Hi Fabrizio,
which Solr version are you using? In more recent versions (starting with 5
I think) you should not use the coreContainer directly but instead go
through the HTTP API (which also supports the swap operation) or use SolrJ.
Best,
Georg
Fabrizio Fortino <ffort...@gilt.com> schrieb am Mo., 29. Aug. 2016 11:53:
> I have a NON-Cloud Solr and I am trying to use the swap functionality to
> push an updated core into production without downtime.
>
> Here are the steps I am executing
> 1. Solr is up and running with a single core (name = 'livecore')
> 2. I create a new core with the latest version of my documents (name =
> 'newcore')
> 3. I swap the cores -> coreContainer.swap("newcore", "livecore")
> 4. I try to unload "newcore" (that points to the old one) and remove all
> the related dirs -> coreContainer.unload("newcore", true, true, true)
>
> The first three operations are OK. But when I try to execute the last one
> the Solr log starts printing the following messages forever
>
> 61424 INFO (pool-1-thread-1) [ x:newcore] o.a.s.c.SolrCore Core newcore is
> not yet closed, waiting 100 ms before checking again.
>
> I have opened an issue on this problem (
> https://issues.apache.org/jira/browse/SOLR-8757) but I have not received
> any answer yet.
>
> In the meantime I have found the following workaround: I try to manually
> close all the core references before unloading it. Here is the code:
>
> SolrCore core = coreContainer.create("newcore", coreProps)
> coreContainer.swap("newcore", "livecore")
> // the old livecore is now newcore, so unload it and remove all the
> related dirsSolrCore oldCore = coreContainer.getCore("newCore")while
> (oldCore.getOpenCount > 1) {
> oldCore.close()
> }
> coreContainer.unload("newcore", true, true, true)
>
>
> This seemed to work but there is some race conditions and from time to time
> I get a ConcurrentModificationException and then an abnormal CPU
> consumption.
>
> I filed a separate issue on this
> https://issues.apache.org/jira/browse/SOLR-9208 but this is not considered
> an issue by the Solr committers. The suggestion is to move and discuss it
> here in the mailing list.
>
> If this is not an issue, what are the steps to swap to cores and unload one
> of them?
>
> Thanks a lot,
> Fabrizio
>
