I'm not sure what's causing this, but it's adding wasted time to the end of every distributed test we have in Solr.

At the very least, we could shut down the various Jettys using a completion service so that we're only blocking for 5 seconds maximum, as opposed to 5 seconds per server.

The cause looks to be that server.join() waits for all threads in the QueuedThreadPool to close themselves, and the MaxIdleTime for the qtp is set to 5 seconds. Should we drop this down?

The cause looks to be that server.join() waits for all threads in the QueuedThreadPool to close themselves, and the MaxIdleTime for the qtp is set to 5 seconds. Should we drop this down?

+1

Good catch!

Simple patch, dropping the 5 second idle timeout to 200ms. Running the full Solr test suite went from taking 16m25s to 12m45s, which is a nice speedup!

I'd like to take a closer look at this as I think it will really affect my patch in SOLR-4509. I've seen a similar speedup, but even with a much higher timeout (60 seconds). Let me poke for a bit and I'll comment again.

Commit 1648387 from Alan Woodward in branch 'dev/trunk'
[ https://svn.apache.org/r1648387 ]

SOLR-6896: Drop SolrJettyRunner max idle time to 200ms to speed up tests

Commit 1648388 from Alan Woodward in branch 'dev/branches/branch_5x'
[ https://svn.apache.org/r1648388 ]

SOLR-6896: Drop SolrJettyRunner max idle time to 200ms to speed up tests

Argh, literally saw your comment the second I'd committed to 5x. Do you want me to revert?

I'm marking this as resolved for now - Mark Miller if you want to make any other changes, shout and I can reopen it.