[ https://issues.jboss.org/browse/WFCORE-1340?page=com.atlassian.jira.plugi... ] Brian Stansberry updated WFCORE-1340: ------------------------------------- Description: Including an EAP 6.x slave in a mixed domain managed by a WF 10 / EAP 7 DC is overly difficult operationally because potentially numerous host configurations need to be manually updated whenever new server groups and profiles are added. An issue with managing mixed domains is the need for the slave's host.xml to include configuration of what domain-wide content should be ignored. This isn't nice as it requires modifying potentially many host configs when new domain-wide content is added (e.g. new server groups or profiles that the legacy slaves won't understand.) Core 2 / Full 10 are better in this regard as they allow "ignore-unused-configuration" where stuff is auto-ignored. But this still has weaknesses: 1) The "ignore-unused-configuration" logic is slave-side and is not present in EAP 6.x slaves. So for those slaves manual configuration is the only option. 2) Extensions are not covered, so new extensions in later releases may need to be manually configured. Idea here is to include config for host-ignores in the domain-wide model, for use by the DC. It's in the domain-wide model, not the DC's host.xml, to ensure that any backup HC has the latest data. The concept will be referred to as a "host-exclude" because what is happening in this case is not the slave ignoring some resources, it's the DC excluding them from the slave's view. Proposed structure: Resources are at address /host-exclude=* Attributes are: * management-major-version * management-minor-version * management-micro-version * host-release These identify the category of slave to which host-ignore data should be applied when a matching slave registers. The first 3 attributes identify the *core management API version* of the slave (not its release version.) The last is a user-friendly *alternative* to the first 3 and is an enum identifying well known releases (e.g. EAP6.2, EAP6.3, EAP6.4, WildFly10.0) from which the api versions can be derived. If management-micro-version is undefined, the meaning is the config applies to all releases of the given major/minor version, excluding any for which a config with a micro version specified is also present. Not specifying a micro is expected to be the norm. The "slave-release" enums will be for minors. In addition to the above scoping attributes, the following attributes will be supported: * excluded-extensions * active-server-groups * active-socket-binding-groups The excluded-extensions attribute is a list of extension names the resources that (/extension=X) the DC should hide from the target hosts. Generally because the hosts will not have the necessary extension modules in their installation. The active-server-groups attribute is a list of server groups names the members of which should be treated as *not* excluded from the target hosts. These are the groups used by the host's servers. The server-group and related profile and socket-binding-group resources will not be hidden; all others of these types will be hidden. This is the same data that a core 2 / WF 10 slave sends when it registers. This JIRA just provides a different mechanism for making the data known to the DC. The active-socket-binding-groups attribute is only meaningful if active-server-groups is set, and it only needs to be set if the set of socket-binding-groups associated with the server groups listed in active-server-groups is not the complete set of sbgs needed on servers running the legacy release. This can happen is the server-config element for some servers overrides the normal socket-binding-group specified in the server-group config and specifies some different sbg. This is expected to be an edge case. Adding a new group to active-server-groups or active-socket-binding-groups will not cause existing slave HCs to get new data sent to them. The slave will need to reconnect to get new data. A reload or restart of the slave or master causes a reconnect. Changing the profile or socket-binding-group associated with a group listed in active-server-groups will not cause existing slave HCs to get new data sent to them. The slave will need to reconnect to get new data. There is other data that could be included in these resources, e.g. fine grained "exclude" information matching what can be configured in in the ignored-resources elements host.xml, but that is out of scope for this first cut, and may never be added if there is no clear demand. If a slave has ignored-resources explicitly configured, that information will be used in managing that slave, in combination with any matching configuration in a host-exclude. However, the related ignore-unused-configuration setting added in WildFly Core (not present in EAP 6) will be handled differently. A WildFly Core 2 or later slave will send its ignore-unused-configuration setting when it registers. If this is set to 'true', the DC will not use any of the domain wide active-server-groups and active-socket-binding-groups data in its handling of that slave. That setting means the slave is taking responsibility for informing the DC as to what should server-groups, profiles and socket-binding-groups be ignored. Mixing this WFCORE-1340 functionality into that creates too much complication. However, the domain-wide 'excluded-extensions' data *will* be used for the slave. An EAP 6.x slave can't set ignore-unused-configuration, so there is no confusion for that use case, which is the primary one. was: Including an EAP 6.x slave in a mixed domain managed by a WF 10 / EAP 7 DC is overly difficult operationally because potentially numerous host configurations need to be manually updated whenever new server groups and profiles are added. An issue with managing mixed domains is the need for the slave's host.xml to include configuration of what domain-wide content should be ignored. This isn't nice as it requires modifying potentially many host configs when new domain-wide content is added (e.g. new server groups or profiles that the legacy slaves won't understand.) Core 2 / Full 10 are better in this regard as they allow "ignore-unused-configuration" where stuff is auto-ignored. But this still has weaknesses: 1) The "ignore-unused-configuration" logic is slave-side and is not present in EAP 6.x slaves. So for those slaves manual configuration is the only option. 2) Extensions are not covered, so new extensions in later releases may need to be manually configured. Idea here is to include config for host-ignores in the domain-wide model, for use by the DC. It's in the domain-wide model, not the DC's host.xml, to ensure that any backup HC has the latest data. The concept will be referred to as a "host-exclude" because what is happening in this case is not the slave ignoring some resources, it's the DC excluding them from the slave's view. Proposed structure: Resources are at address /host-exclude=* Attributes are: * management-major-version * management-minor-version * management-micro-version * host-release These identify the category of slave to which host-ignore data should be applied when a matching slave registers. The first 3 attributes identify the *core management API version* of the slave (not its release version.) The last is a user-friendly *alternative* to the first 3 and is an enum identifying well known releases (e.g. EAP6.2, EAP6.3, EAP6.4, WildFly10.0) from which the api versions can be derived. If management-micro-version is undefined, the meaning is the config applies to all releases of the given major/minor version, excluding any for which a config with a micro version specified is also present. Not specifying a micro is expected to be the norm. The "slave-release" enums will be for minors. In addition to the above scoping attributes, the following attributes will be supported: * ignored-extensions * active-server-groups * active-socket-binding-groups The ignored-extensions attribute is a list of extension names the resources for which (/extension=X) should be treated as ignored by the target hosts. The active-server-groups attribute is a list of server groups names the members of which should be treated as *not* ignored by the target hosts. These are the groups used by the host's servers. The server-group and related profile and socket-binding-group resources will not be ignored; all others of these types will be ignored. This is the same data that a core 2 / WF 10 slave sends when it registers. This JIRA just provides a different mechanism for making the data known to the DC. The active-socket-binding-groups attribute is only meaningful if active-server-groups is set, and it only needs to be set if the set of socket-binding-groups associated with the groups listed in active-socket-binding-groups is not the complete set of sbgs needed on server's running the legacy release. This can happen is the server-config element for some servers overrides the normal socket-binding-group specified in the server-group config and specifies some different sbg. This is expected to be an edge case. Adding a new group to active-server-groups will not cause existing slave HCs to get new data sent to them. The slave will need to reconnect to get new data. Changing the profile or socket-binding-group associated with a group listed in active-server-groups will not cause existing slave HCs to get new data sent to them. The slave will need to reconnect to get new data. There is other data that could be included in these resources, e.g. fine grained "exclude" information matching what can be configured in host.xml, but that is out of scope for this first cut, and may never be added if there is no clear demand. A WildFly Core 2 or later slave will send its ignore-unused-configuration setting when it registers. If this is set to 'true', the DC will not use any of the domain wide active-server-groups and active-socket-binding-groups data in its handling of that slave. That setting means the slave is taking responsibility for informing the DC of what should server-groups, profiles and socket-binding-groups be ignored. Mixing this WFCORE-1340 functionality into that creates too much complication. However, the domain-wide 'ignored-extensions' data *will* be used for the slave. An EAP 6.x slave can't set ignore-unused-configuration, so there is no confusion for that use case, which is the primary one. -- This message was sent by Atlassian JIRA (v6.4.11#64026)