eucalyptus-cc init script doesn't always clear /var/lib/eucalyptus/CC

[Dustin Kirkland ]

For a currently unknown reason, /var/lib/eucalyptus/CC isn't always cleaned out on eucalyptus restart.

I noticed this sometimes when I change the SCHEDPOLICY between POWERSAVE and ROUNDROBIN in /etc/eucalyptus/eucalyptus.conf and restart eucalyptus. Sometimes the changes don't take affect because the CC state isn't cleared.

[Thierry Carrez (ttx)]

Should be cleared if CLEAN=1 is passed to upstart. It's a bug if it's not.

[Dustin Kirkland  (kirkland)]

There seems to be some non-determinism, here.

I can confirm that sometimes when I pass CLEAN=1, it is cleared, and
sometimes it's not.

This needs to be fixed.

[Thierry Carrez (ttx)]

CLEAN=1 is supported in eucalyptus-cc.conf, not in eucalyptus.conf. So maybe it's something like:
stop eucalyptus CLEAN=1 -> FAIL (env not passed to eucalyptus-cc)
stop eucalyptus-cc CLEAN=1 -> PASS
start eucalyptus[-cc] CLEAN=1 -> FAIL (start not supported, bug 491254)
restart eucalyptus CLEAN=1 -> FAIL (env not passed to eucalyptus-cc)

Looks quite non-deterministic from here.
Dustin, if you can confirm the above behavior, in the end it's a matter of what CLEAN commands we want to support. IMHO we should support "restart eucalyptus CLEAN=1" and "stop eucalyptus CLEAN=1", "start eucalyptus CLEAN=1", "stop eucalyptus-cc CLEAN=1" and "start eucalyptus[-cc] CLEAN=1", since we mainly use the eucalyptus script to drive eucalyptus in general, but should also support stopping and starting eucalyptus-cc only.

[Dustin Kirkland  (kirkland)]

Thierry-

I'm somewhat happy you could confirm the non-determinism, though it
means that we really need to get this fixed.

I also agree with your analysis of the supported CLEAN=1 states. I
saw your other bug filed and assigned it to myself.

[Thierry Carrez (ttx)]

I meant "Looks quite deterministic from here".
i.e. stop eucalyptus-cc CLEAN=1 always PASSed