On Fri, 23 Dec 2005, Valery Mitsyn wrote:
> Hi *,
>
> recently, SFT on my site JINR-LCG2 and on a few other sites
> failed in "sft-lcg-rm-rep"=="replication to Central SE".
> At the same time there are sites with success.
>
> Quick analysis show that "lcg-rep" always failed with error:
>
> the server sent an error response: 553 553 Could not determine cwdir:
> Invalid argument.
>
> and it's probably try because in this case lcg-rep for unknown
> reasone ended up with bad url like (ha-ha, classic SE?):
>
> Destination URL for copy:
> gsiftp://castorgrid.cern.ch/storage/dteam/generated/2005-12-23/
>
> On the other sites with success in "sft-lcg-rm-rep", it is always right
> url:
>
> Destination URL for copy:
> gsiftp://castorgrid.cern.ch/castor/cern.ch/grid/dteam/generated/2005-12-23/
>
> What could be a reason for such strange behaviour of lcg-rep?
That is because of this junk in the info system:
------------------------------------------------------------------------------
# castorgrid.cern.ch, ingrid.cism.ucl.ac.be:2119/jobmanager-lcgpbs-dteam, Bel
Grid-UCL, local, grid
dn: GlueCESEBindSEUniqueID=castorgrid.cern.ch,GlueCESEBindGroupCEUniqueID=ingr
id.cism.ucl.ac.be:2119/jobmanager-lcgpbs-dteam,mds-vo-name=BelGrid-UCL,mds-vo
-name=local,o=grid
objectClass: GlueGeneralTop
objectClass: GlueCESEBind
objectClass: GlueSchemaVersion
GlueCESEBindSEUniqueID: castorgrid.cern.ch
GlueCESEBindCEAccesspoint: /storage
GlueCESEBindCEUniqueID: ingrid.cism.ucl.ac.be:2119/jobmanager-lcgpbs-dteam
GlueCESEBindMountInfo: none
GlueCESEBindWeight: 0
GlueSchemaVersionMajor: 1
GlueSchemaVersionMinor: 2
------------------------------------------------------------------------------
The GlueCESEBindCEAccesspoint is _wrong_ and must urgently be fixed to be:
/castor/cern.ch/grid
I have alerted the admins of the site, and their ROC.
This shows that one site can still screw up everybody else (though this
problem only occurs for Classic SEs, which should be going away...).
|