Skip to main content
typo
Source Link
Erwin Brandstetter
  • 182.2k
  • 28
  • 457
  • 620

In our shop we selected repmgr and pgbouncer instead of pgpool. repmgr has some nice tooling to setup and maintain the cluster of replicated database servers. In our case 1 master and 2 slaves (one failover and one live read performance test that can become the failover of the new master). pgpool has issues with changes in the configuration, in most cases you have to restart the service and therefor you have some downtime. This is a problem when you need 24x7x365 availability.

repmgrd (the deamon) helps to select the new master after a failover, you really don't want a split brain situation. We have one virtual ip-address for the master database, the database that is master at that moment. When another server becomes master, this is the only server using this address. Every database server also has it's own ip-address for read only queries.

repmrrepmgr is maintained by the same guys that created streaming replication in the first place, so they know what they talk about. Version 2.0 is about to be released.

Prepare for the worst situation, do some serious testing by pulling some power and network plugs! When something goes wrong, many other things already went wrong and will bite you in the back when you can't afford it.

Replication is one thing, a working failover after some serious problems, is another thing.

In our shop we selected repmgr and pgbouncer instead of pgpool. repmgr has some nice tooling to setup and maintain the cluster of replicated database servers. In our case 1 master and 2 slaves (one failover and one live read performance test that can become the failover of the new master). pgpool has issues with changes in the configuration, in most cases you have to restart the service and therefor you have some downtime. This is a problem when you need 24x7x365 availability.

repmgrd (the deamon) helps to select the new master after a failover, you really don't want a split brain situation. We have one virtual ip-address for the master database, the database that is master at that moment. When another server becomes master, this is the only server using this address. Every database server also has it's own ip-address for read only queries.

repmr is maintained by the same guys that created streaming replication in the first place, so they know what they talk about. Version 2.0 is about to be released.

Prepare for the worst situation, do some serious testing by pulling some power and network plugs! When something goes wrong, many other things already went wrong and will bite you in the back when you can't afford it.

Replication is one thing, a working failover after some serious problems, is another thing.

In our shop we selected repmgr and pgbouncer instead of pgpool. repmgr has some nice tooling to setup and maintain the cluster of replicated database servers. In our case 1 master and 2 slaves (one failover and one live read performance test that can become the failover of the new master). pgpool has issues with changes in the configuration, in most cases you have to restart the service and therefor you have some downtime. This is a problem when you need 24x7x365 availability.

repmgrd (the deamon) helps to select the new master after a failover, you really don't want a split brain situation. We have one virtual ip-address for the master database, the database that is master at that moment. When another server becomes master, this is the only server using this address. Every database server also has it's own ip-address for read only queries.

repmgr is maintained by the same guys that created streaming replication in the first place, so they know what they talk about. Version 2.0 is about to be released.

Prepare for the worst situation, do some serious testing by pulling some power and network plugs! When something goes wrong, many other things already went wrong and will bite you in the back when you can't afford it.

Replication is one thing, a working failover after some serious problems, is another thing.

Source Link
Frank Heikens
  • 23.8k
  • 1
  • 28
  • 20

In our shop we selected repmgr and pgbouncer instead of pgpool. repmgr has some nice tooling to setup and maintain the cluster of replicated database servers. In our case 1 master and 2 slaves (one failover and one live read performance test that can become the failover of the new master). pgpool has issues with changes in the configuration, in most cases you have to restart the service and therefor you have some downtime. This is a problem when you need 24x7x365 availability.

repmgrd (the deamon) helps to select the new master after a failover, you really don't want a split brain situation. We have one virtual ip-address for the master database, the database that is master at that moment. When another server becomes master, this is the only server using this address. Every database server also has it's own ip-address for read only queries.

repmr is maintained by the same guys that created streaming replication in the first place, so they know what they talk about. Version 2.0 is about to be released.

Prepare for the worst situation, do some serious testing by pulling some power and network plugs! When something goes wrong, many other things already went wrong and will bite you in the back when you can't afford it.

Replication is one thing, a working failover after some serious problems, is another thing.