rbp@investor.UUCP (Bob Peirce) (10/10/86)
Here' an interesting problem I am trying to solve. We have certain
jobs which, once started, should not be stopped until they finish.
However, we sometimes get too much of a load and wish these jobs
weren't running. I am trying to use traps in my shell scripts to put
these jobs to sleep for a while and start them up again later.
I have a solution, but it is not very neat. I break the script into
sections which can be killed and restarted without problem. I surround
each section with the following
trap "sleep 3600; T=0" 16
T=0
while [ $T -eq 0 ]
do
T=1
this_section
done
Actually, one trap at the top suffices except I put messages in the trap
to log what is happening.
To drive this I have a "suspend" script which is where things get messy.
The problem is if I send kill -16 to the program but this_section is
another program the signal is ignored until this_section completes. To
get around that I run the output of ps through an awk script to
determine the pid of this_section. I send a kill -16 to the program
followed by a kill -15 to this_section. That works.
What I woiuld like is some way of sending a signal to all of the children
of a process without knowing their pid. At the same time, I DON'T want
to send the signal to the parent. The manuals talk about process groups
and that seemed like a possibility, but I haven't figured out how to
create same let alone whether that is really the solution.
Any ideas would be appreciated.
--
Bob Peirce, Pittsburgh, PA
uucp: ...!{allegra, bellcore, cadre, idis}
!pitt!darth!investor!rbp
412-471-5320
NOTE: Mail must be < 30K bytes/message