From: Gustaf N. <ne...@wu...> - 2017-10-11 09:24:51
|
Am 06.10.17 um 1:47 PM schrieb David Osborne: > That doesn't seem to quite do it... still getting a segfault in the > same place using naviserver built from tip. > Just to let you know: the problem seems to be resolved with the latest commits. The problems were race-conditions with half-initialized connection structures, since ns_server lets inspect the conn structures of all concurrently running threads while these are potentially modifying/freeing these. There is now more locking needed, but performance looks still ok. I get on my notebook in a VM with debian sid still more than 7k requests/sec with ab in a stress test, where a preauth filter calls "ns_server active" for every request. -gn |