[jboss-jira] [JBoss JIRA] (WFCORE-622) DomainOperationContext transformation assumes slave host and slave host's servers receive the same ops

[Brian Stansberry (JIRA)]

    [ https://issues.jboss.org/browse/WFCORE-622?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13055889#comment-13055889 ] 

Brian Stansberry commented on WFCORE-622:
-----------------------------------------

I think there are three cases relevant to this basic problem:

1) An end user invokes an op /host=oldslave/server=a/subsystem=foo:op.

Here, perhaps the user understands that oldslave is using the old API, and therefore doesn't expect transformation. In that case, result transformation is invalid.

But if they do expect transformation, there's no reason to assume that operation transformation is not expected as well.

2) The slave HC itself generates the op, as part of domain rollout.

In this case, operation transformation is invalid, as the generated op is what the slave understands. But is result transformation valid? It may make the response consistent with current user expectations, or it may not, depending on the relationship to the operation transformation that may have gone along with the result transformation.

I think this is not a real case though, as custom result transformers don't make much sense for the add/remove/write-attribute ops generated as part of a domain rollout. Those ops return 'undefined.'

3) A wildcard read (see WFCORE-282). So, /host=*/server=*/subsystem=foo:read-attribute(name=bar). The user would likely expect a consistent response for all servers.

An example custom result transformer might be something that converts types, e.g.

{ "time" => 1, "unit" => "seconds"} becomes 1000L because the attribute type or response value type has changed from complex to a simple ms value.

Another example is https://github.com/wildfly/wildfly/blob/8.x/jmx/src/main/java/org/jboss/as/jmx/JMXExtension.java#L239 which uses a result transformer to provide a value for a newly added attribute.

> DomainOperationContext transformation assumes slave host and slave host's servers receive the same ops
> ------------------------------------------------------------------------------------------------------
>
>                 Key: WFCORE-622
>                 URL: https://issues.jboss.org/browse/WFCORE-622
>             Project: WildFly Core
>          Issue Type: Bug
>          Components: Domain Management
>            Reporter: Brian Stansberry
>            Assignee: Brian Stansberry
>             Fix For: 1.0.0.CR1
>
>
> https://github.com/wildfly/wildfly-core/commit/f5c516f2#diff-dc2038399fc72656590c09fe73305791R151 assumes that the result transformer for the host request is appropriate for use with the subsequent requests to that host's servers during domain rollout.
> That's a shaky supposition, as the server ops generated from a host request have no fixed relationship to the host request itself. Alignment is the norm but not a rule. In particular it can faill apart with composite ops where some steps in the composite are not relevant to all hosts.
> Without this extra transformation, test at [1] fails. This indicates the server op is not getting the transformation registered at [2] applied. But if it's not getting the result transformation applied, wouldn't that mean relevant operation transformation isn't being applied as well? 
> [1] https://github.com/wildfly/wildfly-core/blob/master/testsuite/domain/src/test/java/org/jboss/as/test/integration/domain/suites/OperationTransformationTestCase.java#L226
> [2] https://github.com/wildfly/wildfly-core/blob/master/testsuite/domain/src/test/java/org/jboss/as/test/integration/domain/extension/VersionedExtension2.java#L108



--
This message was sent by Atlassian JIRA
(v6.3.11#6341)