I think they believe this because of what you say above -- they are misinterpreting the call for topology as a handoff where zookeeper "does the rest"...
This info will allow me to straighten out that misunderstanding... Thanks all! On Tue, Nov 29, 2016 at 10:49 AM, Walter Underwood <wun...@wunderwood.org> wrote: > This is easy to prove. Shut down Zookeeper, then send search requests to > different hosts in the Solr Cloud cluster. If they work, then the requests > are > not going through Zookeeper. > > wunder > Walter Underwood > wun...@wunderwood.org > http://observer.wunderwood.org/ (my blog) > > > > On Nov 29, 2016, at 9:44 AM, Erick Erickson <erickerick...@gmail.com> > wrote: > > > > bq: My new place tells me they're sending requests to Zookeeper - and > those are > > getting sent on to Solr by Zookeeper - -- this is news to me if it's > true... > > > > Have you seen the code? Because if they're using SolrJ, then they do, > indeed > > connect to Zookeeper (i.e. CloudSolrClient takes a ZK ensemble) which > they may > > be misinterpreting as sending request to ZK that are forwarded to Solr. > That is > > _not_ what's happening. The ZK connection allows SolrJ to get the > > topology from ZK > > but then CloudSolrClient sends requests to Solr nodes directly based > > on the state.json > > fetched from ZK. And it should cache that. > > > > There have been JIRAs about reducing the number of times any particular > client > > requests state.json from ZK, but there's _no_ code on ZK that forwards > requests. > > ZK is just a bookkeeper, for it to "do the right thing" with a Solr > > request it'd have to > > know about Solr, i.e. we'd have to distribute some custom ZK code > > which we don't. > > > > It would help if you provided the reason they think this. Code like > above? > > Network traffic? They've monitored the ZK activity and it's high? I.e. > > is there any evidence of this? > > > > Best, > > Erick > > > > > > > > On Tue, Nov 29, 2016 at 8:52 AM, Kevin Risden <compuwizard...@gmail.com> > wrote: > >> If using CloudSolrClient or another zookeeper aware client, then a > request > >> gets sent to Zookeeper to determine the live nodes. If indexing, > >> CloudSolrClient can find the leader and send documents directly there. > The > >> client then uses that information to query the correct nodes directly. > >> > >> Zookeeper is not forwarding requests to Solr. The client requests from > >> Zookeeper and then the client uses that information to query Solr > directly. > >> > >> Kevin Risden > >> > >> On Tue, Nov 29, 2016 at 10:49 AM, John Bickerstaff < > j...@johnbickerstaff.com > >>> wrote: > >> > >>> All, > >>> > >>> I've thought I understood that Solr search requests are made to the > Solr > >>> servers and NOT Zookeeper directly. (I.E. Zookeeper doesn't decide > which > >>> Solr server responds to requests and requests are made directly to > Solr) > >>> > >>> My new place tells me they're sending requests to Zookeeper - and > those are > >>> getting sent on to Solr by Zookeeper - -- this is news to me if it's > >>> true... > >>> > >>> Is there any documentation of exactly the role(s) played by Zookeeper > in a > >>> SolrCloud setup? > >>> > >
