Re: [Evolution-hackers] Composer failure ...

[Michael Meeks <michael ximian com>]

Hi guys,

On Fri, 2002-10-18 at 18:27, Havoc Pennington wrote:
> Maybe some kind of /tmp scanner/cleaner-upper process like tmpwatch?

	Ah; thanks Hans / Havoc - this turns out to be more than likely I
think.

> Alternatively, don't we do some kind of cleanup of orbit sockets
> somewhere? (Maybe I'm just thinking of the atexit())

	Nope - well, there's 'linc-cleanup-sockets' but that does the right
thing, and doesn't cause the problem. Neither did the atexit's run in my
massive strace. So ... it's clearly something external like tmpwatch.

	The problem I suppose could arise from several places; firstly I have
'noatime' as an option on my '/' mount - although quite how that
happened I don't remember [ RH default ? ], if so that's bad - since we
rather need atime updates on /tmp for tmpwatch to be useful. Otherwise
it could be that the socket name doesn't have it's atime updated
correctly. Lastly, I notice we do a utime (0) on /tmp/orbit-$USER to
hide information about our random number generator, perhaps that causes
problems.

	I suppose an easy solution would be to do a utime every half hour or so
on the (single) unix socket name that we open per process, perhaps that
would fend off tmpwatch.

	Ideas ?

		Michael.

-- 
 mmeeks@gnu.org  <><, Pseudo Engineer, itinerant idiot