ignoring the seed, and just trying the test with "-Dtests.nightly=true" i've only seen this test pass once (and i might have had a typo in that nightly param – it was the first time i tried it and i didn't have a shell log).
Unless i'm missing something...
- BaseDistributedSearchTestCase.createServers initializes the following pairwise...
- protected List<JettySolrRunner> jettys
- protected List<SolrServer> clients
- TestDistributedSearch.doTest then...
- copies those lists into local upJettys and upClients instances and maintains a list of "upShards"
- iteratively shutsdown some number of jetty instances, removing from upJettys, upShards, and upClients
- passes upShards and upClients to queryPartialResults
- TestDistributedSearch.queryPartialResults ...
- does some random quering of upShards and upClients
- if stress is non-zero (which it is if it's nightly) then it also spins up a bunch of threads using a client from the original "clients" list
...which seems fundamentally flawed to me ... because each "client" knows about a specific jetty instance, and the test has explicitly shut down some jetty instances.
Is this just a typo? are the refs to "clients" in queryPartialResults all just suppose to be "upClients" ?