4 This simple project aims to automate and make easy the online recovery process of a failed pgpool's backend node in master/slave mode.
6 This version is work-in-progress using Centos7 and upstream packages. It doesn't require psmisc package, making Centos7 minimal installation sufficient for scripts to run, since it uses systemd to manage postgresql-9.6 installed in /var/lib/pgsql/9.6/data/
8 Hardware configuration is 2 nodes:
10 10.200.1.61 edozvola-db-01
11 10.200.1.62 edozvola-db-02
13 Deployment script ./t/1-init-cluster.sh assumes that management machine from which it's run is 10.200.1.1
14 which is added in pg_hba.conf as authorized to be able to deploy cluster. It also assumes that management machine
15 has ssh access to nodes of clustetr using ssh keys or you will need to type passwords multiple times.
17 You can run cluster creation it with:
21 This will destroy all databases on all nodes, archive logs, etc, so don't do this if you need your old data later.
23 On the other hand this will also create setup whole cluster, and you can examine it's status using:
27 If you edited local files, push changes to all nodes using:
31 To restart pgpool (and cleanup it's state) do:
35 If you want to see systemd status of services just type:
40 Now you can start './t/80-insert-test.sh' in one terminal to create insert and select load on cluster and
41 kill one of nodes with 'echo b > /proc/sysrq-trigger'
43 For example, kill slave:
45 ssh root@10.200.1.62 'echo b > /proc/sysrq-trigger'
47 pgpool should detect broken back-end and remove it.
48 You can verify that using just 'make' and see that one node is down.
49 To issue online recovery, you can use:
53 now, try to kill master:
55 ssh root@10.200.1.61 'echo b > /proc/sysrq-trigger'
57 FIXME: pgpool is stuck and needs to be restarted
60 If installing on existing streaming replication you will need to tell pgpool where current master is with:
62 echo 0 > /tmp/postgres_master
64 You can also force re-check of nodes by removing status file and restarting pgool:
66 rm /var/log/pgpool_status
67 systemctl restart pgpool
74 There are two requirements to these scripts to work.
76 * The first one is [pgpool-II](http://www.pgpool.net) (v3.6.5) available for [Centos7 from upstream](http://www.pgpool.net/yum/rpms/3.6/redhat/rhel-7-x86_64/pgpool-II-pg96-3.6.5-1pgdg.rhel7.x86_64.rpm).
77 * The second one is obviously Postgres server (v9.6) also for [Centos7 from upstream](https://yum.postgresql.org/9.6/redhat/rhel-7-x86_64/pgdg-redhat96-9.6-3.noarch.rpm)
79 There are several tutorials about setting up pgpool2 and postgres servers with [Streaming Replication](http://wiki.postgresql.org/wiki/Streaming_Replication) and this readme is far to be a howto for configuring both of them.
81 Installation and configuration
82 ==============================
83 What about the given scripts and config files ?
85 **pgpool.conf** : This is a sample config file for pgpool that activates master/slave mode, loadbalancing, backends health check, failover, ...
87 **postgresql.conf.master** : A config file for postgres master node.
89 **postgresql.conf.slave** : A config file for postgres slave node.
91 **recovery.conf** : A config file used by postgres slave for streaming replication process.
93 **failover.sh** : This script will be executed automatically when a pgpool's backend node (postgres node) fails down. It'll switch the standby node (slave) to master (new master).
95 **online-recovery.sh** : This is the bash script which you'll execute manually in order to :
96 * Reboot, sync and reattach slave node to pgpool if it fails.
97 * Setup new master and new slave, sync and reattach them to pgpool if current master fails.
98 This script will invoque remotely the script streaming-replication.sh (in the new slave node) to start the [online recovery process](http://www.postgresql.org/docs/8.1/static/backup-online.html) within the standby node.
99 PS : When a node (master or slave) fails, pgpool still running and DBs remain available. Otherwise, pgpool will detach this node for data consistancy reasons.
101 **streaming-replication.sh** : This script can be executed manually to synchronize a slave node with a given master node (master name/ip must be passed as argument to streaming-replication.sh). Otherwise, this same script is triggred be online-recovery.sh via ssh during failback process.
106 The installation steps are simple. You just need to copy provided bash scripts and config files as follow.
109 * Copy pgpool.conf to /etc/pgpool-II/. This is an optional operation and in this case you have to edit the default pgpool.conf file in order to looks like the config file we provided.
110 * Copy failover.sh into /etc/pgpool-II/ and online-recovery.sh to same directory or another directory that will be easily accessible.
112 **In the master and slave postgres nodes** :
113 * Copy streaming-replication.sh script into /var/lib/pgsql/ (postgres homedir).
114 * Copy postgresql.conf.master and postgresql.conf.slave files to /var/lib/pgsql/9.6/data/.
115 * Finally copy recovery.conf into /var/lib/postgresql/9.1/main/.
117 PS : All similar old files must be backed up to be able to rollback in case of risk (e.g: cp -p /etc/pgpool-II/pgpool.conf /etc/pgpool-II/pgpool.conf.backup).
119 - All scripts are executable and owned by the proper users.
120 - /var/lib/pgsql/9.6/archive directory is created (used to archive WAL files). This folder must be owned by postgres user !
121 - Do not forge to edit pg_hba.conf in each postgres server to allow access to cluster's nodes.
123 Not enough ! It remains only the configuration steps and we'll be done :)
128 To do, just follow these steps :
130 1- First of all make sure you have created a postgres user in pgpool node with SSH access to all Postgres nodes. All cluster's nodes have to be able to ssh each other. You can put "config" file with "StrictHostKeyChecking=no" option under .ssh/ directory of postgres user. This is a best practice (essencially when automating a bunch of operations) that allows postgres to ssh remote machine for the first time without prompting and validating Yes/No authorization question.
132 2- In Pgpool node set up pgpool.conf file for instance the parameters :
134 # Controls various backend behavior for instance master and slave(s).
135 backend_hostname0='master.foo.bar'
138 backend_data_directory0 = '/var/lib/postgres/9.1/main/'
139 backend_flag0 = 'ALLOW_TO_FAILOVER'
140 backend_hostname1='slave.foo.bar'
143 backend_data_directory1 = '/var/lib/postgres/9.1/main/'
144 backend_flag1 = 'ALLOW_TO_FAILOVER'
146 num_init_children = 32
148 # Master/Slave and load balancing (replication mode must be off)
149 load_balance_mode = on
150 master_slave_mode = on
151 master_slave_sub_mode = 'stream'
152 #Health check (must be set up to detecte postgres server status up/down)
153 health_check_period = 30
154 health_check_user = 'postgres'
155 health_check_password = 'postgrespass'
156 # - Special commands -
157 follow_master_command = 'echo %M > /tmp/postgres_master'
159 failover_command = '/path/to/failover.sh %d %H %P /tmp/trigger_file'
161 3- In failover.sh script, specify the proper ssh private key to postgres user to access new master node via SSH.
163 ssh -i /var/lib/postgresql/.ssh/id_rsa postgres@$new_master "touch $trigger_file"
165 4- Idem for online-recovery.sh you have juste to change if needed the postgres's private key, the rest of params is set automatically when the script runs. Magic hein ! :)
167 5- Change the primary_conninfo access parameters (to master) in recovery.conf file in slave side :
169 primary_conninfo = 'host=master-or-slave.foo.bar port=5432 user=postgres password=nopass'
171 6- Rename recovery.conf to recovery.done in master side.
173 7- Setup postgres master node (after backup of postgresql.conf) :
175 cp -p postgresql.conf.master postgresql.conf
176 /etc/init.d/postgresql restart
178 8- Setup postgres slave node (after backup of postgresql.conf) :
180 cp -p postgresql.conf.slave postgresql.conf
182 9- Start first slave synchronisation with master by executing streaming-replication.sh as postgres user :
186 ./streaming-replication.sh master.foo.bar
190 /etc/init.d/pgpool2 restart
192 At his stage slave node is connected to master and both of them are connected to pgpool. If the master fails down, pgpool detach it from the pool and perform failover process (slave become master) automatically.
197 Test PCP interface (as root) :
199 #retrieves the node information
200 pcp_node_info 10 localhost 9898 postgres "postgres-pass" "postgres-id"
201 #detaches a node from pgpool
202 pcp_detach_node 10 localhost 9898 postgres "postgres-pass" "postgres-id"
203 #attaches a node to pgpool
204 pcp_attach_node 10 localhost 9898 postgres "postgres-pass" "postgres-id"
206 After starting pgpool, try to test this two scenarios :
208 **1. When a slave fails down** :
210 Open pgpool log file 'journalctl -u pgpool -f'
212 Stop slave node 'sudo systemctl stop postgresql-9.6'
214 After exceeding health_check_period, you should see this log message :
216 [INFO] Slave node is down. Failover not triggred !
218 Now, start slave failback process (as root) :
220 # ./online-recovery.sh
222 **2. When a master fails down** :
224 Idem, open pgpool log file.
226 Stop master node '/etc/init.d/postgres stop'.
228 After exceeding health_check_period, you should see this log message :
230 [INFO] Master node is down. Performing failover...
232 Start failback process (as root) to switch master(new slave) and slave(new master) roles :
234 # ./online-recovery.sh