Deploy on multiple availability zones (AZ)

During the deployment to hardware/VMs, it is important to spread all the database copies (Juju units) to different hardware servers, or even better, to the different availability zones (AZ). This will guarantee no shared service-critical components across the DB cluster (eliminate the case with all eggs in the same basket).

This guide will take you through deploying a PostgreSQL cluster on GCE using 3 available zones. All Juju units will be set up to sit in their dedicated zones only, which effectively guarantees database copy survival across all available AZs.

Summary

• Set up GCE on Google Cloud
• Deploy PostgreSQL with Juju zones constraints
  • Simulation: A node gets drained
• Additional resources

Set up GCE on Google Cloud

Let’s deploy the PostgreSQL Cluster on GKE (us-east4) using all 3 zones there (us-east4-a, us-east4-b, us-east4-c) and make sure all pods always sits in the dedicated zones only.

Log into Google Cloud and bootstrap GCE on Google Cloud:

gcloud auth login
gcloud iam service-accounts keys create sa-private-key.json  --iam-account=juju-gce-account@[your-gcloud-project-12345].iam.gserviceaccount.com
sudo mv sa-private-key.json /var/snap/juju/common/sa-private-key.json
sudo chmod a+r /var/snap/juju/common/sa-private-key.json

juju add-credential google
juju bootstrap google gce
juju add-model mymodel

Deploy PostgreSQL with Juju zones constraints

Juju provides the support for availability zones using constraints. Read more about zones in Juju documentation.

The command below demonstrates how Juju automatically deploys Charmed PostgreSQL VM using Juju constraints:

juju deploy postgresql -n 3 \
  --constraints zones=us-east1-b,us-east1-c,us-east1-d

After a successful deployment, juju status will show an active application:

Model    Controller  Cloud/Region     Version    SLA          Timestamp
mymodel  gce         google/us-east1  3.5.4      unsupported  00:16:52+02:00

App         Version  Status  Scale  Charm       Channel    Rev  Exposed  Message
postgresql  14.12    active      3  postgresql  14/stable  468  no       

Unit           Workload  Agent  Machine  Public address  Ports     Message
postgresql/0   active    idle   0        34.148.44.51    5432/tcp  
postgresql/1   active    idle   1        34.23.202.220   5432/tcp  
postgresql/2*  active    idle   2        34.138.167.85   5432/tcp  Primary

Machine  State    Address        Inst id        Base          AZ          Message
0        started  34.148.44.51   juju-e7c0db-0  ubuntu@22.04  us-east1-d  RUNNING
1        started  34.23.202.220  juju-e7c0db-1  ubuntu@22.04  us-east1-c  RUNNING
2        started  34.138.167.85  juju-e7c0db-2  ubuntu@22.04  us-east1-b  RUNNING

and each unit/vm will sit in the separate AZ out of the box:

> gcloud compute instances list
NAME           ZONE        MACHINE_TYPE  PREEMPTIBLE  INTERNAL_IP  EXTERNAL_IP    STATUS
juju-a82dd9-0  us-east1-b  n1-highcpu-4               10.142.0.30  34.23.252.144  RUNNING  # Juju Controller
juju-e7c0db-2  us-east1-b  n2-highcpu-2               10.142.0.32  34.138.167.85  RUNNING  # postgresql/2
juju-e7c0db-1  us-east1-c  n2-highcpu-2               10.142.0.33  34.23.202.220  RUNNING  # postgresql/1
juju-e7c0db-0  us-east1-d  n2-highcpu-2               10.142.0.31  34.148.44.51   RUNNING  # postgresql/0

Simulation: A node gets lost

Let’s destroy a GCE node and recreate it using the same AZ:

> gcloud compute instances delete juju-e7c0db-1 
No zone specified. Using zone [us-east1-c] for instance: [juju-e7c0db-1].
The following instances will be deleted. Any attached disks configured to be auto-deleted will be deleted unless they are attached to any other instances or the `--keep-disks` flag is given and specifies them for keeping. Deleting a disk is 
irreversible and any data on the disk will be lost.
 - [juju-e7c0db-1] in [us-east1-c]

Do you want to continue (Y/n)?  Y

Deleted [https://www.googleapis.com/compute/v1/projects/data-platform-testing-354909/zones/us-east1-c/instances/juju-e7c0db-1].

Model    Controller  Cloud/Region     Version    SLA          Timestamp
mymodel  gce         google/us-east1  3.5.4      unsupported  00:25:14+02:00

App         Version  Status  Scale  Charm       Channel    Rev  Exposed  Message
postgresql  14.12    active    2/3  postgresql  14/stable  468  no       

Unit           Workload  Agent  Machine  Public address  Ports     Message
postgresql/0   active    idle   0        34.148.44.51    5432/tcp  
postgresql/1   unknown   lost   1        34.23.202.220   5432/tcp  agent lost, see 'juju show-status-log postgresql/1'
postgresql/2*  active    idle   2        34.138.167.85   5432/tcp  Primary

Machine  State    Address        Inst id        Base          AZ          Message
0        started  34.148.44.51   juju-e7c0db-0  ubuntu@22.04  us-east1-d  RUNNING
1        down     34.23.202.220  juju-e7c0db-1  ubuntu@22.04  us-east1-c  RUNNING
2        started  34.138.167.85  juju-e7c0db-2  ubuntu@22.04  us-east1-b  RUNNING

Here we should remove the no-longer available server/vm/GCE node and add a new one. Juju will create it in the same AZ us-east4-c:

> juju remove-unit postgresql/1 --force --no-wait
WARNING This command will perform the following actions:
will remove unit postgresql/1

Continue [y/N]? y

The command juju status shows the machines in a healthy state, but PostgreSQL HA recovery is necessary:

Model    Controller  Cloud/Region     Version    SLA          Timestamp
mymodel  gce         google/us-east1  3.5.4      unsupported  00:30:09+02:00

App         Version  Status  Scale  Charm       Channel    Rev  Exposed  Message
postgresql  14.12    active      2  postgresql  14/stable  468  no       

Unit           Workload  Agent  Machine  Public address  Ports     Message
postgresql/0   active    idle   0        34.148.44.51    5432/tcp  
postgresql/2*  active    idle   2        34.138.167.85   5432/tcp  Primary

Machine  State    Address        Inst id        Base          AZ          Message
0        started  34.148.44.51   juju-e7c0db-0  ubuntu@22.04  us-east1-d  RUNNING
2        started  34.138.167.85  juju-e7c0db-2  ubuntu@22.04  us-east1-b  RUNNING

Request Juju to add a new unit in the proper AZ:

juju add-unit postgresql -n 1

Juju uses the right AZ where the node is missing. Run juju status:

Model    Controller  Cloud/Region     Version    SLA          Timestamp
mymodel  gce         google/us-east1  3.5.4      unsupported  00:30:42+02:00

App         Version  Status  Scale  Charm       Channel    Rev  Exposed  Message
postgresql           active    2/3  postgresql  14/stable  468  no       

Unit           Workload  Agent       Machine  Public address  Ports     Message
postgresql/0   active    idle        0        34.148.44.51    5432/tcp  
postgresql/2*  active    idle        2        34.138.167.85   5432/tcp  Primary
postgresql/3   waiting   allocating  3                                  waiting for machine

Machine  State    Address        Inst id        Base          AZ          Message
0        started  34.148.44.51   juju-e7c0db-0  ubuntu@22.04  us-east1-d  RUNNING
2        started  34.138.167.85  juju-e7c0db-2  ubuntu@22.04  us-east1-b  RUNNING
3        pending                 juju-e7c0db-3  ubuntu@22.04  us-east1-c  starting

Remove GCE setup

Check the list of currently running GCE instances:

> gcloud compute instances list
NAME           ZONE        MACHINE_TYPE   PREEMPTIBLE  INTERNAL_IP  EXTERNAL_IP    STATUS
juju-a82dd9-0  us-east1-b  n1-highcpu-4                10.142.0.30  34.23.252.144  RUNNING
juju-e7c0db-2  us-east1-b  n2-highcpu-2                10.142.0.32  34.138.167.85  RUNNING
juju-e7c0db-3  us-east1-c  n2d-highcpu-2               10.142.0.34  34.23.202.220  RUNNING
juju-e7c0db-0  us-east1-d  n2-highcpu-2                10.142.0.31  34.148.44.51   RUNNING

Request Juju to clean all GCE resources:

juju destroy-controller gce --no-prompt --force --destroy-all-models

Re-check that there are no running GCE instances left (it should be empty):

gcloud compute instances list