DC Suite is SSO-only. Sign-in fails if your account isn't yet in an
organization, or your IdP session expired. Retry; if it persists, ask an
administrator to confirm your account and role.
A button or page is missing — why?
Controls are gated by your role at your scope. If you can't see or do
something, you likely lack that permission. Ask an admin to grant it.
The profile shows "0 nodes available." What do I do?
The fleet currently has no free machines of that profile. Try a smaller node
count, a different profile, or wait for machines to return to the pool. If
you believe machines should be free, an operator may need to recover a
stuck node — see the capacity runbook.
My cluster went to FAILED. Now what?
Open the cluster's detail page — the failure reason is shown. Common causes
are insufficient capacity (lower the count) or a node failing to boot the
image (retry or pin a different image version).
I added my SSH key but can't connect.
Check the Access panel's key status. "Applying/pending" means the node
hasn't applied it yet — wait a moment. "Node agent can't apply keys" means
the node needs operator attention.
My exposed web UI URL doesn't load.
Confirm the port you exposed is the one your app actually listens on, that
public exposure isn't disabled, and that your source IP isn't excluded by a
source restriction. DNS for new hostnames can take a moment to resolve.
The Observability logs panel shows an error.
That's an operator-side backend issue, not something you can fix. Metrics
may still work. Let your operators know — see the
logs runbook.
My cost report shows "unrated hours."
That's real usage with no price attached because a rate card is missing for
that profile. Ask an administrator to add one.
Do I have to deploy the server and console together?
When a change touches both, yes — otherwise the UI may call an endpoint the
server doesn't have yet (or vice versa). See
Deploys & Releases.
A machine shows ERROR but it's in a running cluster. Is it broken?
Usually not. It's typically a stale background validation the fleet couldn't
run while the machine was busy. DC Suite displays such machines as IN_USE
rather than a red error. If it's genuinely idle and stuck, see the
node recovery runbook.
Metrics work but logs don't. Why just logs?
Multi-tenant logs backends require a tenant header on reads that the metrics
gateway often injects automatically for metrics. Set
observability.loki_tenant. See the
logs runbook.
Does the control plane restarting lose in-flight cluster operations?
No. Operations run as durable workflows and resume after a restart.
**What's the difference between a cluster node's state and a bare-metal
machine's state?**
A node state (AVAILABLE, IN_CLUSTER, QUARANTINE, …) is the fleet's
view used for scheduling. A bare-metal state (REGISTERED, READY,
ERROR, …) is the operator's onboarding record for the physical machine.
They're linked but distinct.
What's the difference between a template and an image?
An image is the base OS/machine image nodes boot from. A template
(stack) is software you apply on top of a running cluster.
Can one cluster mix GPU types?
Yes — use multiple node groups, one per profile. That's a heterogeneous
cluster.