LHC Computer Grid - Rollout
> [mailto:[log in to unmask]] On Behalf Of Di Qing said:
> If my understanding is correct, SEType is used by lcg-rep and
> lcg-cr to decide if SE is classic SE or SRM SE.
There are several different things here. The type of the SE (being
careful about how I say that :) has for a long time been stored as part
of the SE name, so you get things like SFU-LCG2:srm_v1 in the GlueSEName
attribute. Like many things that was a quick hack which has so far
lasted about 3 years! At some point a GlueSEType attribute got added,
presumably to try to make this a bit less hacky. However, it was never
an official part of the glue schema, and as far as I know has never been
used.
The new 1.2 glue schema has a separate ControlProtocol object which
carries this type (and which can be multiple, so an SE could be both
classic and SRM, or SRM v1 and v2). However, we aren't yet publishing
that, and even if we were the code doesn't look for it yet as far as I
know.
Separately there is the CESEBind (close SE) information, which is
published by a CE, and which has traditionally carried not only the CESE
binding but also information for classic SEs on where to store the data,
hence all classic SEs need to be bound to a CE. That isn't needed for an
SRM, but it looks like lcg-cr still thinks it is. That's also fixed by
the new schema, but again the code hasn't yet been updated to look for
it.
Stephen
|