Managing StackGres 1.6 HA Clusters Fully with Patroni

Identify Master and Replica Pods

List StackGres pods with label filters:

kubectl get pods -n default -l app=StackGresCluster,stackgres.io/cluster=true

Master only:

kubectl get pods -n default -l app=StackGresCluster -l role=master

Replica only:

kubectl get pods -n default -l app=StackGresCluster,stackgres.io/cluster=true -l role=replica

Alternatively, exec into a Patroni container and run patronictl list to view cluster members, roles, IPs, ports, state, timeline and replication lag:

kubectl exec -it stackgres-0 -c patroni -- bash
patronictl list
+-----------+-------------+------------------+--------+---------+----+-----------+
|  Cluster  |    Member   |       Host       |  Role  |  State  | TL | Lag in MB |
+-----------+-------------+------------------+--------+---------+----+-----------+
| stackgres | stackgres-0 | 10.244.0.11:5433 | Leader | running |  2 |           |
| stackgres | stackgres-1 | 10.244.0.12:5433 |        | running |  2 |       0.0 |
| stackgres | stackgres-2 | 10.244.0.13:5433 |        | running |  2 |       0.0 |
+-----------+-------------+------------------+--------+---------+----+-----------+

Note: Change the -n flag to the appropriate namespace.

Graceful Switchover

Execute Switchover to a Replica

Run patronictl switchover. The tool prompts for the current master, the candidate replica, the execution time (default now) and a confirmation.

patronictl switchover
Master [stackgres-0]:
Candidate ['stackgres-1', 'stackgres-2'] []:
When should the switchover take place (e.g. 2020-01-16T17:23) [now]:
Current cluster topology
+-----------+-------------+------------------+--------+---------+----+-----------+
|  Cluster  |    Member   |       Host       |  Role  |  State  | TL | Lag in MB |
+-----------+-------------+------------------+--------+---------+----+-----------+
| stackgres | stackgres-0 | 10.244.0.11:5433 | Leader | running |  2 |           |
| stackgres | stackgres-1 | 10.244.0.12:5433 |        | running |  2 |       0.0 |
| stackgres | stackgres-2 | 10.244.0.13:5433 |        | running |  2 |       0.0 |
+-----------+-------------+------------------+--------+---------+----+-----------+
Are you sure you want to switchover cluster stackgres, demoting current master stackgres-0? [y/N]: y
2020-01-16 16:26:13.03648 Successfully switched over to "stackgres-1"
+-----------+-------------+------------------+--------+---------+----+-----------+
|  Cluster  |    Member   |       Host       |  Role  |  State  | TL | Lag in MB |
+-----------+-------------+------------------+--------+---------+----+-----------+
| stackgres | stackgres-0 | 10.244.0.11:5433 |        | stopped |    | unknown |
| stackgres | stackgres-1 | 10.244.0.12:5433 | Leader | running |  3 |           |
| stackgres | stackgres-2 | 10.244.0.13:5433 |        | running |    | unknown |
+-----------+-------------+------------------+--------+---------+----+-----------+

Important: Avoid other patronictl operations unless you fully understand their impact, as they may cause data loss or corruption.

Automated Failover

Patroni‑Driven Failover

Delete the leader pod to trigger failover: kubectl delete pod cluster-0 -n stackgres Kubernetes recreates the pod. Depending on the lock expiration timing, either the new pod assumes leadership or the remaining replica ( cluster-1) is promoted.

+ Cluster: cluster (6979461716096839850) ---+---------+----+-----------+
| Member    | Host                | Role    | State   | TL | Lag in MB |
+-----------+---------------------+---------+---------+----+-----------+
| cluster-0 | 192.168.40.142:7433 | Replica | running |  2 |           |
| cluster-1 | 192.168.12.150:7433 | Leader  | running |  2 |         0 |
+-----------+---------------------+---------+---------+----+-----------+

The same steps can be repeated by terminating either the primary or a replica pod.

Replica Reinitialization

Reinitialize a Stale Replica

When a replica cannot catch up and Patroni cannot recover automatically, reinitialize it from the primary using patronictl reinit.

patronictl reinit
Which member do you want to reinitialize [stackgres-1]? []:
Are you sure you want to reinitialize members stackgres-1? [y/N]: y
Success: reinitialize for member stackgres-1

# Verify the new topology
patronictl list
+-----------+-------------+------------------+--------+---------+----+-----------+
|  Cluster  |    Member   |       Host       |  Role  |  State  | TL | Lag in MB |
+-----------+-------------+------------------+--------+---------+----+-----------+
| stackgres | stackgres-0 | 10.244.0.11:5433 | Leader | running |  2 |           |
| stackgres | stackgres-1 | 10.244.0.12:5433 | Replica| running |  2 |       0   |
+-----------+-------------+------------------+--------+---------+----+-----------+

Important: Use reinitialization only when you understand the consequences; improper use can corrupt the cluster.

Reference:

https://github.com/zalando/patroni