Hi Kevin, As Jason mentioned, you can check the shards.preference setting. Instead of distribute the requests by replica type, you can use the replica.leader option to make query requests goes to all non-leader nodes (TLOG or PULL), and leader node(TLOG) dedicated to indexing, exempt from queries. The feature is available since 8.10.0.
https://issues.apache.org/jira/browse/SOLR-15472 replica.leader Prefer replicas based on their leader status, set to either true or false. Consider a shard with two TLOG replicas and four PULL replicas (six replicas in total, one of which is the leader). With shards.preference=replica.leader:false, 5 out of 6 replicas will be preferred. Contrast this with shards.preference=replica.type:PULL where only 4 of 6 replicas will be preferred. Note that the non-leader TLOG replica behaves like a PULL replica from a search perspective; it pulls index updates from the leader just like a PULL replica and does not perform soft-commits. The difference is that the non-leader TLOG replica also captures updates in its TLOG, so that it is a candidate to replace the current leader if it is lost. Thanks, Wei On Tue, Aug 27, 2024 at 1:54 PM Kevin Liang (BLOOMBERG/ 919 3RD A) < klian...@bloomberg.net> wrote: > No worries, that's helpful. Thanks > > From: users@solr.apache.org At: 08/26/24 09:58:00 UTC-4:00To: > users@solr.apache.org > Subject: Re: TLOG/PULL query distribution > > Hey Kevin, > > I hope I'm not replying here too late. > > The best docs on this are in the "SolrCloud Distributed Requests" page > [1]. In short though - by default Solr won't have any preference, it > does "just" a round-robin or random choice among the healthy replicas > for each shard. Users may provide a "shards.preference" parameter to > tell Solr to prefer replicas meeting various criteria, but there's no > default value for that. > > In terms of the CPU you're observing in your perf test - maybe your > perf driver is specifying a "shards.preference" param that prioritizes > PULL replicas? > > Best, > > Jason > > [1] > > https://solr.apache.org/guide/solr/latest/deployment-guide/solrcloud-distributed > -requests.html#shards-preference-parameter > <https://solr.apache.org/guide/solr/latest/deployment-guide/solrcloud-distributed-requests.html#shards-preference-parameter> > > On Thu, Aug 15, 2024 at 5:21 PM Kevin Liang (BLOOMBERG/ 919 3RD A) > <klian...@bloomberg.net> wrote: > > > > Hello, > > > > Given a mix of TLOG/PULL nodes in our solr cloud, the behavior I'm > observing > from metrics is that CPU load is high for the leader TLOG (indexing) and > the > PULL nodes (read queries) however for the most part the other TLOG nodes > remain > very underutilized by comparison. My understanding is that TLOG nodes > replicate > the same way PULL nodes do (+ transaction log), so they should also be > available to service read queries no? I'm wondering if anyone knows how > the > leader distributes read queries, as I wasn't able to find much in the way > of > documentation. Thanks > > > > -Kevin > > >
