Improve error reporting

azimuth_caas_operator/operator.py

@@ -141,21 +141,20 @@ async def cluster_create(body, name, namespace, labels, **kwargs):
     # Wait for Blazar lease to be active
     try:
         flavor_map = await lease_utils.ensure_lease_active(K8S_CLIENT, cluster)
-    except lease_utils.LeaseInError:
+    except lease_utils.LeaseInError as exc:
         # TODO(johngarbutt) we need to tell between these two cases!
+        message = str(exc)
+        if "external service enforcement filter denied the request" in message.lower():
+            message = "Not enough credits to create platform"
         await cluster_utils.update_cluster(
-            K8S_CLIENT,
-            name,
-            namespace,
-            cluster_crd.ClusterPhase.FAILED,
-            error="Cloud is full, or you are out of credits.",
+            K8S_CLIENT, name, namespace, cluster_crd.ClusterPhase.FAILED, error=message
         )
         msg = f"Lease is in Error state for {name} in {namespace}"
         LOG.error(msg)
         # keep polling until the lease is active, but report the error
         raise kopf.TemporaryError(msg, delay=20)
-
-    await cluster_utils.update_cluster_flavors(K8S_CLIENT, cluster, flavor_map)
+    else:
+        await cluster_utils.update_cluster_flavors(K8S_CLIENT, cluster, flavor_map)
 
     # Check for an existing create job
     create_job = await ansible_runner.get_create_job_for_cluster(

azimuth_caas_operator/utils/ansible_runner.py

@@ -9,7 +9,6 @@
 from cryptography.hazmat.primitives import serialization
 
 from easykube import ApiError
-import kopf
 
 from azimuth_caas_operator.models.v1alpha1 import cluster as cluster_crd
 from azimuth_caas_operator.models.v1alpha1 import cluster_type as cluster_type_crd

azimuth_caas_operator/utils/lease.py

@@ -7,12 +7,14 @@
 from azimuth_caas_operator.models.v1alpha1 import cluster as cluster_crd
 
 SCHEDULE_API_VERSION = "scheduling.azimuth.stackhpc.com/v1alpha1"
-FINALIZER = "caas.stackhpc.com"
+FINALIZER = "caas.azimuth.stackhpc.com"
 LOG = logging.getLogger(__name__)
 
+
 class LeaseInError(Exception):
     pass
 
+
 async def _patch_finalizers(resource, name, namespace, finalizers):
     """
     Patches the finalizers of a resource. If the resource does not exist any
@@ -33,7 +35,7 @@ async def ensure_lease_active(client, cluster: cluster_crd.Cluster):
     """
     if not cluster.spec.leaseName:
         LOG.info("No leaseName set, skipping lease check.")
-        return
+        return {}
 
     lease_resource = await client.api(SCHEDULE_API_VERSION).resource("leases")
     lease = await lease_resource.fetch(
@@ -54,13 +56,15 @@ async def ensure_lease_active(client, cluster: cluster_crd.Cluster):
         )
         LOG.info("Added finalizer to the lease.")
 
-    if lease and "status" in lease and lease["status"]["phase"] == "Active":
+    lease_status = lease.get("status", {})
+
+    if lease_status.get("phase", "Unknown") == "Active":
         LOG.info("Lease is active!")
         # return mapping of requested flavor to reservation flavor
-        return lease["status"]["flavorMap"]
+        return lease["status"]["sizeMap"]
 
-    if lease and "status" in lease and lease["status"]["phase"] == "Error":
-        raise LeaseInError("Lease is in Error state.")
+    if lease_status.get("phase", "Unknown") == "Error":
+        raise LeaseInError(lease_status.get("errorMessage", "Error creating lease"))
 
     LOG.info(f"Lease {cluster.spec.leaseName} is not active, wait till active.")
     delay = 60
@@ -81,10 +85,16 @@ async def ensure_lease_active(client, cluster: cluster_crd.Cluster):
 
 async def drop_lease_finalizer(client, cluster: cluster_crd.Cluster):
     lease_resource = await client.api(SCHEDULE_API_VERSION).resource("leases")
-    lease = await lease_resource.fetch(
-        cluster.spec.leaseName,
-        namespace=cluster.metadata.namespace,
-    )
+    try:
+        lease = await lease_resource.fetch(
+            cluster.spec.leaseName,
+            namespace=cluster.metadata.namespace,
+        )
+    except easykube.ApiError as exc:
+        if exc.status_code == 404:
+            return
+        else:
+            raise
     finalizers = lease.get("metadata", {}).get("finalizers", [])
     await _patch_finalizers(
         lease_resource,