Hi Erick, It is Solr 1.41 (a Drupal installation) running on Jetty. How can one get a stack trace? (there is no exception/error)
Could it be that solr does something like this? start delete job cannot find bogus id to delete does some reindex or optimization anyway regardless which takes 80 seconds end delete job Anyway, does it sound like Solr is just waiting 80 seconds for some exclusive lock or is it actually doing something in a background thread?. I do not know what kind of calls drupal is doing. Thanks & Regards Ericz On Sun, Oct 28, 2012 at 3:08 PM, Erick Erickson <erickerick...@gmail.com>wrote: > That is very weird. What version of Solr are you using, and is there > any way you could get a stack trace when this is happening? > > Best > Erick > > On Sun, Oct 28, 2012 at 6:22 AM, Eric Grobler <impalah...@googlemail.com> > wrote: > > Hi > > > > I am a bit confused why the server sometimes takes 80 seconds to respond > > when I specify an id to delete than does not even exist in the index. > > > > If I loop this query and send a bogus id to delete every minute. > > 03:27:38 125 ms <delete><id>bogusidthatdoesnotexist</id></delete> > commit > > 03:28:38 125 ms <delete><id>bogusidthatdoesnotexist</id></delete> > commit > > 03:29:38 69125 ms <delete><id>bogusidthatdoesnotexist</id></delete> > commit > > 03:30:38 124 ms <delete><id>bogusidthatdoesnotexist</id></delete> > commit > > 03:31:38 84141 ms <delete><id>bogusidthatdoesnotexist</id></delete> > commit > > 03:33:38 125 ms <delete><id>bogusidthatdoesnotexist</id></delete> > commit > > 03:34:38 141 ms <delete><id>bogusidthatdoesnotexist</id></delete> > commit > > 03:35:43 55476 ms <delete><id>bogusidthatdoesnotexist</id></delete> > commit > > 03:36:38 141 ms <delete><id>bogusidthatdoesnotexist</id></delete> > commit > > This was at 3am and the server only has about 200,000 documents and is > not > > busy, average query time is a constant < 5ms. > > > > If the server takes 80 seconds when it needs to update the index I would > > understand it. > > *But in this case the id does not exists, so the server should just > return > > immediately?* > > I then must assume that the delete command must be in some low priority > > queue and waits for some exclusive lock? > > When I look at the stats it seems that it was only my loop that did > > cumulative_deletesById every minute. > > > > What settings in the solrconfig.xml would effect this behaviour? > > > > Thank you & Regards > > Ericz >
