On Tue, Mar 13, 2012 at 1:32 PM, Mark Miller <[email protected]> wrote: > > On Mar 13, 2012, at 1:23 PM, Yonik Seeley wrote: > >> Looks like the same underlying cause again... the recovery thread is >> still running when the test ends and hence a core (and it's searcher) >> are still open. > > Yeah, just came to the same likely conclusion - this was also changed in the > recent refactoring - we used to get and close the core on each retry which > may have affected the timing around this.
Definitely - it still would have happened, just with less frequency (the thread still have been running, but grabbing/releasing the core would make this happen more intermittently). >> I thought there was code in the test framework to wait for threads >> that hadn't finished yet? > > No, I don't think so. Part of why we have the wait for searcher open/close > balance. It's already fairly large but perhaps that wait needs to be even > longer, jenkins has recently gotten slower it seems (see those 100ms > connection timeout fails that have started showing up). Yeah, I was wondering if the drastically increased connection timeout fails corresponded at all with the Jetty upgrade, or if it was just Jenkins getting slower. -Yonik lucenerevolution.com - Lucene/Solr Open Source Search Conference. Boston May 7-10 --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
