ServiceManager Guide

This part of the guide is intended to cover the installation and configuration of gCube services that are not mentioned in the Administration guide. Mainly we refer to services that are not Enabling and that can be installed dynamically by the Infrastructure/VO Managers. The list includes also for each component known issues and specific configuration steps to follow.

Search

The installation of a Search Node in gCube is characterised by the installation of 2 web-services ( in the minimal configuration ) :

• SearchSystemService
• ExecutionEngineService

This is the minimal installation scenario but it's possible to enable distributed search as well and this will required the installation and configuration of several ExecutionEngineServices

HW requirements

The minimal installation requirements for a Search node are a Single CPU node with 2GB RAMm but it's more than recommended to have at least 3GB RAM on the node dedicated to the GHN.

Configuration

The SearchSystemService and ExecutionEngineService have to be automatically/manually deployed in a VRE scope. In addition if we want to configure the SearchSystemService to exploit the local ExecutionEngineService to run the queries ( minimal installation) we should configure the jndi service as follows:

• excludeLocal = false
• collocationThreshold = 0.3f
• complexPlanNumNodes = 800000

Known Issues

• The dynamic co-deployment of ExecutionEngine and Search service carries 2 RR configuration jars ( the one for search and the other for ExecutionEngine). The RR configuration for the ExecutionEngine should be removed manually.

• the rrcongprovider-service component is not shipeed with the Search service ( issues with circular dependencies), therefore must be installed manually.

• the search node should run at VRE level, manually we should add the VRE scope to the GHNProfile startScopes property.

Executor and GenericWorker

HW requirements

The minimal installation requirements for an Executor node with a Generic Worker plugin are a Single CPU node with 2GB RAMm but it's more than recommended to have at least 3GB RAM on the node dedicated to the GHN.

Configuration

The following Software should be installed on the VM:

• R version 2.14.1

whit the following components

• coda
• R2jags
• R2WinBUGS
• rjags
• bayesmix
• runjags

Known Issues

• The GenericWorker is exploited by the Statistical Manager service to run distributed computations. Given that the SM use the root scope to discover instances of the GenericWorker. the plugin must be deployed at root scope level

• Given that the GenericWorker plugin depends on the Executor Service, when dynamically deploying the plugin the Executor Service is also deployed.

DTS

HW requirements

The minimal installation requirements for an DTS node with a Generic Worker plugin are a Single CPU node with 2GB RAMm but it's more than recommended to have at least 3GB RAM on the node dedicated to the GHN.

Configuration

DTS uses Execution Engine to run the transformations so at least one Execution Engine should be deployed in the same scope as DTS and the related GHNLabels.xml file should contain:

<Variable>
      <Key>dts.execution</Key>
      <Value>true</Value>
</Variable>

Known Issues

none

ForwardIndexNode

The ForwardIndexNode service needs to be codeployed with an instance of CouchBase service

HW requirements

Given codeployment with Couchbase it's recommended at least a VM with 4GB RAM and 2 CPUs.

Configuration

The installation of Couchbase should be performed manually and it depends on the OS ( binary package, rpm, debs).

It's recommended to put an higher limit of the open files on the VM ( 32000 min).

The configuration for the FWIndexNode that should be customized (jndi file):

• couchBaseIP = IP of the server hosting Couchbase ( so the same as the GHN)
• couchBaseUseName = the username set when configuring Couchbase
• couchBasePassword = the password set when configuring Couchbase

Once configured it's needed to initialize Couchbase using the cb_initialize_node.sh script contained into the service configuration folder.

Known Issues

• Sometimes the cb_initialize_node.sh script fails, it could mean that there is not enough memory to inizialize the data bucket , try to reduce the value of ramQuota in the jndi file.