xcat-core/xCAT-OpenStack/chef-cookbooks/grizzly-xcat/cookbooks/postgresql/README.md

Description

Installs and configures PostgreSQL as a client or a server.

Requirements

Platforms

• Debian, Ubuntu
• Red Hat/CentOS/Scientific (6.0+ required) - "EL6-family"
• Fedora
• SUSE

Tested on:

• Ubuntu 10.04, 11.10, 12.04
• Red Hat 6.1, Scientific 6.1, CentOS 6.3

Cookbooks

Requires Opscode's openssl cookbook for secure password generation.

Requires a C compiler and development headers in order to build the pg RubyGem to provide Ruby bindings in the ruby recipe.

Opscode's build-essential cookbook provides this functionality on Debian, Ubuntu, and EL6-family.

While not required, Opscode's database cookbook contains resources and providers that can interact with a PostgreSQL database. This cookbook is a dependency of database.

Attributes

The following attributes are set based on the platform, see the attributes/default.rb file for default values.

• node['postgresql']['version'] - version of postgresql to manage

• node['postgresql']['dir'] - home directory of where postgresql data and configuration lives.

• node['postgresql']['client']['packages'] - An array of package names that should be installed on "client" systems.

• node['postgresql']['server']['packages'] - An array of package names that should be installed on "server" systems.

• node['postgresql']['contrib']['packages'] - An array of package names that could be installed on "server" systems for useful sysadmin tools.

• node['postgresql']['enable_pgdg_apt'] - Whether to enable the apt repo by the PostgreSQL Global Development Group, which contains newer versions of PostgreSQL.

• node['postgresql']['enable_pgdg_yum'] - Whether to enable the yum repo by the PostgreSQL Global Development Group, which contains newer versions of PostgreSQL.

The following attributes are generated in recipe[postgresql::server].

• node['postgresql']['password']['postgres'] - randomly generated password by the openssl cookbook's library. (TODO: This is broken, as it disables the password.)

Configuration

The postgresql.conf and pg_hba.conf files are dynamically generated from attributes. Each key in node['postgresql']['config'] is a postgresql configuration directive, and will be rendered in the config file. For example, the attribute:

node['postgresql']['config']['listen_address'] = 'localhost'

Will result in the following line in the postgresql.conf file:

listen_address = 'localhost'

The attributes file contains default values for Debian and RHEL platform families (per the node['platform_family']). These defaults have disparity between the platforms because they were originally extracted from the postgresql.conf files in the previous version of this cookbook, which differed in their default config. The resulting configuration files will be the same as before, but the content will be dynamically rendered from the attributes. The helpful commentary will no longer be present. You should consult the PostgreSQL documentation for specific configuration details.

See Recipes config_initdb and config_pgtune below to auto-generate many postgresql.conf settings.

For values that are "on" or "off", they should be specified as literal true or false. String values will be used with single quotes. Any configuration option set to the literal nil will be skipped entirely. All other values (e.g., numeric literals) will be used as is. So for example:

node.default['postgresql']['config']['logging_collector'] = true
node.default['postgresql']['config']['datestyle'] = 'iso, mdy'
node.default['postgresql']['config']['ident_file'] = nil
node.default['postgresql']['config']['port] = 5432

Will result in the following config lines:

logging_collector = 'on'
datestyle = 'iso,mdy'
port = 5432

(no line printed for ident_file as it is nil)

The pg_hba.conf file is dynamically generated from the node['postgresql']['pg_hba'] attribute. This attribute must be an array of hashes, each hash containing the authorization data. As it is an array, you can append to it in your own recipes. The hash keys in the array must be symbols. Each hash will be written as a line in pg_hba.conf. For example, this entry from node['postgresql']['pg_hba']:

{:comment => '# Optional comment',
:type => 'local', :db => 'all', :user => 'postgres', :addr => nil, :method => 'md5'}

Will result in the following line in pg_hba.conf:

# Optional comment
local   all             postgres                                md5

Use nil if the CIDR-ADDRESS should be empty (as above). Don't provide a comment if none is desired in the pg_hba.conf file.

Note that the following authorization rule is supplied automatically by the cookbook template. The cookbook needs this to execute SQL in the PostgreSQL server without supplying the clear-text password (which isn't known by the cookbook). Therefore, your node['postgresql']['pg_hba'] attributes don't need to specify this authorization rule:

# "local" is for Unix domain socket connections only
local   all             all                                     ident

(By the way, the template uses peer instead of ident for PostgreSQL-9.1 and above, which has the same effect.)

Recipes

default

Includes the client recipe.

client

Installs the packages defined in the node['postgresql']['client']['packages'] attribute.

ruby

NOTE This recipe may not currently work when installing Chef with the "Omnibus" full stack installer on some platforms due to an incompatibility with OpenSSL. See COOK-1406. You can build from source into the Chef omnibus installation to work around this issue.

Install the pg gem under Chef's Ruby environment so it can be used in other recipes. The build-essential packages and postgresql client packages will be installed during the compile phase, so that the native extensions of pg can be compiled.

server

Includes the server_debian or server_redhat recipe to get the appropriate server packages installed and service managed. Also manages the configuration for the server:

• generates a strong default password (via openssl) for postgres (TODO: This is broken, as it disables the password.)
• sets the password for postgres
• manages the postgresql.conf file.
• manages the pg_hba.conf file.

server_debian

Installs the postgresql server packages and sets up the service. You should include the postgresql::server recipe, which will include this on Debian platforms.

server_redhat

Manages the postgres user and group (with UID/GID 26, per RHEL package conventions), installs the postgresql server packages, initializes the database, and manages the postgresql service. You should include the postgresql::server recipe, which will include this on RHEL/Fedora platforms.

config_initdb

Takes locale and timezone settings from the system configuration. This recipe creates node.default['postgresql']['config'] attributes that conform to the system's locale and timezone. In addition, this recipe creates the same error reporting and logging settings that initdb provided: a rotation of 7 days of log files named postgresql-Mon.log, etc.

The default attributes created by this recipe are easy to override with normal attributes because of Chef attribute precedence. For example, suppose a DBA wanted to keep log files indefinitely, rolling over daily or when growing to 10MB. The Chef installation could include the postgresql::config_initdb recipe for the locale and timezone settings, but customize the logging settings with these node JSON attributes:

"postgresql": {
  "config": {
    "log_rotation_age": "1d",
    "log_rotation_size": "10MB",
    "log_filename": "postgresql-%Y-%m-%d_%H%M%S.log"
  }
}

Credits: This postgresql::config_initdb recipe is based on algorithms in the source code for the PostgreSQL initdb utility.

config_pgtune

Performance tuning. Takes the wimpy default postgresql.conf and expands the database server to be as powerful as the hardware it's being deployed on. This recipe creates a baseline configuration of node.default['postgresql']['config'] attributes in the right general range for a dedicated Postgresql system. Most installations won't need additional performance tuning.

The only decision you need to make is to choose a db_type from the following database workloads. (See the recipe code comments for more detailed descriptions.)

• "dw" -- Data Warehouse
• "oltp" -- Online Transaction Processing
• "web" -- Web Application
• "mixed" -- Mixed DW and OLTP characteristics
• "desktop" -- Not a dedicated database

This recipe uses a performance model with three input parameters. These node attributes are completely optional, but it is obviously important to choose the db_type correctly:

• node['postgresql']['config_pgtune']['db_type'] -- Specifies database type from the list of five choices above. If not specified, the default is "mixed".

• node['postgresql']['config_pgtune']['max_connections'] -- Specifies maximum number of connections expected. If not specified, it depends on database type: "web":200, "oltp":300, "dw":20, "mixed":80, "desktop":5

• node['postgresql']['config_pgtune']['total_memory'] -- Specifies total system memory in kB. (E.g., "49416564kB".) If not specified, it will be taken from Ohai automatic attributes. This could be used to tune a system that isn't a dedicated database.

The default attributes created by this recipe are easy to override with normal attributes because of Chef attribute precedence. For example, if you are running application benchmarks to try different buffer cache sizes, you would experiment with this node JSON attribute:

"postgresql": {
  "config": {
    "shared_buffers": "3GB"
  }
}

Note that the recipe uses max_connections in its computations. If you want to override that setting, you should specify node['postgresql']['config_pgtune']['max_connections'] instead of node['postgresql']['config']['max_connections'].

Credits: This postgresql::config_pgtune recipe is based on the pgtune python script developed by Greg Smith and other pgsql-hackers.

contrib

Installs the packages defined in the node['postgresql']['contrib']['packages'] attribute. The contrib directory of the PostgreSQL distribution includes porting tools, analysis utilities, and plug-in features that database engineers often require. Some (like pgbench) are executable. Others (like pg_buffercache) would need to be installed into the database.

Also installs any contrib module extensions defined in the node['postgresql']['contrib']['extensions'] attribute. These will be available in any subsequently created databases in the cluster, because they will be installed into the template1 database using the CREATE EXTENSION command. For example, it is often necessary/helpful for problem troubleshooting and maintenance planning to install the views and functions in these [standard instrumentation extensions] (http://www.postgresql.org/message-id/flat/4DC32600.6080900@pgexperts.com#4DD3D6C6.5060006@2ndquadrant.com):

node['postgresql']['contrib']['extensions'] = [
  "pageinspect",
  "pg_buffercache",
  "pg_freespacemap",
  "pgrowlocks",
  "pg_stat_statements",
  "pgstattuple"
]

Note that the pg_stat_statements view only works if postgresql.conf loads its shared library, which can be done with this node attribute:

node['postgresql']['config']['shared_preload_libraries'] = 'pg_stat_statements'

apt_pgdg_postgresql

Enables the PostgreSQL Global Development Group yum repository maintained by Devrim Gündüz for updated PostgreSQL packages. (The PGDG is the groups that develops PostgreSQL.) Automatically included if the node['postgresql']['enable_pgdg_apt'] attribute is true. Also set the node['postgresql']['client']['packages'] and node['postgresql']['server]['packages'] to the list of packages to use from this repository, and set the node['postgresql']['version'] attribute to the version to use (e.g., "9.2").

yum_pgdg_postgresql

Enables the PostgreSQL Global Development Group yum repository maintained by Devrim Gündüz for updated PostgreSQL packages. (The PGDG is the groups that develops PostgreSQL.) Automatically included if the node['postgresql']['enable_pgdg_yum'] attribute is true. Also use override_attributes to set a number of values that will need to have embedded version numbers. For example:

node['postgresql']['enable_pgdg_yum'] = true
node['postgresql']['version'] = "9.2"
node['postgresql']['dir'] = "/var/lib/pgsql/9.2/data"
node['postgresql']['client']['packages'] = ["postgresql92"]
node['postgresql']['server']['packages'] = ["postgresql92-server"]
node['postgresql']['server']['service_name'] = "postgresql-9.2"
node['postgresql']['contrib']['packages'] = ["postgresql92-contrib"]

You may set node['postgresql']['pgdg']['repo_rpm_url'] attributes to pick up recent PGDG repo packages.

Resources/Providers

See the database for resources and providers that can be used for managing PostgreSQL users and databases.

Usage

On systems that need to connect to a PostgreSQL database, add to a run list recipe[postgresql] or recipe[postgresql::client].

On systems that should be PostgreSQL servers, use recipe[postgresql::server] on a run list. This recipe does set a password for the postgres user. If you're using chef server, if the attribute node['postgresql']['password']['postgres'] is not found, the recipe generates a random password and performs a node.save. (TODO: This is broken, as it disables the password.) If you're using chef-solo, you'll need to set the attribute node['postgresql']['password']['postgres'] in your node's json_attribs file or in a role.

On Debian family systems, SSL will be enabled, as the packages on Debian/Ubuntu also generate the SSL certificates. If you use another platform and wish to use SSL in postgresql, then generate your SSL certificates and distribute them in your own cookbook, and set the node['postgresql']['config']['ssl'] attribute to true in your role/cookboook/node.

Chef Solo Note

The following node attribute is stored on the Chef Server when using chef-client. Because chef-solo does not connect to a server or save the node object at all, to have the password persist across chef-solo runs, you must specify them in the json_attribs file used. For Example:

{
  "postgresql": {
    "password": {
      "postgres": "iloverandompasswordsbutthiswilldo"
    }
  },
  "run_list": ["recipe[postgresql::server]"]
}

That should actually be the "encrypted password" instead of cleartext, so you should generate it as an md5 hash using the PostgreSQL algorithm.

• You could copy the md5-hashed password from an existing postgres database if you have postgres access and want to use the same password:
  select * from pg_shadow where usename='postgres';
• You can run this from any postgres database session to use a new password:
  select 'md5'||md5('iloverandompasswordsbutthiswilldo'||'postgres');
• You can run this from a linux commandline:
  echo -n 'iloverandompasswordsbutthiswilldo''postgres' | openssl md5 | sed -e 's/.* /md5/'

License and Author

• Author:: Joshua Timberman (joshua@opscode.com)
• Author:: Lamont Granquist (lamont@opscode.com)
• Author:: Chris Roberts (chrisroberts.code@gmail.com)
• Author:: David Crane (davidc@donorschoose.org)

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.