michaeldexter
Replace the `ps ax | grep ...` call with pkill
Using the pipes has the nasty side effect of killing the service command itself. Having a guest called openbsd0 these are the processes killed:
root@i5:~ # ps ax | grep openbsd0
33916 2 Is+ 0:00.01 csh -c /bin/sh /usr/local/etc/rc.d/vm onerun openbsd0
33920 2 I+ 0:00.01 /bin/sh /usr/local/etc/rc.d/vm onerun openbsd0
33955 2 S+ 0:13.20 bhyve: openbsd0 (bhyve)
34114 3 S+ 0:00.00 grep --color openbsd0
Killing the grep process stops the execution of the f_stop function before cleaning up /dev/vmm and the tap interfaces.
root@i5:~ # tmux ls
openbsd0: 1 windows (created Fri Oct 31 01:14:20 2014) [159x69]
root@i5:~ # service vm onestop openbsd0
Stopping openbsd0
Sending ACPI shutdown to openbsd0
Terminated
root@i5:~ # pgrep -lf bhyve
root@i5:~ # ls /dev/vmm
openbsd0
This issue was not present when running the stop directive without an argument (service vm onestop).
After this patch, and using the pkill command I didn't notice any
leftovers after the stop command, and it didn't change the behavior of
service vm onestop, though I only have one guest on this host, and I
didn't try the loop.
By the way, maybe I should use pkill -TERM -f "bhyve: $1" if that is better. That way the vm script won't be killed either.
On 10/31/14 9:51 AM, vrachnis wrote:
By the way, maybe I should use
pkill -TERM -f "bhyve: $1"if that is better. That way the vm script won't be killed either.
There was definitely some risky placeholder code. I have added new routines this last two days and have some additional ones commented out. I welcome your feedback on these.
They are in the snapshot at: http://vmrc.bsd.lv/vmrc-snapshot.tar and should be easy to find in 'vm' as there are few instances of 'kill'.
Thank you for testing this and suggesting ideas.
Michael
Indeed, pkill -f is equivalent to grep | grep -v as Wikipedia notes.
On the other hand, I was also thinking about a completely different approach. Right now, the stop call will attempt to terminate all the guests with the given name. If it is acceptable to assume that the only bhyve guests running in the system are managed by vmrc, maybe the bhyve pid could be stored in some place like /usr/local/vmrc/vm0/vm0.pid and then use that file with the kill command to send the ACPI signal.
Are you willing to pull and try this again? I'm not dead, it's just been stable enough for production use.
I had actually forgotten about this PR. Since vm_pid is now detected by running ps axwww | grep -v grep | grep "bhyve: $vmname (bhyve)" | awk '{ print $1 }' , the issue that I was trying to address is now gone.
In theory, we could replace all those pipes with pgrep -f "bhyve: $vmname (bhyve)", but that's a preferential and extremely minor change. So, as far as I am concerned, we can close this PR.