Bug 1999297

For scenario 2 I see the following payload in the UI

```
{"api_vip":"192.168.127.52","ingress_vip":"192.168.127.51","ssh_public_key":"ssh-rsa AAAAB[...]W+b6wp5c=","vip_dhcp_allocation":false,"network_type":"OVNKubernetes","user_managed_networking":false,"cluster_networks":[{"cidr":"10.128.0.0/14","host_prefix":23}],"service_networks":[{"cidr":"172.30.0.0/16"}],"machine_networks":[{"cidr":"192.168.127.0/24"}]}
```

Modifying it and sending manually via curl as PATCH with machine_networks removed, succeeds

```
'{"api_vip":"192.168.127.52","ingress_vip":"192.168.127.51","ssh_public_key":"ssh-rsa AAA[...]wp5c=","vip_dhcp_allocation":false,"network_type":"OVNKubernetes","user_managed_networking":false,"cluster_networks":[{"cidr":"10.128.0.0/14","host_prefix":23}],"service_networks":[{"cidr":"172.30.0.0/16"}]}'
```

The succeeding payload can as well contain *empty* list of "machine_networks", i.e.

```
'{"api_vip":"192.168.127.52","ingress_vip":"192.168.127.51","ssh_public_key":"ssh-rsa AAAA[...]W+b6wp5c=","vip_dhcp_allocation":false,"network_type":"OVNKubernetes","user_managed_networking":false,"cluster_networks":[{"cidr":"10.128.0.0/14","host_prefix":23}],"service_networks":[{"cidr":"172.30.0.0/16"}],"machine_networks":[]}'
```

For scenario 1 the error message "IP allocation from the DHCP server timed out" is coming from the following validator function

```
func isDhcpLeaseAllocationTimedOut(c *clusterPreprocessContext) bool {
	return c.cluster.MachineNetworkCidrUpdatedAt.String() != "" && time.Since(c.cluster.MachineNetworkCidrUpdatedAt) > DhcpLeaseTimeoutMinutes*time.Minute
}
```

Those values are updated via

```
func UpdateMachineCidr(db *gorm.DB, cluster *common.Cluster, machineCidr string) error {

[...]

		return db.Model(&common.Cluster{}).Where("id = ?", cluster.ID.String()).Updates(map[string]interface{}{
			"machine_network_cidr":            machineCidr,
			"machine_network_cidr_updated_at": time.Now(),
		}).Error

```

++++++++++++++++++
+++ SCENARIO 1 +++
++++++++++++++++++

Those are payloads used to configure respective options

+++ Cluster-Managed Networking, disabling "Allocate virtual IPs via DHCP server", manually providing VIPs

`--data '{"api_vip":"192.168.127.201","ingress_vip":"192.168.127.202","vip_dhcp_allocation":false,"network_type":"OVNKubernetes","user_managed_networking":false,"cluster_networks":[{"cidr":"10.128.0.0/14","host_prefix":23}],"service_networks":[{"cidr":"172.30.0.0/16"}]}'`

+++ Cluster-Managed Networking, enabling "Allocate virtual IPs via DHCP server", machine network #1

`--data '{"vip_dhcp_allocation":true,"network_type":"OVNKubernetes","user_managed_networking":false,"cluster_networks":[{"cidr":"10.128.0.0/14","host_prefix":23}],"service_networks":[{"cidr":"172.30.0.0/16"}],"machine_networks":[{"cidr":"192.168.127.0/24"}]}'`

+++ Cluster-Managed Networking, enabling "Allocate virtual IPs via DHCP server", machine network #2

`--data '{"vip_dhcp_allocation":true,"network_type":"OVNKubernetes","user_managed_networking":false,"cluster_networks":[{"cidr":"10.128.0.0/14","host_prefix":23}],"service_networks":[{"cidr":"172.30.0.0/16"}],"machine_networks":[{"cidr":"192.168.145.0/24"}]}'`

------------------

Now the observations

- disabling auto-allocation just works, no more comments
- enabling auto-allocations causes `GET /api/.../clusters/UUID | jq '.api_vip'` to return NULL for about 15 seconds after sending the request; after this time, the IP is returned consistently
- sending any PATCH request containing "machine_networks" (even if the value does not change) causes GET to once again return NULL for about 15 seconds

From this I believe what happens is the following

- whenever params.ClusterUpdateParams.MachineNetworks is not empty, we are triggering a reallocation of the VIP
- it does not matter whether params.ClusterUpdateParams.MachineNetworks is equal to cluster.MachineNetworks or not

What should happen

- reallocation should be triggered only when params.ClusterUpdateParams.MachineNetworks != cluster.MachineNetworks

Additional observation, not impacting the fundamental flaw of the reallocation logic described above, but something that causes the issue to be visible - when I created a cluster and then using the UI I go to the Networking tab, the browser keeps sending PATCH requests recurrently even though I'm not touching the UI at all.

I would have expected PATCH to be sent only when I explicitly click something in the UI that changes the underlying value, but I observe them being sent even when I don't touch the browser.

The current state is believed to be

- fix for scenario (1) in https://github.com/openshift/assisted-service/pull/2527
- fix for scenario (2) in https://github.com/openshift-assisted/assisted-ui-lib/pull/767

https://github.com/openshift/assisted-service/pull/2512, as not directly affecting the issue, should not be considered as a scope of this BZ.

Verified both scenarios on Staging
UI 1.5.35
BE v1.0.25.3

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory (Moderate: OpenShift Container Platform 4.9.0 bug fix and security update), and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://access.redhat.com/errata/RHSA-2021:3759