[jira] [Commented] (ACE-169) Refactor the management agent so support running multiple instances

["Marcel Offermans (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/ACE-169?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13821292#comment-13821292 ] 

Marcel Offermans commented on ACE-169:
--------------------------------------

With the complete rewrite of the management agent, this issue needs some new thought.

The original description already highlights the dangers of taking the route of using multiple, isolated deployment packages. This has not changed. However, our new agent is much more "configurable" and even though we don't directly support instantiating it more than once, such a feature can quite easily added by using for example multi-tenancy (multiple, isolated deployment packages coming from different sources sounds like a typical use case for a multi-tenant setup of an agent). Whilst ACE itself provides no way to run a multi-tenant agent, the Amdatu project does. A description for that module can be found here: http://amdatu.org/components/multitenancy.html

I therefore propose to resolve this issue, and its related sub-issues in a way that enables others to run the agent in a multi-tenant way, but not implementing this in the ACE codebase itself.

> Refactor the management agent so support running multiple instances
> -------------------------------------------------------------------
>
>                 Key: ACE-169
>                 URL: https://issues.apache.org/jira/browse/ACE-169
>             Project: ACE
>          Issue Type: Task
>          Components: Management Agent
>            Reporter: Marcel Offermans
>            Assignee: Marcel Offermans
>
> Currently, there can be only one instance of a management agent in an OSGi framework. For a couple of use cases it makes sense to support multiple instances, each with their own identity, talking to their own discovered server(s). One use case is to have multiple management agents each deploy subsets of software that are completely isolated from each other (respecting the constraints that having multiple deployment packages impose). Of course there are downsides to this use case, as conflicts can arise that can never be detected beforehand on the server, so using multiple instances in this use case is strongly discouraged unless you can be sure you won't run into these issues. Another use case is when a management agent is being used to "import" software from one provisioning server to another, and you want to get software from multiple sources. This can be used in scenarios where you want to have a controlled merge between software from multiple sources, giving you control over how they are merged and allowing validation before sending the result to a target.
> In short, changes that need to be made are converting everything that is currently implemented as a "singleton" service into a managed service factory, adding a property to each service that can be used to identify the management agent it belongs to, and filtering the audit logs in such a way that each log sees the difference between "their" deployment package and "other" deployment packages.



--
This message was sent by Atlassian JIRA
(v6.1#6144)