Hello, Steve. Tried to sweep all remainings of MiniDFSCluster https://jira.apache.org/jira/browse/SOLR-13630. Once again I need it for tests, just to make sure that everything stop before starting new test that's not what's absolutely necessary for real life usage. Here's the list what's keep running after shutdown: - o.a.h.hdfs.server.datanode.fsdataset.impl.BlockPoolSlice.addReplicaThreadPool is supposed to be shutdown via jvm shutdown hook - o.a.h.fs.FileSystem.Statistics.STATS_DATA_CLEANER never stops - o.a.h.hdfs.DFSClient.getLeaseRenewer().interruptAndJoin() is never called. - o.a.h.hdfs.ClientContext.getPeerCache().close() isn't called as well - o.a.h.ipc.ProtobufRpcEngine.CLIENTS aren't closed - o.a.h.ipc.Client.clientExcecutorFactory keeps .Client.ClientExecutorServiceFactory.clientExecutor. It's expected to be stopped via ref counting, but when I've checked it with debugger after MiniDFSCluster.shutdown() ref counter was behind hundred. Seems like no way. I think it's far from being fully shutdown.
On Wed, Jul 3, 2019 at 5:27 PM Steve Loughran <ste...@cloudera.com.invalid> wrote: > if its not a UGI token renewer, it's probably a bug > > On Wed, Jul 3, 2019 at 4:08 PM Mikhail Khludnev <m...@apache.org> wrote: > > > Hello, > > There are integration tests between Solr and HDFS. Turns out even calling > > MiniDFSCluster.shutdown() leaves some HDFS threads running. It's probably > > ok, but it upsets Solr's tests since they assert that there's nothing > > running after test is over to verify against leaks. > > Are there any total shutdown helper for test? Does it make sense to > raise a > > jira for it? > > > > Thank you. > > -- > > Sincerely yours > > Mikhail Khludnev > > > -- Sincerely yours Mikhail Khludnev
