Broaden Svc Selector

[eberlep]

No description provided.

[eberlep]

Another idea: For standby databases, select the first pod directly and don't use and spilo-role label. In a perfect world, standby databases should consist of a single instance anyway (which, by definition, will always be called ...-0). If there happen to be more instances, we would not round robin between those (as would be the case when simply selecting application=spilo pods), and if one would downscale the standby, existing connections to it would be kept open.

[eberlep]

Tested with Database with 2 Pods

 $ kgp -o wide
NAME                                 READY   STATUS    RESTARTS   AGE   IP            NODE           NOMINATED NODE   READINESS GATES   SPILO-ROLE
pgfits-philipptest5b769db3d-0        3/3     Running   0          39m   10.244.2.38   shoot--t6zvn   <none>           <none>            replica
pgfits-philipptest5b769db3d-1        3/3     Running   0          34m   10.244.2.39   shoot--t6zvn   <none>           <none>            master
postgres-operator-67565dcbc6-6dqs2   1/1     Running   4          21h   10.244.2.7    shoot--t6zvn   <none>           <none> 

 $ k exec -c postgres -ti pgfits-philipptest5b769db3d-0 -- su postgres -c "patronictl list"                                                                                                                                                                                                                         
+ Cluster: pgfits-philipptest5b769db3d (7138085200722174034) --+---------+----+-----------+                                                                                                                                                                                                                                                                                                    
| Member                        | Host        | Role           | State   | TL | Lag in MB |                                                                                                                                                                                                                                                                                                    
+-------------------------------+-------------+----------------+---------+----+-----------+                                                                                                                                                                                                                                                                                                    
| pgfits-philipptest5b769db3d-0 | 10.244.2.38 | Sync Standby   | running | 18 |         0 |                                                                                                                                                                                                                                                                                                    
| pgfits-philipptest5b769db3d-1 | 10.244.2.39 | Standby Leader | running | 18 |           |                                                                                                                                                                                                                                                                                                    
+-------------------------------+-------------+----------------+---------+----+-----------+

With ENABLE_LEGACY_STANDBY_SELECTOR set to true:

Pointing to pod -1, which internally is the StandbyLeader

$ k describe svc pgfits-philipptest5b769db3d-external 
[..]
Selector:                 application=spilo,cluster-name=pgfits-philipptest5b769db3d,spilo-role=master,team=pgfits
[...]
Endpoints:                10.244.2.39:5432

With ENABLE_LEGACY_STANDBY_SELECTOR set to false / removed:

Pointing to pod -0, which in this case is not the StandbyLeader (if it were a Replica, there could even be some lag). When downscaling, no reconnect is happening. After downscaling, all is correct.

 $ k describe svc pgfits-philipptest5b769db3d-external 
[..]
Selector:                 application=spilo,cluster-name=pgfits-philipptest5b769db3d,statefulset.kubernetes.io/pod-name=pgfits-philipptest5b769db3d-0,team=pgfits
[..]
Endpoints:                10.244.2.38:5432

After promotion:

Pointing to pod -1, statefulset.kubernetes.io/pod-name successfully removed from selector.

 $ k describe svc pgfits-philipptest5b769db3d-external 
[..]
Selector:                 application=spilo,cluster-name=pgfits-philipptest5b769db3d,spilo-role=master,team=pgfits
[..]
Endpoints:                10.244.2.39:5432