Cluster configuration v23
tpaexec configure command generates a YAML cluster configuration
file that is required by subsequent stages in the provision/deploy/test
This command will create a directory named
generate a configuration file named
config.yml that follows the
layout of the architecture named M1 (single primary, N replicas).
It will create a git repository in the new directory and make an initial
commit containing the generated
The command also accepts various options (some specific to the selected architecture or platform) to modify the configuration, but the defaults are sensible and intended to be usable straightaway. You are encouraged to read the generated config.yml and fine-tune the configuration to suit your needs. (Here's an overview of configuration settings that affect the deployment.)
It's possible to write config.yml entirely by hand, but it's much easier to edit the generated file.
The first argument must be the cluster directory, e.g.,
~/clusters/speedy (the cluster will be named speedy in both cases).
We recommend that you keep all your clusters in a common directory,
~/clusters in the example above.
Next, you must specify a flavour and version of Postgres to install.
The arguments above are always mandatory. The rest of the options described here may be safely omitted, as in the example above; the defaults will lead to a usable result.
tpaexec help configure-options for a list of common options.
The architecture you select determines what other options are accepted. Typically, each architecture accepts some unique options as well as the generic options described below.
For example, with M1 you can use
--num-cascaded-replicas 3 to create
a cluster with three cascaded replicas. Please consult the
documentation for an architecture for a list of options that it accepts
(or, in some cases, requires).
An architecture may or may not support a particular platform. If not, it will fail to configure the cluster.
The choice of platform affects the interpretation of certain options.
For example, if you choose aws, the arguments to
--region <region> and
must be a valid
AWS region name
EC2 instance type
respectively. Please refer to the platform documentation for more details.
If you do not explicitly select a platform, the default is currently aws.
Note: TPA fully supports creating clusters with instances on
different platforms, but
tpaexec configure cannot currently generate
such a configuration. You must edit config.yml to specify multiple
--owner <name> to associate the cluster (by some
platform-specific means, e.g., AWS tags) with the name of a person
responsible for it. This is especially important for cloud platforms. By
default, the owner is set to the login name of the user running
(You may use your initials, or "Firstname Lastname", or anything else that identifies you uniquely.)
--region <region> to select a region.
This option is meaningful only for cloud platforms. The default for AWS is eu-west-1.
Note: TPA fully supports creating clusters that span multiple
tpaexec configure cannot currently generate such a
configuration. You must edit config.yml to specify multiple regions.
By default, each cluster will be configured with a number of randomly selected
/28 subnets from the CIDR range
10.33.0.0/16, depending on the selected
--network 192.168.0.0/16 to assign subnets from a different network.
Note: On AWS clusters, this corresponds to the VPC CIDR. See aws documentation for details.
--subnet-prefix 26 to assign subnets of a different size, /26 instead
of /28 in this case.
--no-shuffle-subnets to allocate subnets from the start of the
network CIDR range, without randomisation, e.g.
10.33.0.16/28 and so on.
--exclude-subnets-from <directory> to exclude subnets that are
already used in existing cluster config.yml files. You can specify this
argument multiple times for each directory.
Note: These options are not meaningful for the "bare" platform, where TPA will not alter the network configuration of existing servers.
--instance-type <type> to select an instance type.
This option is meaningful only for cloud platforms. The default for AWS is t3.micro.
--root-volume-size 64 to set the size of the root volume in
GB. (Depending on the platform, there may be a minimum size required for
the root volume.)
--postgres-volume-size <size> and
--barman-volume-size <size> options are available to set the sizes
of the Postgres and Barman volumes on those architectures and platforms
that support separate volumes for Postgres and Barman.
None of these options is meaningful for the "bare" platform, where TPA has no control over volume sizes.
tpaexec configure will randomly select as many hostnames
as it needs from a pre-approved list of several dozen names. This should
be enough for most clusters.
--hostnames-from <filename> to select hostnames from a file
with one name per line. The file must contain at least as many valid
hostnames as there are instances in your cluster. Each line may contain
an optional IP address after the name; if present, this address will be
set as the
ip_address for the corresponding instance in
--hostnames-pattern '…pattern…' to limit the selection to
lines matching an egrep pattern.
--hostnames-sorted-by="--dictionary-order" to select a sort(1)
option other than
--random-sort (which is the default).
--hostnames-unsorted to not sort hostnames at all. In this case,
they will be assigned in the order they are found in the hostnames file.
This is the default when a hostnames file is explicitly specified.
Hostnames may contain only letters (a-z), digits (0-9), and '-'. They
may be FQDNs, depending on the selected platform. Hostnames should be
in lowercase; any uppercase characters will be converted to lowercase
internally, and any references to these hostnames in config.yml (e.g.,
upstream: hostname) must use the lowercase version.
--distribution <name> to select a distribution.
The selected platform determines which distributions are available, and which one is used by default.
In general, you should be able to use "Debian", "RedHat", "Ubuntu", and "SLES" to select the right images.
This option is not meaningful for the "bare" platform, where TPA has no control over which distribution is installed.
TPA can enable any 2ndQuadrant or EDB software repository that you have access to through a subscription.
By default, it will install the 2ndQuadrant public repository (which does not need a subscription) and add on any product repositories that the architecture may require (e.g., the PGD repository).
More detailed explanation of how TPA uses 2ndQuadrant and EDB repositories is available here
--2Q-repositories source/name/maturity … or
--edb-repositories repository … to specify the complete list of
2ndQuadrant or EDB repositories to install on each instance in addition
to the 2ndQuadrant public repository.
If any EDB repositories are specified, any 2ndQuadrant ones will be ignored.
Use this option with care. TPA will configure the named repositories with no attempt to make sure the combination is appropriate.
To use these options, you must
export EDB_SUBSCRIPTION_TOKEN=xxx before you run tpaexec.
You can get a 2ndQuadrant token from the 2ndQuadrant Portal under
"Company info" in the left menu, then "Company". You can get an EDB
token from enterprisedb.com/repos.
--enable-local-repo to create a local package repository from
which to ship packages to target instances.
In environments with restricted network access, you can instead use
--use-local-repo-only to create a local repository and disable all
other package repositories on target instances, so that packages are
installed only from the local repository.
The page about Local repository support has more details.
TPA supports PostgreSQL, EDB Postgres Extended, and EDB Postgres Advanced Server (EPAS) versions 11 through 16.
You must specify both the flavour (or distribution) and major version of Postgres to install, for example:
--postgresql 14will install PostgreSQL 14
--edb-postgres-extended 15will install EDB Postgres Extended 15
--edb-postgres-advanced 15 --redwoodwill install EPAS 15 in "Redwood" mode
--edb-postgres-advanced 15 --no-redwoodwill install EPAS 15 in non-Redwood mode
If you are installing EPAS, you must specify whether it should operate
--no-redwood mode, i.e., whether to enable or
disable its Oracle compatibility features.
Installing EDB Postgres Extended or Postgres Advanced Server requires a valid EDB repository subscription.
By default, we always install the latest version of every package. This is usually the desired behaviour, but in some testing scenarios, it may be necessary to select specific package versions using any of the following options:
You may use any version specifier that apt or yum would accept.
If your version does not match, try appending a
* wildcard. This
is often necessary when the package version has an epoch qualifier
You may also specify
--extra-packages p1 p2 … or
--extra-postgres-packages p1 p2 … to install additional packages.
The former lists packages to install along with system packages, while
the latter lists packages to install later along with postgres packages.
(If you mention packages that depend on Postgres in the former list, the
installation will fail because Postgres will not yet be installed.) The
arguments are passed on to the package manager for installation without
--extra-optional-packages p1 p2 … option behaves like
--extra-packages, but it is not an error if the named packages
cannot be installed.
Please note that the use of wildcards in
*_package_version when added
config.yml, can result in unexpected updates to
installed software during
tpaexec deploy on nodes with RHEL 8 and
above (or derivative OSs which use dnf such as Rocky Linux).
When deploy runs on an existing cluster that already has packages
installed ansible may be unable to match the full package version.
For example, if the value for
bdr_package_version was set to
then ansible would not be able to match this to an installed version of
PGD, it would assume no package is installed, and it would attempt to
install the latest version available of the package with the same name
in the configured repository, e.g. 3.7.
We are aware of this limitation as an ansible dnf module bug and hope to address this in a future release of TPA.
If you specify
--install-from-source postgres, Postgres will be
built and installed from a git repository instead of installed from
2ndqpostgres instead of
postgres to build and
install 2ndQPostgres. By default, this will build the appropriate
You may use
--install-from-source 2ndqpostgres pglogical3 bdr3 to
build and install all three components from source, or just use
--install-from-source pglogical3 bdr3 to use packages for
2ndQPostgres, but build and install pglogical v3 and PGD v3 from source.
By default, this will build the
master branch of pglogical and PGD.
To build a different branch, append
:branchname to the corresponding
argument. For example
--install-from-source 2ndqpostgres:dev/xxx, or
You may not be able to install packages that depend on a package that you chose to replace with a source installation instead. For example, PGD v3 packages depend on pglogical v3 packages, so you can't install pglogical from its source repository and PGD from packages. Likewise, you can't install Postgres from source and pglogical from packages.
You may optionally specify
--overrides-from a.yml … to load one or
more YAML files with settings to merge into the generated config.yml.
Any file specified here is first expanded as a Jinja2 template, and the result is loaded as a YAML data structure, and merged recursively into the arguments used to generate config.yml (comprising architecture and platform defaults and arguments from the command-line). This process is repeated for each additional override file specified; this means that settings defined by one file will be visible to any subsequent files.
For example, your override file might contain:
These settings will augment
would otherwise be in config.yml. Settings are merged recursively, so
cluster_tags will end up containing both the default Owner tag as
some_tag. Similarly, the
will override that variable, leaving other
(if any) unaffected. In other words, you can set or override specific
subkeys in config.yml, but you can't empty or replace
or any other hash altogether.
The merging only applies to hash structures, so you cannot use this
mechanism to change the list of
instances within config.yml. It is
most useful to augment
common settings for your environment.
That said, the mechanism does not enforce any restrictions, so please exercise due caution. It is a good idea to generate two configurations with and without the overrides and diff the two config.yml files to make sure you understand the effect of all the overrides.
--tower-git-repository options to
create a cluster adapted for deployment with Ansible Tower. See Ansible
Tower for details.
By default, a git repository is created with an initial branch named
after the cluster, and a single commit is made, with the configure
options you used in the commit message. If you don't have git in your
$PATH, tpaexec will not raise an error but the repository will not be
created. To suppress creation of the git repository, use the
option. (Note that in an Ansible Tower cluster, a git repository is
required and will be created later by
tpaexec provision if it does not
Let's see what happens when we run the following command:
There is no output, so there were no errors. The cluster directory has been created and populated.
The cluster configuration is in config.yml, and its neighbours are links to architecture-specific support files that you need not interact with directly. Here's what the configuration looks like: