[
https://issues.apache.org/jira/browse/OAK-2265?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14219312#comment-14219312
]
Marcel Reutegger commented on OAK-2265:
---------------------------------------
I think this is an interesting idea, but I'm also not sure if this makes sense
in general. I tend to agree with Michael that it might be better to just throw
a RepositoryException if there is a temporary network issue or a MongoDB
process dies. In many cases these interruptions should be temporary and e.g.
MongoDB would elect a new primary. I'm not sure bouncing the entire stack is
the best option is this case.
If we do want to support this behavior, I would rather prefer an external
system to perform this monitoring an controlling of services. One example is
[Pacemaker|http://clusterlabs.org/wiki/Main_Page].
> DocumentNodeStore service should go away if MongoDB becomes unavailable
> -----------------------------------------------------------------------
>
> Key: OAK-2265
> URL: https://issues.apache.org/jira/browse/OAK-2265
> Project: Jackrabbit Oak
> Issue Type: Bug
> Reporter: Bertrand Delacretaz
>
> Stopping the MongoDB backend that a Sling instance is using via the
> DocumentNodeStore service should cause that OSGi service to be unregistered.
> Currently the service stays registered even if MongoException$Network is
> thrown when trying to access the MongoDB backend.
> The code that manages the service should ping the backend and
> register/unregister the DocumentNodeStore to follow the backend's
> availablity. This will allow other services that depend on DocumentNodeStore
> to restart correctly once MongoDB is available again.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
