This documents available plugin roles. 1) list of plugin roles | Role | Class | Status | |---------------+--------------------+---------------| | Generator | Generator | done | | Structure | Structure | done | | Pull | PullSource | class defined | | Metadata | Metadata | done | | Connector | Connector | class defined | | Probing | Probing | done | | Decision | Decision | done | | Remote | Remote | none | | Statistics | Statistics | class defined | | Structure Val | StructureValidator | done | | Goals Val | GoalValidator | class defined | | Syncing | Syncing | none | |---------------+--------------------+---------------| 2) Interactions between plugins and the core * Metadata Construction ** Get Base Metadata from (single) Metadata plugin instance ** Get additional data from each Connector plugin instance ** Merge in additional connector data into single ClientMetadata instance * 3) Configuration of plugins Plugin configuration will be simplified substantially. Now, a single list of plugins (including plugins of all capabilities) is specified upon startup (either via bcfg2.conf or equivalent). This mechanism replaces the current split configuration mechanism where generators, structures, and other plugins are listed independently. Instead, all plugins included in the startup list will be initialized, and each will be enabled in all roles that it supports. This will remove a current source of confusion and potential configuration errors, wherein a plugin is enabled for an improper set of goals. (ie Cfg enabled as a structure, etc) This does remove the possibility of partially enabling a plugin for one of its roles without activating it across the board, but I think this is a corner case, which will be poorly supported by plugin implementers. If needed, this use case can be explicitly supported by the plugin author, through use of a config file directive.