Deploying PostgreSQL for high availability with Patroni on Debian or Ubuntu¶

This guide provides instructions on how to set up a highly available PostgreSQL cluster with Patroni on Debian or Ubuntu.

Preconditions¶

For this setup, we will use the nodes running on Ubuntu 20.04 as the base operating system and having the following IP addresses:

Node name	Public IP address	Internal IP address
node1	157.230.42.174	10.104.0.7
node2	68.183.177.183	10.104.0.2
node3	165.22.62.167	10.104.0.8
HAProxy-demo	134.209.111.138	10.104.0.6

Note

In a production (or even non-production) setup, the PostgreSQL nodes will be within a private subnet without any public connectivity to the Internet, and the HAProxy will be in a different subnet that allows client traffic coming only from a selected IP range. To keep things simple, we have implemented this architecture in a DigitalOcean VPS environment, and each node can access the other by its internal, private IP.

Setting up hostnames in the `/etc/hosts` file¶

To make the nodes aware of each other and allow their seamless communication, resolve their hostnames to their public IP addresses. Modify the /etc/hosts file of each node as follows:

node 1	node 2	node 3
`127.0.0.1 localhost node1 10.104.0.7 node1 10.104.0.2 node2 10.104.0.8 node3`	`127.0.0.1 localhost node2 10.104.0.7 node1 10.104.0.2 node2 10.104.0.8 node3`	`127.0.0.1 localhost node3 10.104.0.7 node1 10.104.0.2 node2 10.104.0.8 node3`

The /etc/hosts file of the HAProxy-demo node looks like the following:

127.0.1.1 HAProxy-demo HAProxy-demo
127.0.0.1 localhost
10.104.0.6 HAProxy-demo
10.104.0.7 node1
10.104.0.2 node2
10.104.0.8 node3

Install Percona Distribution for PostgreSQL¶

Follow the installation instructions to install Percona Distribution for PostgreSQL on node1, node2 and node3.
Remove the data directory. Patroni requires a clean environment to initialize a new cluster. Use the following commands to stop the PostgreSQL service and then remove the data directory:

$ sudo systemctl stop postgresql
$ sudo rm -rf /var/lib/postgresql/12/main

Configure ETCD distributed store¶

The distributed configuration store helps establish a consensus among nodes during a failover and will manage the configuration for the three PostgreSQL instances. Although Patroni can work with other distributed consensus stores (i.e., Zookeeper, Consul, etc.), the most commonly used one is etcd.

The etcd cluster is first started in one node and then the subsequent nodes are added to the first node using the addcommand. The configuration is stored in the /etc/default/etcd file.

Install etcd on every PostgreSQL node using the following command:
```
$ sudo apt install etcd
```

Modify the /etc/default/etcd configuration file on each node.

On node1, add the IP address of node1 to the ETCD_INITIAL_CLUSTER parameter. The configuration file looks as follows:

ETCD_NAME=node1
ETCD_INITIAL_CLUSTER="node1=http://10.104.0.7:2380"
ETCD_INITIAL_CLUSTER_TOKEN="devops_token"
ETCD_INITIAL_CLUSTER_STATE="new"
ETCD_INITIAL_ADVERTISE_PEER_URLS="http://10.104.0.7:2380"
ETCD_DATA_DIR="/var/lib/etcd/postgresql"
ETCD_LISTEN_PEER_URLS="http://10.104.0.7:2380"
ETCD_LISTEN_CLIENT_URLS="http://10.104.0.7:2379,http://localhost:2379"
ETCD_ADVERTISE_CLIENT_URLS="http://10.104.0.7:2379"
…

On node2, add the IP addresses of both node1 and node2 to the ETCD_INITIAL_CLUSTER parameter:

ETCD_NAME=node2
ETCD_INITIAL_CLUSTER="node1=http://10.104.0.7:2380,node2=http://10.104.0.2:2380"
ETCD_INITIAL_CLUSTER_TOKEN="devops_token"
ETCD_INITIAL_CLUSTER_STATE="existing"
ETCD_INITIAL_ADVERTISE_PEER_URLS="http://10.104.0.2:2380"
ETCD_DATA_DIR="/var/lib/etcd/postgresql"
ETCD_LISTEN_PEER_URLS="http://10.104.0.2:2380"
ETCD_LISTEN_CLIENT_URLS="http://10.104.0.2:2379,http://localhost:2379"
ETCD_ADVERTISE_CLIENT_URLS="http://10.104.0.2:2379"
…

On node3, the ETCD_INITIAL_CLUSTER parameter includes the IP addresses of all three nodes:

ETCD_NAME=node3
ETCD_INITIAL_CLUSTER="node1=http://10.104.0.7:2380,node2=http://10.104.0.2:2380,node3=http://10.104.0.8:2380"
ETCD_INITIAL_CLUSTER_TOKEN="devops_token"
ETCD_INITIAL_CLUSTER_STATE="existing"
ETCD_INITIAL_ADVERTISE_PEER_URLS="http://10.104.0.8:2380"
ETCD_DATA_DIR="/var/lib/etcd/postgresql"
ETCD_LISTEN_PEER_URLS="http://10.104.0.8:2380"
ETCD_LISTEN_CLIENT_URLS="http://10.104.0.8:2379,http://localhost:2379"
ETCD_ADVERTISE_CLIENT_URLS="http://10.104.0.8:2379"
…

On node1, add node2 and node3 to the cluster using the add command:

$ sudo etcdctl member add node2 http://10.104.0.2:2380
$ sudo etcdctl member add node3 http://10.104.0.8:2380

Restart the etcd service on node2 and node3:
```
$ sudo systemctl restart etcd
```

Check the etcd cluster members.

$ sudo etcdctl member list

The output resembles the following:

21d50d7f768f153a: name=node1 peerURLs=http://10.104.0.7:2380 clientURLs=http://10.104.0.7:2379 isLeader=true
af4661d829a39112: name=node2 peerURLs=http://10.104.0.2:2380 clientURLs=http://10.104.0.2:2379 isLeader=false
e3f3c0c1d12e9097: name=node3 peerURLs=http://10.104.0.8:2380 clientURLs=http://10.104.0.8:2379 isLeader=false

Set up the watchdog service¶

The Linux kernel uses the utility called a watchdog to protect against an unresponsive system. The watchdog monitors a system for unrecoverable application errors, depleted system resources, etc., and initiates a reboot to safely return the system to a working state. The watchdog functionality is useful for servers that are intended to run without human intervention for a long time. Instead of users finding a hung server, the watchdog functionality can help maintain the service.

In this example, we will configure Softdog - a standard software implementation for watchdog that is shipped with Ubuntu 20.04.

Complete the following steps on all three PostgreSQL nodes to load and configure Softdog.

Load Softdog:

$ sudo sh -c 'echo "softdog" >> /etc/modules'

Patroni will be interacting with the watchdog service. Since Patroni is run by the postgres user, this user must have access to Softdog. To make this happen, change the ownership of the watchdog.rules file to the postgres user:
```
$ sudo sh -c 'echo "KERNEL==\"watchdog\", OWNER=\"postgres\", GROUP=\"postgres\"" >> /etc/udev/rules.d/61-watchdog.rules'
```
Remove Softdog from the blacklist.
- Find out the files where Softdog is blacklisted:
```
$ grep blacklist /lib/modprobe.d/* /etc/modprobe.d/* |grep softdog
```
In our case, modprobeis blacklisting the Softdog:
```
/lib/modprobe.d/blacklist_linux_5.4.0-73-generic.conf:blacklist softdog
```
- Remove the blacklist softdog line from the /lib/modprobe.d/blacklist_linux_5.4.0-73-generic.conf file.
- Restart the service
```
$ sudo modprobe softdog
```
- Verify the modprobe is working correctly by running the lsmodcommand:
```
$ sudo lsmod | grep softdog
```
The output will show a process identifier if it’s running.
```
softdog                16384  0
```
Check that the Softdog files under the /dev/folder are owned by the postgresuser:

$ ls -l /dev/watchdog*

crw-rw---- 1 postgres postgres  10, 130 Sep 11 12:53 /dev/watchdog
crw------- 1 root     root     245,   0 Sep 11 12:53 /dev/watchdog0

Tip

If the ownership has not been changed for any reason, run the following command to manually change it:

$ sudo chown postgres:postgres /dev/watchdog*

Configure Patroni¶

Install Patroni on every PostgreSQL node:
```
$ sudo apt install percona-patroni
```
Create the patroni.yml configuration file under the /etc/patroni directory. The file holds the default configuration values for a PostgreSQL cluster and will reflect the current cluster setup.

Add the following configuration for node1:

scope: stampede1
name: node1

restapi:
  listen: 0.0.0.0:8008
  connect_address: node1:8008

etcd:
  host: node1:2379

bootstrap:
  # this section will be written into Etcd:/<namespace>/<scope>/config after initializing new cluster
  dcs:
    ttl: 30
    loop_wait: 10
    retry_timeout: 10
    maximum_lag_on_failover: 1048576
#    primary_start_timeout: 300
#    synchronous_mode: false
    postgresql:
      use_pg_rewind: true
      use_slots: true
      parameters:
        wal_level: replica
        hot_standby: "on"
        logging_collector: 'on'
        max_wal_senders: 5
        max_replication_slots: 5
        wal_log_hints: "on"
        #archive_mode: "on"
        #archive_timeout: 600
        #archive_command: "cp -f %p /home/postgres/archived/%f"
        #recovery_conf:
        #restore_command: cp /home/postgres/archived/%f %p

  # some desired options for 'initdb'
  initdb:  # Note: It needs to be a list (some options need values, others are switches)
  - encoding: UTF8
  - data-checksums

  pg_hba:  # Add following lines to pg_hba.conf after running 'initdb'
  - host all all 10.104.0.7/32 md5
  - host replication replicator 127.0.0.1/32 trust
  - host all all 10.104.0.2/32 md5  
  - host all all 10.104.0.8/32 md5 
  - host all all 10.104.0.6/32 trust
#  - hostssl all all 0.0.0.0/0 md5

  # Additional script to be launched after initial cluster creation (will be passed the connection URL as parameter)
# post_init: /usr/local/bin/setup_cluster.sh
  # Some additional users users which needs to be created after initializing new cluster
  users:
    admin:
      password: admin
      options:
        - createrole
        - createdb
    replicator: 
      password: password    
      options:
        - replication 
postgresql:
  listen: 0.0.0.0:5432
  connect_address: node1:5432
  data_dir: "/var/lib/postgresql/12/main"
  bin_dir: "/usr/lib/postgresql/12/bin"
#  config_dir:
  pgpass: /tmp/pgpass0
  authentication:
    replication:
      username: replicator
      password: password
    superuser:
      username: postgres
      password: password
  parameters:
    unix_socket_directories: '/var/run/postgresql'

watchdog:
  mode: required # Allowed values: off, automatic, required
  device: /dev/watchdog
  safety_margin: 5

tags:
    nofailover: false
    noloadbalance: false
    clonefrom: false
    nosync: false

Patroni configuration file

Let’s take a moment to understand the contents of the patroni.yml file.

The first section provides the details of the first node (node1) and its connection ports. After that, we have the etcd service and its port details.

Following these, there is a bootstrap section that contains the PostgreSQL configurations and the steps to run once the database is initialized. The pg_hba.conf entries specify all the other nodes that can connect to this node and their authentication mechanism.

Create the configuration files for node2 and node3. Replace the reference to node1 with node2 and node3, respectively.
Enable and restart the patroni service on every node. Use the following commands:
```
$ sudo systemctl enable patroni
$ sudo systemctl restart patroni
```

When Patroni starts, it initializes PostgreSQL (because the service is not currently running and the data directory is empty) following the directives in the bootstrap section of the configuration file.

Troubleshooting Patroni

To ensure that Patroni has started properly, check the logs using the following command:

$ sudo journalctl -u patroni.service -n 100 -f

The output shouldn’t show any errors:

…

Sep 23 12:50:21 node01 systemd[1]: Started PostgreSQL high-availability manager.
Sep 23 12:50:22 node01 patroni[10119]: 2021-09-23 12:50:22,022 INFO: Selected new etcd server http://10.104.0.2:2379
Sep 23 12:50:22 node01 patroni[10119]: 2021-09-23 12:50:22,029 INFO: No PostgreSQL configuration items changed, nothing to reload.
Sep 23 12:50:22 node01 patroni[10119]: 2021-09-23 12:50:22,168 INFO: Lock owner: None; I am node1
Sep 23 12:50:22 node01 patroni[10119]: 2021-09-23 12:50:22,177 INFO: trying to bootstrap a new cluster
Sep 23 12:50:22 node01 patroni[10140]: The files belonging to this database system will be owned by user "postgres".
Sep 23 12:50:22 node01 patroni[10140]: This user must also own the server process.
Sep 23 12:50:22 node01 patroni[10140]: The database cluster will be initialized with locale "C.UTF-8".
Sep 23 12:50:22 node01 patroni[10140]: The default text search configuration will be set to "english".
Sep 23 12:50:22 node01 patroni[10140]: Data page checksums are enabled.
Sep 23 12:50:22 node01 patroni[10140]: creating directory /var/lib/postgresql/12/main ... ok
Sep 23 12:50:22 node01 patroni[10140]: creating subdirectories ... ok
Sep 23 12:50:22 node01 patroni[10140]: selecting dynamic shared memory implementation ... posix
Sep 23 12:50:22 node01 patroni[10140]: selecting default max_connections ... 100
Sep 23 12:50:22 node01 patroni[10140]: selecting default shared_buffers ... 128MB
Sep 23 12:50:22 node01 patroni[10140]: selecting default time zone ... Etc/UTC
Sep 23 12:50:22 node01 patroni[10140]: creating configuration files ... ok
Sep 23 12:50:22 node01 patroni[10140]: running bootstrap script ... ok
Sep 23 12:50:23 node01 patroni[10140]: performing post-bootstrap initialization ... ok
Sep 23 12:50:23 node01 patroni[10140]: syncing data to disk ... ok
Sep 23 12:50:23 node01 patroni[10140]: initdb: warning: enabling "trust" authentication for local connections
Sep 23 12:50:23 node01 patroni[10140]: You can change this by editing pg_hba.conf or using the option -A, or
Sep 23 12:50:23 node01 patroni[10140]: --auth-local and --auth-host, the next time you run initdb.
Sep 23 12:50:23 node01 patroni[10140]: Success. You can now start the database server using:
Sep 23 12:50:23 node01 patroni[10140]:     /usr/lib/postgresql/12/bin/pg_ctl -D /var/lib/postgresql/12/main -l logfile start
Sep 23 12:50:23 node01 patroni[10156]: 2021-09-23 12:50:23.672 UTC [10156] LOG:  redirecting log output to logging collector process
Sep 23 12:50:23 node01 patroni[10156]: 2021-09-23 12:50:23.672 UTC [10156] HINT:  Future log output will appear in directory "log".
Sep 23 12:50:23 node01 patroni[10119]: 2021-09-23 12:50:23,694 INFO: postprimary pid=10156
Sep 23 12:50:23 node01 patroni[10165]: localhost:5432 - accepting connections
Sep 23 12:50:23 node01 patroni[10167]: localhost:5432 - accepting connections
Sep 23 12:50:23 node01 patroni[10119]: 2021-09-23 12:50:23,743 INFO: establishing a new patroni connection to the postgres cluster
Sep 23 12:50:23 node01 patroni[10119]: 2021-09-23 12:50:23,757 INFO: running post_bootstrap
Sep 23 12:50:23 node01 patroni[10119]: 2021-09-23 12:50:23,767 INFO: Software Watchdog activated with 25 second timeout, timing slack 15 seconds
Sep 23 12:50:23 node01 patroni[10119]: 2021-09-23 12:50:23,793 INFO: initialized a new cluster
Sep 23 12:50:33 node01 patroni[10119]: 2021-09-23 12:50:33,810 INFO: no action. I am (node1) the leader with the lock
Sep 23 12:50:33 node01 patroni[10119]: 2021-09-23 12:50:33,899 INFO: no action. I am (node1) the leader with the lock
Sep 23 12:50:43 node01 patroni[10119]: 2021-09-23 12:50:43,898 INFO: no action. I am (node1) the leader with the lock
Sep 23 12:50:53 node01 patroni[10119]: 2021-09-23 12:50:53,894 INFO: no action. I am (node1) the leader with the

A common error is Patroni complaining about the lack of proper entries in the pg_hba.conf file. If you see such errors, you must manually add or fix the entries in that file and then restart the service.

Changing the patroni.yml file and restarting the service will not have any effect here because the bootstrap section specifies the configuration to apply when PostgreSQL is first started in the node. It will not repeat the process even if the Patroni configuration file is modified and the service is restarted.

If Patroni has started properly, you should be able to locally connect to a PostgreSQL node using the following command:

$ sudo psql -U postgres

The command output should be similar to the following

psql (12.8)
Type "help" for help.

postgres=#

Configure HAProxy¶

HAProxy node will accept client connection requests and route those to the active node of the PostgreSQL cluster. This way, a client application doesn’t have to know what node in the underlying cluster is the current primary. All it needs to do is to access a single HAProxy URL and send its read/write requests there. Behind-the-scene, HAProxy routes the connection to a healthy node (as long as there is at least one healthy node available) and ensures that client application requests are never rejected.

HAProxy is capable of routing write requests to the primary node and read requests - to the secondaries in a round-robin fashion so that no secondary instance is unnecessarily loaded. To make this happen, provide different ports in the HAProxy configuration file. In this deployment, writes are routed to port 5000 and reads - to port 5001.

Install HAProxy on the HAProxy-demo node:
```
$ sudo apt install haproxy
```

The HAProxy configuration file path is: /etc/haproxy/haproxy.cfg. Specify the following configuration in this file.

global
    maxconn 100

defaults
    log global
    mode tcp
    retries 2
    timeout client 30m
    timeout connect 4s
    timeout server 30m
    timeout check 5s

listen stats
    mode http
    bind *:7000
    stats enable
    stats uri /

listen primary
    bind *:5000
    option httpchk /primary 
    http-check expect status 200
    default-server inter 3s fall 3 rise 2 on-marked-down shutdown-sessions
    server node1 node1:5432 maxconn 100 check port 8008
    server node2 node2:5432 maxconn 100 check port 8008
    server node3 node3:5432 maxconn 100 check port 8008

listen standbys
    balance roundrobin
    bind *:5001
    option httpchk /replica 
    http-check expect status 200
    default-server inter 3s fall 3 rise 2 on-marked-down shutdown-sessions
    server node1 node1:5432 maxconn 100 check port 8008
    server node2 node2:5432 maxconn 100 check port 8008
    server node3 node3:5432 maxconn 100 check port 8008

HAProxy will use the REST APIs hosted by Patroni to check the health status of each PostgreSQL node and route the requests appropriately.

Restart HAProxy:
```
$ sudo systemctl restart haproxy
```

Check the HAProxy logs to see if there are any errors:

$ sudo journalctl -u haproxy.service -n 100 -f

Testing¶

See the Testing PostgreSQL cluster for the guidelines on how to test your PostgreSQL cluster for replication, failure, switchover.

Contact Us

For free technical help, visit the Percona Community Forum.

To report bugs or submit feature requests, open a JIRA ticket.

For paid support and managed or consulting services , contact Percona Sales.

Last update: December 5, 2022
Created: December 15, 2021