There is no guarantee that sending an update to a non-leader node is slower. It certainly seems like a bad idea, but forwarding a document is fast and indexing a document is slow, so it might not even be measurable.
We’ve indexed a million docs per minute by sending all updates to the load balancer for the cluster, ignoring shards or leaders. wunder Walter Underwood wun...@wunderwood.org http://observer.wunderwood.org/ (my blog) > On Mar 1, 2019, at 9:34 AM, Jason Gerlowski <gerlowsk...@gmail.com> wrote: > > Hi Ganesh, > > I'm not an expert on pysolr, but from a quick scan of their update > code, it does look like pysolr attempts to send update requests to _a_ > leader node for a particular collection. But that's all it does. It > doesn't check which shard the document(s) will belong to and try to > pick the _correct_ leader. If your collections only have 1 shard, this > is still pretty great. But if your collections have multiple shards > (and multiple leaders), then this will perform worse than SolrJ. > > (This is based on what I gleaned from the code here: > https://github.com/django-haystack/pysolr/blob/master/pysolr.py#L1268 > . Happy to be corrected by someone with more context.) > > Best, > > Jason > > On Tue, Feb 26, 2019 at 1:50 PM Ganesh Sethuraman > <ganeshmail...@gmail.com> wrote: >> >> We are using Solr Cloud 7.2.1. Is there a leader aware python client (like >> SolrJ for Java), which can send the updates to the leader and it its highly >> available? >> I see PySolr https://pypi.org/project/pysolr/ project, not able to find any >> documentation if it supports leader aware updates. >> >> Regards >> Ganesh
