I don't know the ValueSourceParser from a hole in my head, but it looks like it has access to the solrcore with fp.req.getCore?
If so, it's easy to get the zk stuff core.getCoreDescriptor.getCoreContainer.getZkController(.getZkClient). From memory, so perhaps with some minor misname. - Mark On Mar 25, 2013, at 6:03 PM, Timothy Potter <thelabd...@gmail.com> wrote: > I have a custom ValueSourceParser that sets up a Zookeeper Watcher on some > frequently changing metadata that a custom ValueSource depends on. > > Basic flow of events is - VSP watches for metadata changes, which triggers > a refresh of some expensive data that my custom ValueSource uses at query > time. Think of the data in Zookeeper as a pointer to some larger dataset > that is computed offline and then loaded into memory for use by my custom > ValueSource. > > In my ValueSourceParser, I connect to Zookeeper using an instance of the > SolrZkClient class and am receiving WatchedEvents when my metadata changes > (as expected). > > All this works great until core reload happens. From what I can tell, > there's no shutdown hook for ValueSourceParsers, so what's happening is > that my code ends up adding multiple Watchers and thus receives multiple > update events when the metadata changes. > > What I need is either > > 1) a shutdown hook in my VSP that allows me to clean-up the SolrZkClient > instance my code is managing, or > > 2) access to the ZkController instance owned by the CoreContainer from my > VSP. > > For me #2 is better as I'd prefer to just re-use Solr's instance of > SolrZkClient. > > I can go and hack either of these in pretty easily but wanted to see if > someone knows a better way to get 1 or 2? > > In general, it might be handy to allow plug-ins to get access to the > Zookeeper client SolrCloud is using. > > Thanks. > Tim
