It is probably more of a design flaw than an outright bug, and I think it was addressed with https://issues.apache.org/jira/browse/SOLR-15300 (released in 8.9).
As explained in that issue, that `state` flag was only meant to reflect the state during SPLITSHARD operations, but that’s actually a really rare thing and if we’re going to have a `state` flag it would be better if it reflected the real state on an ongoing basis. Cassandra On Jul 15, 2021, 4:52 AM -0500, Bram Van Dam <[email protected]>, wrote: > Howdy, > > Solr 7.7, in Cloud, with ~400 cores spread over 6 chunky servers. > > We've noticed that CLUSTERSTATUS occasionally contains incorrect > information. This mostly seems to happen when things are pear shaped > (couple of servers down, shards recovering, that sort of thing). > > Here's an example where a shard is supposedly active, while its only two > replicas are down. Because ... two wrongs .. make a right? Ahem. > > "foobar01":{ > "range":null, > "state":"active", ========> WTF? > "replicas":{ > "core_node362":{ > "core":"foobar_foobar01_replica_n361", > "base_url":"http://1.2.3.4:8180/solr";, > "node_name":"1.2.3.4:8180_solr", > "state":"down", ========> WTF? > "type":"NRT", > "force_set_state":"false" > }, > "core_node364":{ > "core":"foobar_foobar01_replica_n363", > "base_url":"http://5.6.7.8:8180/solr";, > "node_name":"5.6.7.8:8180_solr", > "state":"down", ========> WTF? > "type":"NRT", > "force_set_state":"false" > } > } > } > > Is this a bug? Couldn't find anything relevant on Jira. Is there any way > we can force solr to compute a correct cluster status? I don't think we > can reproduce this reliably, but it's nevertheless an annoyance. > > Thanks, > > - Bram
