TPC problems replicating from dCache to xrootd sites

[sfayer]

Hi,

We're seeing TPC transfer problems when replicating from dCache (SRM protocol) to xrootd (root protocol) sites with the error "The local source file does not exist or is a directory". DIRAC version is v7r2p28 with DIRACOS v1r27.

What appears to be happening is that the __getSingleTransportURL function is initially called for the source with the protocol 'xroot': dCache (7.2) happily responds to this and provides a TURL that starts xroot://hostname...

This source URL is eventually passed to GFAL2_XROOTStorage.__putSingleFile but the XROOTStorage module only lists "root" as an INPUT_PROTOCOL, not "xroot", so presumes it's a local path and trips up trying to stat it.

Could we please add xroot to the input protocol list, or add a filter here[1] to rewrite the URL string to just root:// if it starts with xroot:// before it's returned? (I tried the latter and it seemed to fix the problem).

[1] https://github.com/DIRACGrid/DIRAC/blob/2e353c23514379074d9040cd455f59dbf2fec439/src/DIRAC/Resources/Storage/GFAL2_SRM2Storage.py#L167-L168

(I was going to submit a patch directly, but as there were at least two possible fixes to choose from I thought you might want to discuss it first...).

Regards, Simon

[chaen]

Hi @sfayer

I checked with the xroot developers, and although historically there was a difference, nowadays the protocols are in principle aliases. So I'd prefer the option of adding xroot to the list of INPUT_PROTOCOL. (Note that as a test, you can override this list in your CS)

I am very puzzled by dCache not accepting root in the query, but that's another story

bash-4.2$ gfal-xattr -D "SRM PLUGIN:TURL_PROTOCOLS=root" srm://lhcbsrm-disk-kit.gridka.de:8443/srm/managerv2?SFN=/pnfs/gridka.de/lhcb/LHCb_USER/lhcb/user/c/chaen/zozo.xml user.replicas
gfal-xattr error: 95 (Operation not supported) - srm-ifce err: Operation not supported, err: [SE][PrepareToGet][SRM_NOT_SUPPORTED] httpg://lhcbsrm-disk-kit.gridka.de:8443/srm/managerv2: Protocol(s) not supported: [root]

bash-4.2$ gfal-xattr -D "SRM PLUGIN:TURL_PROTOCOLS=xroot" srm://lhcbsrm-disk-kit.gridka.de:8443/srm/managerv2?SFN=/pnfs/gridka.de/lhcb/LHCb_USER/lhcb/user/c/chaen/zozo.xml user.replicas
xroot://lhcbsrm-kit.gridka.de:1094/pnfs/gridka.de/lhcb/LHCb_USER/lhcb/user/c/chaen/zozo.xml

However, I am a bit curious about what you are doing. Are you replicating interactively, or with the RMS, or with FTS ?

[chaen]

And btw, kudos to walk your way through these dark corners of DIRAC :-p

[sfayer]

Thanks, I've asked the end user to try it out with a modified input protocol list and report back.

The use case here is just replications using dirac-dms-replicate-lfn initially: They do want to automate all of this, but it's being done manually for now as we haven't redeployed/configured/tested all of the components needed for the full RMS/FTS3/transformation set-up yet.

I was going to include my theories on the dCache xroot/root stuff, but I see you've already opened a dCache ticket about that :-)

Regards, Simon

[chaen]

To be honest, with the latest trends (multihop and other geniuses ideas), dirac-dms-replicate-lfn will work less and less often... Actually it will, it will just do a local copy first...

For completeness, here is the dCache ticket :-) https://github.com/dCache/dcache/issues/6372

[guiguem]

Hi, I added xroot in https://github.com/DIRACGrid/DIRAC/blob/2e353c23514379074d9040cd455f59dbf2fec439/src/DIRAC/Resources/Storage/GFAL2_XROOTStorage.py#L30 as suggested by Simon, and it fixed the issue about the duplication between the two sites. Thanks a lot!

Concerning the dirac-dms-replicate-lfn future, what is the preferred method to replace this script? Local copying is annoying because it is doubling the transfered data but fine to some extend.

[chaen]

I've just added xroot here https://github.com/DIRACGrid/DIRAC/pull/5755

The preferred method to replace the script is unfortunately either a local copying or the full fledged machinery Simon mentioned

[fstagni]

Closing as included in v7.3.15.