On 8/7/24 17:46, Claudio Fontana wrote: > Hi, > > On 8/7/24 15:37, Stefan Hajnoczi wrote: >> On Sat, Aug 03, 2024 at 11:23:21AM +0200, Claudio Fontana wrote: >>> Hello Stefan, >>> >>> did not know where to report this, but the man page mentions you specifically so here I am. >> >> Hi Claudio, >> I wrote the man page for kvm_stat(1) but am not the maintainer of the >> tool. I have CCed the KVM mailing list although it's possible that no >> one actively maintains the tool :). >> >>> >>> >>> There seems to be an issue when kvm_stat is run with: >>> >>> kvm_stat -p xxx -d -t -s yy -c -L FILENAME.csv & >>> >>> specifically due to the ampersand (&), thus running in the background. >>> >>> >>> It seems that kvm_stat gets the interrupt signal (SIGINT), and does write as a result the output to disk, >>> but then instead of terminating, it just hangs there forever. >> >> That is strange. The only signal handler installed by kvm_stat is for >> SIGHUP, so Python should perform the default behavior for SIGINT and >> terminate. I'm not sure why the process would hang. >> >>> >>> So to avoid ending up with a large number of kvm_stat processes lingering on the system, >>> we needed to put a random sleep, and then send a SIGTERM to terminate the kvm_stat processes. >>> >>> Just sending a SIGTERM (without the SIGINT) does terminate the kvm_stat processes, but NO DATA is written to disk (the files show as 0 size). >> >> That makes sense since kvm_stat does not handle SIGTERM. The default >> SIGTERM behavior is to terminal and any output in Python's I/O buffers >> may not have been written to the file. >> >> Maybe kvm_stat should catch SIGINT and SIGTERM. That would give it a >> chance to write out the log before terminating. Do you want to try >> implementing that?I > > I think SIGINT would be enough, sure I can give it a try, It seemed to make sense to process SIGINT and SIGTERM the same way in the end. I sent a patch to the mailing list, Paolo and you: https://marc.info/?l=kvm&m=172305139224516&w=2 There was some local problem with email, so the email might have reached the list but not you specifically also. Thanks, Claudio > > Thanks, > > C > >> >>> >>> This is the workaround script that we currently have: >>> >>> ---- >>> >>> #! /bin/bash >>> >>> VM_PIDS=`pgrep qemu-system-` >>> >>> for VM_PID in ${VM_PIDS} ; do >>> # warning: kvm_stat is very fragile, change with care >>> kvm_stat -p ${VM_PID} -d -t -s 1 -c -L kvm_stat_${VM_PID}.csv & >>> done >>> >>> if test "x${VM_PID}" != "x" ; then >>> echo "launched kvm_stat processes, capturing 10 seconds..." >>> sleep 10 >>> echo "signaling all kvm_stat processes to write to disk..." >>> pkill -INT -P $$ >>> sleep 5 >>> sync >>> echo "signaling all kvm_stat processes to die..." >>> pkill -TERM -P $$ >>> echo "waiting for kvm_stat processes to exit..." >>> while pgrep -P $$ > /dev/null; do >>> sleep 2 >>> echo "still waiting for kvm_stat processes to exit..." >>> done >>> fi >>> >>> echo "Done." >>> >>> ---- >>> >>> Feel free to forward to the appropriate mailing list if needed, >>> >>> thanks! >>> >>> Claudio >>> >>> -- >>> Claudio Fontana >>> Engineering Manager Virtualization, SUSE Labs Core >>> >>> SUSE Software Solutions Italy Srl >>> >
