Peter Palfrader wrote: > reopen 520179 > thanks > > The version of bind9 running on merkel with the patch by dannf still > does not solve the issue. > > While it seems bind9 now does not crash almost instantly it still does > not survive for very long.
Hi, weasel: Are you still experiencing this bug, or did you end up replacing ia64/bind9 with amd64/unbound? ;-) It looks like upstream fixed several bugs in the ia64 atomics implementation in 2012: https://source.isc.org/cgi-bin/gitweb.cgi?p=bind9.git;a=commitdiff;h=339f08ee1eea16401bb9fbd9232a336cf6f52a8d;hp=24c7ce73cfe75d10620e7e4c842bce6fc7033944 The changelog message is: "Fixed Read-After-Write dependency violation for IA64 atomic operations. [RT #25181]". Unfortunately, I no longer have access to the ISC bug tracker, so I can't check the details. The upstream diff looks very similar to dannf's original proposed patch, except that it changes both isc_atomic_xadd() and isc_atomic_cmpxchg(), whereas dannf's patch only changes isc_atomic_xadd(). I suspect the first hunk in upstream's atomic.h patch is equivalent to dannf's patch, but I am not an ia64 assembly programmer. Your original logs show crashes in rbtdb.c, which indirectly relies on isc_atomic_cmpxchg() via the rwlock code, so it at least seems plausible that this upstream change is the full fix for this bug. The upstream change was released in 9.9.2rc1, so I suspect we could tentatively mark this bug as fixed in Debian version 1:9.9.2.dfsg-1. Thanks for reading this far. Here is an animated GIF of Patrick Stewart yelling: http://i.imgur.com/alZ9VUm.gif. -- Robert Edmonds edmo...@debian.org
