Thanks for chiming in Mikhail! > Then, ping has that healthcheckFile feature which lets one turn a node off > from the pool
You mean turn a "core" off, right? Not an entire node, unless I'm missing something? If this was a hidden LB API we could use to cordon particular nodes prior to restart, etc, I'd be all for keeping it. But AFAICT the healthcheck file that gets created is put inside of the specified core's data directory, so it only affects traffic to that specified core. (And of course, to take advantage of that at all users would need to have some sort of core-aware proxy/LB.) I absolutely agree though that it'd be really nice if Solr had some sort of node-level cordoning API. I swear I read something about that in a recent SIP, but can't find a pointer at the moment... Jason On Mon, Jun 26, 2023 at 3:31 PM Mikhail Khludnev <m...@apache.org> wrote: > > Hello Jason, > I have to say my production experience is quite limited. > Then, ping has that healthcheckFile feature which lets one turn a node off > from the pool via Solr app level. > Nowadays, it should be done via LB API, but ping allows to organise a > rolling recycle via primitive bash script. > > Please don't consider it as an opinion, it's just a reminder. My vote is: 0 > > On Mon, Jun 26, 2023 at 9:57 PM Jason Gerlowski <gerlowsk...@gmail.com> > wrote: > > > Hey all, > > > > Is Solr's "ping" API still useful for folks today? > > > > "/admin/ping" was initially designed as a healthcheck for individual > > SolrCore instances. The idea was that it could be checked > > periodically by an external loadbalancer (that, presumably, was aware > > not just of Solr nodes but of the topology of cores across the > > cluster). > > > > This made sense in 2012 when /admin/ping was introduced, but I'm not > > so sure it still does today. > > > > AFAICT, over time Solr usage has come to prefer node-level > > healthchecks, and other APIs have sprung up to meet that need. (e.g. > > the /admin/info/system and /admin/info/health endpoints commonly used > > as liveness/readiness probes in Kubernetes environments). > > > > Even SolrJ's "load-balancing" client (used by both cloud and > > "standalone" deployments) relies on node-level checks. And of course > > the advent and prevalence of SolrCloud have even further reduced the > > need to track per-core health outside of Solr/ZK. > > > > Does /admin/ping still have usecases that can't be met by other APIs? > > Or should we consider deprecating and removing it in 10.0? > > > > Best, > > > > Jason > > > > --------------------------------------------------------------------- > > To unsubscribe, e-mail: dev-unsubscr...@solr.apache.org > > For additional commands, e-mail: dev-h...@solr.apache.org > > > > > > -- > Sincerely yours > Mikhail Khludnev --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@solr.apache.org For additional commands, e-mail: dev-h...@solr.apache.org