neutron-keepalived-state-change doesnt reap zombies if rootwrap-daemon is used

[Fabian Zimmermann]

I just tested the current stable/pike-branch of neutron.

After some time our monitoring alerted us about increasing amount of zombie-processes. After taking a look, I found

commit 91c26f56586e88a75e942cd06c8f4539acfb4963

which seems to introduces some rootwrap-helper changes. I reverted the commit but the problem still remains, so maybe the commit is not the reason, but triggers the problem more likely.

Steps to reproduce:

* configure neutron-rootwrap-daemon
* start l3-agents
* wait for (ha-)routers to be setup
* kill a rootwrap-daemon
* sudo is now a zombie which is not reapd by neutron-keepalived-state-change.

[Bernard Cafarelli (bcafarel)]

I can reproduce this on current devstack:
stack 20640 2.1 0.6 466476 108520 ? Ss 10:23 7:32 /usr/bin/python /usr/bin/neutron-dhcp-agent --config-file /etc/neutron/neutron.conf --config-file /etc/neutron/dhcp_agent.ini
root 23941 0.0 0.0 272332 5632 ? S 10:23 0:00 \_ sudo /usr/bin/neutron-rootwrap-daemon /etc/neutron/rootwrap.conf
root 23944 0.0 0.1 467628 22496 ? Sl 10:23 0:01 \_ /usr/bin/python /usr/bin/neutron-rootwrap-daemon /etc/neutron/rootwrap.conf

After killing process 23944:

stack 20640 2.1 0.6 466476 108520 ? Ss 10:23 7:32 /usr/bin/python /usr/bin/neutron-dhcp-agent --config-file /etc/neutron/neutron.conf --config-file /etc/neutron/dhcp_agent.ini
root 23941 0.0 0.0 0 0 ? Z 10:23 0:00 \_ [sudo] <defunct>

Was this working before? Thinking at when it may got broken

[Fabian Zimmermann (dev-faz)]

I switched back from stable/pike-branch to tag 11.0.6 and Im no longer having any zombie-processes after running the complete tempest-test-suite.

Im still having the Zombies if I kill the neutron-rootwrap-daemon manually, but I think the daemon normally shouldnt terminate.

So in my opinion the question is: Why is neutron-rootwrap-daemon terminating (more often) in versions >11.0.6?