You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 3c5d9f4
Browse filesBrowse the repository at this point in the historyBrowse files
docs(rabbitmq): scope the runbook check and close the narrower-variant question
The confirmation step told the operator to wait for an empty namespace,
while the guard it confirms deliberately ignores objects naming another
release. A tenant running two brokers in one namespace would wait for
something that never happens. Say the check is release-scoped, in the
README and in the refusal.
Closing AMQP through additionalConfig while keeping the operator on
plaintext management was recorded as an open question. It is answerable:
cluster-operator pins the readiness probe to the amqp port unless
disableNonTLSListeners is set, so that configuration leaves every pod
failing readiness. Stated instead of left open.
Also repoint one template reference at the heading that exists, and name
the runbook test after what it compares.
Assisted-by: LLM
Signed-off-by: Aleksei Sviridkin <[email protected]>
Copy file name to clipboardExpand all lines: packages/apps/rabbitmq/README.md
+2-2Lines changed: 2 additions & 2 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -53,7 +53,7 @@ This chart can add TLS listeners for AMQP (port 5671) and the management HTTPS i
53
53
54
54
**Plaintext cannot be closed on the public side alone.** There is no operator-supported way to drop a port from the published Service by itself: `spec.override.service.spec.ports` is applied as a strategic merge patch keyed on `port`, so it can add or mutate entries but never remove them, and the operator regenerates its default port set on every reconcile. The only switch this chart exposes is `tls.disableNonTLSListeners`, described below, and it closes plaintext everywhere at once rather than only on the public side — in-cluster clients included.
55
55
56
-
A narrower variant — closing AMQP at the broker while leaving the operator's own management path on plaintext, which would keep chart-managed `users` and `vhosts` working — looks reachable through `spec.rabbitmq.additionalConfig`, but this chart does not wire it and it has not been validated here. Treat it as an open question rather than a supported option.
56
+
A narrower variant, closing AMQP at the broker while leaving the operator's own management path on plaintext so that chart-managed `users` and `vhosts`keep working, looks reachable through `spec.rabbitmq.additionalConfig`. It is not. The operator pins the readiness probe to the `amqp` port unless `spec.tls.disableNonTLSListeners` is set, so writing `listeners.tcp = none` through `additionalConfig` alone leaves every pod failing readiness for good. The flag is the only supported way to close the plaintext listeners, and it closes all of them at once.
57
57
58
58
**Setting `tls.enabled: true` on a running instance changes three things.** Plan for them before flipping it.
59
59
@@ -83,7 +83,7 @@ Clearing `users` and `vhosts` and setting `tls.disableNonTLSListeners: true` in
83
83
Do it as two upgrades instead:
84
84
85
85
1. Remove `users` and `vhosts` from the values, along with any hand-written topology object in the namespace, and let that upgrade finish.
86
-
2. Confirm the namespace is clear: `kubectl get bindings.rabbitmq.com,exchanges.rabbitmq.com,federations.rabbitmq.com,operatorpolicies.rabbitmq.com,permissions.rabbitmq.com,policies.rabbitmq.com,queues.rabbitmq.com,schemareplications.rabbitmq.com,shovels.rabbitmq.com,topicpermissions.rabbitmq.com,users.rabbitmq.com,vhosts.rabbitmq.com --namespace <tenant-namespace>` must return nothing. Those are the twelve kinds `messaging-topology-operator` drives through one reconciler, and every one of them carries the same finalizer, so checking only the three the chart creates leaves the other nine to wedge. Objects in `Terminating` still count — a finalizer that has not come off yet means the operator has not finished with the broker.
86
+
2. Confirm the namespace is clear: `kubectl get bindings.rabbitmq.com,exchanges.rabbitmq.com,federations.rabbitmq.com,operatorpolicies.rabbitmq.com,permissions.rabbitmq.com,policies.rabbitmq.com,queues.rabbitmq.com,schemareplications.rabbitmq.com,shovels.rabbitmq.com,topicpermissions.rabbitmq.com,users.rabbitmq.com,vhosts.rabbitmq.com --namespace <tenant-namespace>` must return nothing bound to this release. Those are the twelve kinds `messaging-topology-operator` drives through one reconciler, and every one of them carries the same finalizer, so checking only the three the chart creates leaves the other nine to wedge. A tenant namespace can hold several RabbitMQ applications, so read `spec.rabbitmqClusterReference.name` on whatever comes back: an object naming a sibling release is reached through that broker, whose listeners this flag does not touch, and the chart's own check ignores it too. Objects in `Terminating` still count — a finalizer that has not come off yet means the operator has not finished with the broker.
87
87
3. Set `tls.disableNonTLSListeners: true`.
88
88
89
89
A release already wedged this way recovers without data loss: unset `tls.disableNonTLSListeners`, wait for the finalizers to clear once the operator can reach the broker again, and then start over from step 1.
Copy file name to clipboardExpand all lines: packages/apps/rabbitmq/templates/rabbitmq.yaml
+2-2Lines changed: 2 additions & 2 deletions
Original file line number
Diff line number
Diff line change
@@ -51,7 +51,7 @@
51
51
{{- end }}
52
52
{{- end }}
53
53
{{- if $pending }}
54
-
{{- fail (printf "tls.disableNonTLSListeners cannot be turned on while this release still has topology objects in namespace %s: %s. Remove them in one upgrade, confirm `kubectl get %s --namespace %s` returns nothing, then set the flag in a second upgrade. Setting both at once leaves those objects stuck in Terminating. See README \"Removing the topology objects is its own upgrade\"." $.Release.Namespace (join ", " $pending) (include "rabbitmq.topology.resources" $) $.Release.Namespace) }}
54
+
{{- fail (printf "tls.disableNonTLSListeners cannot be turned on while this release still has topology objects in namespace %s: %s. Remove them in one upgrade, confirm `kubectl get %s --namespace %s` returns nothing bound to this release, then set the flag in a second upgrade. Setting both at once leaves those objects stuck in Terminating. See README \"Removing the topology objects is its own upgrade\"." $.Release.Namespace (join ", " $pending) (include "rabbitmq.topology.resources" $) $.Release.Namespace) }}
55
55
{{- end }}
56
56
{{- end }}
57
57
{{- end }}
@@ -74,7 +74,7 @@ metadata:
74
74
75
75
Expect this key on tenant-visible objects: ReconcileAndFilterAnnotations
76
76
strips only kubernetes.io / k8s.io keys, so the operator copies it onto every
77
-
child it builds. See README "The plaintext management listener". */}}
77
+
child it builds. See README "TLS scope". */}}
78
78
annotations:
79
79
rabbitmq.com/operator-connection-uri: "http://{{ .Release.Name }}.{{ .Release.Namespace }}.svc.{{ include "rabbitmq.clusterDomain" . }}:15672"
0 commit comments