Guides / Operate
Find out why something stays Pending
When an object does not become Ready, or the API refuses a request, start with the object's status. Every object has a status.phase and conditions written by its controller, and they usually say what is wrong.
1. Read the status
kubectl ... get vpcs,accounts,dnsservices,dhcpservices # phase column
kubectl ... get account laptop -o jsonpath='{.status.phase}{"\n"}{.status.conditions}{"\n"}'
kubectl ... wait --for=condition=Ready vpc/lan --timeout=120s
In the console, open the object. The structure view shows the phase and each condition with its reason and message. The overview lists everything that needs attention.
2. Look up the reason
Match the reason in the condition against this table.
| You see | Meaning and fix |
|---|---|
DuplicatePublicKey on an Account | another Account on the same gateway uses that key. Generate a new key for the new member. |
DuplicateService on DHCP or DNS | a VPC can have one of each; the older wins. Delete or edit the other. |
NameInUse on a diagram block | an object with that name exists that the diagram did not create. Rename the block. |
DHCP or DNS service stays Pending | it waits for its VPC to be Ready. Check the VPC first. |
| address pool exhausted or requested address taken | the Account is Failed with a reason; pick another address or a larger cidr (the cidr itself cannot be changed: create a new VPC). |
Check whether a device is really connected
To see whether a machine is talking to the gateway, open the VPC in the console and look at Network at the bottom.
- Never connected means there has been no WireGuard handshake. Check the key, the endpoint and any firewall in between.
- Idle means the last handshake is more than three minutes old. The member went away or its keepalive stopped.
The neighbor table shows whether the gateway has learned the member's MAC address and IP address (ARP for IPv4, NDP for IPv6). The same data is in the Device objects of the VPC: kubectl get devices -l network.etorm.com/vpc=lan.
3. Read the error code
| Code | Usually |
|---|---|
401 | missing, expired or revoked token |
403 | your role does not allow it, a quota is reached, or (on delete or get) the object no longer exists and your access is limited to a billing account |
404 | wrong name, or the project does not exist (the endpoint of an unknown project is 404) |
409 | already exists, edited by someone else (resourceVersion), or still referenced (a VPC with accounts) |
422 | the object is invalid: the message names the field |
Errors are Kubernetes Status objects, so kubectl prints the server's message. Unknown fields are rejected rather than ignored, so a typo in a field name fails loudly instead of being dropped.
4. Find out who changed it
Open Audit log in the console, or filter it by outcome and principal. A refusal there tells you which principal tried what and why it was denied.