monasca-vagrant/utils/cluster
Tim Kuhlman cd13fb8d26 noted errors with the cluster setup on a bleeding edge hLinux box 2014-06-30 14:27:19 -06:00
..
data_bags Updating data bags for new Keystone URL 2014-06-30 08:59:01 -06:00
roles The storm cookbook is behaving differently on chef 11 2014-06-30 14:18:35 -06:00
README.md noted errors with the cluster setup on a bleeding edge hLinux box 2014-06-30 14:27:19 -06:00
__init__.py Moved code out of __init__.py 2014-06-30 13:48:59 -06:00
setup.py Moved code out of __init__.py 2014-06-30 13:48:59 -06:00

README.md

Setup of a test cluster

The goal of this fabric script is to setup a test cluster on baremetal leveraging some tools from mini-mon.

##Steps

  • Before running first setup the following settings for your test cluster:
    • keystone host and project_id in data_bags/mon_agent/mon_agent.json
    • keystone host(serverVip) in data_bags/mon_api/mon_credentials.json
    • wsrep address in the Mon-Node role
    • servers in data_bags/zookeeper/mon.json
    • servers in data_bags/kafka/mon.json
    • vertica data bags in data_bags/vertica
      • ssh_key.json with two fields, public/private corresponding to public/private ssh keys
      • nodes data bag
  • From the utils directory (or specifying that fabfile) start the install script
    • fab cluster.setup -H host1,host2,host3
    • 3 hosts is the minimum needed for a fully HA cluster.
    • All 3 machines will run the Mon-Node role, host1 will also run the Thresh-Nimbus role and remaining hosts the Thresh-Supervisor role
  • After the cluster is up and running some additional configuration steps are needed to setup kafka topics and databases.
    • On one of the machines run the recipes:
      • kafka::create_topics
      • mini-mon::mysql_schema
      • vertica::create_db

Optional Configuration

  • Add in the Vertica Console to one of the machines. This can be done with vertica::console recipe

Known Issues

  • runit used for the storm cookbook needs a newer version for chef 11 and even the newer version has errors on hLinux.
  • If a client connects to kafka before the topics are made in some situations the topics are automatically created and can end up as the wrong size or without replicas.
  • The percona cluster can be fickle regarding how to start it up properly this at times interferes with the cookbook. A workaround is to initially set the wsrep_cluster_address to just gcomm://
  • Various cookbooks were pinned to older versions to accomodate the old chef used in the precise vagrant image, these can be unpinned.