Check platform health
Check health in layers: operator reconciliation, workload readiness, service responses and a representative job/activation/content operation. The conditions and endpoints described below were verified through the platform gateway on the Koios test platform on 2026-11-18. They are checks to repeat for your deployment, not a guarantee that all work succeeds whenever an endpoint returns 200.
Set the diagnostic scope
Choose the authorized cluster context and platform namespace explicitly. The examples use Bash variables; replace the example values before running any commands:
export KAP_CONTEXT='<context>'
export KAP_NAMESPACE='<platform-namespace>'
export KAP_NAME='<platform-name>'
Use oc in place of kubectl on OpenShift; the --context and -n scope remains explicit.
Do not switch the global current context as part of these examples.
1. Inspect KoiosPlatform status
Read the public address, release and condition summary:
kubectl --context "$KAP_CONTEXT" -n "$KAP_NAMESPACE" get koiosplatform "$KAP_NAME" \
-o jsonpath='{.status.version}{"\n"}{.status.URL}{"\n"}{range .status.conditions[*]}{.type}{"="}{.status}{" reason="}{.reason}{" message="}{.message}{"\n"}{end}'
After installation or a change, wait for Successful=True and Failure=False. Read the reason
and message to distinguish progress from failure. During the reconciliation lifecycle, messages
can include Awaiting next reconciliation; that text alone is not evidence of a failed platform.
On the healthy test platform, Running=True had reason Successful while the other two
conditions were successful/non-failing. Do not require Running to be False as a readiness rule.
A success condition records reconciliation, not every service request or downstream job. Check idle mode and any restore hold before treating scaled-down components as unexpected failures.
Inspect selected component fields, without retrieving Secret values:
kubectl --context "$KAP_CONTEXT" -n "$KAP_NAMESPACE" get koiosplatform "$KAP_NAME" \
-o jsonpath='gatewayImage={.status.gatewayImage}{"\n"}controller={.status.controller}{"\n"}eda={.status.eda}{"\n"}hub={.status.hub}{"\n"}postgresStatefulSet={.status.postgresStatefulSet}{"\n"}databaseVersions={.status.databaseVersions}{"\n"}redisMode={.status.redisMode}{"\n"}'
status.version identifies the operator release that reconciled the platform; it does not
independently verify every component image or the success of an upgrade. Component-name fields
can be empty when an optional service is disabled. An external database has no managed PostgreSQL
StatefulSet to inspect.
2. Check services through the platform address
Use status.URL, with HTTPS certificate verification enabled. Set the URL for your deployment:
export KAP_URL='https://platform.example.com'
The verified endpoint behavior is:
Service / endpoint |
Authentication in the verified setup |
What to inspect |
|---|---|---|
Gateway |
No authentication; 200 |
|
Gateway |
401 without authentication; 200 with an authorized account |
|
Controller |
No authentication; 200 |
|
EDA |
No authentication; 200 |
|
Hub |
No authentication; 200 |
|
The EDA endpoint is status/. /api/eda/v1/ping/ returned 404 in the verified setup; do not
use that nonexistent endpoint as an EDA readiness test.
curl --silent --show-error --include "$KAP_URL/api/gateway/v1/ping/"
curl --silent --show-error --include "$KAP_URL/api/controller/v2/ping/"
curl --silent --show-error --include "$KAP_URL/api/eda/v1/status/"
curl --silent --show-error --include "$KAP_URL/api/galaxy/pulp/api/v3/status/"
For the authenticated gateway summary, use an account allowed to inspect service status. Supplying only the username makes curl prompt for its password rather than putting it in the command line:
curl --silent --show-error --include -u '<platform-user>' "$KAP_URL/api/gateway/v1/status/"
This password method requires gateway Basic authentication to be enabled. Other authentication
methods are described in API authentication. A 401 from this
protected endpoint with no credentials is expected, not evidence that the gateway is down.
For a private CA, configure client trust or use --cacert /path/to/platform-ca.crt; do not make
skipping TLS verification the health-check procedure.
3. Inspect workloads and pods
Use the component map and actual child names from status:
kubectl --context "$KAP_CONTEXT" -n "$KAP_NAMESPACE" get deployments,statefulsets
kubectl --context "$KAP_CONTEXT" -n "$KAP_NAMESPACE" get pods -o wide
kubectl --context "$KAP_CONTEXT" -n "$KAP_NAMESPACE" get pvc
Review desired/ready replicas, pod readiness/restarts, pending pods and PVC binding. Identify the owner of an unexpected pod before acting: a job pod, activation pod, database pod and operator pod have different lifecycles. A completed job pod need not stay Running; a pending activation may need worker or registry diagnosis rather than a gateway restart.
For the operators, select their installation namespace, which may differ from the platform’s:
export KAP_OPERATOR_NAMESPACE='<operator-namespace>'
kubectl --context "$KAP_CONTEXT" -n "$KAP_OPERATOR_NAMESPACE" get deployments
Find the fixed package names described in Operator deployments.
Do not construct them by adding -gateway-operator-controller-manager to the platform name.
4. If something is not ready
Start with events and the identified pod/PVC, rather than changing generated resources:
kubectl --context "$KAP_CONTEXT" -n "$KAP_NAMESPACE" get events --sort-by=.metadata.creationTimestamp
kubectl --context "$KAP_CONTEXT" -n "$KAP_NAMESPACE" describe pod '<affected-pod>'
kubectl --context "$KAP_CONTEXT" -n "$KAP_NAMESPACE" describe pvc '<affected-claim>'
describe can include arguments and environment configuration. Inspect locally and redact before
sharing; do not describe or dump Secret objects. Use only needed status fields and Secret names.
Match the failed layer to its owner: the gateway operator manages KoiosPlatform and shared
resources; the controller, EDA and hub operators manage the corresponding children. Use
Logs and diagnostics to select the correct container and operator, then follow the
runbook for the symptom. After correction, repeat the status, workload and service
checks, followed by a representative application operation.