Troubleshooting

Organized by what you're seeing, not by component. If your question is conceptual rather than broken, see the FAQ instead.

Installation

`curl http://127.0.0.1:18099/readyz` doesn't return `{"status":"ready"}`

Check the pods actually came up:

    kubectl -n consize-system get pods

If a pod is `CrashLoopBackOff` or `Pending`, check its logs and events:

    kubectl -n consize-system logs <pod-name>
    kubectl -n consize-system describe pod <pod-name>

If the pods are Running but readyz still fails, the API usually can't reach its dependencies yet. Check the Prometheus secret is actually set:

    kubectl -n consize-system get secret consize-store -o jsonpath='{.data.prometheus-url}' | base64 -d

and confirm that URL is reachable from inside the cluster, not just from your machine.

If the port-forward itself hangs or refuses the connection, confirm the service name matches your release name (consize-api assumes you installed the chart as consize, see Production Installation):

    kubectl -n consize-system get svc

`helm install` fails with a permissions or namespace error

Confirm the namespace exists and your current kubectl context has permission to create resources in it:

kubectl config current-context
kubectl auth can-i create deployments -n consize-system

Cronjobs exist but never seem to run

kubectl -n consize-system get cronjobs
kubectl -n consize-system get jobs

If jobs aren't being created at all, check the cronjob's schedule and that the Kubernetes cronjob controller isn't suspended cluster-wide. If jobs are created but fail, check their pod logs the same way as above.

Recommendations

No recommendations are showing up

Consize needs a minimum amount of historical data before it will generate a recommendation (CONSIZE_MIN_DATA_DAYS, 5 days by default, see Configuration). A freshly installed instance won't have recommendations yet, this is expected, not a bug.

A recommendation looks smaller than I expected

Sizing is based on p95 (requests) and p99 (limits) usage with headroom, not a plain average, so recommendations are intentionally more conservative than "reduce to what's used right now." See why percentiles, not averages.

Verification and rollback

A change was applied but verification never completes

The verification window scales with the apply step (1 hour for step 1, 2 for step 2, and so on), see Observability. Check how far into that window you are before assuming something's stuck. If the window has clearly passed, check the verifier is running:

kubectl -n consize-system get cronjobs

A change rolled back and I don't understand why

Check the verification run for that apply, it records which signal breached its threshold (restarts, OOM kills, evictions, CPU throttling, or an app-level SLI if you configured one) and for how long, see Observability for the signal list and The Safety Net for the decision matrix. Rollbacks are byte-identical to the pre-apply state on purpose, so this shouldn't have caused a second incident on top of the first.

Cloud integrations

AWS/GCP metrics aren't showing up

Confirm CONSIZE_DBMETRICS is actually set (it's none by default) and that the credentials you configured have the permissions listed in Production Installation. A silent empty result is almost always a permissions or region/project mismatch, not a connectivity failure.

GitHub pull requests aren't being created

Confirm the consize-github secret's token has repo write access to the target repository and hasn't expired, see Production Installation.

Still stuck?

See Support for where to ask, include your Consize version and the output of kubectl -n consize-system get pods when you do, it's usually the first thing anyone will ask for.

Next steps

Was this page helpful?

Suggest edits Raise issue