This is a collection of playbooks I hacked together, to set up a Greenplum
test cluster, using the sources directly from the GIT repository.

Usage
=====

To go from scratch to a running cluster:

0. Set ANSIBLE_CONFIG
---------------------

To save some typing, first do:

  export ANSIBLE_CONFIG=`pwd`/ansible.cfg

1. Set up a servers
-------------------

I used Centos 7. To test locally, you can use the scripts in
virt-manager-scripts directory to create a bunch of local VMs. Or you can
use e.g. Amazon AWS.

If you're using AWS, make sure the Security Group settings allow the
servers can communicate freely with each other.

2. Edit Ansible config
----------------------

Open "ansible-hosts" in a text editor. Enter the IP addresses of the master
and segment hosts in "ansible-hosts" file.

3. Set up SSH keys

Copy SSH key needed to log in to the servers to the same directory, in a
file called "gpdb-ssh-key.pem". Put the corresponding public key in
gpdb-ssh-key.pub

For AWS, the SSH key is the key you created when you launched the instances.
Use "ssh-keygen -y -f gpdb-ssh-key.pem > gpdb-ssh-key.pub" to extract the
public key.

Note: The private key is copied to the master VM, to allow connecting from
master to segments without password. Don't use a valuable key that you use
for other purposes, best to create a new one.

3. Set up the cluster

Run:

ansible-playbook ansible-preparehosts.yml
ansible-playbook ansible-compile.yml
ansible-playbook ansible-distribute-binaries.yml
ansible-playbook ansible-initcluster.yml

If all went well, the cluster is now ready, and you can connect to the
master on port 5432 with psql. However, the 'gpadmin' database user has no
password set up. You can run psql as gpadmin on the master, which doesn't
require a password, or you can set up a password with:

ansible -u gpadmin gpdb-test-master -a "psql postgres -c \"alter user gpadmin password 'gpdbtestpw' \""




Playbook descriptions
=====================

ansible-preparehosts.yml, ansible-compile.yml and
ansible-distribute-binaries.yml are idempotent, which means that you can
run and re-run them repeatedly. If one of them fails, you can fix it, and
try again.

ansible-initcluster.yml runs "gpinitsystem", so it's not idempotent. You can
run ansible-delcluster.yml first and then retry, however.

ansible-preparehosts.yml
------------------------
Sets up gpadmin user accounts, firewall rules, hosts files, and other such
OS-level setup of the hosts. This is the first playbook to run on host systems.


ansible-compile.yml
-------------------
Clones the GPDB repository from github, compiles it, and installs to master.

Alternatively, you download and run the official binary installers

ansible-distribute-binaries.yml
-------------------------------

Copies /usr/local/gpdb directory from master to all segments. Use this after
compiling the binaries on master.

There's a built-in script in Greenplum, gpseginstall, that does the same,
but I found this easier.

ansible-initcluster.yml
-----------------------

Initializes a new cluster, with gpinitsystem

ansible-delcluster.yml
-----------------------

Forcibly stops all running GPDB processes, and removes the data directories
on all hosts.