while attempting to reproduce CR6588979 on bostonb.sfbay (which was running
the liverpool high stress tests), i needed to regularly run "pkill -9 inetd" to
induce inetd failure. after a while i noticed svc.startd was not running - the
error message displayed was "svc:/network/rpc/gss:default : queue is full"
(originating from ln405 from startd/protocol.c it seems). the only remedy was
to restart inetd by hand (setting SMF_FMRI in the environment prior to running
inetd so that it would not refuse to run outside of SMF). perhaps it is not
relevant, but the system was also out of file descriptors due to the stress
testing:
Aug 16 19:25:16 bostonb inetd[5183]: [ID 702911 daemon.error] Too many open files
Work Around
# export SMF_FMRI=svc:/network/inetd:default
# /usr/lib/inet/inetd start
# /lib/svc/bin/svc.startd
If it happens during an "init 6", it is still possible to log in and type 'reboot'.