I said that the cluster prop was *not* related to SSL:  "adding a cluster
property  (albeit wasn't SSL; I forget what for)".  For SSL in 10.1+, just
set solr.ssl.enabled=true or SOLR_SSL_ENABLED=true

I think I now remember it was a replica placement plugin configuration:
https://solr.apache.org/guide/solr/latest/configuration-guide/cluster-plugins.html
 I believe that such things should be possible to set in solr.xml but only
one of the plugin types can live there today, if I recall.

Uploading the configset before Solr starts can be important in an upgrade
of plugins because the cores on the node should be loaded using the
configsets & plugins getting uploaded.  I say "upgrade" but don't mean an
upgrade of Solr itself, I mean new/updated plugins (JARs) and an updated
configset specifies classes in those JARs.  Thus the configset & Solr
software must be launched as a whole, definitely not Solr first using an
older ConfigSet in ZooKeeper that might contain references to
classes/configuration that no longer work.  IMO users should become more
familiar with FileSystemConfigSetService, which I helped make pluggable to
address this conundrum.  It needs more documentation, and there are
trade-offs (like mutable configsets / config-edit API will no longer be
practical).  It's not going to be appropriate for everyone.  If Solr's CLI
scripts cease supporting direct ZK communication, a user could use the ZK
CLI capability built-into zookeeper.jar that we used to provide a simple
shell script for.  I don't know what happened to that; I just looked for it
without success.

Solr can upload one configset on startup very early (before cores load)
-Dboostrap_confdir=/path/myname/conf/ -Dcollection.configName=myname (9.x
names; have changed in 10.x) which I recall isn't documented.  I'm using
that now to avoid an explicit upload step for a Solr docker image used in
testing.


On Wed, May 13, 2026 at 6:56 AM David Eric Pugh via dev <[email protected]>
wrote:

>  Gus, when you say "node for solr", are you referencing the ZK Root [1]?
>  That is also created at runtime I believe as well.
> Jan, thanks for sharing the operator perspective!  Security.json doesn't
> have to be there before Solr starts, though of course we've seen our
> travails on that, and of course maybe not having it in ZK at all.   The SSL
> cluster prop is another example of something thing we have to deal with.
> David, was the cluster prop around what http protocol to use?  I do know
> that configsets don't have to be loaded first, just start up empty solr,
> add them, and then generate collections!
> What I am hearing is that there are some cluster level settings that we
> have to get right, but that we already seem to have pretty robust support
> for collection level configuration.
> I would very much like to see ZK become a implementation detail!   The
> work on "solrConnection" string support will remove the need to specify
> zkHost in many places (streaming, cli), and help wean us off of thinking
> about ZK as something we interact with directly.
> Maybe a better way to ask this question: In the Solr CLI, what functions
> can/must operate directly on ZK (and require you to have direct access)
> versus being mediated through a Solr REST API?  We have a hodge podge of
> different access mechnisms.   "bin/solr zk" commands do seem like they
> should go to ZK, but outside of that?
>
>
>
>
> [1]
> https://solr.apache.org/guide/solr/latest/deployment-guide/taking-solr-to-production.html#zookeeper-chroot
>     On Tuesday, May 12, 2026 at 08:22:14 PM EDT, David Smiley <
> [email protected]> wrote:
>
>  In recent years, I recall
> 1. adding a cluster property  (albeit wasn't SSL; I forget what for)
> 2. adding configsets
>
> On Tue, May 12, 2026 at 8:15 AM David Eric Pugh via dev <
> [email protected]>
> wrote:
>
> > Does anyone stand up ZooKeeper first, and then use the Solr CLI to
> > interact with it, for example, by uploading configsets or maybe setting
> > some Cluster Properties, before any Solr nodes are started?
> > I ask because as we have matured the V2 API and added more SolrJ http
> > apis, there has been some interest in making Solr CLI use HTTP instead of
> > talking direct to ZooKeeper.
> > There is also some interesting work in
> > https://github.com/apache/solr/pull/4320 that moves us in that direction
> > as well.
> > I struggle in my direct experience to think about where I have configured
> > ZooKeeper independent of Solr, and indeed it feels like an anti-pattern
> > since the Solr APIs are where we have lots of validation etc.  And having
> > ZooKeeper exposed directly to the CLI seems also to have it's own issues
> > from a security perspective.
> > So was curious if this is a real need?
> > Eric
> >
>

Reply via email to