedGeneration and Ready describe reconciliation and server Deployment availability. /health reports
process health, not data presence or subscription availability. Metrics are exposed at Pod port 9090 under /metrics.
Monitor hit rate, errors, latency, and memory alongside peer health. See observability and sizing.
The chart generates a one-year serving certificate and a ten-year CA. Renewal is manual. Before the serving certificate
expires, increment webhook.certificateRevision through Helm:
helm upgrade swytch ./operator/charts/swytch-operator \
-n swytch-system --reuse-values \
--set-string webhook.certificateRevision=2 --wait
Use a new revision value for each renewal. Helm reuses the CA and rolls the operator. Use Helm against the cluster; repeatedly applying offline renders is not a supported renewal workflow. Plan CA replacement before its expiry. See the operator lifecycle reference.