Hi Andy, On Thu, Sep 02, 2004 at 06:29:45PM -0700, Andrew Hanushevsky wrote: > > In this case "client" = xcp (or xrdcp, you ducked the naming question), is > > that correct? > Yes, I ducked it. I would vote for xrdcp to avoid name clashes. Ok, done. We'll call it 'xrdcp'. > > (a) Simply stops with an error message if one tries to do such a wildcarded > > copy via a load balancer instead of an actual dataserver. Can the > > client application determine this? > Yes, but that leads to non-determinsitic behaviour. One could shoose a > compromise and allow directory copies but not full wildcard copies. This > keeps the client somewhat simple because doing full wildcard copies is not > trivial. Could you clarify what you mean by "directory copies"? (Including example syntax for the xrdcp command.) > > (b) Gives me what is actually there on disk instead of what might be > > out there in tertiary storage someplace. > True, but again not completely satisfying unless you are quite aware of > what you're doing. But there is no way to avoid "not being completely satisfying", is there? > > Since I may be going via the load-balancer to write the file, how do I > > create the "/some/path"? I have no idea in advance to which server I will be > > redirected. ("I" in this case being xcp/xrdcp.) > Writes are problematic. Generally, you don't need to know ahead of time. > You create the path once you get there. So the server doesn't take care of this itself, but 'xrdcp' _could_ be setup to create the paths automatically. Is that the proposal? Pete ------------------------------------------------------------------------- Peter Elmer E-mail: [log in to unmask] Phone: +41 (22) 767-4644 Address: CERN Division PPE, Bat. 32 2C-14, CH-1211 Geneva 23, Switzerland -------------------------------------------------------------------------