[
https://issues.apache.org/jira/browse/QPID-7477?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15628730#comment-15628730
]
Keith Wall edited comment on QPID-7477 at 11/2/16 12:00 PM:
------------------------------------------------------------
The problem is that whilst {{BrokerImpl#performActivation}} is being run by the
Broker Config thread, the Subject attached to the thread erroneously includes
the {{VirtualHostPrincipal}}. This is causing the log messages that pertain to
the Broker to be written to the wrong place.
The problem arises as follows:
Broker is being started.
# VHN tries to attain its desired state
# VHN recovers the VH
# VHN opens the VH asynchronously (AbstractStandardVirtualHostNode.java:117).
Note that the *VH open future* is returned as the *VHN's state change future*.
# VH opens successfully
# VH open future completes (which is the VHN state future)
# Guava's future listening mechanics (SettableFuture/ExecutionList) which is
*running on the VH config thread* causes the future callback
(AbstractConfiguredObject.java:1030) to be submitted to the Broker-Config
thread. TaskExecutorImpl.java:397 constructs a Subject for the task, but as
the thread's current context includes the VH principal, so does the Subject
used for the execution of the Broker task *(defect)*.
# The VHN state future now complete, so other Broker objects proceed to attain
their desired state too,
# The Broker children now have all attained state, so
BrokerImpl#performActivation is run. This is still the same task and still
carries the erroneous VH Principal.
was (Author: k-wall):
The problem is that whilst {{BrokerImpl#performActivation}} is being run by the
Broker Config thread, the Subject attached to the thread erroneously includes
the {{VirtualHostPrincipal}}. This is causing the log messages that pertain to
the Broker to be written to the wrong place.
The problem arises as follows:
Broker is being started.
# VHN tries to attain its desired state
# VHN recovers the VH
# VHN opens the VH asynchronously (AbstractStandardVirtualHostNode.java:117).
Note that the *VH open future* is returned as the *VHN's state change future*.
# VH opens successfully
# VH open future completes (which is the VHN state future)
# Guava's future listening mechanics (SettableFuture/ExecutionList) which is
*running on the VH config thread* cause the future callback
(AbstractConfiguredObject.java:1030) to be submitted to the Broker-Config
thread. TaskExecutorImpl.java:397 constructs a Subject for the task, but as
the thread's current context includes the VH principal, so does the Subject
used for the execution of the Broker task.
# The VHN state future now complete, so other Broker objects proceed to attain
their desired state too,
# The Broker children now have all attained state, so
BrokerImpl#performActivation is run. This is still the same task and still
carries the erroneous VH Principal.
> [Java Broker] Recovery of Broker preferences should happen without holding
> VirtualHostPrincipal
> -----------------------------------------------------------------------------------------------
>
> Key: QPID-7477
> URL: https://issues.apache.org/jira/browse/QPID-7477
> Project: Qpid
> Issue Type: Bug
> Components: Java Broker
> Affects Versions: qpid-java-6.1
> Reporter: Lorenz Quack
> Fix For: qpid-java-6.1, qpid-java-6.2
>
>
> If a {{VirtualHostLogger}} is configured it will log the recovery of Broker
> preferences on the Broker-Config thread. It should not do this.
> The cause seems to be that during execution of
> {{BrokerImpl#performActivation}} the {{Subject}} contains the
> {{VirtualHostPrincipal}}.
> I think this is related to the fact that the call stack shows that this code
> is called from the {{onSuccess}} callback of {{VirtualHostNode#doActivation}}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
