LHC Computer Grid - Rollout
> [mailto:[log in to unmask]] On Behalf Of Alessandra Forti
said:
> I removed the MON box from everywhere and the site didn't publish
> correctly anymore. I must have missed from the docs that the
> BDII needs
> to be added instead. It eludes me why we need two sets of the
> same info,
> one that propagates and the other that doesn't and why the one that
> propagates depends from a service, but I will not ask.
Well, I'll answer anyway as it may help others ... the site BDII
shouldn't have needed to be added instead, it should already have been
configured that way. However, it seems not uncommon that when people
move a site BDII from one node to another they don't update the site
publishing configuration, and for a while it seems to work, but
eventually it runs into trouble as you've found.
The general way the info system works is that every service node -
including site and top bdiis - has a resource bdii, i.e. it publishes
its own services under mds-vo-name=resource,o=grid. The site bdii should
then be configured to collect information from every resource bdii at
the site, *including its own* (and including the one on a top bdii if
you have one). That collected information is of course published in a
different branch, mds-vo-name=<site-name>,o=grid, so there is no
recursion.
The site information (GlueSite object) is somewhat anomalous because
it refers to the whole site rather than a particular service. The
standard configuration includes it in the resource bdii attached to the
site bdii, since by definition every site should have one of those.
However it still works if the site information is published by a
different node, as you have found - but it may be confusing. The site
info is just a piece of static LDIF created by YAIM, so once it's
present on a node it's likely to stay there unless you delete it
explicitly.
Another thing I see from time to time is that the site info is present
in two different resource bdiis. In that case the site BDII will take
whichever one it sees first and reject any duplicate(s). Again that may
appear to work fine, until you change the site information and wonder
why the changes don't show up in the site BDII.
Stephen
|