Hi Rob,
I'm happy to leave it any format you like, these are the sort of issues
I was trying to avoid. Is it possible that we can have an approval
process defined ?
In regards to the spawn_cache issue, maybe it was just my
implementation. Can anybody else verify that polw cache fails to start
after a power failure or unclean shutdown ?
cheers
On Fri, 2009-10-30 at 10:38 +0100, Robert Felber wrote:
> On Thu, Oct 29, 2009 at 07:37:55AM +0100, Robert Felber wrote:
> > On Thu, Oct 29, 2009 at 12:20:40PM +1100, Morgan Weetman wrote:
> > > Hi,
> > >
> > > I wasn't sure what the process was to submit updates to polw, if you
> > > could please let me know.
> >
> > You can release it on sourceforge and I'll include it on policyd-weight.org.
>
>
> First,
>
> is it okay to keep line-breaks as is, resp. to format it for 80-char
> displays? It's otherwise a hard reading. Also, as suggestion: format the
> code at a black-bg,white-fg-terminal without syntax-highlighting. That's
> why I kept those long
>
> #
> #
> #############################################################################
> #
>
> lines.
>
>
>
> Second,
>
> code:
>
> I don't see the reason for the massive change.
>
> If no cache proecess exists, every process is allowed to create one.
> In the start-phase of policyd-weight or in case of a death of a cache
> this could mean a couple of hundreds cache-process trying to start up. If you
> let them sit around by sleeping 2 seconds, you will cause a fork problem
> (imagine 20 or 30 smtp requests per second).
> They have to return undef, and non verbose instantly if they detect that other
> caches are ahead.
>
> As soon as a cache is successfully forked, it does delete the lockfile.
>
> A stale socket will always be deleted.
>
>
> In order to make sure that there is no stale lock-file at the beginning of
> the world we could remove an existing lock-dir before (like):
>
> line: 1152
> + # a cache-lock-file shouldn't be there yet
> + if( -d $LOCKPATH.'/cache_lock )
> + {
> + unlink $LOCKPATH.'/cache_lock;
> + }
> cache_query("start"); # pre-launch cache
>
>
> Rationale: the master/child have to control when to start caches. As such
> it is their, in this case the masters, responsibility to make precautions
> for a clean environment.
>
> If a cache crashes between 'mkdir lock' and 'rmdir lock' then it
> has to be logged.
>
>
> Also, for such things that concern security and robustness, I'd suggest that
> we talk first about it. I haven't included it on policyd-weight.org because
> of the 2-seconds-of-vague-sleep, which really shouldn't be there.
>
> We should also start to sign the changes made in changes.txt.
> Or maybe use the sourceforge SVN.
>
____________________________________________________________
Policyd-weight Mailinglist - http://www.policyd-weight.org/
