Live Logs for Pods with Sidecars
- Open a Pod with a sidecar, such as Istio or Linkerd, and see its app logs
right away. Versus picks the container the waykubectl logsdoes: the
default-container annotation first, otherwise the first app container that is
not a sidecar. - See "Waiting for logs…" while the Pod is still loading, never a raw error
code. - Workload logs now include Pods that run more than one container.
Sync Status That Keeps Your Data
- Keep your data on screen when a sync fails. The top bar says "Sync failed —
showing data from 14:05" or "Sync degraded — showing data from 14:05". - Hover or focus the status to see each failing kind, its reason and how long
it has failed, with a Retry sync button. - See "Cannot connect to this cluster" when a new cluster's first sync fails,
and "Cannot read this cluster right now" when a cluster that synced before has
nothing to show, instead of an endless loading screen.
Least-Privilege Kubernetes Readers
- Remove
secretsfrom the reader role, as the docs recommend, and keep a steady
view. Secret is shown as not readable, everything else stays on screen, and
history is saved on every cycle. Before, the whole view switched to "Cluster
sync failed" about once a minute and history saves were refused. - Leave out the Gateway API grant, or install the Gateway API, without the
cluster flapping. A kind Versus may not read is shown as not readable, never
as unsupported. - One slow or failing kind no longer marks the whole cluster as failed. The
cluster shows as partial and names the kind. - A real API server outage still shows "Sync failed" within about 30 seconds.
- Widen a reader role from namespaces to the whole cluster and see every kind
ready within two sync cycles. - The top bar no longer flashes "Updates disconnected" every 30 seconds.
Install in Your Cluster in Two Minutes
- Install with
helm install ... --set connectors.kubernetes.enabled=true, or
with onekubectl applyof
examples/kubernetes/k8s-connector-in-cluster-enabled.yaml, and see your
cluster. The flag turns on the Kubernetes connector within_clusteraccess
and creates a read-only reader role for it. The chart also generates a
sign-in secret and keeps it across upgrades. - Follow the new Kubernetes quickstart step
by step. On a small cluster, the Overview is ready in under 30 seconds. - The default role reads no Secrets. It covers Gateway, HTTPRoute, Argo CD,
Argo Rollouts and Flux. Secret shows as "Not collected by configuration", not
as an error. The Helm tab says it needs Secret access and links to how to
allow it. - Information about kinds Versus does not read moves out of the page and into
the top bar, next to "Last synced". - Use a namespace-only role with
connectors.kubernetes.namespaces. You get a
steady view of those namespaces, without history. - Setup errors say what went wrong and how to fix it. An unknown connector key
is named in the error. - Live updates no longer flash "Updates disconnected" on their planned
reconnect. - See "CPU and memory need metrics-server" when your cluster has no metrics
server.
Upgrade Notes
- Go 1.26.11 (GO-2026-6859, an HTTP/1 server crash in
net/http). - A namespace-only reader role shows a partial live view and saves no Kubernetes
history. Use the cluster-wide reader role, withoutsecrets, when you need
history. - The Kubernetes connector stays off by default. With
connectors.kubernetes.enabled=true, the chart also creates the read-only
reader ClusterRole. SetkubernetesReaderRBAC.external: trueif you manage
RBAC yourself. kubernetesReaderRBAC.gatewayAPIis replaced byoptionalKinds, which is on
by default. Installs with the old key fail until you remove it.- The chart defaults to 1 replica. More than one replica needs Postgres
storage. queue.debug_bodyandalert.debug_bodynow default tofalse.- The Helm chart no longer uses the Bitnami Redis subchart. With
redis.enabled=trueit runs its own Redis (officialredis:7.4, Service
<fullname>-redis) on a new 1Gi volume. Moveredis.master.persistence.*to
redis.persistence.*and dropredis.architecture, then upgrade with
--reset-then-reuse-values -f <values> --set redis.master=null --set redis.architecture=null
(or only-f <values>). If your old values setredis.image(for example the
bitnamilegacy/redisoverride), also add--set redis.image=null; the chart
refuses Bitnami images. Pending on-call escalations in the old Redis are not
carried over. Delete the old volumeredis-data-<release>-redis-master-0
afterwards. - The webhook endpoint answers HTTP 201 once the incident is stored, even when a
channel or on-call send fails; the response and the incident carry
notify_status(sent,partial,failed,none),notify_errorand
oncall_error. HTTP 500 now means only that the incident could not be stored,
so Alertmanager no longer retries a delivery failure into duplicate incidents. - With AI turned off, or on with no usable key, Versus makes no AI provider call
in detection, Analyze or the Enterprise SLO advisor. - Delivery and on-call error text no longer includes channel URLs or tokens.
What's Changed
- feat(kubernetes): live logs for Pods with sidecars by @hoalongnatsu in #409
- fix(kubernetes): least-privilege reader roles keep a steady view and saved history by @hoalongnatsu in #410
- feat(kubernetes): one-flag in-cluster install that works with no errors by @hoalongnatsu in #412
- docs: restructure the docs around quickstarts, and fix what the following them found by @hoalongnatsu in #413
Full Changelog: v1.4.35...v1.4.36