From 5c40b607e609f93842d9dd3e9d4b2a70b45f844d Mon Sep 17 00:00:00 2001
From: Dmitriy Rabotyagov <noonedeadpunk@gmail.com>
Date: Fri, 29 Nov 2024 13:17:55 +0100
Subject: [PATCH] Add option to handle absent cross_az_attach

At the moment volumes for workers are gonna be spawn in a default
AZ completely disregarding user request for Availability Zone.

In a design where cross_az_attach is disabled, an attempt
to add a volume to worker has high failure percentage due to fallback
to the default scheduling zone, unless allow_availability_zone_fallback
is disabled.

This patch adds a configuration option `cross_az_attach` which is set
to True by default to align with Nova typical behavior. It will define
AZ to be set to `nova` according to a CSI default [1].

When `cross_az_attach` is set to False, AZ for the volume will be set to
the cluster AZ value.
This ensures that volume will be created in a same zone as workers are
preventing failures.

[1] https://github.com/kubernetes/cloud-provider-openstack/blob/d228854cf58e7b4ed93d5e7ba68ab639450e3449/docs/cinder-csi-plugin/using-cinder-csi-plugin.md#supported-parameters

Relates-to: #366
---
 docs/user/configs.md            | 11 +++++++++++
 magnum_cluster_api/conf.py      | 14 ++++++++++++++
 magnum_cluster_api/resources.py | 10 ++++++++++
 3 files changed, 35 insertions(+)

diff --git a/docs/user/configs.md b/docs/user/configs.md
index 9cb1c5a9..7553e90f 100644
--- a/docs/user/configs.md
+++ b/docs/user/configs.md
@@ -108,3 +108,14 @@ Options under this group are used for configuring Openstack authentication for C
 :   If set, then the server's certificate will not be verified.
     **Type**: `boolean`
     **Default value**: `False`
+
+## cinder
+Options under this group are used for configuring OpenStack Cinder behavior.
+
+`cross_az_attach`
+
+:   When set to False, Cluster Availability Zone will be used to create a volume.
+    For that Availability Zone names in Cinder and Nova should match.
+    Otherwise, default `nova` Availability Zone will be used for volumes.
+    **Type**: `boolean`
+    **Default value**: `True`
diff --git a/magnum_cluster_api/conf.py b/magnum_cluster_api/conf.py
index 3edfadb9..cd844350 100644
--- a/magnum_cluster_api/conf.py
+++ b/magnum_cluster_api/conf.py
@@ -21,6 +21,8 @@
     name="capi_client", title="Options for the Cluster API client"
 )
 
+cinder_group = cfg.OptGroup(name="cinder", title="Options for the Cinder client")
+
 manila_client_group = cfg.OptGroup(
     name="manila_client", title="Options for the Manila client"
 )
@@ -98,6 +100,17 @@
     ),
 ]
 
+cinder_opts = [
+    cfg.BoolOpt(
+        "cross_az_attach",
+        default=True,
+        help=_(
+            "In case of multiple availability zones, allows to create and "
+            "attach volumes from random AZ to worker nodes. When set to False "
+            "Availability Zone names in Cinder and Nova should match."
+        ),
+    )
+]
 
 manila_client_opts = [
     cfg.StrOpt(
@@ -157,6 +170,7 @@
     (auto_scaling_group, auto_scaling_opts),
     (capi_client_group, capi_client_opts),
     (capi_client_group, common_security_opts),
+    (cinder_group, cinder_opts),
     (manila_client_group, manila_client_opts),
     (manila_client_group, common_security_opts),
     (proxy_group, proxy_opts),
diff --git a/magnum_cluster_api/resources.py b/magnum_cluster_api/resources.py
index 7c917f96..d4858b31 100644
--- a/magnum_cluster_api/resources.py
+++ b/magnum_cluster_api/resources.py
@@ -296,6 +296,15 @@ def get_object(self) -> pykube.ConfigMap:
         if cinder.is_enabled(self.cluster):
             volume_types = osc.cinder().volume_types.list()
             default_volume_type = osc.cinder().volume_types.default()
+            # Default is set in accordance to the CSI:
+            # https://github.com/kubernetes/cloud-provider-openstack/blob/d228854cf58e7b4ed93d5e7ba68ab639450e3449/docs/cinder-csi-plugin/using-cinder-csi-plugin.md#supported-parameters
+            # If allow_availability_zone_fallback is set to False in Cinder
+            # and "nova" AZ is not present in Cinder, CSI request will fail.
+            volume_availability_zone = "nova"
+            if not CONF.cinder.cross_az_attach:
+                volume_availability_zone = self.cluster.labels.get(
+                    "availability_zone", volume_availability_zone
+                )
             data = {
                 **data,
                 **{
@@ -333,6 +342,7 @@ def get_object(self) -> pykube.ConfigMap:
                             "provisioner": "cinder.csi.openstack.org",
                             "parameters": {
                                 "type": vt.name,
+                                "availability": volume_availability_zone,
                             },
                             "reclaimPolicy": "Delete",
                             "volumeBindingMode": "Immediate",