Hi Zhu, You can ignore the init process part of message since the hang was seen during the shutdown not reboot as I thought. Sorry for confusion.
Steve Steve Peng wrote: > Zhu Xiao wrote: >> Hi All, >> I was running a powercycle test on huron. My script boots,then >> shutdowns solaris by issuing a "shutdown" command right after solaris >> boots. So there were fewer services were stopping. >> >> # shutdown -i0 -g0 -y >> ... >> # svc.startd: The system is coming down. Please wait. >> svc.startd: 74 system services are now being stopped. >> <------ (hang occurs here) >> >> One in hundreds of cycles, the solaris shutdown hang. Solaris kernels >> looked fine by examining the coredumps. Looking through /var/svc/log, >> the following services started, but didn't log the "stop" method in >> /var/svc/log. >> >> milestone-devices:default.log: >> milestone-single-user:default.log: >> network-initial:default >> network-routing-setup:default >> network-rpc-bind:default >> network-rpc-keyserv:default >> system-cryptosvc:default >> system-device-fc-fabric:default >> system-filesystem-local:default >> system-identity:domain >> system-manifest-import:default >> system-sysevent:default >> >> Any suggestion is greatly appreciated. >> >> Thanks, >> >> -Zhu >> _______________________________________________ >> smf-discuss mailing list >> smf-discuss at opensolaris.org >> > What you can do is to turn on svc.startd debug logging by doing the > followings: > > /usr/sbin/svccfg -s system/svc/restarter:default > svc:/system/svc/restarter:default> addpg options application > svc:/system/svc/restarter:default> setprop options/logging = astring: > debug > svc:/system/svc/restarter:default> exit > > And reboot. You can discard the log of good boots and only keep the > one that looks suspicious and hopefully the log can tell some clues on > what's going on when the system is hung. Coredump with user pages > helps as well. You can set dump device up by doing: > > dumpadm -c all > boot kmdb > enter kmdb when you see the hang > systemdump > > and exam the svc.startd process. But get the dump with user pages is > sort of tricky and depends on how early the hang occurs. > If the hang occurs pretty early then we may need to do something trick > on the init process. First I would suggest that you run > 'boot -m verbose' to see how early the hang is and of course with the > debug logging turned on. Hope this helps. > > Steve > > > > > >
