On Mon, 2007-05-07 at 21:05 +0200, Ruediger Pluem wrote:
>
> On 05/07/2007 05:56 PM, Mladen Turk wrote:
>
> >
> > I think we could use a simple two use case situation.
> > To be able to backport that to the 2.2 branch I propose
> > a following patch.
> >
> > It adds additional struct member sticky_path that is
> > set to the sticky so if someone has in the config
> >
> > ... stickysession=JSESSIONID ... both sticky and
> > sticky_path will be the same.
> > However if there is a stickysession with
> > ... stickysession=JSESSIONID|;jsessionid ...
> > will make w->sticky=JSESSIONID and w->sticky_path=;jsessionid
> >
> > I'm not aware of any other way without introducing additional
> > directives that could be backported to 2.2
> >
> > The question is, is that backportable as well?
>
> I think this is backportable in general if we do a minor bump (which
> is possible for backports). Currently I see the following problems
> with the patch which seem to be solvable:
>
> 1. Documentation.
> We definitely need a patch for the documentation that describes how
> to handle the servlet case with the above patch. Otherwise people keep
> falling into the pit.
>
> 2. Apart from route search balancer->sticky is used in several additional
> locations for
>
> - Status pages (mod_status hook, loadbalancer manager)
> - Some notes / environment variables are set to balancer->sticky which is
> wrong if balancer->sticky_path was used to find the actual route.
> Furthermore it might be worth adding an additional environment variable
> (e.g. BALANCER_SESSION_STICKY_TYPE) that indicates whether
> BALANCER_SESSION_STICKY contains balancer->sticky or
> balancer->sticky_path.
>
> Adjusting sessionstickyness via the loadbalancer manager might not be
> needed as this
> currently does not work anyway since the balancer configuration is not
> stored in shared
> memory (Maybe we should turn off the possibility to change the balancer
> configuration via
> the manager until this is fixed as this can lead to weird results with
> different
> processes using different configurations).
Yes we should avoid adjust sessionstickyness via the loadbalancer
manager because it won't work with prefork and worker mpm.
>
> 3. The minor bump is missing which is needed due to the change of the struct.
>
> 4. I am confused by the following part of the patch :-):
>
> + balancer->sticky = balancer->sticky_path = apr_pstrdup(p, val);
> + if ((path = strchr(balancer->sticky, '|'))) {
> + *path++ = '\0';
> + balancer->sticky_path = path;
> + }
> + else
> + balancer->sticky_path = balancer->sticky;
>
> Why setting balancer->sticky_path twice in both cases (string with or without
> |)
Right no need of the else here.
Cheers
Jean-Frederic
>
>
> Regards
>
> Rüdiger
>
>
