Just thinking out loud here about something spooky...

I've noticed some relatively rare and seemingly random SolrCloud tests that
fail to create a collection because an expected configSet resource isn't
found (isn't in Zookeeper where it ought to be).  The initial exception
will have "Can't find resource" then the name of the resource -- sometimes
it's solrconfig.xml or other times some language specific stop-word file or
something.  This happens when a collection is being created, but fails
because a core/replica won't load because the configSet is faulty because a
required file isn't there.  I do *not* believe this is some sort of
visibility race of the configSet wherein a sleep/wait before creating the
collection would help, because I've seen an entire test suite (test class
file) of many tests (test methods) all fail for the same reason. In that
case the MiniSolrCloudCluster was simply initialized in beforeClass and
thus ought to have "_default" ready to be used before the test methods, yet
every test method failed for the same reason.   This is very strange; the
code that uploads the _default configSet seems sound.  Yet there's some
sporadic bug so I guess *somewhere *there's either some race or there's an
underlying upload failure that is being ignored*.*

I think I ought to compile a list of tests that have shown this error and
the dates(s) / build info in which it occurred.  Maybe there's a pattern /
something they have in common.
-- 
Lucene/Solr Search Committer, Consultant, Developer, Author, Speaker
LinkedIn: http://linkedin.com/in/davidwsmiley | Book:
http://www.solrenterprisesearchserver.com

Reply via email to