lexicons and xrpc
lexicon
com.publicdomainrelay.temp.cluster.createCluster
The gateway creates the named cluster of the calling account and leaves it RUNNING, in one call. This is the same create-on-read path getKubeconfig walks - the account workspace, its home, the Cluster record and the mounted child - followed by the same desired-state flip controlPlaneUp performs, so a door that wants a cluster it can immediately reach does not have to read a kubeconfig it will throw away and then wake the control plane in a second round trip. Idempotent in both halves: a second call finds the tree it made and patches a desired state that is already true. Requires a service-auth JWT from the CALLER's own PDS; the account is the VERIFIED token's iss and never read from this input. The response is the phase as of this call, so a caller that sees Initializing knows the tree exists and the control plane is still coming up, and can poll controlPlaneUp - which is where the 429 and its Retry-After live, because that is the verb a client-go driven caller retries on.
input
| field | type | description |
|---|---|---|
name * | string | The cluster label, unique within the calling account. RFC-1123 label, <=63 characters. |
output
| field | type | description |
|---|---|---|
phase * | string | The cluster's phase as of this call, after the desired state was set to running. Initializing means the control plane is still coming up; the cluster exists either way. knownValues ["Unavailable","Initializing","Ready","Draining"] |
errors
| error | description |
|---|---|
InvalidRequest | Name absent, or not a legal RFC-1123 label within 63 characters. |
Unauthorized | Missing or invalid service-auth token. Also returned when the token's aud is not this service or its lxm is not this method. |
IdentityRevoked | The caller's identity carries a deny record for this account. The deny is written before any deletion, so it refuses while the revocation is still propagating - and it refuses BEFORE anything is created, which is what keeps create-on-read from re-arming a revoked account. |
ClusterNotFound | The tree was created and the Cluster was not readable immediately after. Deliberately indistinguishable from a name that belongs to another account. |
ClusterDraining | The cluster is being torn down and its desired state is frozen. Not retryable: a Draining cluster cannot be un-drained, by design - a cancellable teardown gate is a gate an operator clears to make the symptom go away. |
ClusterUnavailable | 503: the gateway could not resolve the token issuer's signing key in time, could not read the caller's deny record, or could not prepare the cluster. Retry. |
InternalError | 500: a failure the gateway maps to no named error. |
used in
no page of these docs names it yet.
raw schema: lexicons/com/publicdomainrelay/temp/cluster/createCluster.json
{
"lexicon": 1,
"id": "com.publicdomainrelay.temp.cluster.createCluster",
"defs": {
"main": {
"type": "procedure",
"description": "The gateway creates the named cluster of the calling account and leaves it RUNNING, in one call. This is the same create-on-read path getKubeconfig walks - the account workspace, its home, the Cluster record and the mounted child - followed by the same desired-state flip controlPlaneUp performs, so a door that wants a cluster it can immediately reach does not have to read a kubeconfig it will throw away and then wake the control plane in a second round trip. Idempotent in both halves: a second call finds the tree it made and patches a desired state that is already true. Requires a service-auth JWT from the CALLER's own PDS; the account is the VERIFIED token's iss and never read from this input. The response is the phase as of this call, so a caller that sees Initializing knows the tree exists and the control plane is still coming up, and can poll controlPlaneUp - which is where the 429 and its Retry-After live, because that is the verb a client-go driven caller retries on.",
"input": {
"encoding": "application/json",
"schema": {
"type": "object",
"required": [
"name"
],
"properties": {
"name": {
"type": "string",
"description": "The cluster label, unique within the calling account. RFC-1123 label, <=63 characters."
}
}
}
},
"output": {
"encoding": "application/json",
"schema": {
"type": "object",
"required": [
"phase"
],
"properties": {
"phase": {
"type": "string",
"knownValues": [
"Unavailable",
"Initializing",
"Ready",
"Draining"
],
"description": "The cluster's phase as of this call, after the desired state was set to running. Initializing means the control plane is still coming up; the cluster exists either way."
}
}
}
},
"errors": [
{
"name": "InvalidRequest",
"description": "Name absent, or not a legal RFC-1123 label within 63 characters."
},
{
"name": "Unauthorized",
"description": "Missing or invalid service-auth token. Also returned when the token's aud is not this service or its lxm is not this method."
},
{
"name": "IdentityRevoked",
"description": "The caller's identity carries a deny record for this account. The deny is written before any deletion, so it refuses while the revocation is still propagating - and it refuses BEFORE anything is created, which is what keeps create-on-read from re-arming a revoked account."
},
{
"name": "ClusterNotFound",
"description": "The tree was created and the Cluster was not readable immediately after. Deliberately indistinguishable from a name that belongs to another account."
},
{
"name": "ClusterDraining",
"description": "The cluster is being torn down and its desired state is frozen. Not retryable: a Draining cluster cannot be un-drained, by design - a cancellable teardown gate is a gate an operator clears to make the symptom go away."
},
{
"name": "ClusterUnavailable",
"description": "503: the gateway could not resolve the token issuer's signing key in time, could not read the caller's deny record, or could not prepare the cluster. Retry."
},
{
"name": "InternalError",
"description": "500: a failure the gateway maps to no named error."
}
]
}
}
}