M1 v23
A Postgres cluster with one or more active locations, each with the same number of Postgres nodes and an extra Barman node. Optionally, there can also be a location containing only a witness node, or a location containing only a single node, even if the active locations have more than one.
This architecture is suitable for production and is also suited to testing, demonstrating and learning due to its simplicity and ability to be configured with no proprietary components.
If you select subscription-only EDB software with this architecture it will be sourced from EDB Repos 2.0 and you will need to provide a token. See How TPA uses 2ndQuadrant and EDB repositories for more detail on this topic.
Application and backup failover
The M1 architecture implements failover management in that it ensures that a replica will be promoted to take the place of the primary should the primary become unavailable. However it does not provide any automatic facility to reroute application traffic to the primary. If you require, automatic failover of application traffic you will need to configure this at the application itself (for example using multi-host connections) or by using an appropriate proxy or load balancer and the facilities offered by your selected failover manager.
The above is also true of the connection between the backup node and the
primary created by TPA. The backup will not be automatically adjusted to
target the new primary in the event of failover, instead it will remain
connected to the original primary. If you are performing a manual
failover and wish to connect the backup to the new primary, you may
simply re-run tpaexec deploy
. If you wish to automatically change the
backup source, you should implement this using your selected failover
manager as noted above.
Cluster configuration
Overview of configuration options
An example invocation of tpaexec configure
for this architecture
is shown below.
You can list all available options using the help command.
The tables below describe the mandatory options for M1 and additional important options. More detail on the options is provided in the following section.
Mandatory Options
Parameter | Description |
---|---|
--architecture (-a ) | Must be set to M1 . |
Postgres flavour and version (e.g. --postgresql 15 ) | A valid flavour and version specifier. |
One of: --failover-manager {efm, repmgr, patroni} --enable-efm --enable-repmgr --enable-patroni | Select the failover manager from efm , repmgr and patroni . |
Additional Options
Parameter | Description | Behaviour if omitted |
---|---|---|
--platform | One of aws , docker , bare . | Defaults to aws . |
--location-names | A space-separated list of location names. The number of active locations is equal to the number of names supplied, minus one for each of the witness-only location and the single-node location if they are requested. | A single location called "main" is used. |
--primary-location | The location where the primary server will be. Must be a member of location-names . | The first listed location is used. |
--data-nodes-per-location | A number from 1 upwards. In each location, one node will be configured to stream directly from the cluster's primary node, and the other nodes, if present, will stream from that one. | Defaults to 2. |
--witness-only-location | A location name, must be a member of location-names . | No witness-only location is added. |
--single-node-location | A location name, must be a member of location-names . | No single-node location is added. |
--enable-haproxy | 2 additional nodes will be added as a load balancer layer. Only supported with Patroni as the failover manager. | HAproxy nodes will not be added to the cluster. |
--patroni-dcs | Select the Distributed Configuration Store backend for patroni. Only option is etcd at this time. Only supported with Patroni as the failover manager. | Defaults to etcd . |
More detail about M1 configuration
You may also specify any of the options described by
tpaexec help configure-options
.