On Fri, Jun 28, 2019 at 09:49:51AM +0000, [email protected] wrote: > Every so often (roughly once in a thousand runs or so it seems), a server will > crash in test050 on master. This test replicates cn=config which is not > officially supported yet. > > Some output from a gdb session from the core is uploaded here: > ftp://ftp.openldap.org/incoming/test050-crash-master-20190628.txt > > It's not clear to me how attrs_dup has reached the attribute at 0x170b6a8 > unless > there was another thread messing with the same structure. str2entry2 is > running > in another thread, but I wouldn't think it runs on the same entry? Stuff is > optimised out in this one.
Another crash, this time with better logs and an unoptimized binary: ftp://ftp.openldap.org/incoming/test050-crash-20190702.txz Again, attribute is being accessed that seems to have been cleaned and released and another thread is running str2entry2 (on cn=config as well, just with a different pointer), might still be a red herring. Regards, -- OndÅej KuznÃk Senior Software Engineer Symas Corporation http://www.symas.com Packaged, certified, and supported LDAP solutions powered by OpenLDAP
